Productivity: Evidence from Case and Argument Structure in Icelandic (Constructional Approaches to Language)

Productivity

Constructional Approaches to Language The series brings together research conducted within different constructional models and makes them available to scholars and students working in this and other related fields. The topics range from descriptions of grammatical phenomena in different languages to theoretical issues concerning language acquisition, language change, and language use. The foundation of constructional research is provided by the model known as Construction Grammar (including Frame Semantics). The book series publishes studies in which this model is developed in new directions and extended through alternative approaches. Such approaches include cognitive linguistics, conceptual semantics, interaction and discourse, as well as typologically motivated alternatives, with implications both for constructional theories and for their applications in related fields such as communication studies, computational linguistics, AI, neurology, psychology, sociology, and anthropology. This peer reviewed series is committed to innovative research and will include monographs, thematic collections of articles, and introductory textbooks.

Editors Mirjam Fried

Institute for the Czech Language Czech Academy of Sciences Prague, Czech Republic

Jan-Ola Östman

Department of Scandinavian Languages & Literature, University of Helsinki Helsinki, Finland

Advisory Board Peter Auer

Seizi Iwata

Hans C. Boas

Paul Kay

William Croft

Knud Lambrecht

Charles J. Fillmore

Michael Tomasello

Adele E. Goldberg

Arnold M. Zwicky

University of Freiburg, Germany University of Texas at Austin, USA University of New Mexico, USA Int. Computer Science Institute, Berkeley, USA Princeton University, USA

Osaka City University, Japan University of California, Berkeley, USA University of Texas at Austin, USA Max Planck Institute for Evolutionary Anthropology, Germany Stanford University, USA

Volume 8 Productivity: Evidence from Case and Argument Structure in Icelandic by Jóhanna Barðdal

Productivity Evidence from Case and Argument Structure in Icelandic

Jóhanna Barðdal University of Bergen

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Jóhanna Barðdal, 1969  Productivity: Evidence from Case and Argument Structure in Icelandic / Jóhanna Barðdal.        p. cm. (Constructional Approaches to Language, issn 1573-594X ; v. 8) Includes bibliographical references and index. 1.  Icelandic language--Syntax. 2.  Icelandic language--Morphology. 3.  Icelandic language--Verb. 4.  Productivity (Linguistics)  I. Title. PD2423.J64  2008 439'.695--dc22 isbn 978 90 272 1830 8 (Hb; alk. paper)

2008041942

© 2008 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Til Harðar/To Hardy

Table of contents Preface chapter 1 Introduction Aims and objectives  1 1.1 1.2 Construction Grammar  4 1.3 Overview  6 chapter 2 Productivity 2.1 Senses and synonyms of productivity  10 2.2 The concepts of productivity  19 2.3 Definitions and measurements of productivity  24 2.3.1 Existing definitions and measurements  24 2.3.2 Predicting productivity  27 2.4 Syntactic productivity  29 2.5 Type frequency, schematicity and degrees of productivity  34 2.6 Summary  53 chapter 3 New verbs in Icelandic: A general outline 3.1 Case and argument structure in Icelandic  56 3.1.1 Type frequency  57 3.1.2 The semantics of the Nominative subject construction  62 3.1.2.1 Nom-Acc  63 3.1.2.2 Nom-Dat  69 1.2.3 Nom-Gen  75 3.1.3 The productivity of the Nominative subject construction  76 3.2 New verbs in Icelandic  78 3.2.1 Recent borrowings  78 3.2.2 Analogy and token frequency  89 3.3 Summary  96

xi

1

9

55

 Productivity

chapter 4 Nonce verbs: A psycholinguistic experiment 4.1 The experiment  102 4.2 The findings  104 4.3 The semantics of the Dative subject construction  111 4.4 Summary  116

101

chapter 5 119 New verbs of communication: A questionnaire 5.1 The Transfer, the Caused-motion and the Ditransitive constructions  120 5.1.1 The Transfer and the Caused-motion constructions  120 5.1.2 The Ditransitive construction  126 5.2 The questionnaire survey  128 5.3 The findings  128 5.4 Analogy vs. high type frequency  132 5.5 Summary  135 chapter 6 137 Old and Modern Icelandic: A frequency comparison 6.1 Structural vs. lexical case  139 6.2 The present cognitive-functional CxG approach  143 6.2.1 Swedish  147 6.2.2 English  148 6.2.3 German  150 6.2.4 Icelandic  151 6.2.5 Vocabulary, language contact, type frequency and loss of case  156 6.3 Dative substitution in the history of Icelandic  158 6.4 The “blended” construction in English, Swedish and Faroese  164 6.5 Historical productivity  166 6.6 Summary  167 chapter 7 Synthesis

171

References

183

Appendix A: Case and argument structure constructions in the text corpora and the predicates which instantiate them

191



Table of contents 

Appendix B: Recent borrowings in Icelandic

199

Appendix C: The questionnaire

201

Name index

203

Subject index

205

Constructions index

209

Preface When working on my dissertation on morphological case in Icelandic (Barðdal 2001a), my main research interest was focused on the function or motivation of this morphological category in that language. This work brought me to productivity since one of the methods I used was to investigate the productivity of the morphological cases in Icelandic. In fact, two of the empirical studies presented in this book originate in my dissertation: First, my original study of new and borrowed verbs in Modern Icelandic and the morphological case they assign to their arguments (Barðdal 2001a: Ch. 5) inspired me to carry out an additional study, for the purpose of this book, focusing on borrowed verbs from the area of information technology. Second, Chapter 7 from my dissertation, comparing the frequencies of different case and argument structure constructions in Old and Modern Icelandic, has undergone severe changes and advancements from its original version and occurs as Chapter 6 in this monograph. One of the fundamental questions I was confronted with during the final period of my dissertation project had to do with the nature of productivity. I found that certain definitions of productivity could as well be regarded as measurements of productivity, and not necessarily as prerequisites for productivity. If so, then productivity had to be some underlying phenomenon, if only I could put my fingers on that. I also found that not only were there different definitions of productivity figuring in the literature, but also that there were different concepts of productivity around. When faced with the question of revising my dissertation into a commercial linguistic monograph, it felt natural to extend my work on morphological case and use the dissertation data to cast light on a completely different research question, namely the one on the nature of productivity. I have, thus, complemented the two studies from my dissertation with two more studies on the productivity of morphological case and argument structure: One is a psycholinguistic experiment on nonce verbs and which case and argument structure constructions they are assigned by Icelandic speakers, both children and adults (cf. Barðdal 2000). The other is a replication of an earlier questionnaire study, using a revised questionnaire, on the acceptability of new verbs of communication when instantiated by certain Icelandic case and argument structure constructions (cf. Barðdal 2003). This last study, furthermore, provides some insight into the social and cultural

 Productivity

factors at work when new verbs are either coined or borrowed in a speech community. In this book I will present my view and understanding of productivity as it is at the moment. I will argue for a definition of syntactic productivity based on type frequency and semantic coherence and an inverse correlation between the two. I will report on the four empirical studies, mentioned above, as supporting my view. My aim is to participate in a discussion on productivity in general, but more importantly I want to contribute to a commencing discussion on syntactic productivity, by which I mean the productivity of case and argument structure constructions and how they are extended to either already existing verbs or novel verbs in the process of being coined or of entering the language. This book has not emerged out of the blue; I feel grateful to Laurie Bauer for his monograph on morphological productivity (2001), both because it contains an extensive overview of research on productivity but also because of his profound understanding of the subject. His book has saved me a lot of time, search and research. I am also grateful to Bill Croft for his theoretical and intellectual contribution, his work on productivity, and guidance during the initial stages of this project. The earlier work of Langacker, as well as the work of Bybee, Clausner and Croft, and Goldberg, on productivity within the usage-based model has been a source of inspiration. For particular support I thank Michela Cennamo, Eystein Dahl, Östen Dahl, Thórhallur Eythórsson, Kjersti Fløttum, Sandra Halverson, Willem Hollmann, Gard Jenset, Christer Johansson, Torodd Kinn, Kristian E. Kristoffersen, Joan Maling, Valéria Molnár, Carita Paradis, Christer Platzack, Halldór Ármann Sigurðsson, Hanne Gram Simonsen, Koenraad de Smedt, Andreas Sveen, Beatrice Warren and Jordan Zlatev. I also thank all my other colleagues at the Scandinavian department in Lund, the Linguistics department in Manchester, the English department in Denton, TX, the Scandinavian department in Bergen, the Linguistics department in Berkeley, CA, and the Department of Linguistic, Literary and Aesthetic Studies in Bergen, where I have spent the last years, among other things, working on this project. I thank the editors of CAL, Jan-Ola Östman and Miriam Fried, and the editor at Benjamin’s, Anke de Looper, for their help and for welcoming this book into their series. I have presented this work at conferences, workshops and colloquia in Aalborg, Austin, Bergen, Berkeley, Bloomington, Bran, Helsinki, Logroño, Lund, Madison, Manchester, Marseille, München, Naples, Oakland, Oslo, Prague, Stockholm, Stuttgart and Tokyo, and I hereby thank the audiences at these talks. I am also indebted to the following scholars who have read and commented on earlier or current versions of parts of the manuscript: Hans C. Boas, Bert Cappelle, Ellen H. Toft, and Annette Rosenbach; and to Thórhallur Eythórsson, Gard Jenset, Jordan Zlatev, and an anonymous reviewer for comments on the manuscript in its



Preface 

entirety. Their comments have improved this work considerably. I am also indebted to Gard Jenset and Christer Johansson for help with the statistics, although they are not to blame for how I followed their advise. Jan-Ola Östman deserves the greatest thanks for his enthusiasm, editorial comments and his support in general. Finally, I thank my partner, Hans Heinrich Marxen, for having given my life a whole new (non-linguistic) meaning, and I dedicate this book to my son, Hörður, whom I love the most.

chapter 1

Introduction 1.1 Aims and objectives Productivity has hitherto mostly been investigated from a morphological and phonological angle. Unlike previous approaches, the aim of this monograph is to study productivity from the conspicuously neglected syntactic perspective, in particular the productivity of case and argument structure constructions. By syntactic productivity, or syntactic creativity as it is sometimes also referred to in the syntactic literature, I do not refer to speakers’ ability to generate and understand sentences they have never encountered before, as is common within the generative literature (cf. Chomsky 1965). Rather, by syntactic productivity, I refer to an argument structure construction’s ability to attract new or existing lexical items, i.e. a construction’s extensibility. A major debate within the syntactic literature has to do with the nature of speakers’ ability to produce and understand sentences never encountered before. In my view, this is not a question of productivity but rather is it a token of speakers’ nonmysterious knowledge of syntactic constructions, the lexemes which can instantiate these constructions, speakers’ know-how to combine these together into meaningful utterances, and their ability to categorize previously unencountered utterances as instances of these constructions (see Section 2.4). This is of course not so say that this categorizing ability of speakers is not worthy of research, even though I do not regard it as being a matter of productivity. This ability, however, is not the same as speakers’ ability to extend existing argument structure constructions to new or existing predicates of a language, although both may ultimately reside in speakers’ ability to categorize. The goal of this monograph is not to explore this issue, but rather to stake out a new field of syntactic research, namely the extensibility of case and argument structure constructions, which is a research field that has been gradually emerging for the last decade or so, most prominently within cognitive-functional linguistics. The goal of this monograph is to contribute to research on productivity as extensibility and to provide, at the same time, an account of how the term productivity has come to mean so many different things to so many different scholars. As already noted by several scholars of linguistics (see, for instance, Bauer 2001: 1), there is little consensus in the literature on how to define productivity, and different scholars elaborate with different notions of productivity. This has



Productivity

made it necessary to delve into the literature on productivity, which in turn has revealed that the meta-concept of productivity is a multi-facetted one, with at least three subconcepts, namely those of extensibility, generality and regularity, and several different senses related to each of these subconcepts (Sections 2.1–2.2). Speakers’ ability to understand and generate sentences never encountered before is a manifestation of the regularity concept, not the extensibility concept, I argue in Section 2.4. It is, of course, crucial that the research community has a shared understanding of the term productivity, if we are to avoid useless pseudodebates and quasi-discussions in our work and if we aim to go forward with our research and with our contributions to this research field. Such a shared understanding does not exist at present. Yet another issue within the productivity literature is the topic of gradience. That is, is productivity (in terms of extensibility) an all-or-nothing phenomenon or is it gradient? Productivity is traditionally defined in terms of rules generating new forms, with the output determining the level of productivity for that particular rule. The concept of a rule, however, does not allow for gradience in productivity at all, as a rule is either operative or not. This is a major problem for all rulebased approaches to productivity (Section 3.2.2). It seems clear, nevertheless, that linguistic categories show variation in their ability to attract new items, which again means that productivity must be gradient. In fact, I will argue in Chapters 3–5 that productivity and analogy are two sides of the same coin, and that the difference between “full” productivity and analogy is a difference in degree, not in kind, although it may appear as a difference in kind if the relevant factors behind productivity are not properly identified. These, I argue, are type frequency and coherence, with full productivity at the top of the productivity cline and analogical extensions at the bottom of the productivity cline (see Section 2.3). There are not only several different definitions of productivity found in the literature, but also various suggestions on how to measure it. In the cognitivefunctional literature, Bybee’s view (1995:  430) that productivity is a function of type frequency and coherence stands out. In practice, however, Bybee argues for the role of type frequency, as opposed to the role of schematicity, in determining productivity. In this monograph I will take this hypothesis, that productivity is a function of type frequency and coherence, to its logical conclusion, as a serious evaluation of these two factors and the interaction between them has not been carried out in the literature before, as far as I know. Such an enterprise entails a close examination of the premises and predictions of this hypothesis of productivity on syntactic structures, in particular on case and argument structure constructions. Analogy is also a vague and problematic concept in the linguistic literature (cf. Itkonen 2005), used differently by different scholars to capture different linguistic processes. By the term analogy I refer to analogical extensions, i.e. extensions of a



Chapter 1.  Introduction

grammatical pattern based on only one model item, because of structural or semantic similarities between the two items. Another dichotomy found in the literature on morphological productivity is the distinction between creativity and productivity. The term creative is no less problematic than the term analogy, as it is used to denote different linguistic phenomena by different scholars. Bauer (2001: 64), for instance, gives one definition as “the extension of non-productive patterns”. Such extensions, however, are usually taken to be the result of analogy, and I subscribe to that view here. Moreover, as I argue that “full” productivity and analogy are two sides of the same coin, there can be no extensions of non-productive patterns, only different degrees of productivity and a scale from high-level to low-level extensions. This follows directly from the view that productivity is extensibility and that regularity and generality have come to be linked with productivity by association. The generative distinctions between lexical and structural case, on the one hand, and thematic and idiosyncratic case, on the other, make clear predictions about productivity (Sections 3.2.1, 4.2 and 6.1), particularly about the productivity of argument structure constructions. Regarding the first dichotomy, I show that only structural case should be productive, as structural case is assigned on the basis of the structure in the sentence, while lexical case should be unproductive, as it has to be specified specifically for the relevant set of lexical entries. This is exactly parallel to regular and irregular verbs within the generative paradigm, where regular verbs are assumed to be produced by a rule, while irregular verbs are assumed to be learned by rote (Pinker 1999). Lexical case has, moreover, also been divided into thematic and idiosyncratic case (Zaenen, Maling and Thráinsson 1985, Yip, Maling and Jackendoff 1987), where thematic case is assigned on the basis of a particular thematic role, while idiosyncratic case is not. This dichotomy predicts that thematic case assignment should be productive within its thematic domain, while idiosyncratic case assignment should not be productive. It turns out that these predictions are not borne out by the data found in two of the data sets discussed here, i.e. the borrowed verbs and the nonce verbs, and hence that the predictions of these dichotomies are not borne out by the productivity of case and argument structure constructions in Icelandic. There are several questions to be asked and answered when working on productivity, precisely because of the vagueness of the term and the general confusion in the literature. These are questions like: What is productivity? Which factors determine if a category is productive or not? When I first started thinking seriously about productivity approximately ten years ago, I was confronted with the claim that the productivity of a linguistic category can be determined by its type frequency (Bybee 1985: 132–133, Goldberg 1995: 134). To me it seemed equally reasonable to argue that type frequency is a consequence of productivity, and not necessarily a prerequisite for





Productivity

productivity. In the mean time I have adopted the more nuanced view that productivity is a function of both type frequency and coherence, and on such a view, type frequency can clearly be both a predictor and a consequence of productivity. In this monograph I will argue that productivity is an emergent feature of the structure of the grammatical system (broadly defined), and that productivity falls directly out from our linguistic model, if we adopt a usage-based cognitive-functional CxG approach to language and grammar (cf. Section 2.5). This means that specific theories of productivity are not needed, and can as such be dispensed with from linguistic theorizing. On the present approach, productivity is argued to follow directly from the organization of constructions into lexicality–schematicity hierarchies, where information about type frequency and semantics are coded at and through the different levels of schematicity (see Section 2.5). The current approach to syntactic productivity, moreover, focusing on the extensibility concept, offers a unified account of productivity, as it captures different degrees of productivity, ranging from highly productive patterns to various intermediate degrees of productivity to low-level analogical extensions. The regularity and the generality concepts can also be derived from this approach, as such establishing the theoretical primacy of the extensibility concept for productivity. 1.2 Construction Grammar The basic tenet within Construction Grammar (CxG) is that the language system is made up of form–meaning or form–function correspondences, i.e. the construction. These form–meaning correspondences can either have a general or specific meaning (cf. Tomasello 1998: 481–482, Croft and Cruse 2004: 253–254), i.e. their meaning can either be derived from the meaning of their parts (general), or their meaning cannot be derived from the meaning of their parts (specific). The latter type of construction is also discussed in the literature as being idiomatic or having non-compositional semantics. The most radical version of the constructional position entails that all form– meaning correspondences are constructions of their own, be they general or specific, and not only the specific, non-compositional ones are regarded as having a constructional status (Croft 2001, Barðdal 2001a–b, inter alia). This position also entails that language is simply made up of constructions existing at all levels of language, as is shown in Table 1.1, where form–meaning correspondences are specified for several linguistic levels, for instance the morphological and the syntactic level. One major advantage of this CxG view of language and grammar is that all linguistic objects are treated in a uniform way, namely as form–function correspondences. This is what Table 1.1 is intended to illustrate.



Chapter 1.  Introduction

Table 1.1  The syntax-lexicon continuum. (Croft and Cruse 2004: 255) Construction type

Traditional name

Examples

Complex and (mostly) schematic Complex, substantive verb Complex and (mostly) substantive Complex but bound Atomic and schematic Atomic and substantive

syntax subcategorization frame idiom morphology syntactic category word/lexicon

[SBJ be-TNS V-en by OBL] [SBJ consume OBJ] [kick-TNS the bucket] [NOUN-s], [VERB-TNS] [DEM], [ADJ] [this], [green]

Syntactic constructions in CxG can be divided into two types: 1) sentence-type constructions, and 2) argument structure constructions. Sentence-type constructions are, for instance, active and passive sentences, questions, imperatives, topicalizations, and other sentence types which display word order variation. Argument structure constructions, also referred to as subcategorization frames in the literature, are intransitives, transitives, ditransitives, etc. In languages with case marking, these argument structure constructions can be divided into further subconstructions, based on case marking properties (see Chapter 3.1 below for an overview of intransitive and transitive case and argument structure constructions in Icelandic). One could view argument structure constructions as corresponding to derivation in morphology and sentence-type constructions as corresponding to inflection. Observe, also, that all utterances are made up of both types of syntactic constructions, in that argument structure constructions always occur in language use in either the active or the passive, indicative or subjunctive, statements or questions, etc., exactly like new derivational items are always inflected during real usage events. The cognitive-functional CxG approach that I am advocating here is also usage-based, in that frequencies are regarded as essential for grammatical description (for a discussion and overview of differences between various constructional approaches to language and grammar, see Goldberg 2006). Argument structure constructions, like other linguistic categories, are organized in a lexicality–schematicity hierarchy, based on their type frequency (Croft 2003, Barðdal 2001b, 2006a–b, Iwata 2008). At the top of the hierarchy we find schematic constructions, while at the bottom of the hierarchy we find the verb-specific constructions. At the intermediate levels we find verb-subclass-specific and verb-class-specific constructions. All argument structure constructions always exist at the lowest, most concrete level as verb-specific constructions. The intermediate levels are, furthermore, postulated on the bases of verb classes and verb subclasses instantiating the construction. In contrast to the prevalent view in the cognitive-functional linguistic literature, I argue in Chapter 2 below that not all constructions exist at the





Productivity

highest most schematic level of the hierarchy. Even though linguists assume such a schematic level in their linguistic descriptions, it is not a given that such a level has any psycholinguistic validity (i.e. model-theoretical existence vs. ontological existence). It seems obvious that a fully schematic level for all constructions is a matter of empirical investigation, and in fact, of the Icelandic case and argument structure constructions, discussed in the following chapters, only the transitive Nom-Acc construction is attested as a schematic construction at the highest, most abstract level of the hierarchy. All the other constructions exist only as high as verb-subclass-specific, verb-class-specific or event-type-specific constructions. This, I argue, has consequences for syntactic productivity, as it restricts the productivity domains of the relevant constructions. 1.3 Overview The content of this monograph is two-fold: First I present a theoretical approach to the concept of productivity and then I describe four data-driven investigations on case and argument structure in Icelandic. The first three studies are synchronic, while the last one is diachronic. In Chapter 2 I discuss the use of the terms productive and productivity in the linguistic literature and present on the basis of that a semantic analysis of the highly structured metaconcept of productivity. I put forward the hypothesis that productivity is a function of type frequency and coherence, and the inverse correlation between the two, which in turn entails gradient productivity. I also discuss the relation between type frequency, schematicity and degrees of productivity. In Chapter 3 I give an overview of case and argument structure constructions in Modern Icelandic, i.e. the Nominative subject construction and the Oblique subject construction. The first can be divided into Nom-Acc, Nom-Dat and NomGen, and the latter into Accusative, Dative and Genitive subject constructions. The three subconstructions of the Nominative subject construction are analyzed in terms of type frequency and semantic coherence, and their productivity domains determined on the basis of that. An investigation of verbs borrowed into Icelandic, mostly from the area of information technology, reveals that the predictions of the cognitive-functional CxG approach advocated here are borne out. The productivity of the three constructions can be determined from their highest level of schematicity, even though the actual case and argument structure constructions are clearly assigned on the basis of low-level verb-specific analogical extensions. The importance of token frequency for low-level analogies is also established. Chapter 4 reports on a psycholinguistic experiment of nonce verbs, with forty randomly-selected Icelandic speakers, twenty adults and twenty children. They



Chapter 1.  Introduction

were presented with five nonce verbs and corresponding synonymous verbs in Icelandic, in order to elicit their use of case and argument structure constructions with the nonce verbs. The results of the experiment suggest that speakers assign a case and argument structure construction to an unknown verb either on the basis of a synonymous low-level verb-specific construction or on the basis of a highly schematic abstract construction. In Chapter 5 I describe a questionnaire survey on the acceptability of three new verbs of communication, faxa ‘fax’, (e)meila ‘email’ and smsa ‘text’, in three constructions in Icelandic, the Transfer construction, the Caused-motion construction and the Ditransitive construction. It is expected that the Transfer construction with an accusative object will be judged acceptable, as this is the construction senda ‘send’, the most prototypical verb of sending, occurs in. The Caused-motion construction with a dative object is also judged acceptable with the three verbs, clearly on the basis of a small cluster of verbs of non-translational motion. This is thus an obvious case of constructional competition between a verb-specific and a verb-subclass-specific construction in Icelandic. In Chapter 6 I point out the predictions of the present cognitive-functional CxG approach on historical productivity, namely that over time the high type frequency constructions should attract new and existing verbs, while low type frequency constructions should go further down in type frequency. I argue that the breakdown of the case system in English and Swedish, and the reduction in the alignment system in German and Icelandic, is due to synonymy in argument structure constructions. This development should coincide in time with language contact, as periods of contact entail substantive borrowing of new verbs. As expected, a correlation is found between the density and time span of the language contact and the time span of the breakdown of the case and alignment system in the relevant Germanic languages. I also discuss the onset of dative substitution during the latter part of the 19th century and the emergence of the ‘blended’ construction in the history of English, Swedish and Faroese. Chapter 7 synthesizes the main claims and conclusions on syntactic productivity put forward in this monograph, amongst other things arguing that specific theories of productivity are not needed and can thus be dispensed with, as facts of syntactic productivity fall directly out from the usage-based cognitive-functional approach to language and grammar adopted here.



chapter 2

Productivity Productivity has presented one of the most indistinct and puzzling phenomena for linguistic research over the decades. Several definitions of it can be found in the literature, and a close inspection of the existing discussion reveals an exceedingly multi-facetted concept of productivity. In this chapter I will present my view and definition of productivity, in particular syntactic productivity, and place it in a proper context within the wider concept of productivity. I begin by giving an overview of the various senses and uses of the terms productive and productivity encountered in the linguistic literature (Section 2.1). These include both formal technical senses and more informal lax uses of the terms. A fairly comprehensive analysis of these reveals that the concept of productivity is a highly complex one, involving at least three subconcepts and various types of sense relations between the subconcepts themselves and the different senses (Section 2.2). This partly explains why productivity has been such a diffuse and debated concept in morphology. After a brief introduction of the definitions and measurements of morphological productivity suggested in the field (Section 2.3), I present the hypothesis advocated in this book, namely that type frequency, semantic coherence and an inverse correlation between the two adequately predicts the productivity of syntactic constructions (Section 2.3.1). This means that constructions high in type frequency need not show a high degree of semantic coherence in order to be productive, while constructions low in type frequency must show a high degree of semantic coherence in order to be productive. Before elaborating on this hypothesis, however, it is essential to discuss how I define syntactic productivity and how my definition relates to the prevalent view of productivity within syntax (Section 2.4). Such an outline reveals some unexpected parallels between morphological and syntactic structures that are relevant for the application of the concept of productivity within these areas. The hypothesis of productivity advocated here makes the following predictions: Productivity is a gradient phenomenon, ranging from high productivity, based on high type frequency and a high degree of schematicity, to low productivity, based on low type frequency and a high degree of specificity (Section 2.5). This presupposes that high type frequency constructions are general and schematically open, while low type frequency constructions can vary in their degree of specificity,



Productivity

with only semantically coherent constructions yielding some degree of productivity. At one end of the productivity cline we find full productivity, while at the other end we find analogical extensions based on a high degree of token similarity. In other words, highly schematic argument structure constructions with the least semantic specifications are located at the upper end of the cline. At the lower end of the cline, verb-specific lexically-filled constructions are found, some of which are concrete instantiations of more schematic constructions and some are item-specific islands. The various intermediate levels of schematicity, at which constructions can be assumed to exist in speakers’ minds, represent degrees and domains of syntactic productivity. The productivity cline argued for in this study thus corresponds to the lexicality–schematicity continuum assumed within cognitive-functional linguistics and usage-based versions of construction grammar, according to which all linguistic constructions are categorized and organized. The chapter closes with a summary of its main content and conclusions (Section2.6). 2.1 Senses and synonyms of productivity In this section I argue that the term productive is being used in various senses in current linguistic writings. I will enlist these senses and document their occurrence in the literature. This discussion will thus form the basis for my analysis of the structure of the productivity concept, laid out in Section 2.2 below. The most concrete meaning of the adjective productive is probably ‘being produced’. Although I have come across such a use of the term, most notably within corpus-based work (Baayen 2001: 156, Hallan 2001: 109) and experimental work (Tomasello 2000: 210), this sense of productive is not the predominant one in the literature on linguistic productivity. Generally within the linguistics literature, I have found the adjective productive being used synonymously with the following: 1. ‘frequent’ 2. ‘regular’ 3. ‘rule-based’ 4. ‘operative’ 5. ‘easily combinable’ 6. ‘transparent or compositional’ 7. ‘having a wide coverage’ 8. ‘default’ 9. ‘schematically open or non-restricted’ 10. ‘schematic (vs. listable)’



Chapter 2.  Productivity

11. ‘lexically and semantically modifiable’ 12. ‘syntactically manipulatable/flexible (vs. frozen or idiomatic)’ 13. ‘having a meaning/function (vs. being historical relics)’ 14. ‘developing new functions’ 15. ‘occurring or existing’ 16. ‘occurring with new/novel/nonce items’ 17. ‘attracting already existing items’ 18. ‘living (vs. dead)’ 19. ‘deviating from adult language’ This non-exhaustive list of nineteen usages supports the general conception found within the field of morphology that the notion or concept of productivity is not a simple one (Zimmer 1964, Aronoff 1976, Mayerthaler 1980, Plag 1999, Baayen 2001, Bauer 2001, amongst others). I will argue later in this chapter that there are at least three concepts of productivity around, and that the relation between them can be regarded as an inferential implicational relation in the sense of Traugott and Dasher (2002). Before that, however, it is important to show that the nineteen different senses of the adjective productive are indeed found in the literature. The following exposition is not meant to give a full account of the notion of productivity, as defined by the scholars quoted, but only to illustrate the different senses and the polysemy of the terms productive and productivity. This has been done through a fine-grained content analysis of meaning nuances in the relevant citations below and the surrounding text. The most informal usage of the adjective productive is probably with the meaning ‘frequent’. Such a usage is, for instance, found in Fleischer’s discussion of the distribution of the German prefix haupt‑ ‘main’:

(1) Das präfix haupt‑ ... ist heute außerordentlich produktiv zur Bezeichnung der Hervorhebung eines Begriffes ... (Fleischer 1982: 219 [Emphasis original])

‘The prefix haupt‑ ... is extraordinarily productive today when characterizing the salience of a concept ...’ [Translation, JB] After this statement, Fleischer presents a long list of words in German that can occur with the prefix. In this particular passage, therefore, productive refers to the high frequency of the prefix, as Fleischer is not discussing new words entering the language, but already existing German words. Baayen (2001: 154) also asserts that “morphological patterns can be ‘productive’ … in the sense that many rather than few types have been observed”.





Productivity

The most frequently found morphological patterns are also often the most regular ones. Pinker (1999: 19), for instance, contrasts regularity and productivity with irregularity and non-productivity:

(2) The theory that regular forms are generated by rule and irregular forms are retrieved by rote is pleasing ... because it explains the differences in productivity between the two patterns ...

Pinker thus equates productivity with regularity, assuming that regular linguistic behavior is productive behavior, being the output of grammatical rules represented in our mental grammars. Leonard (2000), in his conclusions on language production of children with specific language impairment (SLI), states:

(3) It may prove to be the case that some individuals with SLI are truly operating without a productive system. (Leonard 2000: 350 [Emphasis added])

Leonard, exactly like Pinker, opposes productive with learned by rote, which in turn is usually taken to be the converse of ‘generated by rule’ or ‘rule-based’. Therefore, in Leonard’s discussion, productive is used synonymously with ‘rule-based’ and a productive system refers to a ‘rule-based system’. Delsing (2002) discusses differences in the distribution of the linking element ‑s and the genitive ‑s in first parts of compounds in Modern Swedish. He states:

(4) Övriga förleder tycks inte styras av några produktiva regler i nutida svenska utan bör förklaras historiskt. (Delsing 2002: 77 [Emphasis added])

‘The remaining first parts do not seem to be governed by any productive rules in Modern Swedish but have to be given a historical explanation.’ [Translation, JB] In this paragraph Delsing is discussing the distribution of ‑s in already existing words in Swedish. Thus, productive here does not refer to the formation of new words. Rather, productive seems to mean ‘operative’, since it refers to a grammatical rule assumed to be at work (or not) in the synchronic grammar of Modern Swedish speakers. Productivity is defined by O’Grady, Archibald, Aronoff and Rees-Miller (2001: 157) in the following way:

(5) ... productivity – the relative freedom with which they [i.e. affixes] can combine with bases of the appropriate category. [Emphasis original]

In this work O’Grady, ArchiBald, Aronoff and Rees-Miller go on to describe the differences in how easily inflectional affixes are combined as opposed to derivational



Chapter 2.  Productivity

affixes. Therefore, on their account, the term productivity seems to mean ‘ease of combinability’. Demuth (2000), in her work on the nominal declension in the Bantu languages, assumes a dichotomy between grammatical productivity and semantic productivity. For her, grammatical productivity is at issue when nominal classification is based on morphological and/or phonological factors while semantic productivity is at issue when nominal classification is based on semantic factors. About the lack of early overgeneralizations in child language acquisition, Demuth (2000: 284) says:

(6) Thus, it appears that very young children do not have access to the productive semantics of the noun class system, but treat it rather as a formal grammatical system. [Emphasis added]

Since the semantics of a given nominal declension cannot be productive in and of itself, productive semantics here must mean ‘semantically transparent’ as it refers to a subclass of the noun declension which is semantically, as opposed to morphologically or phonologically, defined. Consider now the following quote from Fromkin and Rodman (1995):

(7) The rule that adds an ‑er to verbs in English ... appears to be a very productive morphological rule; most English verbs accept this suffix ... (Fromkin and Rodman 1995: 51 [Emphasis added])

Fromkin and Rodman refer to the high number of verbal bases available to the nominalizing suffix ‑er in order to clarify what they label “a productive morphological rule”. Thus, in the passage above, productive seems to mean ‘having a wide coverage’. Basbøll (2003) discusses the Danish stød, a syllabic prosody found in only a subset of the existing vocabulary, and its occurrence with new words in the Danish language. Consider Basbøll’s explanation of his notion of “full productivity” of certain morphological endings:

(8) ... as well as some endings which are ‘productive’ in the sense that they are added to new words as default ... (Basbøll 2003: 10)

Obviously, for Basbøll, productivity is closely linked with default status. Durie (1997), and following him Enfield (2002), discusses verb serialization and the fact that not all verbs are equally accepted in such constructions. On this matter, Durie (1997: 322) states:

(9) [T]he productivity of verb serialisation is constrained in such a way that a large variety of syntactically well-formed verb combinations will be rejected by native speakers ...





Productivity

The fact that only a limited set of verbs can occur in verb serialization constructions in certain languages is taken by Durie as evidence for the limited productivity of such constructions. Thereby, productivity is equated with the lack of restrictions on, or with the openness of, the relevant constructional schema. This means that the more verbs there are that can occur in a syntactic construction, the more open or general the construction is, and thus, the more productive Durie considers it. In this discussion, therefore, the term productive clearly means ‘schematically open or non-restricted’, in other words, ‘general’. This is also the sense in which the term productivity is used by Goldberg (1995: Ch. 5) in her discussion of the ditransitive construction in English and its “partial productivity”, which is manifested by the fact that only certain semantically well-defined clusters of verbs can instantiate the ditransitive construction while the occurrence of other verbs is excluded (for the same use, see also Goldberg and Jackendoff 2004 and Boas 2005 on the resultative construction). The notion of productivity is treated in a similar fashion, albeit not identically, in the literature comparing the syntax of idiomatic expressions with general syntactic rules. For Kay and Fillmore, for instance, productivity seems to be the same as schematicity: (10) ... a construction-based approach appears to provide promise of accounting both for the relatively idiomatic and for the abstract and more fully productive aspects of language. (Kay and Fillmore 1999: 31 [Emphasis added]) Kay and Fillmore view idiomatic expressions as instances of more general syntactic constructions or schemas. Such schemas or constructions are by most usagebased construction grammarians taken to exist at different levels of schematicity, ranging from highly abstract schematic and thus general constructions to specific lexically-filled concrete instantiations of the more general schema, with miscellaneous constructions and subconstructions found at various intermediate levels (Langacker 1988, 2000, Bybee 1995, Clausner and Croft 1997, Croft 2003, Barðdal 2001b, 2004, 2006a–b, Iwata 2008, Eckhoff 2009). This will be discussed further in Section 2.5 below. However, on this view of language and grammar, the highest level of abstraction is also the most schematic one and thus the most general level with the least restrictions. So, as opposed to the ‘schematically open or general’sense of the term productivity discussed in the preceding paragraph, in the quote in (10) the adjective productive is associated with schematicity itself, and not the fact that full abstractness or schematicity also entails the least lexical and/or semantic specifications.



Chapter 2.  Productivity

That productivity is equated with schematicity in Kay and Fillmore’s work is confirmed by the following quote from Fillmore, Kay and O’Connor, where productivity is opposed with listedness: (11) In spite of the fact that it is host to a large number of fixed expressions the form [the X‑er the Y‑er] has to be recognized as fully productive. Its member expressions are in principle not listable. (Fillmore, Kay and O’Connor 1988: 507 [Emphasis added]) It thus seems, according to Fillmore, Kay and O’Connor, that categories which are so small that they are in principle listable do not exist as abstract schematic entities and can thus not be regarded as exhibiting productivity. In their paper “Semantic productivity and idiom comprehension”, McGlone, Glucksberg and Cacciari (1994) discuss what they call variant idioms. Certain idiomatic expressions are characterized by a high degree of analyzability and can thus undergo both lexical and semantic modification, which again leads to a variant form of the original idiom. McGlone, Glucksberg and Cacciari give break the ice as an example of an analyzable idiom and shatter the ice as a variant form, which in this case has been lexically altered with a subsequent modification of the meaning. The substitution of shatter for break entails that an uncomfortable social situation has not only been broken down but also that it has happened “flamboyantly in one fell swoop”. McGlone, Glucksberg and Cacciari (1994: 170) speak about such modified idiom variants as “semantically productive idiom variants”. Thereby, semantically productive has received the meaning ‘semantically modifiable’ on their approach. Another related use of the term productivity can also be found in work on idioms. Consider the following quote from Gibbs (1995: 98): (12) Many linguists have also noted that the noncompositional nature of idioms explains why idioms tend to be limited in their syntactic and lexical productivity. For example, one cannot syntactically transform the phrase John kicked the bucket into a passive construction (i.e. *The bucket was kicked by John) without disrupting its nonliteral meaning. [Emphasis added] By syntactic productivity Gibbs is here referring to syntactic ‘flexibility’ or ‘manipulatability’ as opposed to syntactic ‘frozenness’, which in turn refers to the fact that many idioms cannot be altered syntactically but seem to be stored with their rigid word order.





Productivity

In a similar vein, A. Smith (2001) contrasts rigidity or frozenness with productivity by equating frozenness with non-productivity: (13) This last point accounts for why we find relic constructions of non-productive morphological and syntactic patterns of high token frequency. (A. Smith 2001: 378 [Emphasis added]) Consider also the quote in (14) below, taken from Bybee and Thompson’s discussion of the lexically-distributed differences found with the Indicative and the Subjunctive in Canadian French, originally studied by Poplack (1992, 1995): (14) ... we also have to bear in mind that not all contrasts and distributions are meaningful or functional. Some patterns represent a lexically-arbitrary residue of formerly productive patterns. (Bybee and Thompson 1997: 384 [Emphasis added]) Here meaningfulness/functionality is indirectly linked with productivity, while historical residue serves as an intermediate (possibly antonymic) link connecting the two. A better example of productivity being directly associated with functionality can be found in the following quote from Michaelis and Ruppenhofer: (15) If we assume that the productivity of a form is evidence of a specific function ... (Michaelis and Ruppenhofer 2001: 5) This statement refers to the idea that productivity is the materialization of a semantic/pragmatic function of a particular form. Yet another sense of the term productive, namely ‘developing new functions’ is attested. In Ritz and Engel’s work (2003) on the present perfect in Australian English, productivity is defined in the following way: (16) Productivity is understood here as the use of the form in new contexts such that new meanings are created. Ritz and Engel show that in Australian English the present perfect can be used in non-standard functions, for instance, to denote certain past events and events of particular salience in narratives. Ritz and Engel argue that the present perfect in Australian English has expanded functionally, since these functions are unattested in earlier periods. Such a functional expansion is by them regarded as an example of the category’s productivity. Sandøy discusses the development of, and the variance in, the distribution of the nominal suffixes ‑ing and ‑ning in North Germanic and Modern Norwegian. It turns out that the suffix ‑ning can only be found in a limited set of words in Modern Norwegian, and it is not clear in all instances whether the initial ‑n in ‑ning belongs to the suffix or the base. Because of that, Sandøy argues that it is more



Chapter 2.  Productivity 

correct to say that it is the ‑ing suffix that is found in Modern Norwegian and not the ‑ning suffix: (17) ... men noko klårt skilje mellom dei to suffiksa finst ikkje lenger, slik at det i dag kan vere rettast å seie at det er berre ing‑suffikset som er produktivt av dei to. (1997: 161) ‘... but a clear distinction between the two suffixes does not exist anymore, so today it is probably most appropriate to say that it is only the ‑ing suffix which is the productive one of the two.’ [Translation, JB] Therefore, since Sandøy is discussing already existing Norwegian vocabulary and not the introduction of new lexical items, the term productive is here used in the meaning ‘occurring or existing in a language’. Haspelmath (2002:  98) begins his discussion of the productivity of derivational patterns in the following way: (18) ... a morphological rule or pattern is said to be productive if (and to the extent that) it can be applied to new bases and new words can be formed with it. This is the standard definition of productivity found within morphology. Hare and Elman (1995), in one of their papers on connectionist networks, give an overview of the development of strong to weak verbs throughout the history of English. About this change, they say: (19) For one thing, many strong verbs moved to other strong classes rather than to weaks, showing that at least some of the strong classes were semiproductive. (Hare and Elman 1995: 64) Here, a minor class is regarded as being partially productive on the basis of attracting already existing words from other classes. Thus, productive refers to the ability to attract, not new items, but already existing items in a language. Aronoff (1983: 168) discusses the fact that in his frequency counts of competing suffixes there is not a single example of a word containing the suffix ‑ibleness, while words containing the suffix ‑ibility are easily found. He relates this to differences in productivity: (20) The only explanation for this data is that Xibleness is not merely less productive than Xibility, it is dead, not a viable W[ord] F[ormation] P[attern] of the language. [Emphasis added] This quote shows that Aronoff equates productivity with ‘living’ and non-productivity with ‘non-living’ or ‘dead’.



Productivity

Finally, the following quote from Tomasello (2000: 211) shows how the term productivity is sometimes used within research on language acquisition: (21) ... both of which focus on children’s productivity, that is, their use of language in ways that go beyond what they have heard from adults. Observe that this definition of productivity is ambiguous between two readings: a) deviating from adult language, b) consistent with adult language but unheard or unencountered by the child. The first reading covers cases where children produce past-tense forms like throwed and transitive sentences like I disappeared him, which are hardly found in the language children are exposed to. The second reading covers forms that children may not have heard but which are nevertheless completely grammatical and regular; as for instance when they produce regular inflectional forms of highly infrequent words or nonce verbs which have only been presented to them in the infinitive. It needs to be stressed that the term productive, with the meanings listed above, has not necessarily been used in any technical sense in the sources quoted. The degree of conventionalization found for each sense probably also varies. It is not uncommon that the same author uses the term in several different senses. The quote in (21), for instance, gives Tomasello’s explicit definition of productivity as children’s use of language that goes beyond what they have heard from adults. However, Tomasello also uses the adjective productive with two other meanings in the very same article: on p. 210 he speaks of “lack of productivity” when referring to lack of children’s language production. On p. 236 he discusses the Incredulity construction in English, exemplified by instances like Him be a doctor!, and states that this construction is “highly productive in the sense that any fluent speaker of English can generate innumerable further exemplars”. This last sense of productive refers to the construction’s high degree of schematicity and is therefore clearly different from the ‘deviating from adult language’- and the ‘being produced’-senses also employed by Tomasello. Also Bybee (1995) uses the term productive synonymously with both ‘regular’ and ‘default’, even though she explicitly defines productivity as “the ability to be extended to new items” (1995: 426). This is evident from the following two citations from her 1995 article: (22) The main issue to be discussed is the role of type frequency in the determination of regularity or productivity. (Bybee 1995: 426 [Emphasis added]) (23) I will argue that default status or productivity is not necessarily associated with source-oriented rules. (Bybee 1995: 444 [Emphasis added])



Chapter 2.  Productivity 

In (22) above, Bybee is not stating that she will discuss the role of type frequency for either regularity or productivity but instead she is equating the two. The same goes for (23); here, Bybee takes default status to be the same as productivity. The various senses and meanings of the terms productive and productivity, discussed above, are probably located at several intermediate points on the continuum of semantic change, as it is outlined by Traugott and Dasher (2002: 34–40). According to their model, The Invited Inferencing Theory of Semantic Change, meaning variants start out as contextually invited inferences, or factual entailments, in certain local situations. When that happens, the meaning variant is only an inference and the original meaning is maintained; thus a certain token-specific meaning has arisen. If the meaning variant gains symbolic value and spreads to new contexts, a type-specific meaning can be said to have emerged. In such instances, both the original and the new meaning variant are equally plausible in the original context and the generalized inference has become pragmatically polysemous. Finally, generalized invited inferences can develop into separately coded senses or meanings of their own, giving rise to a semantically (as opposed to pragmatically) polysemous meaning. When that happens, pragmatic ambiguity has disappeared and semantic polysemy is detained. My primary goal here, however, is not to pinpoint the location of the various senses and uses of the term productive on the cline of semantic development. Rather, my aim is to show that the terms productive and productivity are used in various senses in the literature. I will argue in next section that these senses and uses reflect three different subconcepts of the productivity concept in linguistics. In this section, I have shown that the terms productive and productivity have been used fairly loosely in the linguistic literature, with several distinct, although possibly related, meanings. Not all of the afore-mentioned senses are conventionalized or used as technical terms in the works cited here. As we saw in the examples from Tomasello (2000) and Bybee (1995), even scholars who employ the term as a technical term can be found to use it in non-technical, possibly even nonconventionalized, senses. And this is true of the way many linguists use these terms. In next section I will put forth an analysis of the possible relations and interrelations between the different senses and synonyms of productivity surveyed in this passing section. 2.2 The concepts of productivity In this section I propound that the metaconcept of productivity in linguistics is structured as to contain three concepts of productivity which either entail, or are concomitant with, each other in naturally occurring linguistic categories. This



Productivity

proposal is based on a semantic analysis of the nineteen senses and uses of the term productive and productivity discussed in the preceding section. When taking together all the uses and senses of productive and productivity documented in the section above, three concepts of productivity can be discerned: 1. Productivity as generality 2. Productivity as regularity 3. Productivity as extensibility However, before addressing the issue of how the productivity concept became so complex and why it is structured as it is, let us first consider the question of how the various senses and uses of the terms productive and productivity, documented in 2.1 above, are related to these three subconcepts of productivity and/or to each other. Several of the senses are elaborations or further specifications of one of the subconcepts. The following senses can, for instance, all be regarded as elaborations of the extensibility concept: (14) ‘developing new functions’, (16) ‘occurring with new/novel/nonce items’, (17) ‘attracting already existing items’, and (19) ‘deviating from adult language’. These are all further specifications of how extensibility can be manifested in language, and are labeled with E for the Elaboration Relation in Figure 2.1, in which the three concepts, the senses and the sense relations are graphically laid out. Recall that the sense in (19) ‘deviating from adult language’ can have two readings. Only the first reading ‘differing from adult use’ is an elaboration of the extensibility concept. The second reading ‘used in a regular fashion although unheard from adults’ is an elaboration of the regularity concept. In addition to the Elaboration Relation, other types of sense relations can also be discerned, like the Entailment Relation (Ent), the Inferentially Implied Co-occurrence Relation (Co), Metaphorical Relation (M), and the Antonym-to-Antonym –> Synonym Relation (AAS). Beginning with the Entailment Relation, sense (15) ‘occurring or existing’ is entailed by sense (16) ‘occurring with new/novel/ nonce items’, since occurrence with new items presupposes occurrence in general in a language. For the AAS Relation, consider the following facts: Extensibility can have many opposites, of which one is ‘rigidity’, another is ‘frozenness’, and a third is ‘historical remains’. These opposites of extensibility can again function as opposites to other senses, like ‘flexible’ which can be the opposite of ‘rigid’ or ‘frozen’, and ‘meaningful’ which can be the opposite of ‘historical remains’. Such opposites of opposites can ultimately result in “synonymous” uses of certain lexical items. This is, I believe, how sense (12) ‘flexible vs. rigid or frozen’ and (13) ‘meaningful vs. historical relics’ have come to be used synonymously with productive.

E

Co

Transparent or compositional

Regularity

E

Rule-based

Ent

Schematic vs. listable

E

E

Frequent

Modifiable

Co

E

Figure 2.1  Making sense of the different senses of productive

E

AAS

Extensibility

Flexible vs rigid or frozen

E

Deviating from adult language

Default

AAS

Co

Having a wide coverage

Generality

E

M = Metaphorical relation, E = Elaboration relation, Ent = Entailment relation, CO = Co-occurrence relation, AAS = Antonym to antonym –> Synonym relation

Easily combinable

E

Operative

Ent

Schematically open or non-restricted

E

E

M

Meaningful vs. historical relics

E

Occurring

Living vs. dead

Developing new functions

Attracting existing items

Occurring with new items

Ent

Chapter 2.  Productivity 



Productivity

Turning to the generality concept, the following senses can be regarded as elaborating it: (7) ‘having a wide coverage’, (8) ‘default’, (9) ‘schematically open or nonrestricted’ and (10) ‘schematic (as opposed to listable)’. All these senses are further specifications, or elaborations, of how the generality concept can be realized in linguistic categories. The regularity concept has the following elaboration senses: (2) ‘regular’, (3) ‘rule-based’, (5) ‘easily combinable’ and (6) ‘transparent or compositional’. The sense in (4) ‘operative’ entails sense (3) ‘rule-based’ since ‘operative’ presupposes a rule-based system. Also, the sense in (11) ‘lexically and semantically modifiable’ entails sense (6) ‘transparent or compositional’ since modification presupposes transparency/compositionality. In other words, an opaque category will not be analyzed as being compositional and thus no parts of a non-composite structure can be replaced with other parts. A transparent/compositional category, in turn, invites the possibility of replacing certain parts with new parts. In that way, the sense (11) ‘modifiable’ entails sense (6) ‘transparent or compositional’. The sense in (1) ‘frequent’ stands in an Inferentially Implied Co-occurrence Relation to all three subconcepts of productivity, since high frequency is concomitant with a high degree of generality, a high degree of regularity and a high degree of extensibility. Finally, the synonym of productive in (18) ‘living’, as opposed to ‘dead’, can be regarded as a metaphor for the whole metaconcept of productivity, invoking the image of language as a living organism. Turning now to the three subconcepts of productivity, regularity, generality and extensibility, the relation between them can be characterized as that of entailment and co-occurrence: General categories are always extendable while extendable categories need not be general.  General categories may or may not be regular, as regular categories may or may not be general. And finally, regular categories may or may not be extendable, exactly as extendable categories may or may not be regular. The first relation, the one between generality and extensibility, is an entailment relation, as generality entails extensibility. The two others, however, do not involve entailment but are relations of possible co-occurrence. In other words, regularity does not necessarily entail extensibility but the two may well be concomitant in certain categories in certain languages. The metaconcept of productivity is illustrated in Figure 2.2, where Ent stands for an Entailment Relation, symbolized with a one-way arrow, and Co for an Inferentially Implied Co-occurrence (or Concomitance) Relation, shown with two-way arrows.



Chapter 2.  Productivity 

Generality Co

Ent

Regularity

Extensibility Co

Figure 2.2  The three subconcepts of productivity

To exemplify, the category of weak or regular verbs in English is highly general since there are no phonological or morphological restrictions on its members. In other words, a verb does not need to comply with any phonological and/or morphological requirements to form a past tense with the ‑ed dental suffix. Therefore, the category of weak verbs is a very extendable one; most, if not all, new verbs entering the English language take on the behavior of weak verbs. However, extensibility is not restricted to generality. Highly specific categories are also extendable although the range of such extensions is much more limited. The category of strong or irregular verbs, for example, is not a general category, but a highly specific one. The irregular past tense in English can be extended to new or already existing verbs but only if the candidates fulfill certain morphological and/or phonological requirements (cf. Bybee and Slobin 1982). Thus, generality leads to extensibility while extensibility in itself is not dependent on generality. Regarding regularity, for many well-known languages, the most general categories are typically also fairly regular, again like weak verbs in English. In this case, however, regularity does not follow logically from generality, and vice versa, but is simply a matter of non-necessitated co-occurrence. Consider, strong or irregular verbs again. For several of the subclasses of strong verbs there is a systematic vowel alternation which defines the pattern. An example is the vowel alternation found for the so-called strung-verbs (see Section 2.5 below), where for instance /i/ may alternate systematically with /æ/ or /%/, as in sing–sang, across several verbs. Strong or irregular verbs are thus not necessarily irregular, but can be quite regular, even though this is not suggested by the term irregular which clearly is a misnomer. With this subclass of verbs we get regularity but not generality, as the pattern is phonologically highly restricted. The same is true for the relation between regularity and extensibility; one of the concepts does not necessarily entail the other, but for many categories in several languages, regularity and extensibility are simply concomitant with each other,



Productivity

perhaps because regular categories are often the most general ones anyway. These entailments and factual co-occurrences are, in my opinion, the cause of the complexity of the productivity concept in linguistics. Were it not for the fact that regularity is concomitant with both generality and extensibility in the largest and most productive categories in many languages, regularity would probably not have been associated with productivity by linguists in the first place. My aim in this section has been to show that the variation in the use of the terms productive and productivity in linguistics reflects at least three underlying concepts of productivity which are related in different ways. The disentangling of the various subconcepts of productivity is essential both for the individual scholar and his/her research on productivity, and hence for the furtherance of this book, and also for the general theoretical debate on productivity. Since terminological and conceptual confusion only leads to pseudo-debates and no real understanding of the issues at stake, it is important to distinguish between the subconcepts and the senses of productivity that exist in the “concepticon” of us linguists so that clarity and real scholarly insight can be gained. To summarize this section, I have suggested that the metaconcept of productivity within linguistics can be divided into three main subconcepts, namely those of generality, regularity and extensibility. All the senses or uses of the terms productive and productivity documented in 2.1 above can be related to these three subconcepts in one way or another. The sense relations proposed here are Elaboration, Co-occurrence, Entailment, Antonym-to-Antonym –> Synonym and Metaphorical Extension. Finally, I have argued for the necessity of dismantling the metaconcept of productivity into subconcepts in order to set the ground for a fruitful discussion of productivity. I will now give a compendious overview of the definitions and measurement of productivity suggested in the literature, before I turn to the main subject of this study, namely syntactic productivity. 2.3 Definitions and measurements of productivity In this section I briefly present the definitions of productivity found in the literature on morphology before I introduce my own hypothesis, namely that syntactic productivity, as extensibility, is a function of a construction’s type frequency, semantic coherence and an inverse correlation between the two. 2.3.1 Existing definitions and measurements The majority of all linguistic work on productivity has been carried out within the field of morphology. This accounts for the fact that most definitions of productivity,



Chapter 2.  Productivity 

and the methods to measure it, have been suggested so as to handle word formation and derivational patterns. In spite of the fact that the use of the terms productive and productivity is not always very precise, as demonstrated in 2.1 above, there still exists a reasonably strong consensus on how morphological productivity should be defined technically. Aronoff, for instance, defines it as in (i) below and Bybee (1995) and Bybee and Thompson (1997) give the definition in (ii). Both definitions are meant to capture the extensibility of a word formation pattern, i.e. whether, and to which degree, the word formation patterns of a given language are available when new words come into existence. As evident from (iii), Bauer (2001) takes this definition one step further by adding a usage-based dimension to it. i. The probability of a word formation pattern accepted as the model for new/ potential words (Aronoff 1983: 163, fn. 1) ii. The likelihood that a pattern will apply to new forms (Bybee 1995: 430, Bybee and Thompson 1997: 384) iii. The interaction between the potential of a morphological process to generate repetitive non-creative forms and the degree to which it is utilized in language use to yield new lexical items (Bauer 2001: 211) The main debate within morphology does not, however, evolve around how productivity should be defined, but rather how it should be measured and operationalized. Several methods, more or less refined, have been suggested as measurements of morphological productivity (see Rainer 1987, Plag 1999 and Bauer 2001 for overviews and critical discussions). As early as in 1976 Aronoff suggested that the proportion between actual lexical items and potential lexical items, generated by a linguistic rule, could serve as a method to measure productivity. To exemplify, the nominal suffixes ‑ness and ‑ity both combine with adjectives to form abstract nouns. If ‑ness is found in more existing words than ‑ity, then it is also more productive than ‑ity. A few years later Aronoff (1983) suggested another procedure, namely to compare the token frequency of a lexical item to the mean token frequency of other items in a corpus. The items with the lower token frequency should have been formed by productive rules since infrequent words, as opposed to high-frequent words, are generated on the fly and not stored in memory. For instance, if words containing the nominal suffix ‑ivity are higher in token frequency than words containing the suffix ‑iveness, that can be considered to reflect the higher degree of lexicalization of ‑ivity words. The lower token frequency of ‑iveness words, by contrast, can be interpreted as showing that these words are actively generated on-line and do not exist in speakers’ minds as lexicalized unanalyzed forms. A third method to measure productivity is to count the number of new words coined according to a particular morphological pattern during a specific period of



Productivity

time. Anshen and Aronoff (1989) have carried out such a count for four different time periods in Modern and Early Modern English. By comparing the number of new words listed in dictionaries from 1600–1900 they found that ‑iveness words are always much higher in type frequency than ‑ivity words, ranging from 66 to 91% of the total number of types found for the two patterns. Lieber (1981: 114–15), however, suggests that productivity should be measured in terms of the number of bases a morphological process may be applied to. The more words or bases an affix may be attached to, the more productive is the rule generating the potential output. To give an example, the English nominal suffix ‑th is much less productive than the suffix ‑ness because it can only be combined with a limited set of adjectival bases (long, strong, wide, etc.), while ‑ness can combine with considerably many more. This measurement is thus based on the input category of a given pattern and not the output category, as are the measurements discussed above and suggested by Aronoff (cf. Plag 1999). Baayen and Lieber (1991), in contrast, argue that productivity can be measured by calculating the ratio between the overall token frequency of words of a particular morphological pattern in a given corpus, and the number of words of the same pattern which only occur once in the corpus. Such once-occurring tokens are referred to in the literature as hapaxes. Thus, if the ratio between ‑ness hapaxes and the total token frequency of ‑ness words in a corpus is higher than the ratio been ‑ity hapaxes and the total token frequency of ‑ity words, it can be concluded that ‑ness is more productive than ‑ity. This is a consequence of the fact that hapaxes are on-line formations while high-frequent tokens are more often lexicalized, as originally pointed out by Aronoff (1983). This measure has been called the category-conditioned degree of productivity (cf. Baayen 2001: 157). A revised version of this method has been suggested by Baayen and Lieber (1991) and Baayen (1992). They take as a starting point the number of hapaxes found in a corpus, of various morphological patterns, and compare them with the type frequency of each morphological pattern in that same corpus. This procedure, termed global productivity by Baayen, measures the ratio between type frequency and hapaxes of a given pattern rather than the ratio between overall token frequency and hapaxes (see next subsection on type vs. token frequency). Baayen (1993) suggests a third formula which calculates the ratio between the number of hapaxes of a particular morphological pattern in a corpus and the total number of hapaxes in the corpus. Thus, the number of hapaxes for ‑ity and ‑ness are here being compared with each other and not with the overall type or token frequency of each pattern. This measure goes under the name hapax-conditioned degree of productivity. Baayen has also elaborated with other measurements, such as the unconditioned degree of productivity and simple type frequency (cf. Baayen 2001, Hay and Baayen 2002).



Chapter 2.  Productivity 

These measures reflect different aspects of productivity and how it can be realized in texts, dictionaries and larger corpora. For instance, all the measures based on hapaxes take it as a starting point that items occurring only once in a large enough corpus are semantically decomposable, and thus that they are also morphologically compositional. Thereby, hapaxes reflect the productivity (in the sense of ‘combinability’) of their particular word formation patterns. Lexical items with high token frequency, on the other hand, are more likely to have developed specialized meanings and are as such less analyzable semantically and morphologically. 2.3.2 Predicting productivity The hypothesis that I will be pursuing in this study is that the type frequency of a schema and its coherence appropriately predict a schema’s productivity (in the sense of ‘extensibility’). In other words, productivity is a function of both type frequency and coherence. By type frequency I mean the total number of types which can instantiate a construction. For morphological constructions, like the ‑th and the ‑ness abstract nominal constructions, the ‑th construction is low in type frequency since there are so few bases that can occur together with it. The ‑ness construction, on the contrary, has a much higher type frequency since there are many more bases that combine with it. For a syntactic construction, like for instance the ditransitive construction, all predicates which can instantiate the ditransitive construction together make up its type frequency. Type frequency should not be mixed up with token frequency; the latter refers to the total occurrences of either one or all the types of a construction in a text or corpus. That is, if one and the same verb occurs very often in the same argument structure construction, which perhaps is low in type frequency, then the total token frequency of the construction is high despite its low type frequency. Token frequency is sometimes also referred to as text frequency. By schema coherence I mean the internal consistency found between all the members of each schema or a category. Schemas or constructions can be morphologically, phonologically and/or semantically defined. I take syntactic constructions to be primarily semantically, sentence-functionally or pragmatically defined, although morphophonetics can play a role in a syntactic construction’s delimitation. To exemplify, an English verb cannot occur in the ditransitive construction unless its semantics is compatible with the semantics of the construction. However, verbs of Latinate origin, like for instance donate, cannot occur in the ditransitive in spite of being of the right semantic type (Pinker 1989, Gropen, Pinker, Hollander, Goldberg and Wilson 1989, Goldberg 1995:  128, amongst others). Therefore, the ditransitive construction in English is semantically defined with



Productivity

certain morphophonetic restrictions (for a comparison of the ditransitive construction in English and Icelandic and the lexical predicates instantiating it, see Section 5.1.2 below and Barðdal 2007). Morphological constructions are usually morphologically and/or phonologically defined, although a semantic common denominator can sometimes be found for the members of a particular morphological category. Demuth (2000: 280–82), for instance, shows that the subclasses in the nominal declension of certain Bantu languages, which are productively assigned to new words and borrowings, are usually morphophonologically defined except for one declension class whose assignment is based on semantic criteria. This means that a new word in Sesotho, a Bantu language spoken in South Africa, is assigned to class 3/4 if its initial consonant is /m/, to class 5/6 if it is /l/, etc., while new words denoting human beings are declined according to the pattern in 1/2 even though they begin with either /m/ or /l/. Class 1/2 is the class for human or animate objects. Therefore, the noun declension in Sesotho is mostly morphophonetically defined although one declensional subclass is clearly defined by its semantic properties. The idea that productivity can be determined from a construction’s type frequency and its coherence is not new. This is, in fact, the most prevalent view of productivity found within cognitive-functional linguistics and usage-based approaches to language and grammar. It has been argued for by Bybee (1995: 430), Goldberg (1995), Bybee and Thompson (1997), Clausner and Croft (1997), Barðdal (2001a, 2006b), Clausner (2002), Croft and Cruse (2004), amongst others. That type frequency is an important indicator of productivity has been observed by many morphologists (Aronoff 1976, Anshen and Aronoff 1989, Bauer 2001, Hay and Baayen 2002). The role of semantic coherence for productivity has also been pointed out in the morphological literature (Zimmer 1964, Aronoff 1976, Plag 1999). However, a serious evaluation of the interaction of these two factors, type frequency and semantic coherence, has not been systematically carried out for different linguistic categories. That is, the exact premises and predictions of this hypothesis of productivity have not been laid out in the literature. In this work I will argue that the type frequency and semantic coherence of a syntactic construction accurately predicts its degree of productivity. Unlike previous work, I emphasize that it is not only a question of both type frequency and semantic coherence, but also the inverse correlation found between them. In the remainder of this book I will elaborate further on this hypothesis of productivity for case and argument structure constructions and present empirical data supporting it from a language that still employs morphological case as an inseparable part of all its argument structure constructions, namely Icelandic. Before that, however, a short discussion of the concept of syntactic productivity is in place.



Chapter 2.  Productivity 

2.4 Syntactic productivity In this section I analyze two existing notions of syntactic productivity and argue that they represent the regularity and the extensibility concepts, respectively. These two existing notions of syntactic productivity correspond grossly to the inflectional–derivational dichotomy within morphology and interact with the type– token distinction. The notion of syntactic productivity, or syntactic creativity as it is sometimes also termed in the literature, referring to the language user’s ability to generate and understand novel sentences never encountered before is documented as early as in Zimmer (1964) and Chomsky (1965). It refers to the language user’s linguistic competence and is based upon the idea that knowing one’s language is all about knowing the rules of one’s language. This notion of syntactic productivity, however, is an elaboration of the regularity concept of productivity and not the extensibility concept (cf. 2.2 above). The extensibility concept is based on the extensibility of a pattern to new types, not new tokens. In syntax, therefore, the notion of productivity must refer to how extendable already existing argument structure constructions of a language are to new types in order to qualify as elaborating the extensibility concept. A type is regarded here as a verb or a predicate instantiating an argument structure construction. So, the more verbs that can occur in a construction, the more types are associated with it and the higher its type frequency is. However, the generation of sentences never encountered before only entails generating new tokens of already existing types. Consider now the quote from Langacker in (24) below. Observe that his use of the term productivity is ambiguous between a reading involving the regularity and the extensibility concepts: (24) Productivity is a matter of how available a pattern is for the sanction of novel expressions. (Langacker 2000: 26 [Emphasis added]) The ambiguity of (24) is caused by the vagueness of the phrase novel expressions. It can either refer to tokens or types. As token reference (24) would have to be understood as an elaboration of the regularity concept, while a type reference would yield an elaboration of the extensibility concept. Thereby, in syntactic research, both the regularity and the extensibility concept of productivity have been explicated. The traditional generative definition given in (i) is based on syntactic regularity while the more recent definition in (ii) aims at capturing syntactic extensibility: i. The language user’s ability to generate and understand sentences never encountered before



Productivity

ii. The ability of a syntactic pattern or argument structure construction to be extended to new lexical verbs. I believe, moreover, that these two notions are the syntactic correspondence to the inflectional–derivational dichotomy within morphology. As argued for instance by Josefsson (1997) and Zlatev (1997: 274–75) there are striking similarities between sentences and words in terms of internal structure, collocation patterns, co-occurrence restrictions, etc. In addition to this, I argue, words and sentences are also similar with regard to the type–token distinction, manifested in the application of the notion of productivity within both areas. This is laid out in Table 2.1. The division between inflection and derivation is based on a long-standing tradition in linguistics. This division, however, does not necessarily entail that the mechanisms behind inflection and derivation are very different, since in language production all words, both inflected and derived, are equally well produced by speakers, presumably irrespective of the inflectional–derivational dichotomy. The difference between inflection and derivation, therefore, corresponds in a way to the difference between tokens and types. Inflection entails that new forms or tokens are generated from already existing types, whereas derivation yields new types. Thus, derivation adds to the type frequency of the vocabulary while inflection adds to the text/token frequency of a derivational type. In that sense, inflected forms of already existing words categorize as regular performance, while the derivation of new words according to a particular word formation pattern, categorizes as extension. This is given in the first column in Table 2.1. Syntactic constructions like argument structure constructions are, in language use, always instantiated by some lexical material, for instance the predicates which occur in them. This is parallel to morphological constructions or words which are instantiated by their morphemes. That is, argument structure constructions are syntactic patterns, instantiated by a specific number of verbal types, exactly like word formation patterns are morphological patterns, instantiated by a certain Table 2.1  Regular performance vs. extensibility in morphology and syntax

Inflection = Regular Performance New Tokens Derivation = Extensibility New Types

Morphology

Syntax

The application of a morphological pattern to create new word forms of already existing words

The application of a syntactic pattern to create new instances of already existing syntactic patterns

The extension of a morphological pattern to create new words

The extension of a syntactic pattern or argument structure to new verbs



Chapter 2.  Productivity

number of morphologically-based types. Therefore, a speaker’s generation of a new sentence, never encountered before, is just a new token of a compositional syntactic pattern or argument structure construction, regularly formed according to the inventory of syntactic constructions of the language in question, instantiated by a known verbal type and other lexical material suitable for that particular communicative purpose. This is the regularity concept of productivity being manifested in syntax. When a new verb, however, enters a language, it always occurs as an integrated part of some argument structure construction. When that happens, a new type has been added to the list of verbs associated with that particular syntactic pattern, which has thereby been extended. This is the extensibility concept of productivity being manifested in syntax. The application of the regularity and the extensibility concepts of syntax are given in the second column in Table 1. In addition to this, new constructions can also evolve, both morphological and syntactic (cf. Croft 2000: 117–40). To give an example from morphology, a new suffix ‑oholic has arisen through a reanalysis of the form–meaning correspondence of the word alcoholic (cf. Smyk 2003). In addition to the etymologically correct analysis alchohol+ic, English speakers have also (re)analyzed the morpheme boundaries as alch+o+holic. This has given rise to word formations like chocoholic ‘one that is addicted to chocolate’, bloggoholic ‘one that is always blogging’, HarryPotteroholic ‘a devoted fan of Harry Potter’, and others like that. Similarly, new syntactic constructions can also come into existence through the remapping of form and meaning. Israel (1996) documents and describes the development of one syntactic construction, the way-construction in English, which successively arises through the repeated use of the lexical item way with motion verbs selecting for optional possessive path arguments. This gradually led to the mapping of that form with a function of its own, yielding Modern English examples like the following: “My daughter giggled her way through the story”, where no concrete motion is involved. The development of new morphological and syntactic constructions, however, does not represent the extensibility concept of productivity, as I use the term here, since it results in the emergence of new patterns, not in the extension of already existing patterns to new types. However, the emergence of new patterns illustrates yet another parallelism between morphological and syntactic constructions, which in turn suggests that we are here dealing with a tripartite division, approximately corresponding to tokens, types and patterns, found for both morphological and syntactic categories. This is laid out in Table 2.2. The first column refers to language use, in which both morphological and syntactic tokens are produced. The second column refers to extensions of morphological and syntactic patterns to new types, and the third refers to the emergence of new morphological and syntactic patterns.





Productivity

Table 2.2  Tokens, types and patterns in morphology and syntax New tokens of existing patterns

New types of existing patterns

New patterns

Morphology

Word forms

‑oholic

Syntax

Sentences

New lexical items of specific morphological patterns New lexical verbs of specific argument structure constructions

way-construction

Recasting this into the terminology of historical linguistics and language change, the extension of morphological and syntactic categories to new types exemplifies spread while the emergence of new morphological and syntactic patterns exemplifies innovation (cf. Croft 2000). A possible source of confusion, however, lies in the fact that, for morphological constructions, types and patterns coincide to a larger degree than they do for syntactic constructions. This means that the verb stabilize, for instance, both exemplifies a type and the whole morphological word formation pattern, while new verbs instantiating syntactic patterns will never make up the whole pattern but only a small part of it. The extension of regular patterns to already existing linguistic types can also be found in language history, usually labeled regularization, generalization, analogical leveling or productivity. Table 2.3 gives two examples of this, one from morphology and the other from syntax The morphological genitive in the Old Scandinavian languages had five allomorphs, i.e. ‑s, ‑ar, ‑a, ‑u and zero ‑ø, depending on declension classes. In Modern Table 2.3  The extensibility of regularity in morphology and syntax Modern Mainland Scandinavian

Old Mainland Scandinavian

Morphology

Syntax

sg. gen ‑s ‑ar ‑a ‑u ‑Ø Nom-Acc Nom-Dat Nom-Gen

–>

sg. gen ‑s

–>

Nom-Acc



Chapter 2.  Productivity 

Mainland Scandinavian only one of the forms, ‑s, has survived. Throughout history, ‑s has piecemeal spread to contexts where the other forms used to be found, gradually becoming the only genitive marker in Modern Mainland Scandinavian (Delsing 1991, Norde 1997, 2006). Similarly, in Old Scandinavian, several twoplace argument structure constructions existed, including Nom-Acc, Nom-Dat and Nom-Gen, to mention only the most prototypical case patterns. In the modern languages, only the Nom-Acc construction (or its equivalent in a non-case language) exists; the others have gradually been lost through history (Delsing 1991, Falk 1997, Barðdal 2001a: 196–208, and Ch. 6 below). It has been argued that there is a profound difference between morphology and syntax with regard to the productivity of rules, since syntactic rules are always productive while morphological rules can be unproductive (cf. Haspelmath 2002: 40). This claim, however, is based on an undifferentiated view of the concept of syntactic productivity, I argue. As soon as we acknowledge that syntactic productivity can be realized as either regular performance or extensibility we find that unproductive syntactic constructions can also be identified in languages. Low type frequency, and semantically non-coherent, case and argument structure constructions qualify as candidates for being unproductive syntactic constructions. The Nom-Gen transitive construction in both Icelandic and German, for instance, does not attract any new items to it and can thus, with all certainty, be regarded as an unproductive syntactic construction (see the discussion of the Icelandic Nom-Gen construction in Section 3.1.2.3 and 3.2.1 below). In this sense, syntactic constructions can be as unproductive as morphological constructions, as the Nom-Gen construction is on par with the English ‑th schema for abstract nouns. However, the view that syntax consists of form–meaning correspondences like constructions, and a differentiated notion of productivity, is a prerequisite for discerning the existing parallels between morphology and syntax that I have laid out here. To summarize, in this section I have shown that there is a parallelism between morphological and syntactic constructions, with regard to the regularity and the extensibility concepts of productivity. I have argued that the inflectional– derivational dichotomy within morphology is similar to a token–type distinction, corresponding to the two notions of productivity within syntax, i.e. speakers’ ability to generate sentence tokens never encountered before, and argument structure constructions’ ability to be extended to new verbal types. In the remainder of this book I will focus on the extensibility concept of productivity within syntax.



Productivity

2.5 Type frequency, schematicity and degrees of productivity In this section I outline my approach to syntactic productivity, namely that it is a function of a construction’s: i. type frequency ii. semantic coherence iii. the inverse correlation between the two I also discuss full productivity, degrees of productivity and analogy, and the fact that the regularity and the generality concepts fall out naturally from the present definition. I then compare my approach to the traditional cognitive-functional linguistic approach, that productivity as a derivative of a construction’s most entrenched level, and argue in contrast for the standpoint that productivity is a derivative of a construction’s highest level of schematicity. Thereby, high type frequency constructions are assumed to exist at higher levels of schematicity, and at more intermediate levels of schematicity, than do low type frequency constructions. Their productivity domains also vary accordingly. Finally, I argue that high token frequency also contributes to productivity, although only at the lower end of the productivity cline, which involves analogy, as high token frequency makes individual lexical items more entrenched and thus more suitable as models for analogical extensions. On such an account, full productivity and analogical formations are located at the opposite ends of the productivity cline, representing two different sides of the same coin. As stated in Section 2.4 above, syntactic productivity is here defined as the extension of syntactic patterns or argument structure constructions to new types or verbs. Productivity is further regarded as the function of a construction’s type frequency and semantic coherence (cf. Section 2.3.1 above) and an inverse correlation between the two. This means that the higher the type frequency of a construction, the lower the degree of semantic coherence is needed for a construction to be productive. Conversely, the lower the type frequency of a construction, the higher degree of semantic coherence is needed for a construction to be extendable. This correlation is graphically laid out in Figure 2.3. To exemplify, consider transitive and ditransitive verbs in Icelandic. The class of transitive verbs is by far the higher one of the two in type frequency. The higher a category is in type frequency the less coherence can be found between its items. This is an immediate consequence of the fact that there are limits to how much internal consistency there can be between items of a large category, given an even distributional pattern of the items. I have counted 2,156 transitive verbs/predicates in Icelandic (see Section 3.1 below) but only ca. 150 ditransitive verbs (see Barðdal 2007 and Section 5.1.2 below). On the assumption that the vocabulary is fairly



Chapter 2.  Productivity 

Type Frequency High

Semantic Coherence Low

High

Figure 2.3  The inverse correlation between type frequency and semantic coherence

evenly distributed across semantic fields, a contention which may well turn out to be wrong, the 2,156 transitive verbs are bound to occupy larger regions of semantic space than the ca. 150 ditransitive predicates. In other words, we would not expect 2,156 transitive verbs to be dividable into the same number of narrowly defined semantic subclasses as 150 ditransitive verbs are, as that would entail that each semantic subclass of transitive verbs would exhibit a disproportionally high type frequency. This is a simple mathematical task; if we divide the 150 ditransitive verbs in Icelandic into seventeen semantic classes (cf. Barðdal 2007), which is considerably more classes than have been assumed for English (Pinker 1989, Goldberg 1995, Croft 2003), the average size of each class will be approximately nine verbs (given that most verbs belong to only one class). The average size of each semantic class of transitive predicates would however be 127 if we divide the 2,156 verbs into the same number of subclasses, i.e. seventeen. Now, obviously we would never expect there to be 127 verbs of consumption (eat, drink, sip, etc.), for instance, or 127 verbs of creation (bake, sew, paint, etc.) in any language, while the number of verbs in each class may well be somewhere around nine on average. Therefore, stated differently, the semantic scope of the transitive construction is bound to be much wider than the semantic scope of the ditransitive construction, as there must simply



Productivity

be more classes of transitive verbs in a language than classes of ditransitive verbs. Hence, because of the higher type frequency and the higher degree of generality of the category of transitive verbs in Icelandic, it is also expected that the category is more extendable. The statistical chances of a new verb matching the semantic criteria of the transitive construction are higher than the chances that it will match the criteria for the ditransitive, since ditransitives are, due to their much lower type frequency, distributed across smaller regions in semantic space (or vice versa, due to their local semantics they are much lower in type frequency). Such a small category can only be extended within its limited semantic range. Therefore, the category of transitive predicates is located in the upper left corner of the cline in Figure 2.3, due to its high type frequency and high degree of schematicity, while ditransitives are somewhere in the lower right field of the cline, where the more semantically restricted low-type-frequency constructions are situated. Observe that the cline in Figure 2.3 predicts that productivity is not an all-ornothing phenomenon but gradient, an assumption that is widely accepted within linguistics today (Bybee 1995, Plag 1999, Baayen 2001, Bauer 2001). However, not all linguists agree that productivity is gradable. It is, for instance, not clear from the discussion in Fillmore, Kay and O’Connor (1988) and Kay and Fillmore (1999), cited in Section 2.1 above, whether they regard productivity as an all or nothing phenomenon, i.e. whether it is only a property of full schematicity, or whether it is gradient. As I will argue below, the key idea of constructions existing at various levels of schematicity entails gradability and, thus, it implies that productivity is a gradient phenomenon. Kay (2002) discusses the difference in productivity between, on the one hand, highly regular and fully general constructions, and, on the other, what he (following Fillmore 2002), labels “patterns of coining”. Kay takes certain idiomatic expressions, as well as some more established constructions like the Caused-motion construction, to represent patterns of coining. He argues that even though novel expressions of such patterns can be found, these novel instances are not generated through productive processes but are simply coined as analogies. He further argues that only productive generative patterns should be assigned construction status, i.e. should be regarded as constructions, whereas patterns of coining should be regarded as extra-grammatical and located outside of grammar proper. I doubt that anybody would disagree with Kay that there are certain lexical and/or semantic restrictions on which verbs can occur in the Caused-motion construction in English. However, the fact that the construction is not completely open schematically does not necessarily entail (a) that the Caused-motion construction should not be assigned constructional status, or (b) that it is unproductive. The first issue, the constructional status of the Caused-motion construction, is clearly a matter of theoretical preferences. That is, it depends on whether one



Chapter 2.  Productivity 

espouses a version of CxG where all form–meaning correspondences are assumed to be the basic units of language or not. The second issue, the productivity of the Caused-motion construction, is a matter of definition. It is clear from Kay’s discussion that, in his view, only completely general, fully schematic constructions count as productive. This is confirmed by the following quote from Kay (2002: 3): (25) The somewhat severe view of grammar adopted here excludes partially productive processes ... Fillmore (2002), however, assumes that productivity is a gradient phenomenon, which a grammar needs to be able to account for, but he nevertheless maintains the traditional assumption that there is a qualitative difference between instantiations of patterns being “generated” or “coined”: (26) We can distinguish two kinds of “creativity” in language. In one case there is the ability of speakers, using existing resources in the language, to produce and understand novel expressions. In the other case, the one for which we use the term coining, a speaker uses existing patterns in the language for creating new resources. This quote reveals that Fillmore makes a clear distinction, albeit with different terminology, between the regularity and the extensibility concepts. His term “coined” can also refer to the emergence of new patterns/constructions (cf. Section 2.4 above). Kay’s term “generated pattern”, on the other hand, refers to the regularity/generality concepts while “coined” refers to the extensibility concept. This means that when Kay argues against the productivity of the Caused-motion construction it is not that he argues that the pattern cannot be extended but rather that it is not completely and fully schematic. Therefore, for Kay productivity is regularity/generality and not extensibility, as I understand it. This conceptual and/ or terminological mix-up shows clearly that we need to tease apart the various concepts kept under the label of productivity in order to be able to have a fruitful discussion of it. Moreover, the arguments Kay presents against productivity as a gradient phenomenon are not valid since his whole point is to argue for the primacy of the regularity/generality concepts at the exclusion of the extensibility concept when defining productivity. Kay’s model, however, and his arguments, is not relevant for my concerns, as it is the extensibility concept and its gradient character that are being investigated in this book. Kay’s arguments are also based on a rule–analogy distinction, a distinction which will be rejected in Section 3.2.2 below. To return to the graph in Figure 2.3, observe that this representation of the extensibility concept not only entails that productivity is gradient but also that the regularity and the generality concepts can be derived from it. This is shown



Productivity

Type Frequency High Regularity–Generality–Open Schema

Different Degrees of Productivity

Analogy

Semantic Coherence Low

High

Figure 2.4  Different aspects of the cline of productivity

in Figure 2.4. At the upper end of the productivity cline we find full extensibility. In addition, both full regularity and full generality occupy that sphere, since the highest possible type frequency is bound to yield both full generality and full regularity. Full generality is the corollary of least specificity, which goes hand in hand with high type frequency. Full regularity, however, only co-occurs with extensibility, generality and high type frequency by incidence; it is not concomitant with these by any logical necessity (cf. Section 2.2). Therefore, the extensibility concept can be regarded as being of theoretical primacy with regard to productivity, while both regularity and generality are of secondary status as theoretical entities. The upper end of the productivity cline thus accounts for all three subconcepts of productivity, i.e. regularity, generality and extensibility. The lower the type frequency is of a category, the lower on the vertical axis is it located. The lower the degree of coherence, the more a category is to the left on the horizontal axis. Such a category will be far away from the productivity cline, which is an inverse cline from high type frequency and a high degree of schematicity to low type frequency and a low degree of schematicity. This means that a category A with low type frequency will only be productive if its degree of coherence is high, while a category B with intermediate type frequency has to exhibit intermediate degree of semantic coherence to yield new types. A category C of relatively low type frequency and little internal consistency will be located in the lower left-most sphere



Chapter 2.  Productivity 

Type Frequency High D

B

C

A Coherence

Low

High

Figure 2.5  The location of different categories on the productivity cline

of the graph, a category that will hardly show any signs of productivity at all. Conversely, a category D with very high type frequency and the highest possible degree of semantic coherence will, if such a category exists at all, be located at the upper right-most sphere of the graph. This is illustrated in Figure 2.5. Observe that the hypothetical category D exhibits simultaneously the highest possible type frequency and the highest possible degree of coherence. The question arises as to whether such a category can exist in a language, since it presupposes that the majority of the instances of a morphophonological category, for instance, will have the same or least a very similar morphophonological structure, or that the majority of the instances of a semantic category will have a very similar meaning. Now obviously, as discussed above, we would not expect all verbs in a language to be verbs of consumption or verbs of creation. Similarly, we would not expect all words in a language to be two-syllabic words, beginning in ma‑ and ending in ‑ta, to choose a random example. Therefore, the categories relatively close to the cline will show most signs of productivity, while categories further away from the cline will be either low/nonproductive (lower left-most sphere) or non-existing (upper right-most sphere). Further, the categories that are located close to the cline exhibit full productivity but only within their respective domains. And since the productivity domains of



Productivity

the categories at the lower right-most sphere of the cline are much smaller than the productivity domains of the categories at the upper left-most sphere, the “profit” will also vary accordingly. Consider now a real linguistic example: Past-tense formation in Icelandic is similar to English in that the overwhelming majority of verbs are weak (regular), while only a small number are strong (irregular). According to Ragnarsdóttir, Simonsen and Plunkett (1999) the proportions are very similar to the proportions in English, namely 96% are weaks and 4% are strongs. The strong verbs are neither semantically nor morphophonologically different from the weak verbs. The productivity cline for the past-tense formation in Icelandic can thus be modeled as in Figure 2.6, with w for weak verbs and s for strong verbs. Since the base/stem of weak and strong verbs in Icelandic do not differ in their phonological make up, they both constitute open schemas and only differ with regard to their type frequency. Only weak verbs then are situated at the productivity cline and should show signs of productivity. This is confirmed by the fact that all new verbs in Icelandic are inflected according to the weak paradigm. However, as Ragnarsdóttir, Simonsen and Plunkett point out, matters are not so simple. Not only is the category of strong verbs dividable into at least eleven subclasses but also weak verbs can be divided into four subclasses, depending on the Type Frequency High

W

Phonological

S

Coherence Low

High

Figure 2.6  The productivity of past-tense formation in Icelandic



Chapter 2.  Productivity 

Table 2.4  Weak and strong verbs in Icelandic Past Tense W1 W2 W3 W4 SLP SHP SM

‑aði -ti -di -ði [a] [øy:] [j/ou:, a:, je:, au:]

N 3845 525 355 210 30 32 62

F 75% 10% 7% 4% 0.58% 0.62% 1.2%

Phon. Delimitation None None None None Inf: [e, i, ja, jau] Inf: [j/ou:, j/u:] None

phonological shape of the past-tense suffix. Ragnarsdóttir, Simonsen and Plunkett have carried out an elicitation task in which they investigate the knowledge of four-, six- and eight-year old Icelandic children of the past-tense formation of a representative class of Icelandic verbs. The past-tense formation, the type frequency in real figures and percentages, and the phonological make-up of the classes they investigated are given in Table 2.4. Table 2.4 shows that one of the weak classes W1 is a majority class containing 75% of all Icelandic verbs, while the three other weak classes cover 4, 7 and 10% of the types. Ragnarsdóttir, Simonsen and Plunkett specifically investigated two of the largest classes of strong verbs, which differ with regard to their phonological coherence. One of them, labeled SLP, is of low phonological coherence with the stem vowel in the infinitive divided across two front vowels and two variants of the central vowel [a]. The second subclass, labeled SHP, shows more phonological coherence in that the stem vowel in the infinitive is divided across two variants of the back [ou:] and [u:], respectively. Ragnarsdóttir, Simonsen and Plunkett only investigated a very few instances of some of the other strong verb classes, which they subsume under one class of miscellaneous verbs, here labeled SM. According to the revised numbers of the type frequency of the various past-tense formations in Icelandic in Table 2.4, their location relative to the productivity cline should be modeled as in Figure 2.7 rather than as in Figure 2.6. According to Figure 2.7 only the largest class of weak verbs should be productive since the other three weak classes are located far away from the productivity cline, due to the openness of their schemas. The same is true for the strong class of miscellaneous verbs. The two other classes of strong verbs should be situated closer to the productivity cline but since neither of them exhibits a very high degree of phonological coherence they should not be located near the cline, according to the principles laid out here. The SHP class, for instance, is not coherent with regard to the consonant make-up, although the degree of vowel similarity is high. In this respect the SHP class in Icelandic differs from the sprung class in English, discussed



Productivity

Type Frequency High W1

W2 W3 W4 SM Low

Phonological SLP

SHP

Coherence High

Figure 2.7  A modified version of the productivity of past-tense formation in Icelandic

below (see Table 2.5), of which the vowels, initial consonants and final consonants show an exceptionally high degree of consistency across its members. The results of Ragnarsdóttir, Simonsen and Plunkett show that Icelandic children generalize mostly into the largest weak class. They do, however, also generalize into the other classes, both weak and strong, the degree to which depends on their age. The youngest children generalize most into the largest class, while the older children generalize less into it. The amount of over-extensions into the smaller weak classes and the strong classes are, however, in proportion to the size and homogeneity of the classes. There are fewer over-extensions into the strong classes than into the smaller weak classes, and of the strong classes there are more over-extensions into the larger more homogenous classes than into the smaller miscellaneous classes. It is interesting, however, that the over-extensions into both the smaller weak classes and the strong classes are always based on rhyme, while no rhyming effect is found for over-extensions into the large weak class. This suggests that the generalizations into the smaller weak classes are not based on a schema but on the concrete lexically-filled verb-specific constructions themselves. These findings raise the question of how large a category needs to be before speakers, including language learners, generalize across its members to form a schema. Although, the smaller weak classes are relatively large, it is not a given that the children partici-



Chapter 2.  Productivity 

pating in the experiment have come across all instances of each class during their short lifes. Therefore, the type frequency figures in Table 2.4 above, of these classes in Modern Icelandic, need not correspond to the type frequency and the size of these classes in the minds of language learners. To explore whether type frequency and schematicity can be teased apart, as argued by Bybee (1995), consider Figure 2.8 which demonstrates some points on the cline from high to low type frequency vs. high to low degree of schematicity. The left-most box represents the largest weak class in Icelandic, of which the type frequency is so high that the instances are bound to occupy and be evenly scattered across the whole phonological space. This is a completely open schema. The middle box represents the strong SHP class of which a high degree of vowel similarity can be discerned between the instances. Such a situation yields a low type frequency class with a restricted schema. The box to the right, however, symbolizes the strong verb class of miscellaneous types. I remind the reader that we are still being faced with a low type frequency class but as opposed to the category in the middle box, the instances of the category in the right-most box are fairly evenly scattered across phonological space, resulting in an open schema. The problem here is that we would hardly regard this category as forming an abstract schema. The instances are so few compared to their distribution that it becomes difficult to generalize across them. Instead, what we have is a limited number of isolated verbislands, scattered across phonological space. Observe that the fourth possibility, a restricted schema with high type frequency, is not found with past-tense formation in Icelandic, thus supporting my earlier contention that categories of type D in Figure 2.5 are hard or impossible to find. Now, having discussed the upper, intermediate and lower spheres of the productivity cline, the only part left is the right-most bottom part. At this point on the cline, the lowest type frequency possible is one and the highest semantic coherence possible is full synonymy. Observe, however, that this is exactly how analogy is traditionally defined in the linguistic literature. Analogical formations are extensions

High type frequency Open schema

Low type frequency Restricted schema

Low type frequency Open schema

Figure 2.8  The interrelation between type frequency and open schema



Productivity

based on only one model form (cf. Plag 1999: 17–22, Bauer 2001: 75–96, Haspelmath 2002, Boas 2003:  260–77, among others). I will discuss examples of that kind in chapters 3–6 below from the area of case, argument structure and verbal syntax. Those examples support my argument that the view of syntactic productivity advocated here, i.e. as a function of type frequency, semantic coherence and an inverse correlation between the two, in fact predicts not only that productivity is gradient, but also that analogy is the other side of the productivity coin. This means that analogy is located at one of the extreme poles of the productivity cline, while full productivity is located at the other pole. Everything in between is productivity gradient. In earlier work on productivity within cognitive-functional linguistics, Clausner and Croft (1997) compare morphological productivity with the productivity of metaphors, and Clausner (2002) compares metaphorical productivity with the productivity of spatial-graphical expressions. They define productivity as “the proportion of a schema’s range that can be instantiated as expressions” (Clausner and Croft 1997: 263). The range of a schema is in turn defined on the basis of existing expressions instantiating the construction. Therefore, on Clausner and Croft’s approach productivity is, as I interpret it, the degree to which a schema’s range is utilized in language. In practice, therefore, Clausner and Croft’s definition of productivity captures the relation between a construction’s type frequency and its schematicity, on the assumption that their term “expressions” refers to types and not tokens. The question arises, however, as to whether the relation between the two factors, type frequency and schematicity, is not a fixed one or at least a fairly fixed one. That is, does the type frequency of a given construction not correspond to its degree of schematicity? If that is the case, productivity can be reduced to either type frequency or degree of schematicity. However, as I discussed above in relation to past-tense formation in Icelandic, type frequency and schematicity are not necessarily concomitant with each other in small categories. This means that for low type frequency constructions type frequency by itself does not predict a construction’s productivity, since its degree of coherence is the other activating factor. Only very coherent low type frequency constructions are extendable. Similarly, for constructions exhibiting a high degree of schematicity, the high degree of schematicity in and of itself is not enough for a construction to be productive; the level of type frequency is equally important (recall the difference between the rightmost and the leftmost boxes in Figure 2.8 above). If a language exhibits two or more schematically open constructions within the same semantic range, the construction highest in type frequency will be the most productive one. Furthermore, Clausner and Croft’s definition of productivity does not make any reference to the formation/attraction of new items or the extensibility of a construction. It thus seems to be purely production-based, and the distinction between tokens, types and patterns is somewhat blurred. That is, it is not clear



Chapter 2.  Productivity 

whether a metaphor is a type or a pattern, and thus whether the expressions instantiating the metaphors are different types of the same metaphorical patterns or different tokens of the same metaphor types. While I agree with Clausner and Croft in principle, my view of the nature and status of schematicity differs from theirs. They take the traditional cognitive-functional linguistic stand that schematicity is a linguistic concept, needed for the description and organization of grammar. I argue, on the other hand, that schematicity should be treated as a psycholinguistic concept, first and foremost relevant for determining the level of generality at which a construction exists in the minds of real speakers. To clarify, on the traditional cognitive-functional linguistic approach, constructions are assumed to exist at all levels of schematicity (Langacker 1988, 2000, Clausner and Croft 1997, Bybee 2001, Croft 2003) while the most productive level is the level of entrenchment at which the type frequency is highest in proportion to the level’s schematicity or schematic range. My position is more usage-based in that I take the type frequency and the coherence of a schema to determine the actual level of schematicity at which the construction exists in the minds of speakers (ontological but not model-theoretical existence). This level of schematicity, i.e. a construction’s highest level, also determines the construction’s productivity. The higher the degree of schematicity, the more productive the construction is, and, conversely, the lower the degree of schematicity, the less productive the construction is. Thus, on my approach, the extensibility of a construction is a reflection of the highest level of schematicity at which the construction exists in the minds of speakers. Any levels, posited by linguists, higher than can be motivated by the data cannot be assumed to be psycholinguistically real (cf. Barðdal 2000, 2006a–b). On the traditional cognitive-functional linguistics approach, however, it is the most entrenched level of schematicity that determines a construction’s productivity and not its highest level. This is not to say that higher levels, or higher-level generalizations, cannot be posited. The fact that linguists argue for them shows clearly that it is possible to categorize at such high levels. The question, however, remains whether speakers make such high-level generalizations or whether only linguists do. In other words, the highest possible schematic levels perhaps only exist in the meta-analysis of linguists and not in the actual generalizations made by speakers (cf. Haspelmath 2002: 41). I take this to be an open empirical question. To give an example of what I mean by a construction’s highest level of schematicity, consider again the facts of transitive and ditransitive verbs in Icelandic. As I mentioned earlier, ditransitive verbs in Icelandic can be divided into seventeen subclasses, while transitive verbs, because of their higher type frequency, should be dividable into a considerably larger number of subclasses. Each subclass, in turn, consists of a specific number of lexical verbs which, in language use, occur in



Productivity

certain argument structure constructions. Such constructions are lexically-filled instances of higher level constructions and will be referred to here as verb-specific constructions (following a manuscript by Croft from 1999 which later came to be published as Croft 2003). The subclasses, on the other hand, are generalizations across a certain number of semantically-related lexical verbs. They will be referred to here as verb-subclass-specific constructions. Given a high enough number of subclasses of each construction it may be possible for speakers to generalize across them into even higher-level categories, here referred to as verb-class-specific constructions (again following Croft 2003, cf. also Barðdal 2001b, 2006a–b, Barðdal, Kristoffersen and Sveen 2007, Iwata 2008). The number of subclasses determines the highest possible level of schematicity, the highest of which is more or less devoid of any lexical semantic content, at best exhibiting relational meaning (cf. Section 4.2). The hierarchy of constructions that I am assuming here is thus built up in such a way that the lowest-level constructions at the bottom of the hierarchy are lexically-filled concrete instantiations which contain the most semantic information. The next lowest level in the hierarchy contains less semantic information, and so on, until we encounter the highest level with no lexical semantic information at all. At the highest level we thus find a completely schematic construction with only relational content. In this sense the hierarchy of constructions is a continuum from full lexicality to full schematicity and from lexical semantics to grammatical semantics. The principles of this organization are laid out in Figure 2.9. In accordance with a bottom-up view of the organization of language and grammar, I only assume the existence of higher-level constructions for categories that are general enough to be represented at the highest level of schematicity. The ditransitive construction in Icelandic, for instance, is not a very general construction. Abstract/schematic constructions

Verb-class 1

...

Verb subclass 1

...

Verb-specific 1

...

Figure 2.9  Different levels of schematicity

Verb subclass

Verb-specific 1

Verb specific 2



Chapter 2.  Productivity 

Figure 2.10  Different degrees of productivity according to the lexicality–schematicity continuum The highest level of schematicity that can be postulated for the ditransitive construction is a generalization over only seventeen subclasses. That represents a fairly low degree of schematicity and, by the same token, a high degree of semantic specificity. The highest schematic level of the transitive construction, on the other hand, is bound to be considerably more general due to its higher type frequency, and the construction thereby also exists at more intermediate levels of generalization and schematicity. Figure 2.9 illustrates the differences in levels of schematicity and degrees of semantic content found for two hypothetical constructions which differ as much in type frequency as the transitive construction (to the left) and the ditransitive construction (to the right) do in a language like Icelandic. Another way of illustrating the levels of schematicity found for two constructions of different type frequencies, given an equal amount of similarity across the items, is given in Figure 2.10. Since a high type frequency construction exists at a higher level of schematicity than a low type frequency construction, it is both more general and more entrenched as a whole than a low type frequency construction. A low type frequency construction, on the other hand, is both less entrenched, because of its lower type frequency, and its highest level of schematicity is also low, compared to a high type frequency construction. The highest level of schematicity of each construction, thus, corresponds to its productivity domain. Within cognitive-functional linguistics, however, the ditransitive construction to the right in Figures 2.9 and 2.10 would be represented as existing at the same level of schematicity as the transitive construction. That is, its highest level would be regarded as equally schematic as the highest level of the transitive construction. The only difference between them would be that the transitive construction is regarded as more entrenched than the ditransitive due to its higher type frequency (Langacker 1988: 148ff.). Such an account makes the same formal predictions on productivity as my account, the difference between them relates to their validity in terms of psycholinguistic reality. Given a bottom-up approach to grammar,



Productivity

Schema

Instance

Instance

(a) Productive

Schema

...

Instance

Instance

(b) Semi-Productive

Instance (c) Unproductive

Figure 2.11  Gradient productivity, according to Clausner and Croft (1997)

constructions can only be assumed to exist at abstract schematic levels if there are linguistic data in support of such an analysis. A low type frequency construction with a restricted schema, like the ditransitive, does simply not validate an analysis assuming a high degree of schematicity of the construction. Clausner and Croft (1997), in their comparison between morphological and metaphorical productivity, discuss three points on the productivity scale: high productivity, semi-productivity and non-productivity. Figure 2.11 is a simplified version of their illustration of these three points (1997: 271). The left-most unit in Figure 2.11 represents conventional metaphors, such as good is up/bad is down. They are highly productive since new expressions of them can easily be found and created. Therefore, for conventional metaphors this suggests that the higher-level schema is strongly entrenched (denoted with bold lines) while the instances are not. The middle unit in Figure 2.11 shows transparent idioms, such as spill the beans. They are only semi-productive since they are “relatively fixed semantic assemblies of specific concepts” (Clausner and Croft 1997: 266). Thus, transparent idioms are motivated but not predictable. They are also low in number. This is represented with a weakly entrenched schema relative to strongly entrenched instances (again denoted with bold lines and lack thereof). Opaque idioms, such as kick the bucket, are both unpredictable and unmotivated, since there is nothing in their form which is suggestive of their meaning. Therefore, a super-ordinate schema cannot be assumed at all. Instead, opaque idioms are represented as strongly entrenched lexically-specific islands with no generalizable over-arching schema. This is shown with the right-most unit in Figure 2.11. The morphological parallels to metaphorical productivity, argued for by Clausner and Croft, involve the regular past-tense formation of the dental suffix ‑ed, past-tense formation through ablaut and past-tense formation with suppletive morphology. The suffixation of ‑ed to form the regular past tense is highly productive in English. The pattern of irregular verbs forming past tense with vowel (and in some cases consonant) alternation is semi-productive, since such forms exist dialectally, have arisen historically, and can be elicited in nonce-probe tasks.



Chapter 2.  Productivity 

Finally, suppletive paradigms, such as go–went, are non-productive idiosyncrasies. Clausner and Croft’s comparison shows that there are clear parallels between morphological and metaphorical productivity. I believe, however, that the illustration in Figure 2.11 is misleading in certain respects. While I concur with Clausner and Croft that the difference between high productivity and semi-productivity lies in differences in schematic entrenchment, I also assume that schematic entrenchment, or degree of schematicity, partly correlates with type frequency. Thus, their three points on the productivity scale are more adequately represented as in Figure 2.12 which better captures the fact that the differences in type frequency, found for highly productive vs. semi-productive patterns, also entail the existence of intermediate schematic levels for the more productive pattern, which again results in higher degree of schematicity for the highly productive construction but a lower degree of schematicity (the converse of which is a higher degree of specificity) for semi-productive constructions. So the difference between highly productive and semi-productive patterns is not only a difference in the entrenchment of a super-ordinate schema, but rather a difference in both the type frequency of the two schemas and in the degrees of schematicity at which they exist. This last fact, that a schema’s type frequency and its degree of schematicity are to a large extent concomitant with each other, at least for high type frequency constructions, is not adequately represented in Figure 2.11. The final issue I want to discuss before I close this chapter has to do with the role of token frequency for productivity. It has been observed by Aronoff (1983) and others that high token frequency of individual lexical items often correlates with a high degree of lexicalization. It has also been argued by Bybee (1985: 132–34, 1995: 433–35, 2001: 118–26), Bybee and Thompson (1997) and Clausner and Croft (1997: 253–54) that high type frequency contributes to productivity whereas high token frequency detracts from productivity. The logic behind this assumption is that highly entrenched tokens do not contribute to schematization, i.e. they are not considered instantiations of their schemas by speakers, and because of that they do not reinforce their schemas. Such reinforcement is a necessary prerequisite for a schema to become entrenched. This is both a valid and an important observation. However, in my opinion this reasoning only holds true for high productivity and not for low productivity. Obviously, a highly entrenched token may not contribute to a schema’s entrenchment and thus not to its productivity. But a token is also a type, irrespective of its entrenchment, and given a gradient view of productivity on which constructions exist at different levels of schematicity, ranging from highly schematic, and thus highly general, to weakly schematic, and thus highly specific, speakers’ schematization or categorization also takes place at different levels of the continuum from high schematicity to high specificity. Therefore, for constructions that only exist at



Productivity

Schema

Subschema

Instance

Subschema

Instance

(a) Productive

...

...

Schema

Instance

Instance

(b) Semi-Productive

Instance (c) Unproductive

Figure 2.12  A modified version of the illustration of gradient productivity

low levels of schematicity, such as irregular past formation in English, a high degree of similarity, including token similarity, should be of utmost importance. Without token similarity no effect of rhyming on over-extensions should take place, for instance. In Chapter 3 below I will present data in support of my claim that there is a correlation between high token frequency and the choice of model verbs for analogical extensions, and hence that token frequency is important for productivity. To support the argument that high token frequency detracts from productivity, Bybee (1995: 435) compares two classes of strong verbs in English, the strung class and the swept class, which display a similar type frequency (thirteen vs. fourteen types), according to Moder (1992). However, only the first class shows signs of productivity and Bybee argues, based partly on data from Moder (1992), that the extensibility of this class is due to differences in token frequency: the strung class being of medium token frequency but the swept class being of high token frequency. Hence, since the type frequency of these classes is similar, the differences in productivity, according to Bybee, must be due to differences in token frequency with the more productive strung class being of lower token frequency than the less productive swept class. The problem with this argument is that these two subclasses of strong verbs also differ with regard to yet another factor, namely phonological coherence. I have not had access to Moder (1992) so it is not clear to me exactly which predicates are counted as instantiating these classes, given that Bybee and Slobin (1982: 288) list twenty-seven verbs as belonging to the strung class, and Hornby, Cowie and Gimson’s monolingual English dictionary (1974: 1003–06) lists twen-



Chapter 2.  Productivity

ty-one verbs of the strung class and twenty-two of the swept class. Moreover, a quick glance at these two classes, presented in Table 2.5, based on the list in Hornby, Cowie and Gimson, reveals that the strung class is phonologically much more homogenous than the swept class, not only with regard to the vowel change but also with regard to both the initial and the final consonant clusters. The swept class, by contrast, is only defined by its systematic vowel alternation. All the verbs in the input category of the strung class end in a nasal or a nasal cluster. The final consonants in the input category of swept verbs, however, are divided across bilabial and dental stops, labiodental fricatives, liquids, nasals, and one verb ends in a vowel. In addition, the class of initial consonants is also much more homogenous for strung verbs than for swept verbs. As many as thirteen strung verbs out of twenty-one starts with a s‑, while the initial consonants of the swept class are fairly evenly distributed across the consonant space and no one Table 2.5  The strung and swept classes in English strung begin cling dig drink fling ring shrink sing sink sling slink spin spring stick sting stink string swim swing win wring

swept began clung dug drank flung rang shrank sang sank slung slunk spun sprang stuck stung stunk strung swam swung won wrung

bereave bleed bread cleave creep deal dream feed feel flee keep kneel lead lean leap leave mean meet sleep speed sweep weep

bereft bled bred cleft crept dealt dreamt fed felt fled kept knelt led leant leapt left meant met slept sped swept wept





Productivity

consonant (cluster) seems to be more prominent than any other. Bybee’s metonymical use of swept for the swept class is therefore not as motivated as her use of strung for the strung class, since strung is a prototypical instance of the strung class while swept is not a better instance of the swept class than any other verb. In other words, the internal consistency found within the class of strung verbs is much higher than for swept verbs. Swept verbs are only defined in terms of their vowel alternation, while strung verbs are defined in terms of three properties, namely vowel alternation, initial consonant clusters and final consonant clusters. Therefore, it cannot be assumed, as Bybee does, that the higher productivity of the strung class is due to the fact that it is lower in average token frequency when in fact the two classes also differ with regard to how phonologically coherent they are. To summarize this section, I have argued that type frequency, semantic coherence and the inverse correlation between the two are appropriate predictors of syntactic productivity. Such a view entails that productivity is gradient, that the generality and the regularity concepts can be derived from the extensibility concept, and that in order for a low type frequency construction to be productive it has to exhibit a higher degree of (semantic) coherence. This yields a productivity cline with full productivity at the one end and analogical extension at the other end of the cline. However, the correlation between type frequency and schematicity is not completely fixed. It is possible for low type frequency constructions to be schematically open in the sense that their instances are scattered around their delimiting space. In such cases, it is not a given that speakers generalize at all across the instances; it is equally possible that the instances function as distinct item-specific islands with no superordinate schemas. Moreover, within cognitivefunctional linguistics and usage-based versions of CxG the traditional view of productivity has been that the extensibility of a construction is a derivative of its level of schematic entrenchment. I have argued, on the other hand, that extensibility is a derivative of schematicity itself and that the highest level of schematicity at which constructions exist in speakers’ minds, in fact, predicts their level of productivity. In accordance with a bottom-up model of the organization of language and grammar, the level of schematicity assumed for each construction is based on its type frequency and semantic coherence. This means that low type frequency constructions only exist at low levels of schematicity or, put differently, at high levels of specificity. High type frequency constructions, by contrast, exist at much higher levels of schematicity and are thus more productive. Finally, I have suggested that high token frequency can only be assumed to detract from the productivity of higher level constructions but not from the productivity of lower level constructions, since high token frequency makes individual types more entrenched as verb-specific low-level constructions, and thus more suitable as models for analogical extensions.



Chapter 2.  Productivity 

2.6 Summary In this chapter I have presented an overview of the numerous senses and informal uses connected with the term productivity that I have encountered in the linguistic literature. At first sight, this diversity of gross synonyms and non-conventional uses makes an appearance of an unstructured array of possibly related and unrelated senses. However, a closer inspection reveals that the various uses can be made sense of. The analysis of the different uses, outlined in Section 2.2, unmasks a highly structured and systematic metaconcept of productivity of which several inferentially-related subconcepts can be discerned. The general concept of productivity within linguistics can be broken down into three main subconcepts, namely those of (a) generality, (b) regularity and (c) extensibility. All the senses and uses of the terms productive and productivity which I have found in the literature can be traced to these three subconcepts in one way or another. Several senses are elaborations of one of the main concepts, one sense exists as a metaphorical extension of the whole generalized meta-concept, some are contextually and/or factually implied or entailed, and yet others have come to exist as synonyms to productive via antonymy relations. The dismantling of the general concept into subconcepts is, in my opinion, a prerequisite for fruitful research and discussion on productivity within linguistics. After a short overview of the prevalent definitions of productivity in morphology and the methods suggested to measure it, I presented the hypothesis of productivity advocated in this work, namely that productivity is best regarded as a function of a construction’s type frequency and its coherence. The inverse correlation between these two factors is an appropriate predictor of the productivity of syntactic constructions. I have also pointed out that there are two notions of productivity found within syntax, namely: i. speakers’ ability to generate sentences never heard before ii. speakers’ extensions of argument structure constructions to new verbs These two notions, I argued, are reflections of how the regularity and the extensibility concepts, respectively, can be manifested within syntax. A parallel tripartite division of tokens, types and patterns has also been shown to exist in morphology as in syntax. The final subsection of this chapter outlined how syntactic productivity can be derived from the inverse correlation between type frequency and (semantic) coherence. Such a view of productivity entails that productivity is considered as gradient with high type frequency correlating with a high degree of schematicity, while low type frequency presupposes a high degree of semantic coherence in order for a construction to be productive. The notions of type frequency and



Productivity

schematicity can be teased apart, at least for low type frequency constructions, but it is not a given that speakers generalize across the members of a low type frequency category if the instances do not show a high enough degree of similarity between them. I have argued that it is the highest level of schematicity, at which a construction exists in speakers’ minds, that determines its degree of productivity, and not its most entrenched schematic level. The highest level of schematicity varies from construction to construction depending on its type frequency and degree of semantic coherence. Consequently, low type frequency constructions exist at lower levels of schematicity than high type frequency constructions and their degree and domains of productivity vary accordingly. Finally, token frequency can contribute to the productivity of low type frequency constructions, since low-level lexically-filled verb-specific constructions, rather than abstract schematic constructions, are models for analogical extensions. The remaining chapters in this book will address the following questions: i. Where should we draw the line between productivity and analogy, given that there does not seem to be any principled methods to distinguish between the two? ii. Does token frequency affect productivity, and if so, how? iii. How do type and token frequency interact and affect each other in productivity? iv. How should we delimit ourselves in defining type frequency? v. Do we need to make a distinction between the type frequencies obtained from dictionaries and the type frequencies obtained from real usage events, like corpora? vi. How does historical productivity manifest itself in syntax? vii. How does a contact situation contribute to the productivity of argument structure constructions? viii. How high does the type frequency of a category need to be in order for it not to fall into disuse over time? ix. Do we need a general theory of productivity at all or do facts of productivity fall directly from the basic assumptions of a usage-based construction grammar? I begin with an investigation of verbs borrowed into Icelandic and the case and argument structure constructions they are assigned by Icelandic speakers.

chapter 3

New verbs in Icelandic A general outline

The productivity of case and argument structure constructions is a new and emerging research field in linguistics. In my earlier work on case and argument structure in Icelandic I have found that many Icelandic argument structure constructions are productive, i.e. they can be, and are being, extended to new verbs entering the language (Barðdal 1999a–b, 2000, 2001a: Ch. 5, 2003, 2006b). In this chapter I will recapitulate, and complement, my previous findings on the productivity of case in Icelandic. I start with a general section on the type frequency of the different case and argument structure constructions, together with a semantic analysis of the Nominative subject construction and its three subconstructions, Nom-Acc, NomDat and Nom-Gen, and the predictions type frequency and semantic coherence make about the productivity of these constructions (Section 3.1). I argue that the Nom-Acc construction should be the most productive subconstruction of the Nominative subject construction, as it is highest in type frequency and contains the highest number of further subconstructions, i.e. verbclass-specific constructions, verb-subclass-specific constructions and verb-specific constructions, which in turn makes the Nom-Acc the most general construction of the three. The Nom-Dat construction should be less productive, as it is lower in type frequency and subsumes fewer verb-class-specific constructions and/or lower-level constructions. The Nom-Gen construction, by contrast, should be fairly unproductive as it consists of only very few verb-subclass-specific constructions, and the remaining Nom-Gen predicates do not show much internal consistency between them, but are verb-specific islands scattered across semantic space. With regard to the semantics of the Nominative subject construction, there is some semantic overlap between the three subconstructions, which in turn predicts that Icelandic speakers should have a choice between them when assigning case and argument structure to new verbs. It is also expected that the type frequency of new verbs entering the language should mirror the respective type frequency of the existing subconstructions in Icelandic. This is a consequence of the fact that lower type frequency coincides with more restricted semantics than higher type frequency, which coincides with less restricted semantics or a high degree of schematicity.



Productivity

I then present a survey of recently borrowed verbs in Icelandic, mostly from the area of information technology, although some general slang expressions are also included (Section 3.2). I show that the predictions that follow from generative theories which assume a distinction between lexical and structural case do not hold for borrowed verbs in Modern Icelandic. The present approach, however, captures the fact that the semantics of the borrowed Nom-Acc predicates is within the semantic scope of the existing Nom-Acc construction. In the same way, the Nom-Dat construction is being extended to borrowed verbs that have the same meaning as already existing Nom-Dat verbs in Icelandic. This is exactly what one would expect on the account that productivity is a function of a construction’s semantic coherence in combination with its type frequency. A proper investigation of the borrowings also reveals that they are assigned case and argument structure either on the basis of the verb-specific constructions of their translational equivalents, or on the basis of low-level verb-subclass-specific constructions, i.e. via analogical extension. All the examples from the survey of borrowed verbs were of this type, although clear-cut cases can also be found of low-level analogical extensions of predicates which only exist as idiosyncratic set phrases, and are thus not a part of a higher-level schematic construction. The present approach also contradicts the traditional view that there is a qualitative difference between rulebased extensions and analogy, as there is no reason to assume an ontological difference in the status of extensions based on a verb-specific construction vs. those based on a verb-subclass specific construction. Given that, the token frequency of individual verb-specific constructions is clearly of importance for the options speakers have when they choose between model verbs for their low-level analogical extensions. Token frequency thus turns out to be of considerable importance for productivity, contra previous assumptions in the literature. The chapter closes with a summary of its main content and conclusions (Section 3.3). 3.1 Case and argument structure in Icelandic In this section I give an overview of the different case and argument structure constructions and their type frequency in Modern Icelandic (3.1.1). I also present an analysis of the constructional semantics of the three subconstructions of the Nominative subject construction (3.1.2), i.e. Nom-Acc, Nom-Dat and Nom-Gen, with a discussion of the predictions the present analysis makes about productivity (3.1.3).



Chapter 3.  New verbs in Icelandic 

3.1.1 Type frequency Icelandic has maintained most of the original case and argument structure constructions from Proto-Germanic, listed in Table 3.1. The type frequency of these constructions in Modern Icelandic, however, varies considerably. Starting with the least frequent construction, i.e. the Genitive subject construction in the last column in Table 3.1, there are only a couple of frozen expressions which occur with Gen-Nom (of the type Þess er von ..., literally ‘it.gen is expectation.nom’, or ‘It is expected ...’) while Gen-only (i.e. one place predicates with a genitive argument only), Gen-PP and Gen-S(entence) are instantiated by a few more types. By the term Genitive subject construction, I refer to the fact that the genitive argument behaves syntactically as a subject in Icelandic. The same is true for the accusative and dative in the Accusative and Dative subject constructions, respectively. I will not review here the data and the arguments relevant for this analysis, as it is a standard analysis for Icelandic, widely accepted in the literature. I refer the interested reader to Eythórsson and Barðdal (2005), Barðdal (2006a) and Barðdal and Eythórsson (2006) which are the latest contributions to this topic. Returning to Table 3.1, the cross ‘†’ signals that the relevant case constructions are so low in type frequency in Icelandic that they are on the boarder of being extinct. The overall type frequency of the Genitive subject construction is somewhere between ten and fifteen in Modern Icelandic (cf. Table 3.3 below). The Accusative subject construction is higher in type frequency than the Genitive subject construction, counting around 200 predicates in total (cf. Barðdal 2001a: 136). This count includes different lexical entries of the same verbal stems, like the following two with the verb reka, i) reka að landi ‘drift to the shore’ and ii) reka í rokastans ‘be surprised’. The count covers all the subconstructions in the second column in Table 3.1. Two of these subconstructions are extremely rare: There is only one Acc-Nom predicate left in Modern Icelandic, Mig sækir syfja, literally ‘me.acc visits sleepiness.nom’, or ‘I feel sleepy’, while there are around four Acc-Gen verbs in Modern Icelandic, like for instance, Mig fýsir þess, literally ‘me. acc wants it.gen’, or ‘I want (to do) it’. Table 3.1  Case constructions in Icelandic Nom-subject

Acc-subject

Dat-subject

Gen-subject

Nom Nom-Acc Nom-Dat Nom-Gen Nom-PP Nom-S

Acc †Acc-Nom Acc-Acc †Acc-Gen Acc-PP Acc-S

Dat Dat-Nom †Dat-Gen Dat-PP Dat-S

Gen †Gen-Nom Gen-PP Gen-S



Productivity

The Dative subject construction is much higher in type frequency than the Accusative subject construction, with around 700 predicates in total (Barðdal 2001a: 136, 2004: 105, 2006a: 90–94), including different lexical entries of simple verbs, as well as various adjectival predicates of the type Mér er kalt, literally ‘me. dat is cold’, or ‘I’m freezing’. The Dat-Gen subconstruction has become extinct in Modern Icelandic, but it did exist in Old Icelandic, at least with two verbs, batna ‘recover (from an illness)’ and létta ‘abate (of an illness)’ (cf. Barðdal 2001a: 197–98, and footnote 3 in Chapter 6 below). Both of these occur as Dat-Nom verbs in Modern Icelandic. My counts of both accusative and dative subject predicates are based on a comprehensive list of oblique subject predicates in Icelandic, compiled by Jónsson (1998). Finally, the Nominative subject construction is by far the highest one in type frequency of all the case and argument structure constructions in Icelandic. In order to investigate the differences in type frequency between the various subconstructions of the Nominative subject construction, I have counted the verbs listed as instantiating the Nom-Acc, Nom-Dat and Nom-Gen constructions in a bilingual dictionary of Icelandic and English (Hólmarsson, Sanders and Tucker 1989). The Nom-Acc construction occurs with 1,381 predicates, the Nom-Dat construction with 627 verbs, while the Nom-Gen construction only occurs with thirtyseven types. These counts exclude accusative, dative and genitive direct objects of ditransitive verbs. That is, verbs occurring in the ditransitive construction(s) are only included in this count if they can independently occur as transitive verbs with an accusative, dative or genitive object. One class of verbs, not included in the dictionary count, is the class of so-called alternating verbs (cf. Bernódusson 1982, Barðdal 1998, 2001b, Eythórsson and Barðdal 2005: 840–42, 860–65). These are two-place predicates which occur with one dative and one nominative argument, and can alternate between the Nom-Dat and the Dat-Nom argument structure constructions in Icelandic. An earlier count includes 111 predicates of this type (Barðdal 2001b: 54–55). This raises the number of Nom-Dat predicates to 738, a figure very close to the one suggested by Maling (2002: 31) in her count of dative object verbs in Icelandic. These numbers are of course not exhaustive. First of all, the bilingual dictionary (Hólmarsson, Sanders and Tucker 1989) is only of intermediate size and “all” transitive predicates in Icelandic are presumably not listed there. Nevertheless, this dictionary was suitable for my count as its lexical entries for transitive verbs give specifications on object case marking. However, most entries in the dictionary that were specified for case are simple verbs, while compositional predicates like adjectival predicates (e.g. vera trúr e-m, lit. ‘be faithful sby.dat’, or ‘be faithful to sby’), which often select for dative objects in Icelandic, and light verb constructions of various types with various case pattern possibilities, are not systematically



Chapter 3.  New verbs in Icelandic 

listed in their entirety in this dictionary. Therefore, collocations and idiomatic expressions in Icelandic are bound to add to the numbers of both accusative and dative object marking types, but a comprehensive list of such expressions is not available at present. As the 1988 edition of the single-language Icelandic dictionary, Íslensk orðabók (Böðvarsson 1988), contains approximately 8,500 entries for verbal predicates (Kristín Bjarnadóttir, p.c.) and Orðabók Háskólans (the University Dictionary Project in Reykjavík) contains around 16,000 predicate and argument structure constructions from the history of Icelandic (Maling 2002: 31), it is rather obvious that the present count cannot be complete. However, the proportions between the three subconstructions should be fairly accurately represented in this count. This is indeed confirmed by the comparison in Table 3.2, which summarizes the statistics from the discussion above and gives corresponding figures from a corpus of Modern Icelandic texts (Barðdal 2001a). The corpus from which the figures in the right-hand part of the table are extracted was compiled during the period 1999–2001 and it includes texts from six different genres, five written and one spoken genre. It contains 40,000 running words, with ten texts from each of the five written genres, 25,000 words in total, and the first 15,000 words in the recordings of an interactive call-in radio program from the Icelandic National Radio, recorded in 1996–1997. The following genres are represented in the corpus: i. ii. iii. iv. v. vi.

Spoken Icelandic Teenage and children’s literature Icelandic fiction Translated fiction Biographies and memoirs Non-fiction

The corpus is thus a balanced and well-stratified corpus of Modern Icelandic texts (for detailed information about the texts and the compilation of the corpus, see Barðdal 2001a: 76–80 and its bibliographical section). The present corpus was designed according to the same principles as a larger corpus of 500,000 words which was composed during the eighties for the purpose of making an Icelandic word frequency dictionary (Pind 1991). The two corpora contain the same genres, except that there is no spoken part in the larger corpus. Barðdal (2001a: Section 4.2.4) compares the two corpora, including, for instance, frequencies of nouns, personal pronouns and adjectives in different case forms. This comparison revealed similarities in frequency, bordering on identity, for the two corpora. This shows that even though the present corpus is exceptionally small, it is still large enough for research on case marking.

 Productivity

Table 3.2  Type frequency of the subconstructions of the Nominative subject construction in different counts of Icelandic Dictionary Count N Nom-Acc Nom-Dat Nom-Gen Total

1,381 738 37 2,156

Corpus Count

%

N

%

64.1 34.2 1.7 100.0

303 188 24 515

58.8 36.5 4.7 100.0

This is, of course, a consequence of the fact that case and argument structure constructions occur in every sentence, as opposed to, for instance, research on modal verbs where a much larger corpus would be needed, as modal verbs do not occur in every sentence/utterance. In fact, the present corpus should be even more representative of Modern Icelandic language use than the larger corpus, as the larger corpus did not contain any part on spoken Icelandic. The case and argument structure constructions from the text corpus and the predicates instantiating them are listed in Appendix A. The numbers in Table 3.2 show that the ratio between the Nom-Acc and the Nom-Dat construction is a little less than 2:1. That is, Nom-Dat predicates are approximately 34–36% of the total percentage of nominative subject predicates, while the corresponding percentage for the Nom-Acc construction is 59–64%. Nom-Gen is the lowest of the three subconstructions in type frequency, around 2–5%. Observe that there are proportionally more Nom-Dat verbs and fewer Nom-Acc verbs found in language use than is reflected by their “absolute” type frequency in Icelandic, although a larger corpus is needed to establish whether the difference is significant or not. The overall type frequency of all the Icelandic case constructions discussed in this section is summarized in Table 3.3. The Nominative subject construction is by far the most frequently instantiated case construction in Icelandic, followed by the Dative subject construction, then by the Accusative subject construction, and the construction lowest in type frequency is definitely the Genitive subject construction. There are also considerable differences found between the subconstructions of each construction. These differences in type frequency can be represented graphically as in Figures 3.1 and 3.2. Table 3.3  Type frequency of different case constructions in Icelandic Nom-subject 2,156 +

Acc-subject

Dat-subject

Gen-subject

ca. 200

ca. 700

ca. 10–15



Chapter 3.  New verbs in Icelandic 

X-Verb-Y

Nom-V-Y

Nom-Acc

Acc

Nom-Dat

Nom-Gen

Dat

Nom-PP

Gen

Nom-S

Figure 3.1  Differences in entrenchment between case constructions of two-place predicates in Icelandic

The Nominative subject construction is highest in type frequency and therefore also the most entrenched construction of them all. This is represented in Figure 3.1 with a bold-faced box around the Nominative subject construction in Icelandic, and with a bold-faced line from the superordinate transitive construction to the Nominative subject subconstruction.

X-Verb-Y

Dat-V-Y

Acc-V-Y

Gen-V-Y

Gen-PP Acc-Acc

Acc-PP

Gen-S

Acc-S

Dat-Nom

Dat-PP

Dat-S

Figure 3.2  Differences in entrenchment of Oblique subject constructions in Icelandic



Productivity

Figure 3.2 is a detail of Figure 3.1 and shows that the Dative subject construction is highest in type frequency of the three Oblique subject constructions in Icelandic, thus it is the most entrenched construction of the three, as shown with boldfaced lines in Figure 3.2. The differences in type frequency of the various case constructions in Icelandic make certain predictions about the productivity of these constructions, given that the degree of semantic coherence among the types instantiating each construction correlates inversely with that construction’s type frequency, at least for high type frequency constructions (see Section 2.5 above). On such an assumption, the construction highest in type frequency, the Nominative subject construction, should be most productive, the one lowest in type frequency, the Genitive subject construction, should be least productive, whereas constructions of intermediate type frequency, like the Dative subject construction, should show intermediate degrees of productivity. I will deal with the Oblique subject construction in more detail in Sections 4.3 and 6.3 below. In the remainder of this chapter, however, my focus will be on the Nominative subject construction and its three subconstructions. The Nom-Acc should be the construction extended most to new verbs in Icelandic, as it is the highest of the three subconstructions in type frequency, and thus it is also expected to be the most general of them all. The Nom-Gen should be least, if at all, extended to new verbs, while the Nom-Dat subconstruction should show signs of intermediate degrees of productivity. Before setting our sights on new verbs in Icelandic and the case and argument structure constructions they are assigned by Icelandic speakers, let us first consider the semantics of the three subconstructions of the Nominative subject construction. 3.1.2 The semantics of the Nominative subject construction As established above, there is a considerable difference in type frequency between the different subconstructions of the Nominative subject construction in Icelandic. For the present study, I will restrict my analysis to the verbs occurring in the corpus of Modern Icelandic texts, since this data set is smaller than the one from the dictionary. As the number of Nom-Acc predicates occurring in the corpus is nevertheless as high as 303, and for Nom-Dat predicates as high as 188, the data set should be large enough for a relatively fine-grained semantic analysis. The number of Nom-Gen predicates in the corpus is considerably lower, only twentyfour types, but it is still a substantial part of the total of thirty-seven Nom-Gen predicates, listed in the dictionary.



Chapter 3.  New verbs in Icelandic 

3.1.2.1 Nom-Acc The predicates occurring in the text corpus are of all structural types, i.e. simple verbs, particle verbs, compositional predicates, collocations and idiosyncratic expressions (see Appendix A). The meaning of a predicate sometimes overlaps with the meaning of the verbal stem, but in other cases it cannot be derived from the meaning of its (composite) parts. The predicates are therefore either of general or specific, i.e. idiosyncratic, meaning (cf. Tomasello 1998: 481–82, Barðdal 2001b: 62, Barðdal and Molnár 2003: 234–35, Croft and Cruse 2004: 253–54). They all share the property of occurring in the Nom-Acc argument structure construction in Icelandic. The identifiable narrowly circumscribed verb classes are the following (the translations below are based on the occurrence of these predicates in context): 1. Verbs of appearing: gefa (sig) út ‘make appearance’, nálgast ‘approach’, líta (dagsins ljós) ‘emerge’ ... 2. Verbs of attaching and detaching: binda ‘tie’, frelsa ‘free’, handtaka ‘arrest’, laga að ‘adjust’, losa ‘loosen’, læsa ‘lock’, missa ‘lose’, opna ‘open’, reyra saman ‘tie together’, spenna ‘tighten, hitch’, tryggja ‘secure’ ... 3. Verbs of attempting: prófa ‘try’, reyna ‘try’ ... 4. Verbs of building and handicraft: bora ‘drill’, endurreisa ‘rebuild’, lagfæra ‘mend’, naglhreinsa ‘refine wood’, smíða ‘construct’ ... 5. Verbs of choosing and electing: greiða (atkvæði) ‘vote’, kjósa ‘choose, vote’, velja ‘choose’ ... 6. Verbs of cognition and mental activity: athuga ‘check’, ákveða ‘decide’, átta (sig) ‘realize’, bera (kennsl) ‘recognize’, fatta ‘understand’, flækja ‘complicate’, hafa (hug) ‘intend’, hafa hugfast ‘keep in mind’, halda ‘believe’, hugleiða ‘ponder’, kenna ‘recognize’, kunna ‘know how’, lesa ‘read’, læra ‘learn’, margfalda ‘multiply’, muna ‘remember’, nema ‘learn’, sérhæfa ‘specialize’, skilja ‘understand’, skoða ‘observe’, standa e-ð af sér ‘wait for sth to stop’ taka (ákvörðun) ‘decide’, takast (á herðar) ‘shoulder’, telja ‘consider’, uppgötva ‘discover’, venja ‘acquire a habit’, virða ‘view, respect’, vita ‘know’, víla fyrir sér ‘hesitate’, þekkja ‘know’ ... 7. Verbs of commencement: (endur)nýja ‘(re)new’, hefja ‘begin’, stofna ‘found’ ... 8. Verbs of creation and reshaping: græða ‘grow’, (ný)skapa ‘(re)create’, umsteypa ‘reshape’, mynda ‘form’ ... 9. Verbs of (means of) cutting: hluta ‘cut (up in parts)’, höggva ‘hew’, klippa ‘cut (with scissors)’, skera ‘cut (with knife)’, rýja ‘shear’, umskera ‘circumcise’ ... 10. Verbs of decorating: lita ‘color’, prýða ‘decorate’ ... 11. Verbs of (de)limitation: fela í sér ‘entail’, greina ‘distinguish’, marka ‘demarcate’, merkja ‘mark’, útiloka ‘exclude’ ... 12. Verbs of delivering: gefa ‘give’, gifta ‘give away for marriage (get married)’, rétta ‘hand’, selja ‘sell’ ...



Productivity

13. Verbs of destruction: brenna ‘burn’, brjóta ‘break’, óhreinka ‘dirty’, rífa ‘tear’, skemma ‘damage’ slíta ‘rip’, tæta upp ‘tear up’ ... 14. Verbs of discussing: brýna ‘emphasize’, fullyrða ‘claim’, fullvissa ‘assure’, kjafta til ‘convince’, leggja (áherslu) ‘stress’, ræða ‘discuss’, rökstyðja ‘argue for’, skýra ‘explain’, undirstrika ‘emphasize’ ... 15. Verbs of displaying: auglýsa ‘advertise’, sýna ‘show’, vísa (veginn) ‘show the way’ ... 16. Verbs of doing and producing: búa til ‘make’, (endur)gera ‘(re)do, make’, endurtaka ‘repeat’, framleiða ‘produce’, knýja ‘run (machines)’, laga ‘make’ ... 17. Verbs of dwelling and stationary position: ala (manninn) ‘be somewhere for some time’, gista ‘stay the night’, geyma ‘preserve’, nema (land) ‘settle’, skilja eftir ‘leave behind’, skipa (sess) ‘occupy a place’, sofa (nóttina) ‘sleep (through the night), standa vörð ‘stand guard’, toppa ‘be on top of sth’ ... 18. Verbs of (caused) emotion: angra ‘bother’, drífa ‘hurry’, elska ‘love’, festa (ást) ‘fall in love’, finna ‘feel’, heilla ‘charm’, hræða ‘scare’, líða (skort) ‘suffer (privation)’, óttast ‘fear’, stilla (sig) ‘calm down’, trufla ‘disturb’, vilja ‘want’, víla fyrir sér ‘hesitate’, þurfa ‘need’, æsa upp ‘incite’ ... 19. Verbs of feeding and consumption: beita ‘graze’, bíta ‘bite’, borða ‘eat’, drekka ‘drink’, éta ‘eat’, gleypa ‘swallow’, innbyrða ‘consume’, sjúga ‘suck’, tyggja ‘chew’ ... 20. Verbs of formal communication: afgreiða ‘expedite’, ávarpa ‘address’, bera fram ‘make a request’, blessa ‘bless’, boða ‘proclaim, preach’, fella niður ‘cancel, dismiss’, flytja (erindi) ‘give a speech’, fræða ‘enlighten’, halda (erindi) ‘give a speech’, kynna ‘present, introduce’, kæra ‘accuse, bring a charge against sby’, tilkynna ‘declare’, undirrita ‘sign’ ... Verbs of formal communication can be analyzed as subsuming three subclasses, namely ecclesiastic communication (blessa ‘bless’, boða ‘preach’), legal communication (kæra ‘accuse, bring a charge against sby’, undirrita ‘sign’) and a general subclass of formal communication to which the remaining verbs belong. 21. Verbs of funding and finances: borga ‘pay’, greiða ‘pay’, kosta ‘finance’, rukka ‘collect (bills)’, spara ‘save money’ ... 22. Verbs of (means of) gaining: eignast ‘come into possession’, endurheimta ‘regain’, erfa ‘inherit’, fá ‘get’, finna ‘find’, grípa ‘catch’, hafa upp úr krafsinu ‘end up with sth’, hirða ‘pick up from the street’, hljóta ‘receive’, hneppa ‘acquire (by luck)’, hremma ‘snatch’, hrifsa ‘snatch’, kaupa ‘buy’, pikka upp ‘pick up’, snapa ‘obtain (by begging)’, tína upp ‘pick up’, útvega ‘provide’, vekja (athygli) ‘get attention’, veiða ‘fish’, öðlast ‘obtain’ ... 23. Verbs of heating and illumination: brenna ‘burn’, kveikja ‘lighten’, lýsa ‘lighten up’, tendra ‘lighten’, volgra ‘heat’ ... 24. Verbs of human disposition: bera (sig) að ‘go about sth’, eiga til ‘tend to’, venja ‘acquire a habit’ ...



Chapter 3.  New verbs in Icelandic 

25. Verbs of increasing and strengthening: auka ‘increase’, efla ‘strengthen’, hvítta ‘make white’, lengja ‘lengthen’, styðja ‘support’, styrkja ‘strengthen’, þétta ‘make tight’ ... 26. Verbs of letting: láta ‘let’, leyfa ‘allow’ ... 27. Verbs of human manipulation: blekkja ‘deceive’, dylja ‘cover’, fela ‘hide’, heilla ‘charm’, hrífa upp ‘snatch sby up’, hylla ‘enchant, plata ‘con’ ... 28. Verbs of measurement: mæla ‘measure’, vega ‘weigh’ ... 29. Verbs of meeting and uniting: heimsækja ‘visit’, hitta ‘meet’, sameina ‘unite’ ... 30. Verbs of non-translational motion: beygja ‘stoop’, geifla (sig) ‘grimace’, hafa ekki (augun) af ‘not take one’s eye of ’, hnipra (sig) ‘curl up’, hrista ‘shake’, hvessa (augun) ‘give (a sharp look)’, klípa ‘pinch’, kreppa (fæturna) ‘clench (one’s feet)’, naga ‘gnaw’, reka í ‘bump into sth’, reka úr sér tunguna ‘stick one’s tongue out’, ræskja (sig) ‘clear one’s throat’, smjúga ‘penetrate’, sveipa ‘wrap’, tína upp ‘pick up’, teygja ‘stretch’, tvíhenda ‘grasp (a sword) with both hands’, þrýsta ‘squeeze’ ... 31. Verbs of perception and arousal: heyra ‘hear’, sjá ‘see’, snerta ‘touch’, strjúka ‘stroke’, skynja ‘perceive’, örva ‘arouse’ ... 32. Verbs of (direct) physical affectedness: baða ‘bathe’, barna ‘make pregnant’, berja ‘beat’, drepa ‘kill’, keyra ‘push’, klæða ‘dress’, kyssa ‘kiss’, leiða ‘hold hand’, myrða ‘murder’, pína ‘torture’, skjóta ‘shoot’, slá ‘hit’, svíða ‘singe’, vekja ‘awaken’ ... This class can be divided into two subclasses, verbs entailing violence (berja ‘beat’, drepa ‘kill’, keyra ‘push’, myrða ‘murder’, pína ‘torture’, skjóta ‘shoot’, slá ‘hit’, svíða ‘singe’) and a more neutral subclass of verbs of affectedness. 33. Verbs of possession: eiga ‘own, have’, hafa ‘have’ ... 34. Verbs of practicing and attending: annast ‘attend to’, iðka ‘practice’, leggja (stund á) ‘practice’, sækja (skóla) ‘attend (school)’ ... 35. Verbs of preparing: búa undir ‘prepare’, tygja (sig) ‘prepare for departure’ ... 36. Verbs of putting: leggja ‘lay’, setja ‘put’, setja upp ‘put (on scene)’ ... 37. Verbs of recuperation: hvíla ‘rest’, leggja (sig) ‘take a nap’ ... 38. Verbs of slandering: afflytja ‘distort’, rægja ‘slander’, sverta ‘disparage’ ... 39. Verbs of taking and fetching: taka ‘take’, sækja ‘fetch’ ... 40. Verbs of catching and termination: standa að e-u ‘catch redhanded’, stoppa ‘stop’, stöðva ‘stop’ ... 41. Verbs of transfer: bera ‘carry’, draga ‘draw, pull’, flytja ‘transport’, flæma burt ‘drive away’, færa ‘move’, hreyfa ‘move’, senda ‘send’, taka út ‘withdraw’, toga ‘pull’, (út)breiða ‘spread, unfold’ ... 42. Verbs of translational motion and (means of) traveling: aka (Þrengslin) ‘drive (a mountain pass)’, fara (ferð) ‘have a journey’, leggja (land undir fót) ‘travel’, hlaupa (hring) ‘run around’, húkka ‘hitchhike’, rekja ‘trace’, ösla (snjó) ‘wade (snow)’ ...



Productivity

43. Verbs of utilizing: nota ‘use’, nýta ‘utilize’ ... 44. Verbs of (interactive) verbal behavior: biðja ‘ask’, bjóða ‘offer’, eggja ‘spur’, gefa (sig) á tal ‘start conversation’, hafa (spurnir) ‘be told about sth’, hvísla ‘whisper’, kalla ‘call’, kveðja ‘say goodbye’, mæla ‘speak’, nefna (nöfn) ‘name’, orða ‘put into words’, reka upp ‘cry out’, rita ‘write’, rífa kjaft ‘use foul language’, skrá ‘register’, skrifa ‘write’, spyrja ‘ask’, syngja ‘sing’, vara við ‘warn’ ... 45. Verbs of (verbal) creation: dikta ‘compose (a poem)’, semja ‘compose (a melody or text)’, yrkja ‘compose (a poem or song)’, þýða ‘translate’ ... 46. Verbs of warfare and heroism: afvopna ‘disarm’, sigra ‘conquer’, slíðra ‘sheath’, tvíhenda ‘grasp (a sword) with both hands’, umlykja ‘surround’, vinna (þrekvirki) ‘do great feat’, verja ‘defend, protect’, yfirbuga ‘overpower’ ... Irrespective of whether the reader concurs with me on the exact details of the classification above, it seems clear that Nom-Acc verbs in Icelandic motivate a number of small well-defined semantic classes of verbs. The list presented here, containing forty-six narrowly circumscribed verb classes, is made up of only 303 Icelandic Nom-Acc predicates, the ones from the corpus of Modern Icelandic texts. However, as is evident from Table 3.2 above, these 303 predicates are less than one fourth of the 1,381 Nom-Acc predicates listed in the bilingual Icelandic-English dictionary. Therefore, narrowly circumscribed verb classes of Nom-Acc predicates are bound to be considerably more than only forty-six in total. Obviously, a category containing as many members as at least 1,381 cannot be semantically restricted in any way, but must be semantically open or general. In other words, a putative new verb cannot be expected to have to fulfil any specific semantic requirements in order to become a Nom-Acc verb in Icelandic. Most of the forty-six semantic classes of Nom-Acc verbs in Icelandic are subtypes of more abstract categories of basic human event types (cf. Goldberg 1995: 39–43). Specifically for these 303 Nom-Acc predicates, the basic event types of relevance are: i. somebody making something ii. somebody moving something iii. somebody affecting somebody iv. somebody experiencing something v. somebody carrying out a mental or verbal activity vi. something changing from one state to another vii. something being located somewhere The classification below specifies how the more abstract event-type categories encompass the subclasses of Nom-Acc predicates listed above (the numbers in parenthesis refer to the subclasses above).



Chapter 3.  New verbs in Icelandic 



Making and handicraft: (2) verbs of attaching and detaching, (4) verbs of building and handicraft, (9) verbs of cutting, (10) verbs of decorating, (16) verbs of doing and producing, (28) verbs of measurement, (43) verbs of utilizing ...



Physical or metaphorical movement: (12) verbs of delivering, (15) verbs of displaying, (22) verbs of (means) of gaining, (29) verbs of meeting and uniting, (29) verbs of non-translational motion, (36) verbs of putting, (39) verbs of taking and fetching, (41) verbs of transfer, (42) verbs of translational motion and (means) of traveling ...



Physical human affectedness: (19) verbs of feeding and consumption, (32) verbs of (direct) physical affectedness ...



Verbal activity: (14) verbs of discussing, (20) verbs of formal communication, (38) verbs of slandering, (44) verbs of (interactive) verbal behavior, (45) verbs of verbal creation ...

The semantic class of verbs denoting verbal activities, in turn, makes up one of the subclasses of next class:

Cognition, emotion and other mental activities: (14, 20, 38, 44, 45) verbs denoting verbal activity, (3) verbs of attempting, (5) verbs of choosing and electing, (6) verbs of cognition and mental activity, (18) verbs of (caused) emotion, (24) verbs of human disposition, (27) verbs of human manipulation, (26) verbs of letting, (31) verbs of perception, (34) verbs of practicing and attending, (35) verbs of preparing, (37) verbs of recuperation ...



Change of state: (1) verbs of appearing, (7) verbs of commencement, (8) verbs of creation and reshaping, (13) verbs of destruction, (23) verbs of heating and illumination, (25) verbs of increasing and strengthening, (40) verbs of catching and termination ...



Location: (17) verbs of dwelling and stationary position, (33) verbs of possession ...

This overview shows that the forty-six classes of verbs both subsume some subclasses and at the same time they make up the subclasses of yet other more general classes of verbs. These, in turn, can make up subclasses of other highly schematic categories of basic human event types. As the meaning of the Nom-Acc construction is defined by the verbs which instantiate it, the Nom-Acc construction in Icelandic exists at a very high level of schematicity.1 This is illustrated in Figure 3.3, which specifies several levels of schematicity for the Nom-Acc construction in Icelandic.



Productivity

Nominative-Accusative

MAKING

MOVEMENT

AFFECTEDNESS

[Delivering] [Displaying]

[Feeding] [Physical affectedness] [Gaining] [Non-translational motion] [Producing] [Putting] [Measuring] [Taking/Fetching] [Utilizing] [Translational motion] [Transfer] [Uniting] [Attaching] [Building] [Cutting] [Decorating]

COGNITION/EMOTION

CHANGE

[Dwelling] [Appearing] [Commencement] [Possession] [Creation] [Emotion] [Destruction] [Disposition] [Illumination] [Manipulation] [Increasing] [Letting] [Termination] [Verb] ... [Perception] [Attempting] [Choosing] [Cognition]

[Practicing] [Preparing] [Recuperation] [Verbal activity]

[Verb] ...

[Verb] ...

[Verb] ...

[Verb] ...

LOCATION

[Verb] ...

[Discussing] [Formal communication [Slandering] [Interactive verbal behavior] [Verbal dcreation]

[Verb] ...

Figure 3.  The Nom-Acc construction in Icelandic



Chapter 3.  New verbs in Icelandic 

The vertical axis in Figure 3.3 represents the lexicality–schematicity continuum. The lowest level of the Nom-Acc construction is at the foot of the continuum. It is semantically and lexically the most concrete level, consisting of verb-specific constructions. This is shown with the label [Verb] in Figure 3.3 and it stands for all the 303 verb-specific constructions. As the aim of the figure is to represent the overall organization of the Nom-Acc construction and its subconstructions, the details of this level are omitted (see however Boas 2003 for one way of explicating this level, labeled mini-constructions in his work). Together the concrete verb-specific constructions make up a higher and more intermediate level of at least forty-six verb-class specific constructions. Not less than seven of these are subconstructions of the higher-level construction entailing Verbal activity, which again is a subconstruction of one of the basic-event-type constructions, Cognition/Emotion, together with several other verb-class specific constructions. The semantics of the Nom-Acc construction’s next highest level of schematicity consists of such abstract ontological notions as: i. making ii. moving iii. affectedness iv. cognition v. emotion vii. change viii. location This empirical investigation and analysis of the Nom-Acc construction in Icelandic and the verbs instantiating it, shows that a category containing at least 303 types is bound to exist at a fairly abstract level of schematicity. As such the construction is semantically open or unrestricted, which predicts that it should also be fully productive and extendible to new lexical verbs, irrespective of their semantic or lexical content. For a comparison, let us turn to the Nom-Dat construction in Icelandic. 3.1.2.2 Nom-Dat The 188 Nom-Dat predicates from the corpus of Modern Icelandic texts (see Appendix A) can be divided into the following semantic subclasses:2 1. Verbs of attendance and helping: hita ‘warm’, hjálpa ‘help’, hlynna ‘take care of ’, sinna ‘attend to’, skemmta ‘entertain’, þvo ‘wash’ ... 2. Verbs of attributing: eigna ‘attribute’, vera eignaðar ‘be attributed to’, vera lagt í munn ‘be put in sby’s mouth’ ... 3. Verbs of comparison and equality: jafngilda ‘be equal to’, líkja saman ‘compare’, líkjast ‘resemble’, vera áþekkur ‘resemble’, vera jafnhár ‘be as tall’, vera hliðstæð ‘be parallel’ ...



Productivity

4. Verbs of compensating: gefa ‘give’, múta ‘bribe’ ... 5. Verbs of connection: bindast ‘connect with’, kynnast ‘get to know’, skjóta (rótum), ‘settle’, tengjast ‘connect’, tilheyra ‘belong’, vera nákominn ‘be close’, vera nálægt ‘be near’, vera nær ‘be close’, vera tengdur ‘be related’, vera samgróinn ‘be rooted in sth’ ... 6. Verbs of controlling: bjóða ‘order, invite’, leyfa ‘allow’, skipa ‘order, organize’, stjórna ‘control’, stýra ‘govern’, valda ‘cause’ ... 7. Verbs of (perceived) covertness and danger: vera hulið ‘be hidden’, vera hættulegt ‘be dangerous’, vera (ó)kunnugur ‘be (un)known’ ... 8. Verbs of defending: halda uppi (vörnum) ‘continue with defense’ verjast ‘defend’ ... 9. Verbs of destruction: eyða ‘destroy’, fórna ‘sacrifice’, ráða bana ‘kill’ ... 10. Verbs of division: deila ‘share, divide’, skipta ‘divide’ ... 11. Verbs of (caused) emotion: bregða ‘startle’, flýta ‘hurry’, kenna um ‘blame’, koma á óvart ‘surprise’, kunna vel ‘like’, kvíða ‘be anxious’, sæta (undrum) ‘be of surprise’, verða til ama ‘bother sby’, vera til sóma ‘make sby proud’ ... 12. Verbs of ending and finishing: ljúka ‘finish’, loka ‘close’, slíta ‘end’ ... 13. Verbs of eyeing: depla (auga) ‘blink (with the eyes)’, horfa (bænaraugum) ‘beg with the eyes’, líta (augum) ‘view (with one’s eyes)’ ... 14. Verbs of greeting and welcoming: heilsa ‘greet’, taka ‘greet, welcome’, vera (aufúsugestur) ‘be welcome at sby’s house’ ... 15. Verbs of habitude: vera (ó)vanur ‘(not) be used to’, venjast ‘get used to’ ... 16. Verbs of handling: beita ‘apply’, fara (höndum um) ‘handle’ ... 17. Verbs of increase: fjölga ‘increase’, vera (orðum) aukið ‘be exaggerated’ ... 18. Verbs of instructing: hlýða yfir ‘make sby recite’, kenna ‘teach’ ... 19. Verbs of losing: glata ‘lose’, gleyma ‘forget’ ... 20. Verbs of marrying: giftast ‘marry’, kvænast ‘marry’ ... 21. Verbs of (caused) motion: banda frá ‘wave away’, beina ‘direct’, benda ‘direct, point’, berja (hrömmunum) ‘thrash (one’s paws’), blanda ‘mix’, bylta (sér) ‘turn around’, dreifa ‘spread’, feykja ‘blow’, forða ‘get sth away’, fresta ‘postpone’, fylgja ‘follow, accompany’, fyllast ‘be(come) filled with’, henda ‘throw’, hella ‘pour’, hleypa ‘let, allow (sby into sth)’, hliðra ‘shirk’, kasta ‘throw’, klappa ‘pat’, koma (sér) ‘get off, go away’, mjaka ‘budge’, mæta ‘run into sby’, renna ‘glide’, ríða ‘ride’, róa ‘row’, safna ‘gather’, skella (í sig) ‘slam (in oneself), eat’, skila ‘return’, skjóta ‘shoot’, skjóta upp (kollinum) ‘pop up’, skutla ‘give a ride’, slengja ‘throw’, sleppa ‘drop, release’, smeygja (sér) ‘slide’, smjúga (fingrum) ‘slip’, snúa ‘turn’, sparka ‘kick’, spyrna ‘kick’, stilla upp ‘take a position’, stinga ‘insert’, sveipa ‘wrap’, svæla (í sig) ‘puff (in oneself), eat’, troða ‘squeeze (one’s way)’, tylla (sér) ‘take a seat’, varpa ‘throw’, velta ‘roll’, vefja ‘wrap’, víkja ‘make way’, ýta ‘push’, þrengja ‘push’, þrýsta ‘squeeze, thrust’ ...



Chapter 3.  New verbs in Icelandic 

Verbs of caused motion can be divided into two subclasses, namely verbs of intended motion (for instance, banda frá ‘wave away’, beina ‘direct’, benda ‘direct, point’) and verbs of translational motion (henda ‘throw’, hella ‘pour’, kasta ‘throw’, sparka ‘kick’). 22. Verbs of non-translational motion: fletta ‘flick’, gnísta (tönnum) ‘grind (one’s teeth’), halda ‘keep, hold’, halla ‘lean’, hampa ‘hold up’, kinka (kolli) ‘nod’, lyfta ‘lift’, styðja (olnboganum) ‘lean (the elbow)’ ... 23. Verbs of obeying: gangast á hönd ‘surrender’, gegna ‘obey’, lúta ‘obey’, þjóna ‘serve’ ... 24. Verbs of obtaining and maintaining: fá ‘take’, duga ‘last’, halda áfram ‘continue’, koma á fót ‘establish’, koma (e-u í lag) ‘get (sth in order’), ná ‘obtain’, skila (árangri) ‘lead (to result’), veita viðtöku ‘receive’ ... 25. Verbs of organizing: skipa ‘order, organize’, haga ‘arrange, behave’, hegða ‘organize, behave’, koma fyrir ‘arrange’ ... 26. Verbs of persistence and daring: beita (sér) ‘do (one’s utmost)’, fá framgengt ‘get one’s will through’, voga (sér) ‘dare’ ... 27. Verbs of puttering: dunda (sér) ‘putter’, leika (sér) ‘play’, una (sér) ‘be content with what one is doing’ ... 28. Verbs of readiness and equipment: vera búinn ‘be equipped with’, vera klæddur ‘be dressed in’, vera vaxinn ‘be grown with’, vera viðbúinn ‘be ready for’ ... 29. Verbs of stealing: stela ‘steal’, ræna ‘rob’ ... 30. Verbs of thinking and meaning: gera sér (í hugarlund) ‘imagine’, einbeita (sér) ‘concentrate’, geta (sér) til ‘guess’, hugsa (sér) ‘think, imagine’, vera sammála ‘agree’ ... 31. Verbs of trusting: trúa ‘believe’, treysta ‘trust’ ... 32. Verbs of (means of) verbal communication: etja ‘egg’, gera ljóst ‘make clear to sby’, gera (viðvart) ‘warn’, herma ‘tell’, lýsa ‘describe’, lýsa yfir ‘declare’, mótmæla ‘object to’, segja ‘tell’, segja (skrækri röddu) ‘say (with a squeaky voice)’, svara ‘answer’, tilkynna ‘declare’, skrifa ‘write’, spá ‘predict’, sveia ‘curse’, þakka ‘thank’ ... 33. Miscellaneous verbs: búa (búi) ‘farm’, geispa golunni ‘die’, heita (nöfnum) ‘carry names’, skipta (máli) ‘matter’, (um)breyta ‘change’, vera hlaðinn ‘be loaded with’, vera troðinn ‘be stuffed with’, vera vafinn ‘be wrapped with’ ... The Nom-Dat verbs in the text corpus divide across thirty-two narrowly-circumscribed semantic subclasses of verbs, which is a considerably lower number than the forty-six subclasses that the Nom-Acc construction divides across. These thirty-two subclasses again make up the semantic classes of other more general categories of basic human event types. The Nom-Acc predicates are instances of at



Productivity

least six basic human event types, while the Nom-Dat predicates only instantiate three basic event type categories: i. something changing from one state to another ii. somebody carrying out a mental or verbal activity iii. somebody being superior to somebody In addition, a few verb classes do not make up a coherent superordinate class of their own, nor do they fall under the three previously mentioned basic human event types. This is shown with the dashed line in Figure 3.4. The classification below specifies how the more abstract event-type categories encompass the subclasses of Nom-Dat predicates listed above (the numbers in parenthesis refer to the subclasses above).

Change of quantity: (10) verbs of division, (12) verbs of ending and finishing, (17) verbs of increase ...



Verbal activity: (14) verbs of greeting and welcoming, (18) verbs of instructing, (32) verbs denoting (means) of verbal communication ...



Connecting: (5) verbs of connection, (20) verbs of marrying ...

Verbs denoting verbal activity and verbs of connection are two of the subclasses of the more general basic event type category Cognition, emotion and mental activity:

Cognition, emotion and other mental activities: (14, 18, 32) verbs denoting verbal activity, (2) verbs of attributing, (3) verbs of comparison and equality, (5, 20) verbs of connection, (7) verbs of (perceived) covertness and danger, (11) verbs of (caused) emotion, (15) verbs of habitude, (19) verbs of losing, (23) verbs of obeying, (25) verbs of organizing, (26) verbs of persistence and daring, (27) verbs of puttering, (28) verbs of readiness and equipment, (30) verbs of thinking and meaning, (31) verbs of trusting ...



Superiority: (1) verbs of attendance and helping, (4) verbs of compensating, (6) verbs of controlling, (8) verbs of defending, (9) verbs of destruction, (18) verbs of instructing, (23) verbs of obeying, (25) verbs of organizing ...

Verbs of movement can be divided into two subclasses, i.e (21) verbs of (caused) motion and (22) verbs of non-translational motion. Verbs of (16) handling, (24) verbs of obtaining and maintaining, and (29) verbs of stealing, however, do not make up a coherent class of their own. This overview shows that the thirty-two classes of verbs both subsume some subclasses and at the same time make up the subclasses of yet other more general classes of verbs. These, in turn, can make up subclasses of other highly schematic categories of basic human event types. This is illustrated in Figure 3.4, which specifies several levels of schematicity for the Nom-Dat subconstruction.



Chapter 3.  New verbs in Icelandic 

Nominative-Dative

CHANGE

[Handling] [Obtain/Maintaining] [Stealing] [Miscellaneous] [Motion]

COGNITION/EMOTION

SUPERIORITY

[Division] [Ending/Finishing]

[Attributing]

[Attendance/

[Comparison/

[Increase]

Equality [Perceived danger] [Caused emotion] [Habitude] [Losing] [Obeying] [Organizing] [Persistence/Daring] [Readiness/Equipment]

Helping] [Compensating] [Controlling] [Defending] [Destruction] [Instructing] [Obeying] [Organizing]

[Non-translational motion] [Caused motion]

[Verb] ...

[Puttering] [Thinking/Meaning] [Verb] ...

[Trusting] [Connecting] [Verbal activity]

[Intended motion] [Translational motion] [Connection] [Marrying]

[Greeting/Welcoming] [Instructing] [Verbal communication]

[Verb] ... [Verb] ...

Figure 3.4  The Nom-Dat construction in Icelandic

[Verb] ...

[Verb] ...



Productivity

As in Figure 3.2, the vertical axis in Figure 3.3 represents the lexicality–schematicity continuum. The lowest level at the bottom of the hierarchy is the semantically and lexically most concrete level, consisting of verb-specific constructions (labeled [Verb]). All the 188 verb-specific constructions make up the next level above, which contains at least thirty-two verb-class-specific constructions (with intermediate levels for some of the verb-class constructions). These in turn are subconstructions of the basic-event-type constructions denoting: i. change ii. cognition/emotion iii. superiority Since the Nom-Dat construction only subsumes three basic human event-types as opposed to the six found with the Nom-Acc construction, it appears that the highest level of schematicity of the Nom-Dat construction has considerably narrower scope than the highest schematic level of the Nom-Acc construction. This in turn means that the productivity domain of the Nom-Dat construction is also smaller than that of the Nom-Acc construction, hence it is expected that the Nom-Dat construction is less productive. As doubtlessly noted by the reader, there is a certain degree of semantic overlap between the Nom-Acc and the Nom-Dat constructions in Icelandic. Dative verbs of connection (5) are similar to accusative verbs of meeting and uniting (29), dative verbs of ending and finishing (12) share certain semantic features with accusative verbs of commencement (7), dative verbs of obtaining and maintaining (24) are similar to accusative verbs of (means) of gaining (22), dative verbs of puttering (27) partially occupy the same semantic field as accusative verbs of recuperation (37), and verbs denoting destruction can be found among both accusative (13) and dative object verbs (9). Verbs of cognition and (caused) emotion are also found amongst both accusative and dative object verbs, as are verbs of motion, both non-translational and translational motion. There is, however, one very salient difference between accusative and dative verbs of translational motion, namely that dative object verbs of translational motion are much higher in type frequency than accusative verbs of translational motion, i.e. fifty (of 188) vs. ten (of 303), which I will return to in Section 3.2.1 and Chapter 5 below. The fact that there is a certain degree of semantic overlap between the NomAcc and the Nom-Dat constructions in Icelandic predicts that speakers should have a choice between assigning accusative or dative to objects of new verbs belonging to these semantic classes. It also predicts that new verbs from these semantic classes should mirror the proportional type frequency of these classes in Modern Icelandic. As we will see in Section 3.2.1 below, both these predictions are borne out.



Chapter 3.  New verbs in Icelandic 

Table 3.4  Animate vs. inanimate reference of the object in Nominative subject constructions in Icelandic Inanimate object Nom-Acc Nom-Dat Nom-Gen

Animate object

Total

N

%

N

%

N

%

225 103 19

74.2 54.8 79.2

78 85 5

25.8 45.2 20.8

303 188 24

100 100 100

There is one clear statistical difference between Nom-Dat predicates in Icelandic, on the one hand, and the Nom-Acc and Nom-Gen predicates, on the other, in that there is a higher percentage of Nom-Dat predicates that typically occur with a human object. The verbs giftast ‘marry’, heilsa ‘greet’ and ræna ‘kidnap’ are examples of such Nom-Dat verbs, while the verbs barna ‘make pregnant’, heimsækja ‘visit’ and myrða ‘murder’ are examples of such Nom-Acc verbs. This is not to say that Nom-Dat verbs, or at least some of them, cannot occur with an object that does not have an animate referent. They can and do. It is simply that the meaning of the verb (or perhaps sometimes our world knowledge) singles out an animate referent rather than an inanimate one. To give examples of verbs which typically do not occur with an animate object, consider drekka ‘drink’ and segja ‘say’ from the class of Nom-Acc verbs, while henda ‘throw’ and safna ‘collect’ exemplify Nom-Dat verbs. As Table 3.4 reveals, as much as 45% of the Nom-Dat verbs select an object which refers to an animate being, while only 20–26% of Nom-Acc and Nom-Gen verbs select for animate objects. The observed differences in distribution can occur by pure chance with a probability of less than one in a thousand (Pearson ChiSquare, p < 0.001; Cramer’s V = 0.205). 1.2.3 Nom-Gen The twenty-four Nom-Gen predicates occurring in the corpus of Modern Icelandic texts (see Appendix A) can be classified as follows: 1. Verbs of asking and wishing: biðjast (hjálpar) ‘ask for help’, leita (aðstoðar) ‘ask (for help), spyrja (spurninga) ‘ask (questions)’, óska ‘wish’ ... 2. Verbs of cognition: gæta ‘take into consideration’, vera fullviss ‘be sure’, verða áskynja ‘realize’, verða vísari ‘find out’, verða vart ‘be aware of ’ ... 3. Verbs of emotion: njóta ‘enjoy’, sakna ‘miss’, skammast (sín) ‘be ashamed (of oneself)’, vænta ‘expect’, þurfa ‘need’ ... 4. Verbs of social influence: mega (sín) ‘be influential’, njóta (hylli) ‘be popular’ ... 5. Miscellaneous verbs: afla ‘obtain’, bíða ‘await’, ganga duldir ‘hide’, hefjast (handa) ‘begin’, krefjast ‘demand’, nema (staðar) ‘stop’ vekja (máls) ‘mention’, vitja ‘visit’ ...



Productivity

[social influence]

[Verb] ...

[asking/wishing]

[Verb]...

[cognition]

[Verb]...

[emotion]

[Verb]...

[miscellaneous]

[Verb]...

Figure 3.5  The Nom-Gen construction in Icelandic

Of the twenty-four Nom-Gen predicates, sixteen can be divided into four different semantic classes, while the remaining eight predicates do not seem to form a coherent class but are verb-specific islands scattered across semantic space. More­ over, two of these classes, cognition and emotion verbs, are also found among both Nom-Acc and Nom-Dat predicates, and their type frequency is considerably higher in both cases, i.e. twenty-nine and fifteen for Nom-Acc vs. ten and nine for Nom-Dat predicates. Therefore, new verbs of emotion and cognition are expected to be attracted, not by the Nom-Gen construction in Icelandic, but rather by the Nom-Acc or the Nom-Dat construction. Observe also that three of four verbs of asking and wishing are lexicalized set phrases, where the object slot is lexically filled in advance. The same is true for both predicates in the class of verbs of social influence, and four out of eight verbs in the class of miscellaneous verbs. The question thus arises whether Nom-Gen predicates are perceived of as being members of small but coherent semantic classes by Icelandic speakers, or as verb-specific islands dispersed across semantic space. If they are perceived of as verb-specific islands, one would not expect them to show much signs of being productive. At best they would serve as model verbs for lowlevel analogy. Given this, the structure of the Nom-Gen construction can be represented as in Figure 3.5, with only low-level verb-specific constructions and verbsubclass-specific constructions represented and no higher-level constructions. 3.1.3 The productivity of the Nominative subject construction In the sections above I have presented an analysis of the semantic structure of the Nominative subject construction and its three subconstructions in Icelandic, NomAcc, Nom-Dat and Nom-Gen. I have shown that the Nom-Acc construction is highest in type frequency, as well as being schematically the most open construction. The Nom-Dat construction is of intermediate type frequency, and its semantics is also more restricted than that of the Nom-Acc construction. As a consequence, the highest level of schematicity at which the Nom-Dat construction exists is not as elaborate as that of the Nom-Acc construction, as the Nom-Acc subsumes six basic human event types, whereas the Nom-Dat construction only subsumes



Chapter 3.  New verbs in Icelandic 

three such. The semantic scope of the Nom-Dat construction is therefore considerably narrower than the semantic scope of the Nom-Acc construction. The NomGen construction, in contrast, is a low type frequency construction, and even though a few semantic narrowly-circumscribed verb classes can be discerned, they are in part made up of lexically idiosyncratic set phrases, while the remaining Nom-Gen predicates are clearly verb-specific islands evenly dispersed across semantic space. These facts can be represented graphically as in Figure 3.6, where the leftmost box represents the Nom-Acc construction, the intermediate box the NomDat construction, and the rightmost box represents the Nom-Gen construction. As is evident from Figure 3.6, both the Nom-Acc and the Nom-Dat constructions are high enough in type frequency to be productive. Observe that the NomAcc construction contains so many verb clusters that they more or less fill up the semantic space of the construction, yielding a schematically open construction. The Nom-Dat construction, although it is not as open as the Nom-Acc construction, still shows clear cases of verbs clustering together in semantic space, forming identifiable semantic subclasses. As a consequence, the Nom-Acc construction should be the most productive of the three, the Nom-Dat construction should be less productive, but still showing considerable productivity, while the Nom-Gen construction should at best show sporadic traces of productivity. In order to examine whether these predictions are borne out, let us turn to recently borrowed verbs in Icelandic.

High type frequency Open schema

Low type frequency Restricted schema

Low type frequency Open schema

Figure 3.6  The type frequency and schematicity of the subconstructions of the Nominative subject construction in Icelandic



Productivity

3.2 New verbs in Icelandic In this section I examine which case and argument structure constructions Icelandic speakers assign to new verbs entering the language, confining my discussion to borrowed verbs. First I investigate recent borrowings from the area of information technology which exemplify the productivity of the high type frequency Nom-Acc construction and of the intermediate type frequency Nom-Dat construction (3.2.1). The study confirms that the productivity of the two constructions is determined by their type frequency and semantic coherence. It also shows that despite the fact that the two constructions exist at high levels of schematicity, borrowed Nom-Acc and Nom-Dat verbs are still assigned case and argument structure constructions on the basis of full synonymy with the concrete lexically-filled low-level verb-specific constructions of their translational equivalents in Icelandic. In Section 3.2.2 I present further examples of low-level verb-specific analogical extensions, examples that show beyond doubt that not only higher-level schematic constructions are extended on the basis of their individual lower-level verb-specific constructions, but also that there are constructions which only exist as low level constructions, like idiomatic set phrases, that can function as models for analogical extensions. On the present approach, no ontological difference is assumed for extensions based on higher-level schematic constructions or lower-level verb-specific constructions. That is, there is no qualitative difference between analogy and full productivity, only a difference in degree. Given this, the role of token frequency for productivity turns out to be of no less importance than the role of type frequency. 3.2.1 Recent borrowings The 107 borrowed verbs listed below were collected from two sources: (a) a recent on-line slang dictionary (Magnússon 2004) still under construction, and (b) the Icelandic discussion forum for Mac users (www.apple.is/umraedur). The verbs are listed in alphabetical order in Appendix B. It turns out that of the 107 borrowings collected, sixty-eight predicates select for Nom-Acc, thirty-nine for Nom-Dat and none for Nom-Gen. In fact, the proportions between the Nom-Acc and Nom-Dat constructions are more or less the same as in the dictionary and the corpus counts, shown in Table 2 above. For the sake of comparison, I give the statistics from all three counts in Table 3.5.



Chapter 3.  New verbs in Icelandic 

Table 3.5  The differences between Nom-Acc and Nom-Dat in various counts of Icelandic Dictionary Count

Corpus Count

Borrowed Verbs

N

%

N

%

N

%

Nom-Acc Nom-Dat

1,381 738

65.2 34.8

303 188

61.7 38.3

68 39

63.6 36.4

Total

2,119

100.0

491

100.0

107

100.0

The proportion between Nom-Acc and Nom-Dat is 64 vs. 36% of the borrowed verbs, while it is 62 vs. 38% in the corpus count and 65 vs. 35% in the dictionary count. This is unexpected on the assumptions of the classical generative framework which assumes a dichotomy between structural and lexical case (cf. Zaenen, Maling and Thráinsson 1985, Yip, Maling and Jackendoff 1987, Holmberg 1986, Sigurðsson 1989, Allen 1995, Falk 1997, Jónsson 1997–98, 2003, Jónsson and Eythórsson 2005). On such an account, structural case is nominative case on subjects and accusative case on objects, based on the position of the arguments in an ordinary transitive clause. In contrast, lexical case constitutes an exception to this general rule that the subject be in the nominative case and the object in the accusative case. That is, accusative, dative and genitive subjects are lexically case marked, as are dative and genitive objects. Lexical case has been further divided into thematic/semantic case and idiosyncratic case, with thematic case being semantically motivated, such as the dative of experiencers and beneficiaries, whereas idiosyncratic case is neither structurally nor semantically motivated, hence the term ‘idiosyncratic’. An example of idiosyncratic case would be dative and genitive of (most) objects (Zaenen, Maling and Thráinsson 1985, Jónsson 1997–98, 2005). This means that a child acquiring a language like Icelandic only has to learn and memorize which verbs select for lexical case, while the same information does not have to be stored in the child’s lexicon for Nom-Acc verbs. Chomsky and Lasnik (1995: 19–20) argue that regular behavior, such as the formation of past tense -ed for verbs like walk, takes place through a rule, either in the computational system, which means that the form walked is generated on-line when it is processed, or in the lexicon itself. Irrespective of whether one postulates that the form walked is generated in the computational system or in the lexicon, the point here is that it is produced with a rule. The distinction between structural and lexical case is the same as the distinction between regular and irregular verbs, in that structural case is regular and lexical case is irregular. That would mean that structural nominative and structural accusative are generated with a rule and Nom-Acc verbs do therefore not have to be stored in the lexicon as such, while Nom-Dat and Nom-Gen verbs have



Productivity

to be stored in the lexicon with their case frames. This interpretation is supported by the following quote from Pinker (1999: 135):

(1) In a language such as Kivunju or Turkish every word may come in a half a million to several million forms, and speakers could not possibly have memorized them all in childhood.

Pinker is here discussing regular and irregular verb forms but the distinction between structural and lexical case is exactly parallel to that, with structural case being regular and lexical case being irregular. Chomsky and Lasnik (1995) divide grammar into two components: core grammar and the periphery (cf. also Haegeman 1991: 16, Josefsson 1997: 9–10). The difference between the two is that core grammar is the place where regular behavior is generated while irregularities and idiosyncrasies belong to the periphery. I quote Chomsky and Lasnik (1995: 19–20):

(2) ... we make a rough and tentative distinction between the core of a language and its periphery, where the core consists of what we tentatively assume to be pure instantiations of UG and the periphery consists of marked exceptions (irregular verbs, etc.). [emphasis original]

It follows from this that structural case would belong to the core, since it is regular, and lexical case to the periphery, since it is an irregularity. On such a view of language and grammar, structural case should be productive, since it belongs to the core of grammar, whereas lexical case, or at least idiosyncratic case, should be unproductive, since it belongs to the periphery. A quotation from Pinker confirms this (1999: 19):

(3) The theory that regular forms are generated by rule and irregular forms are retrieved by rote is pleasing ... because it explains the differences in productivity between the two patterns ...

Here Pinker assumes that productivity is associated with rules, while lack of productivity is associated with memory. To summarize, only structural and (possibly) thematic case are expected to be productive, not idiosyncratic dative on objects, the Nom-Dat construction in other words.3 Such a prediction is not borne out for Icelandic. According to the figures in Table 3.5 above, approximately 64% of the borrowed transitive verbs assign accusative case to their objects while approximately 36% of them assign dative case to their objects. This figure of 36% is certainly unexpected for lexical case, which should be unproductive, according to the predictions of classical generative grammar that I have just outlined. What is more, it is totally unexpected that the percentages mirror those of existing Nom-Acc and Nom-Dat predicates in



Chapter 3.  New verbs in Icelandic 

Icelandic, as Table 3.5 demonstrates. Even more unpredicted on this account is the fact that the verbs assigned Nom-Acc by Icelandic speakers belong to the same semantic classes as already existing Nom-Acc verbs, and that the verbs assigned Nom-Dat belong to the same semantic classes as already existing Nom-Dat verbs. In other words, not only do the case constructions assigned to borrowed verbs mirror existing frequency distinctions between Nom-Acc and Nom-Dat but the distribution across semantic classes is also mirrored by these. This is shown in the lists below, where all 107 verbs are assorted according to the categorization of the narrowly-circumscribed semantic classes established for the Nom-Acc and the Nom-Dat constructions in Icelandic in Sections 3.1.2.1–3.1.2.2 above (here rendered with the same spelling as in the source documents). Nom-Acc: 2. Verbs of attaching and detaching: peista ‘paste’, plögga ‘plug’, releasa ‘release’ ... 3. Verbs of attempting: testa ‘test’ ... 4. Verbs of building and handicraft: builda ‘build’ ... 6. Verbs of cognition and mental activity: editera ‘edit’, gúggla ‘google’, pinga ‘ping’, offa ‘off ’, resolva ‘resolve’, stúdera ‘study’, verifya ‘verify’ ... 8. Verbs of creation and reshaping: blasta ‘blast’, copy-a ‘copy’, editera ‘edit’, fixa ‘fix’, formatta ‘format’, fótósjoppa ‘photoshop’, modda ‘modify’, peista ‘paste’, patcha ‘patch’, prógrammera ‘program’, resetta ‘reset’, updata ‘update’, upgreida ‘upgrade’ ... 9. Verbs of (means of) cutting: partiona ‘partition’, trimma ‘trim’ ... 11. Verbs of (de)limitation: branda ‘brand’ ... 13. Verbs of destruction: bomba ‘bomb’, erasa ‘erase’, hakka ‘hack’, krakka ‘crack’, skvassa ‘squash, break’ ... 15. Verbs of displaying: mounta ‘mount’, publisha ‘publish’, rendera ‘render’ ... 16. Verbs of doing and producing: compilera ‘compile’, döbba ‘dub’, fótósjoppa ‘photoshop’, prógrammera ‘program’, pródúsera ‘produce’ ... 17. Verbs of dwelling and keeping in place: archive-a ‘archive’ ... 18. Verbs of (caused) emotion: fíla ‘like’, digga ‘dig, like’ ... 19. Verbs of feeding and consumption: mönnsa ‘munch’, tóka ‘smoke hash’ ... 22. Verbs of (means of) gaining: hössla ‘hustle’, krakka ‘crack’, ranta ‘rent’, rippa ‘rip’ ... 25. Verbs of increasing and strengthening: supporta ‘support’ ... 27. Verbs of human manipulation: bojkotta ‘boycut’, hössla ‘hustle’, kidda ‘kid’ ... 29. Verbs of meeting and uniting: paira ‘pair’, synca ‘synchronize’, synkrónisera ‘synchronize’ ...



Productivity

30. Verbs of non-translational motion: flexa ‘flex’, rippa ‘rip’, rokka ‘rock’, skratsa ‘scratch’ ... 32. Verbs of (direct) physical affectedness: biddsslappa ‘bitchslap’, muffa ‘bang’, meisa ‘spray with tear gas’, skratsa ‘scratch’, slamma ‘slam’ ... 33. Verbs of possession: óna ‘own’ ... 40. Verbs of catching and termination: bösta ‘bust’, massa ‘finish with style’ ... 42. Verbs of translational motion and (means of) traveling: browsa ‘browse’, hakka ‘hack’, logga ‘log’, ssh-a ‘ssh’, sörfa ‘surf ’ ... 44. Verbs of (interactive) verbal behavior: battla ‘battle’, tagga ‘write’ ... The borrowed verbs assigned Nom-Acc by Icelandic speakers fall into twentythree of the forty-six narrowly-circumscribed semantic classes established above, while the Nom-Dat predicates fall into eleven out of thirty-two classes. Nom-Dat: 9. Verbs of destruction: deleta ‘delete’, krassa ‘crash’, mökka ‘contaminate’, offa ‘off ’ ... 10. Verbs of division: sjera ‘share’ ... 12. Verbs of starting and finishing: restarta ‘restart’, starta ‘start’ ... 17. Verbs of increase: adda ‘add’, de-multiplexa ‘demultiplex’ statta ‘stat, dublicate’ ... 19. Verbs of losing: droppa ‘drop’, neimdroppa ‘name drop’ ... 21. Verbs of (caused) motion: blasta ‘blast’, bomba ‘bomb’, bundla ‘bundle’, downloada ‘download’, dumpa ‘dump’, droppa ‘drop’, ejecta ‘eject’, expandera ‘expand’, exporta ‘export’, farta ‘fart’, innstalla/innstallera ‘install’, neimdroppa ‘name drop’, parkera ‘park’, peista ‘paste’, poppa ‘deliver when popping by’, pósta ‘post’, publisha ‘publish’, releasa ‘release’, slamma ‘slam’, streyma ‘stream’, uploada ‘upload’, untara ‘unzip’, unzippa ‘unzip’ ... 22. Verbs of non-translational motion: de-multiplexa ‘demultiplex’ tilta ‘tilt’ ... 25. Verbs of organizing: umba ‘be agent for’ ... 29. Verbs of stealing: rippa ‘steal’ ... 32. Verbs of (means of) verbal communication: msna ‘msn’ ... 33. Verbs of changing: convertera ‘convert’, offa ‘off ’, resolva ‘resolve’ ... Not all of these 107 borrowings are conventionalized in the Icelandic language community. However, the degree of conventionalization of these verbs is irrelevant for the purpose of the present research, as it is the motivation behind the case assignment which is of interest here. It is also worth pointing out that the borrowed verbs assigned the Nom-Dat construction are not necessarily verbs whose object has a human referent, as is otherwise the case with 45% of dative object verbs in Icelandic (cf. Section 3.1.2.2



Chapter 3.  New verbs in Icelandic 

above). In fact, only two of the thirty-nine borrowed Nom-Dat verbs are used with a human argument in my sources, namely msna ‘communicate with sby through msn’ and neimdroppa ‘name drop’: (4) Leiðinlegt að msna fólki sem situr við hliðina á mér. boring to msn people.dat who sit with side-the on me ‘It is so uninteresting to msn people who sit beside me.’ (5) Þú sýnir ekki einu sinni sóma þinn í að neimdroppa mér. you show not one time honor your in to name-drop me.dat ‘You don’t even have the decency to name drop me.’ The remaining Nom-Dat verbs select for ordinary non-human objects. Also, since most of these verbs are borrowed from English which does not have a morphological dative, these verbs have not been borrowed together with their source language’ verb-specific argument structure constructions, which would otherwise be a possibility (cf. Barðdal 1999b, 2001a: 132–33). Observe also that some of the borrowed verbs are in fact listed twice, i.e. both as taking Nom-Acc and Nom-Dat. In some cases there is a clear difference in both semantic construal and subconstruction of the relevant argument structure construction. Consider the following two examples of bomba ‘bomb’: (6) Þá voru Kanarnir að bomba Júgóslava. then were Americans to bomb Yugoslavians.acc ‘Then the Americans were bombing the Yugoslavians.’ (7) Mér tókst að bomba tannkremi í augað á mér. I.dat managed to bomb tooth-paste.dat in eye-the on me ‘I managed to bomb toothpaste into my eye.’ In (6) bomba is a verb of destruction (Nom-Acc subconstruction 13 above), whereas in (7) it is used in the Caused-motion construction (Nom-Dat subconstruction 21). Consider also the two examples of the verb publisha in the material: (8) Get ég publishað dagatal þannig að bara ... can I publish calendar.acc such that only ‘Can I publish a calendar so that only ...’ (9) Get ég publishað mínu iCal dagatali á netið ... can I publish my.dat iCal calendar.dat on net-the ‘Can I publish my iCal calendar onto the internet ...’ In example (8) publisha is a verb of displaying (Nom-Acc subconstruction 15) whereas (9) entails “movement” or projection of the object calendar from the computer (or wherever the calendar is located) onto the World Wide Web, i.e. publisha



Productivity

is used in the Caused-motion construction with a dative object. This is manifested by the directional adverbial in (9) á netið as opposed to a locative adverbial á netinu, which would be expected if publisha were used as an ordinary verb of displaying in (9), not instantiating the Caused-motion construction (cf. also Section 5.1.1 below). A similar example is found with the verb farta ‘fart’ which is not a causedmotion verb in itself but a verb of air-emission: (10) Farið að farta þessari plötu út svo ég ... start to fart this.dat record.dat out so I ‘Can’t you start farting this record out so that I ...’ The adverbial út ‘out(wards)’ in (10) above is directional, while its locative variant is úti ‘out’. It is well known in Icelandic that directional adverbials are used with directed motion and locative adverbials with non-directed or no motion (cf. Barðdal 2001a: 151). The co-occurrence of directional adverbials with publisha and farta shows clearly that borrowed non-motion verbs can be used in the Caused-motion construction in Icelandic with a dative object, a fact which I shall return to below. It is striking, moreover, how many of the borrowed verbs occur in the Causedmotion construction with a dative object, i.e. twenty-two, as opposed to five with an accusative object. This, of course, reflects the already existing frequency differences in Icelandic found between ten translational motion verbs in the Nom-Acc construction as opposed to fifty caused-motion verbs in the Nom-Dat construction, pointed out in Section 3.1.2.2 above. Observe, also, that the five borrowed NomAcc verbs are all verbs of locomotion, i.e. they are of the type where it is the subject referent that moves along a path (11) or covers a “surface” (12), not the object. (11) Honum tókst með bellibrögðum að hakka sig inn á síðuna ... he.dat managed with tricks-the to hack himself.acc in on page-the ‘He managed to hack his way onto the page using tricks ...’ (12) Ég var að sörfa netið ... I was to surf net-the ‘I was surfing the net ...’ In this sense, the five Nom-Acc borrowed verbs of motion differ from the twentytwo borrowings occurring in the Caused-motion construction where it is the dative object that is caused to move. Hence, the borrowed Nom-Acc and Nom-Dat verbs of motion are of different nature, as they constitute different subclasses of motion verbs. They therefore instantiate different subconstructions of the NomAcc and the Nom-Dat, respectively.



Chapter 3.  New verbs in Icelandic 

Recall from Section 2.5 above that the general idea within cognitive-functional linguistics is that it is the most entrenched level of a construction which determines its productivity (cf. Clausner and Croft 1997). There can be no doubt here that it is the subconstruction of Caused-motion that represents the most entrenched level of the Nom-Dat construction in Icelandic, with fifty predicates of 188 instantiating it (see Section 3.1.2.2). That is, this is the level of the construction which is instantiated by most types. This predicts, moreover, that the productivity of the Nom-Dat construction in Icelandic should be confined to the Caused-motion construction. Such a prediction is clearly not borne out by the data from Icelandic discussed here, as seventeen of thirty-nine borrowed Nom-Dat predicates do not instantiate the Caused-motion construction. The hypothesis within cognitive-functional linguistics that productivity is a derivative of a construction’s most entrenched level can therefore not be maintained. In contrast, the hypothesis advocated here, that the productivity of a construction can be derived from its most schematic level, as it is this level which defines the semantic range of the construction, can account for the fact that the borrowed verbs assigned Nom-Dat by Icelandic speakers are distributed within the semantic field of the Nom-Dat construction in Icelandic. The more observant reader may have noticed that an overwhelming majority of the borrowed verbs listed and discussed in this section are translational equivalents of the already existing Icelandic verbs listed and discussed in the preceding section. This suggests that Icelandic speakers assign case constructions to borrowed verbs on the basis of concrete lexically-filled low-level verb-specific constructions, and not on the basis of more abstract schematic higher-level constructions. To exemplify, deleta (Nom-Dat) is a translation of eyða (Nom-Dat), sjera (Nom-Dat) of deila (Nom-Dat), convertera (Nom-Dat) of breyta (Nom-Dat), offa (Nom-Dat) of aflýsa (Nom-Dat) and droppa (Nom-Dat) of sleppa (Nom-Dat). The same pattern can be discerned for Nom-Acc verbs: testa (Nom-Acc) is synonymous with Icelandic prófa (Nom-Acc), builda (Nom-Acc) with byggja (Nom-Acc), peista (Nom-Acc) with líma (Nom-Acc) and supporta (Nom-Acc) with styðja (Nom-Acc). These are, in other words, low-level analogical extensions, where a borrowed verb is assigned a case frame on the basis of full synonymy with an already existing Icelandic verb. Notice that this is equivalent to the rhyming effects observed with children’s over-extensions of declensional patterns to verbs, found by Ragnarsdóttir, Simonsen and Plunkett (1999), reported on in Section 2.5 above, with the only difference that their overextensions were based on form while these are based on meaning. Extensions of verb-specific constructions are therefore clearly not restricted to syntax but are also found in morphology. There is one difference, however, between the results of this study and the study by Ragnarsdóttir, Simonsen and Plunkett (1999), namely that they did not find any rhyming effect



Productivity

for the largest weak class, while I have here found that also the Nom-Acc construction is assigned by speakers on a verb-for-verb basis. Even though these examples show beyond doubt that case and argument structure constructions found with borrowed verbs can be assigned on the basis of the case and argument structure construction of a translational equivalent, not all case assignment to borrowed verbs is of that type. For example, the Nom-Acc verbs edita and erasa do not behave according to this pattern. The Icelandic verbs for the English edit and erase are ritstýra and eyða, respectively, and they both select for the Nom-Dat construction in Icelandic. Hence, if edita and erasa had been assigned case and argument structure construction on the basis of their translational equivalents, they should have occurred in the material with the Nom-Dat construction instead of the Nom-Acc construction. However, they do not. Recall, now, that verbs of cognition and mental activity and verbs of destruction are shared across both Nom-Acc and Nom-Dat constructions in Icelandic, and one of the predictions of my analysis, laid out in Section 3.1.2.2 above, was exactly that there should be variation across the subconstructions where semantic overlap is found between the Nom-Acc and the Nom-Dat constructions. The data presented and discussed here and below show that such variation is indeed found in my material, and hence that the predictions of the current analysis are clearly borne out. Three of the subconstructions which are specifically mentioned in Section 3.1.2.2 as exemplifying a semantic overlap between the Nom-Acc and the NomDat constructions are exactly with verbs of cognition, verbs of destruction and verbs of motion. In fact, verbs of cognition and mental activity is one of the largest class of Nom-Acc verbs, subsuming thirty predicates in the corpus count which was the basis for the classifications in 3.1.2.1–3.1.2.2 above. It is therefore clear that the borrowed verb edita has been attracted by a salient cluster of cognition verbs in Icelandic instead of being attracted by its only translational equivalent ritstýra. Also, verbs of destruction count seven Nom-Acc verbs in the corpus count but only three Nom-Dat verbs. Therefore, Nom-Acc verbs of destruction are higher in type frequency than Nom-Dat verbs of destruction, which again explains why erasa has been attracted by the Nom-Acc construction instead of being attracted by its translational equivalent eyða. Similar examples are also found for Nom-Dat verbs. Some of the borrowed verbs which instantiate the Nom-Dat construction do so despite the fact that their closest translational equivalents are Nom-Acc verbs. Several such examples are found in the Caused-motion construction. Consider for instance expandera and exporta which are only found in the Nom-Dat construction in my material. The closest Icelandic translational equivalents of expandera and exporta, however, are the Nom-Acc verbs þenja and flytja út, respectively. The only explanation for the case marking of expandera and exporta is thus that they have been assigned case on



Chapter 3.  New verbs in Icelandic 

the basis of a verb-subclass-specific construction, i.e. a cluster of verbs with causedmotion meaning, instead of being assigned case on the basis of the case marking of the verb-specific constructions of their translational equivalents. This analysis is further supported by the fact that verbs like farta ‘fart’, publisha ‘publish’ and peista ‘paste’ which are not inherently motion verbs can occur in the Caused-motion construction in Icelandic. That is, they are not assigned case on the basis of the case marking of their translational equivalents but on the basis of the fact that they instantiate the Caused-motion construction here, signifying its means of motion. The data presented in this section do not confirm the validity of the distinction between lexical and structural case, as the borrowed Nom-Acc verbs can as easily as the borrowed Nom-Dat verbs be divided into subconstructions on the basis of the semantics of the verbs instantiating the construction. It turns out that eighty-eight of 107 verbs were assigned case and argument structure on the basis of a synonymous Icelandic verb, sixty-three out of sixty-eight Nom-Acc verbs and twenty-five of thirty-six Nom-Dat verbs (cf. Barðdal 2009a). Most of the remaining verbs were clearly assigned case and argument structure on the basis of verbsubclass-specific constructions, as discussed above. This means that since one must assume that the assignment of the Nom-Acc construction to borrowed verbs is based on verb-specific and verb-subclass-specific constructions, the difference between case assignment to accusative and dative objects in Icelandic is not qualitative. The fact that more borrowed verbs are assigned Nom-Acc than Nom-Dat is a difference of quantity not quality, a difference which is expected anyway on the basis of differences in the type frequency of the two constructions. Moreover, if the borrowed Nom-Acc verbs had not been assigned case and argument structure constructions on the basis of their synonymous verb-specific construction in Icelandic, then one would have expected these borrowings to divide evenly across the semantic spaces of the Nom-Acc and Nom-Dat predicates, and not be confined to the semantic space of the Nom-Acc construction. That is, one would have expected a substantial proportion of Nom-Acc verbs not to be assigned case and argument structure constructions on the basis of near-synonymy. This prediction, however, is not borne out. In addition, there are several examples of the so-called idiosyncratic case being productive, again contradicting the predictions of generative grammar. As the Nom-Dat construction counts as idiosyncratic on the generative approach, it should not be productive. Consider, however, the following examples: (13) Svo geturðu convertað lögunum fyrir iTunes ... then can-you convert songs-the.dat for iTunes ‘Then you can convert the songs for iTunes ...’



Productivity

(14) ... að umba útreiknuðu æviverki ... to be-agent-for calculated.dat life-work.dat ‘... to be agent for the estimated life work ...’ The borrowed verb converta ‘convert’ in (13) occurs here in the Nom-Dat construction, presumably on the basis of synonymy with its Icelandic translational equivalent breyta which takes Nom-Dat. No general or semantic rule can be assumed to be responsible for this case assignment. The verb umba ‘be agent for’ in (14) is derived from the noun umboðsmaður ‘agent’ and there is no verb in Icelandic with this or any similar meaning. On the generative approach, case assignment to new verbs on the basis of one model verb, as with converta, and no model verb, as with umba, would be clear examples of the extensibility of idiosyncratic case, and hence of the productivity of idiosyncratic case assignment. Consider yet another example of the extensibility of idiosynctatic case assignment: (15) ... raungreinagramsari að mökka geli við austurbæjarlaug ... real-science-nerd to smoke jelly.dat at east-side-pool ‘... a science nerd smoking jelly at the East Side Pool ...’ The verb mökka ‘smoke’ in (15) is most likely denominal, derived from the Icelandic noun mökkur ‘smoke’. Also, there are no verbs meaning ‘smoke’ in Icelandic which select for the Nom-Dat construction. There are, however, expressions like poppa pillum ‘pop pills’ and droppa sýru ‘drop LSD’, where the object is clearly dative marked. If mökka is assigned the Nom-Dat construction by Icelandic speakers on the basis of expressions like poppa pillum and droppa sýru, then that suggests a dative object marker for situations where drugs are involved. If so, there is a special dative object case in Icelandic confined to the social context of using drugs. Case assignment can hardly get more idiosyncratic than that, I would say. To summarize the content of this section, both the Nom-Acc and the NomDat subconstructions of the Nominative subject construction are productive in Icelandic. The Nom-Acc construction is assigned to 64% of borrowed verbs while the Nom-Dat construction is assigned to 36% of the borrowings. These figures mirror the existing type frequencies of the Nom-Acc and Nom-Dat constructions in Icelandic. A further investigation reveals that the case marking of borrowed verbs seems to be motivated by the case marking of their Icelandic translational equivalents. This means that both the Nom-Acc and the Nom-Dat subconstructions are extended to new verbs on the basis of analogy with concrete lexicallyfilled low-level verb-specific constructions in Icelandic, or on the basis of lowlevel verb-subclass-specific constructions, and not higher-level schematic constructions. In other words, they are either attracted by a synonymous verb or by a verb cluster consisting of predicates with a similar meaning.



Chapter 3.  New verbs in Icelandic 

These results contradict the assumption within cognitive-functional linguistics that the productivity of a construction can be derived from its most entrenched level, and the following prediction that the productivity of the Nom-Dat construction in Icelandic should be confined to the subconstruction of Caused-motion. The results presented here also contradict the predictions of classical generative grammar where a dichotomy between structural and lexical case is assumed, with only structural case being productive while lexical and idiosyncratic case should be unproductive, as it belongs to the periphery of the grammar. The data presented in this section show that the variation between Nom-Acc and Nom-Dat is either found with the subconstructions which overlap semantically between the two case constructions or the variation involves differences in construal. In other words, case and argument structure assignment of borrowed verbs in Icelandic takes place within the productivity domain of each subconstruction, meaning that new Nom-Acc verbs come to belong to the same semantic subclasses as already existing Nom-Acc verbs, and new Nom-Dat verbs come to belong to the same semantic subclasses as already existing Nom-Dat verbs in Icelandic. 3.2.2 Analogy and token frequency In Section 2.5 above I argue extensively that productivity is a function of the type frequency and semantic coherence of a construction, and an inverse correlation between the two. This means that the higher the type frequency of a construction, the lower degree of semantic coherence is needed for that construction to be productive, and vice versa that the lower the type frequency, the higher degree of semantic coherence is needed for that construction to be extended. The inverse correlation between these two factors is graphically represented in Figure 2.4 in Section 2.5 above, with highest type frequency and lowest degree of semantic coherence at the leftmost top of the graph and lowest type frequency and highest degree of semantic coherence at the rightmost bottom of the graph. The graph also represents the productivity cline, i.e. the range from full productivity, via different degrees of productivity to analogical extensions. This is shown in Figure 3.7, with full productivity located at the leftmost top of the productivity cline, different degrees of productivity located at different intervals along the cline and analogy being located at the rightmost bottom of the productivity cline, as analogical extensions are based on lowest possible type frequency, i.e. one, and highest degree of semantic coherence, i.e. full synonymy between the source and the target item. As already stated in Chapter 1, by analogy I refer to analogical extensions, in this case the extensions of a grammatical property, i.e. morphological case, to new items, based on the similarity found between a source item and a target item. With 88% of the borrowed verbs discussed above, we get an extension based on only one

 Productivity

Type Frequency High Full productivity

Different Degrees of Productivity

Analogy

Semantic Coherence Low

High

Figure 3.7  Productivity vs. analogy: A difference in degree

model verb, and as the source and the target verbs are translational equivalents, there is full synonymy between them. This means, furthermore, that the relation between full productivity and analogy is a consequence of the inverse correlation between type frequency and semantic coherence. Analogy can thus be regarded as the other side of the productivity coin, as is predicted by its location at the bottom of the productivity cline in Figure 3.7. This analysis is further supported by claims in the literature that analogy should not only be regarded as a low-level operation, but that it can also operate at higher levels (cf. Itkonen 2005). This analysis contradicts Bauer (2001: 93) who regards analogy as being fundamentally different from rule-based extensions. In his discussion of rule-governed innovation and rule-changing innovation, which on his account are motivated by rules and analogy, respectively, he states the following: (16) Rule-governed innovation is always based on an already existing pattern which is found across at least two items; rule-changing innovation is based on a pattern which is perceived for the first time at the point of innovation and which may exist in only one model. [Emphasis added] Hence, a rule applies if two items are model items, whereas analogy applies of there is only one model item. There is thus a qualitative difference assumed between rule-



Chapter 3.  New verbs in Icelandic 

based extensions and analogy, even though the difference between two and one model items is purely quantitative. However, this view is not particular to Bauer. It is a general view found very frequently in the literature: Analogy is based on one model item, while rule-based extensions are based on two or more model items. From the present cognitive-functional CxG standpoint where constructions are assumed to exist at different levels of schematicity, this view appears as rather ad-hoc, as it is far from clear why the numerical distinction between one and two model items should yield differences in the nature of the extensions. No ontological difference is assumed between a verb-specific construction or a schema, by definition instantiated by only one verb, and a verb-subclass-specific construction or a schema instantiated by two or more verbs. The difference is only quantitative, adding up to various levels of schematicity at which each construction exists in the language system of individual speakers. Therefore, an extension based on a verbspecific construction is not qualitatively different from an extension based on a verb-subclass-specific construction consisting of two verbs. Hence, the difference between full productivity and low-level analogical extension is a difference in degree, not a difference in quality. This follows directly from the view of productivity advocated here, namely that it is a function of both type frequency and coherence and an inverse correlation between the two. The degree of semantic coherence found between the extended construction and the item the contruction is being extended to is thus proportional to the type frequency of the extended construction: For low type frequency constructions, there will be a high degree of semantic similarity found between the construction and the verb which is being attracted by the construction, whereas for high type frequency constructions, the attracted verb only has to be within the general semantic scope of the relevant high-level schematic construction. This is a consequence of the fact that the semantic scope of high-level schematic constructions is much broader than the semantic scope of low-level verb-subclass-specific constructions which is narrower and more specific. As long as there is no qualitative difference between abstract schematic meanings and specific lexical-semantic meanings, and no qualitative difference between low type frequency and high type frequency, there cannot be a difference in quality here, only a difference in degree. One major advantage of the present approach is that it allows for a unified account of productivity and analogy and it does not force us to assume an ontological difference between extensions based on one or two model items, or, in other words, extensions based on constructions instantiated by one or two verbs. All the examples discussed in Section 3.2.1 above involve low-level analogical extensions of verb-specific or verb-subclass-specific constructions which are subconstructions of other higher-level constructions, either the Nom-Acc or the Nom-Dat construction. Such examples are, however, not the only examples of



Productivity

low-level analogical extensions. Examples can also be found where the model verb is not a part of a schematic construction, but is a verb-specific island – an idiomatic expression, in other words. Consider the following example: (17) Vörubíllinn donaði uppi í brekkunni. truck-the dawned up in slope-the ‘The truck ended up/was forgotten in the slope.’ In (17) the English verb dawn has been borrowed into Icelandic and used as a particle verb dona uppi on the basis of the Icelandic particle verb daga uppi which means ‘end up somewhere/be forgotten.’ The Icelandic predicate daga uppi consists of the simple verb daga meaning ‘dawn’ and the locative adverbial uppi ‘up’. It was originally used about night trolls who were caught by daylight. It is well known that if one is a night troll and is caught by daylight, one immediately turns into stone, and thereby does not compose a threat to the surroundings anymore. Hence, one may simply end up somewhere and gradually be forgotten by everybody. Observe, moreover, that the predicate daga uppi ‘be caught by daylight, be forgotten’ is a lexicalized idiom whose meaning is non-compositional and not derivable from the meaning of the parts. It is not at all obvious that a verb meaning ‘dawn’ together with the locative adverbial ‘up’ will have the meaning ‘be forgotten’. It is much more likely that such a predicate would mean that it dawned somewhere, with the locative adverbial closer specifying the location of the dawning event. However, this is far from the meaning of daga uppi in Icelandic which again shows that daga uppi must be regarded as an idiosyncratic set phrase, and that the formation of the predicate dona uppi is therefore not based on a pattern, rule or a super-ordinate schema, but on the already existing idiomatic set phrase daga uppi and that alone. In other words, dona uppi is formed on the basis of the Icelandic expression daga uppi, in which case we find the lowest possible type frequency, i.e. one, and the highest possible degree of semantic coherence, i.e. full synonymy. This is therefore a clear-cut example of analogy, located at the bottom of the productivity cline in Figure 3.7 above. Given a definition of syntactic productivity based on the extensibility concept (cf. Section 2.4 above), where full productivity and analogy are the extreme poles of the productivity cline, the question arises whether earlier assumptions, like those of Bybee (1985, 1995, 2001), Bybee and Thompson (1997), and Clausner and Croft (1997), on the relation between token frequency and productivity hold. They argue that token frequency detracts from productivity, as items that are high in token frequency are more lexicalized than items that are low in token frequency. Thereby, items high in token frequency do not contribute to the type frequency of their higher-level constructions and hence not to their productivity. All these scholars acknowledge and argue that productivity is gradient, and yet they have



Chapter 3.  New verbs in Icelandic 

only considered the role of token frequency for full or intermediate productivity (see however the discussion in Section 2.5 on strung and swept-verbs in English). As I argue, however, in Section 2.5 above, each token also instantiates a type, and if token similarity contributes to analogical extensions, then the token frequency of the possible model verbs should be of utmost importance for the choices a speaker makes when assigning, for instance, case and argument structure constructions to borrowed verbs. Consider now the following (non-fictive) conversation between two native speakers of Icelandic living abroad: (18) A: B: A: B:

Ef þú ætlar að applisera þessari greiningu á efnið þá ... if you intend to apply this.dat analysis on material-the then Þú meinar applisera þessa greiningu. you mean apply this.acc analysis Nei, applisera einhverju eins og beita einhverju. No, apply something.dat like apply something.dat Nei, applisera eitthvað eins og nota eitthvað. No, apply something.acc like use something.acc

Speaker A has borrowed the Scandinavian verb applisera (cognate to the English verb apply) and uses it in the Nom-Dat construction. Speaker B corrects him/her, claiming that the verb should be used in the Nom-Acc construction. Speaker A then explicitly argues that applisera should be used in the Nom-Dat construction like its Icelandic translational equivalent beita, while speaker B associates applisera with the more general verb nota which means ‘use’ and selects for the Nom-Acc construction in Icelandic. How come, now, that two speakers of Icelandic associate the borrowed applisera with two different case and argument structure constructions, either the NomAcc or the Nom-Dat construction? I have argued in Section 3.2.1 above that such variation in case marking results from analogy with either a verb-specific or a verb-subclass-specific construction. As the translational equivalent of applisera in Icelandic is beita which selects for the Nom-Dat construction, this would exemplify a low-level verb-specific analogical extension of the same kind as discussed in 3.2.1 above. The Icelandic verb nota ‘use’ has one synonym in Icelandic, nýta ‘make use of ’ which is etymologically related to nota. Therefore, nota and nýta make up a small verb-subclass-specific construction (Nom-Acc subconstruction 43 above) which is semantically related to beita and which might thus have attracted applisera to the Nom-Acc construction. However, speaker B only mentions the verb nota in his/her argumentation for the accusative case of the object of applisera, not both verbs. It is therefore reasonable to assume that speaker B assigned the Nom-Acc construction to applisera on the basis of the case marking of nota, in which case we have another verb-specific low-level extension.



Productivity

This raises the more general question of why these two speakers associate the borrowed applisera with two different verbs in Icelandic, beita (Nom-Dat) and nota (Nom-Acc), given that only beita is a direct translational equivalent of the English apply and the Scandinavian applisera, while nota is more general.  One possible answer to this question might be that nota collocates more with the noun greining ‘analysis’ from the conversation above than beita does. If so, the speaker’s choice of the noun greining should prompt him/her to assign the case frame of nota to applisera instead of the case frame of beita. However, an Internet search with the strings að nota þessa greiningu ‘to use this analysis’ and að beita þessari greiningu ‘to apply this analysis’, i.e. the infinitive together with the definite noun phrase þessa/þessari greiningu, yields one hit for each of the two strings. As the number of hits is extremely low for both strings, i.e. only one, it is doubtful that either verb collocates with greining, and as there is no difference in the frequency between them, there is no support for the idea that nota collocates more strongly with greining than beita does. Another possibility is that token frequency plays a role here. An Internet search of the infinitives of beita and nota on Icelandic web sites, together with the infinitive marker að ‘to’, yields enormous differences in token frequency of beita and nota, given in Table 3.6. The infinitive að nota occurs six times more often in language use than the infinitive að beita, with less than a quarter of million hits for beita but with 1,2 million hits for nota. The overwhelmingly higher token frequency of nota in Icelandic might thus contribute to speaker B’s choice of nota over beita, which is otherwise unexpected, as it is beita which is the translational equivalent of applisera, not nota. In other words, the high token frequency of nota makes it more entrenched as an item than beita, which in turn makes it a more salient option for speakers assigning case and argument structure construction to applisera. Observe that these frequency figures are based on document frequency, not word frequency, as some documents published on the Internet may be published more than once. However, as there is no reason to assume that the distribution of these documents is skewed in favor of either verb, there is no reason either not to take the difference found between nota and beita seriously. If token frequency is a relevant factor for speakers’ choice of model items in lower-level extensions, then its importance should correlate inversely with the importance of type frequency. That is, high token frequency should be more important for lower-level verb-specific analogical extensions than for the productivity of high type frequency constructions, i.e. the lower down on the productivity cline, the more important should it be. This inverse correlation is shown in Figure 3.8.



Chapter 3.  New verbs in Icelandic 

Table 3.6  Frequency hits of nota and beita in the infinitive Verb Occurrences

að nota

að beita

1,210,000

223,000

Type Frequency High

Token Frequency Low

High

Figure 3.8  The inverse correlation between type and token frequency for productivity

An ultimate issue can now be addressed: If most extensions are low-level extensions, including extensions of higher-level schematic constructions, then it is not the high type frequency of a construction which contributes to its productivity, as argued for instance by Bybee (1985, 1995, 2001), but rather all the individual lower-level item-specific constructions in themselves. And since more lower-level constructions make up high type frequency patterns than low type frequency patterns, high type frequency patterns will of course attract more new items than patterns which only exist at lower levels. If it is the case that it is the low-level verbspecific constructions which are responsible for a construction’s extensibility, token frequency will not be less important for productivity than type frequency. Type frequency will be an indicator of the highest level of schematicity each construction exists at, and hence an indicator of the semantic scope of the construction and its productivity domain, but token frequency will be an important psycholinguistic factor singling out model items for speakers when they extend



Productivity

low-level constructions. Token frequency will thus be quite important for productivity, contra what has hitherto been assumed in the traditional cognitive-functional linguistic literature. To summarize the content of this section, I have argued that there is no qualitative difference between analogical extension and rule-based extension, as the ontological status of an extension based on a verb-specific construction is not different from an extension based on a verb-subclass specific construction. This is a direct derivative of the fact that there is no qualitative difference between higherlevel schematic meaning and lower-level lexical-semantic meaning, only a difference in specificity. The present cognitive-functional CxG approach to syntactic productivity as extensibility thus provides a unified account of full productivity, degrees of productivity and analogy, as these are only located at different points on the productivity cline. Given that low-level analogical extensions seem to be responsible for the productivity of higher-level schematic constructions, the question of which role token frequency plays becomes highly relevant. Obviously, the token frequency of individual model verbs is central for speakers’ choice of models for their low-level verb-specific analogical extensions. Hence, token frequency is much more important for productivity than argued by Bybee (1995), Clausner and Croft (1997), Croft and Cruse (2004) and others. 3.3 Summary An investigation of the different subconstructions of the Nominative subject construction, i.e. Nom-Acc, Nom-Dat and Nom-Gen, yields significant differences in their type frequency in Icelandic. A dictionary count reveals that the Nom-Acc construction is instantiated by more than 1,380 predicates, the Nom-Dat by at least 738 predicates and the Nom-Gen construction by ca. thirty-seven types. In two different counts, a dictionary count and a corpus count, the proportions between the Nom-Acc and the Nom-Dat constructions turn out to be 65 vs. 35% and 62 vs. 38%, respectively. Regarding the semantics of these constructions, the predicates instantiating the Nom-Acc construction consist of forty-six narrowly-circumscribed semantic classes, the predicates instantiating the Nom-Dat construction subsume thirty-two such, while the Nom-Gen predicates make up only four semantic classes. The highest schematic level of the Nom-Acc construction consists of six basic human event types, i.e. making, movement, affectedness, cognition/emotion, change and location. The highest schematic level of the Nom-Dat construction subsumes three basic human event types, i.e. change, cognition/emotion and superiority. The highest schematic level of the Nom-Gen construction is a verb-subclass level, with a high degree of semantic



Chapter 3.  New verbs in Icelandic 

specificity. These facts predict that only the Nom-Acc and the Nom-Dat construction should be productive in Icelandic, and as there is some semantic overlap between the two constructions, it is expected that new verbs having that particular semantics will have a choice between the two. A survey of verbs recently borrowed into Icelandic, mostly from the area of information technology, reveals that 64% of the borrowings are assigned NomAcc by Icelandic speakers and 36% are assigned Nom-Dat. These frequency figures mirror the proportions between existing Nom-Acc and Nom-Dat predicates in Icelandic. This is, moreover, unexpected on the classical generative approach where nominative is assigned to subjects and accusative to objects on the basis of their structure in the sentence, and all other case marking is regarded as lexical. Such a view predicts that only the Nom-Acc construction should be productive while the Nom-Dat construction should be unproductive, as dative case on objects counts as lexical. The present empirical findings clearly contradict the predictions and suppositions of classical generative grammar. The present findings also contradict the assumption within cognitive-functional linguistics that the productivity of a construction can be derived from its most entrenched level. This assumption predicts that the productivity of the NomDat construction should be confined to the Caused-motion construction, as it clearly represents the most entrenched level of the Nom-Dat construction with fifty of 188 Icelandic predicates instantiating it. This prediction is not borne out. A closer inspection of the borrowed verbs reveals that they are all assigned case and argument structure constructions either on analogy with a verb-specific construction or a verb-subclass-specific construction. They thus fall into the same narrowly-circumscribed semantic classes as already existing Nom-Acc and NomDat predicates. Also, some variation is found between the subconstructions which show semantic overlap across the Nom-Acc and the Nom-Dat constructions. The remaining divergence in case marking is due to differences in construal between the subconstructions. However, low-level analogical extensions are not only found with higher-level schematic constructions like the Nom-Acc and the Nom-Dat constructions. They are also documented with constructions which only exist at low levels to begin with. Idiomatic set phrases exemplify that; they have the lowest possible type frequency, i.e. only one, and the meaning of the whole is not a sum of the meaning of the parts, but is non-compositional. The reason such idiomatic expressions function as model verbs is that there is a 100% semantic overlap between the source and the target item. The source and the target expressions are, in other words, fully synonymous. On the present approach, full productivity and analogy are two sides of the same coin, and the difference between them is a difference in degree, not a difference in ontological status. This is a consequence of the fact that higher-level schematic



Productivity

meanings and lower-level lexical-semantic meanings are not qualitatively different from each other; they only differ with regard to their degree of specificity. Given that low-level analogical extensions seem to be responsible for the high productivity of higher-level schematic constructions, and not their high type frequency in and of itself, as previously argued in the literature, the role of token frequency becomes highly relevant. Obviously, since low-level analogical extensions are based on a high degree of token similarity, the frequency of occurrence of the tokens which can function as models for lower-level extensions becomes most important. Clearly, highly entrenched tokens are more easily activated than non-entrenched tokens, and are thus more available as model items than infrequently occurring tokens. Hence, token frequency is clearly no less important for productivity than type frequency, contra what has been assumed in the earlier cognitive-functional linguistic literature. The findings presented in this chapter suggest that it is irrelevant at which level of schematicity a construction exists when it comes to its extensibility or productivity, in that the actual extensions of a construction to new items are still based on low-level analogy, i.e. on either verb-specific constructions or verb-subclass-specific constructions. This of course does not diminish the importance of a construction’s highest level of schematicity, which helps to determine the semantic scope and the productivity domain of the relevant construction. In other words, the overall productivity of a given construction is still a derivative of its type frequency and semantic coherence and the inverse correlation between the two. In the next two chapters I will present and discuss two different case studies of case and argument structure assignment in Modern Icelandic. In chapter 5 I report on a questionnaire survey of the case and argument structure assignment of new verbs of instrument of communication in Icelandic, while in the next chapter I describe an experimental study with nonce verbs. That study shows that in certain circumstances do Icelandic speakers assign case, not only on the basis of lowlevel verb-specific constructions, as argued above, but also on the basis of highlevel schematic constructions, in contrast to what I have shown in this chapter. Notes 1 Not all constructions are semantically defined by the verbs which instantiate them. In particular, narrowly-defined grammatical constructions that are partially lexically-filled can have a specific meaning not derivable from the meaning of the individual parts. One such construction is the inchoative construction in Icelandic (cf. Barðdal 2001c), which has three different subconstructions, lexically-filled with five different auxiliaries of which only two have the lexical meaning ‘begin’ outside of the inchoative. However, when occurring in the inchoative construc-



Chapter 3.  New verbs in Icelandic  tion, all the auxiliaries mean ‘begin’. This fact cannot be derived from the meaning of any of the lexical items or the individual parts of the inchoative construction in Icelandic but has to be attributed directly to the construction itself. 2

See Maling (2002) for a different semantic classification of Nom-Dat verbs.

3 It may be pointed out here that at the time when I wrote my first papers on dative case with objects of motion verbs in Icelandic (Barðdal 1992, 1993), dative on objects was generally considered as being idiosyncratic. After these papers caught the attention of the research community, this type of object case marking has gradually been redefined within the field of generative grammar, as being ‘thematic’ instead of ‘idiosyncratic’ (as in Jónsson 2005: 384). This raises the more general question of whether the whole concept of idiosyncratic case within generative grammar may simply be a consequence of lack of research on the motivation behind the case assignment of low-level verb-subclass-specific constructions. I would be inclined to answer that question in the affirmative.

chapter 4

Nonce verbs A psycholinguistic experiment

In this chapter I report on a psycholinguistic experiment of Icelandic speakers’ case assignment to nonce verbs, i.e. verbs which do not exist as independent lexemes in Icelandic but have been made up for the purpose of the immediate occasion. Earlier experiments on nonce verbs (cf. Braine, Brody, Shalom, Weisberger and Blum 1990) suggest that speakers use the meaning of a verb as the primary indicator of its argument structure construction. In that case, assignment of case and argument structure constructions to nonce verbs should be of exactly the same kind as assignment of case and argument structure constructions to borrowed verbs, i.e. based on low-level analogical extensions. In Section 4.1 I report on the conduction of the experiment which was carried out in Iceland with five nonce verbs and a filler between each verb. Forty Icelandic speakers participated in the experiment, twenty adults and twenty children. The nonce verbs were presented to the participants as synonymous with five existing Icelandic verbs which select for different case and argument structure constructions. Section 4.2 presents the results of the experiment, which clearly confirm the findings of Chapter 3 that speakers assign case and argument structure constructions on the basis of low-level analogical extensions of already existing verbs, and hence that constructions low in type frequency are also extendable to new items, contra the assumptions in the traditional cognitive-functional linguistic literature that high type frequency is a prerequisite for productivity. Moreover, certain facts evolving from the experiment suggest that Icelandic speakers may also assign case and argument structure constructions to nonce verbs on the basis of high-level schematic constructions which are so general that they are almost devoid of lexical semantic content, at best exhibiting relational content. Regarding the productivity of case and argument structure constructions in Icelandic, the Nom-Acc construction was assigned most often to the present nonce verbs, then the Dative subject construction, while the Nom-Dat construction was assigned least often. These findings give rise to a comparison between the semantic range of the Dative subject construction and the Nom-Dat construction, as these two constructions are considered to have approximately the same type frequency in Icelandic. This endeavor is undertaken in Section 4.3, which indeed shows that

 Productivity

the type frequency of the Dative subject construction is considerably lower in language use than the type frequency of the Nom-Dat construction. The semantic range of the Dative subject construction is also considerably narrower than that of the Nom-Dat construction, with a subsequent difference in the productivity domain of these two constructions, and hence in their factual extensions. Section 4.4 contains a short summary of the main findings and conclusions of this chapter. 4.1 The experiment Forty randomly selected monolingual Icelandic speakers participated in the experiment: twenty children age 6–13 and twenty adults.1 The experiment was presented to them in the following way (in Icelandic):

(1) This is a Funny-game. I will be showing you some pictures, and then you will tell me what is happening in the pictures. The only thing is that you are not going to do it in Icelandic, you are going to tell me in Funny-language. Now, Funny-language is almost identical to Icelandic, there are only a few words that are different, and you don’t have to worry about that because I will teach you those words.

The pictures used were five Screen Beans silhouettes, shown in Figures 4.1–4.5, presented in that order to the participants. Each figure was introduced with a formula (in Icelandic) corresponding to the one in (2).

(2) In this picture we see a man. What he is doing is called grilla [‘barbeque’] in Icelandic, but in Funny-language it is called slobba. Could you now tell me what you see happening in the picture?

Figure 4.1



Figure 4.2



Figure 4.3



Chapter 4.  Nonce verbs 

Figure 4.4



Figure 4.5

Table 4.1  Verbs used in the experiment Nonce verb flokast slobba kísa tvíta spofta

Icelandic verb

Gloss

Case and argument structure

leiðast grilla sparka líða illa elska

‘be bored’ ‘barbeque’ ‘kick’ ‘feel bad’ ‘love’

SubjDat V SubjNom V ObjAcc SubjNom V ObjDat SubjDat V SubjNom V ObjAcc

The participants were thus primed lexically and semantically with five already existing Icelandic predicates, one corresponding to each figure used. These are listed in Table 4.1, where the first column gives the nonce verb, the second column states the Icelandic priming verb used in the introductory formula, the third gives the gloss for the Icelandic priming verb, and the last column specifies its case and argument structure construction. Observe that three of the nonce verbs are construed as emotion verbs: the first one flokast ‘be bored’, the fourth one tvíta ‘feel bad’ and the last one spofta ‘love’. Three of the Icelandic priming verbs are transitive: the one corresponding to the second nonce verb slobba ‘barbeque’, the one corresponding to the third one kísa ‘kick’ and the one corresponding to the last one spofta ‘love’. Two of the transitive verbs select for accusative objects, grilla ‘barbeque’ and elska ‘love’, and one for a dative object, sparka ‘kick’. Of the Icelandic priming verbs selecting for accusative objects, one is agentive, grilla ‘barbeque’, and the other is an emotion verb, elska ‘love’. A filler was inserted between each verb, a stimulus involving both nouns and adjectives, to prevent repetitive answers. Note also that the Icelandic priming verbs were only used in the infinitive form in the introductory formula, in order to secure that the participants were not primed to use the case and argument structure constructions that I was eliciting (cf. Barðdal 2000 for further information about the experiment).

 Productivity

Several of the participants made comments during the experiment showing that they believed that the object of the investigation was morphology, more specifically different conjugational classes of verbs. This is presumably a consequence of the fact that several wug-experiments have been carried out in Iceland – and they are thus well known – focusing either on the singular–plural distinction of nouns or the present–past tense distinction of verbs. Some of the participants also asked explicitly afterwards what the object of the investigation was, again confirming that the participants were generally not aware of the goal of the experiment. Hence, their assignment of case and argument structure constructions to the nonce verbs in the experiment was not in response to what they believed was expected of them. 4.2 The findings Out of forty participants, only three assigned the case and argument structure of the priming verbs to the nonce verbs in all five cases, i.e. two adults and one of the oldest children, or 7.5% of the participants. In contrast, eleven participants, or 27.5%, assigned Nom(-Acc) to all five nonce verbs, i.e. five adults and six children. The remaining participants provided mixed answers, with some of the participants correcting themselves from either dative to nominative subject or, more commonly, from nominative to dative subject. The corrections that the participants made are included as observations in the statistics in Tables 4.2–4.3, yielding a higher number of observations than forty for flokast ‘be bored’ and tvíta ‘feel bad’. These findings suggest that speakers assign case and argument structure constructions to unknown verbs on the basis of either a synonymous verb-specific construction or on the basis of a highly schematic Nom-Acc construction with little or no semantic content. The mixed answers suggest that one and the same speaker can even sway between these two case assignment mechanisms. Table 4.2  Case and argument structure assignment to nonce verbs Nom subj

flokast ‘be bored’ slobba ‘barbeque’ kísa ‘kick’ tvíta ‘feel bad’ spofta ‘love’

Dat subj

Acc obj

Dat obj

N

%

N

%

N

%

23 40 40 25 40

52 100 100 58 100

21 – – 18 –

48 – – 42 –

– 40 33 – 39

– 100 82.5 – 97.5

% – – 7 – 1

– – 17.5 – 2.5



Chapter 4.  Nonce verbs 

Consider now the frequencies for individual verbs, given in Table 4.2. The nonce verbs synonymous to Nom-Acc verbs are always assigned the Nom-Acc construction, with one exception: One of the children assigned the Nom-Dat construction to spofta ‘love’, instead of the Nom-Acc construction. There is, in fact, one verb in Icelandic with the meaning ‘love’ selecting for the Nom-Dat construction, unna. However, this verb belongs to a very formal register and it appears as unlikely that it is known to children at this age. This assignment of dative case to the object could of course be motivated by its animacy (cf. Section 3.1.2.2 above), or it could be an instance of structural priming, as this particular participant had assigned the Nom-Dat construction to the preceding transitive verb kísa ‘kick’. By contrast, the nonce verbs synonymous with Icelandic verbs selecting for either the Nom-Dat construction or the Dat-only construction were not always assigned the Nom-Dat and Dat-only constructions, respectively. Such case assignment is expected if speakers use the synonymous Icelandic verbs as model verbs for low-level analogical extensions. The Nom-Dat construction was only assigned to the nonce verb kísa ‘kick’ in 17.5% of the cases, or by only seven participants. In the remaining cases kísa ‘kick’ was assigned the Nom-Acc construction. This is unexpected because verbs of kicking are prototypical instances of caused-motion and an overwhelming majority of caused-motion verbs in Icelandic select for the Nom-Dat construction (cf. Section 3.1.2.2 above). There are, however, a few verbs of transfer which select for the Nom-Acc construction in Icelandic. Ten of those are listed in Section 3.1.2.1 (under subconstruction 41). It is a distant possibility that the thirty-three participants who assigned the Nom-Acc construction to kísa ‘kick’ associated the nonce verb with one of these Nom-Acc verbs of transfer, for instance, flytja ‘transport’, færa ‘move’ or hreyfa ‘move’, in spite of the fact that they were primed with sparka ‘kick’ which is a Nom-Dat verb. It is also possible that the case assignment of slobba ‘barbeque’, which was presented before kísa in the experiment, primed the participants to use the Nom-Acc construction instead of the Nom-Dat construction. Research on structural priming has shown that priming effects can be found after up to ten fillers (Bock and Griffin 2000). A third possibility is that kísa was assigned case and argument structure on the basis of a general schematic Nom-Acc construction, more or less devoid of lexical semantic content. At this point it is difficult to know whether these thirty-three participants assigned the Nom-Acc construction to kísa on: (a) the basis of an analogy with a different verb of transfer, like færa ‘move’, (b) the basis of a high-level schematic Nom-Acc construction, or (c) the basis of a recent activation of the Nom-Acc construction in the working memory of these speakers. It is even possible that different explanations apply for different speakers, so that one speaker may have used the Nom-Acc construction because s/he was primed to do so, while a different speaker may have assigned it on the basis of a high-level schematic Nom-Acc construction.

 Productivity

I leave this issue open for now, but I will return below to case and argument structure assignment on the basis of high-level schematic constructions. With regard to the Dat-only construction, less than half of the responses on flokast ‘be bored’ and tvíta ‘feel bad’, i.e. the nonce verbs presented as synonymous with dative subject verbs in Icelandic, involved dative case assignment to the subject. This is, however, considerably more than the 17.5% of the Nom-Dat assignment to kísa ‘kick’. The difference between the productivity of the Nom-Dat vs. the Dative subject construction in this experiment is unexpected as the type frequency of these two constructions is approximately the same in Icelandic (cf. Section 3.1.1). I return to this issue in Section 4.3 below. Reverting to the issue of the Dat-only construction, in the cases where flokast ‘be bored’ and tvíta ‘feel bad’ were assigned dative to the subject, it is reasonable to assume that the Icelandic priming verbs functioned as model verbs for the case and argument structure assignment of the nonce verb. This assumption is supported by the fact that dative subject verbs in Icelandic are much lower in type frequency than nominative subject verbs, and thus do not exist at a high schematic level (Barðdal 2004 and 2006a). As already mentioned in Chapter 3, the standard monolinguial Icelandic dictionary (Íslensk orðabók handa skólum og almenningi 1988) contains approximately 8,500 predicates (Kristín Bjarnadóttir p.c.), and dative subject verbs are only ca. 700 of these (cf. Table 3.3 above). Therefore, it is reasonable to assume that a low-level verb-specific construction has been activated and that the case and argument structure constructions assigned to flokast and tvíta originate in the verb-specific constructions of the priming verbs leiðast ‘be bored’ and líða illa ‘feel bad’. This conclusion is further supported by the fact that the last nonce verb spofta ‘love’, also construed as an emotive verb but primed with a nominative subject verb, was never assigned dative case to its subject. If flokast ‘be bored’ and tvíta ‘feel bad’ were assigned the Dative subject construction irrespective of their Icelandic priming verbs, but on the basis of an intermediate or high-level Dative subject construction, the subject of spofta ‘love’ should also have been assigned dative case on the basis of the same intermediate or high-level Dative subject construction. Recall that the other half of the responses on flokast and tvíta involved assignment of the nominative to the subject of these expected “dative subject” nonce verbs. Clearly for flokast ‘be bored’, the assignment of the nominative to the subject argument is not due to structural priming, as flokast was the first nonce verb presented in the experiment. However, the higher percentage of nominative case assignment to the subject argument of the latter verb tvíta (52% for flokast vs. 58% for tvíta) could be due to structural priming, as two Nom-Acc verbs are introduced in between flokast and tvíta in the experiment. The differences between the number of observations of nominative and dative on the subjects of flokast and tvíta are, however, not significant.



Chapter 4.  Nonce verbs 

It is also a fact that there are no nominative subject verbs in Icelandic with the meaning ‘be bored’ and ‘feel bad’, which suggests that the nominative subject may have been assigned on the basis of a high-level schematic construction. There are, of course, emotion verbs in Icelandic that select for nominative subjects, but however none close in meaning to ‘be bored’ and ‘feel bad’; thus no low-level verb-specific construction can have been a model for the assignment of the nominative here. It is also a possibility that the nominative assignment to the subjects of flokast and tvíta is based on an intermediate verb-class-specific construction of emotion verbs and not a high-level schematic construction with little or no lexical semantic content. There is, however, one argument for the assumption that a high-level schematic construction has been at work here, namely that the response time seemed noticeably lower for both the nonce verbs assigned the Nom-Acc construction, and for the group of eleven participants who categorically assigned Nom-Acc to all verbs. The experiment was not recorded so no real-time figures can be presented here to corroborate this claim. However, the fact remains that several of the participants who assigned dative subjects to flokast and tvíta took an unusually long time to think through their responses, and that some of the participants who assigned Nom(-Acc) to all the nonce verbs rattled off their responses in no time. It makes perfect sense, of course, that assignment of case and argument structure on the basis of a high-level schematic construction will take shorter time, as little or no lexical meaning is involved, whereas assignment of case and argument structure on the basis of a low-level verb-specific construction involves retrieval of a concrete form–meaning mapping and the assignment of a form on the basis of the meaning component. Therefore, this response-time argument, although both anecdotal and impressionistic, is still an argument for the assumption that at least the Nom(-Acc) construction was in some cases assigned to nonce verbs on the basis of a high-level schematic construction without recourse to much semantic content. Consider now the figures from Table 4.2, recasted in Table 4.3, with the responses from the children separated from those of the adults. There is one clear difference between these two groups of participants, namely that adults have dative subjects with flokast and tvíta in 52–60% of the cases, while the children do so in only 32– 36% of the cases. It is well known from the literature that low type frequency constructions which can be elicited from adults are not as easily elicited from children (Bybee and Slobin 1982, Bybee and Moder 1983), and the younger the children are, the more difficult the elicitation becomes. This is presumably a consequence of the fact that the younger children have not received enough input of the low type frequency construction in question to be able to use it “productively” (cf. Ragnarsdóttir, Simonsen and Plunkett 1999). This assumption is supported by the higher mean age for the children assigning dative case than for the ones assigning nominative case to the subject of flokast and tvíta, 9.47 vs. 8.03 years of age, respectively. The differ-

 Productivity

ence in age between the two groups of child participants is highly significant, it can occur by pure chance with a probability of less than thirteen in a thousand (Independent Samples T-test, p < 0.007, Mann-Whitney, p < 0.013). Observe, moreover, that the nonce verbs primed with nominative subject verbs in Icelandic were never assigned dative case to the subject. This means, for instance, that the subject of spofta ‘love’ is never assigned the Dative subject construction in spite of spofta ‘love’ being an emotion verb. Also, nonce verbs synonymous to accusative object verbs in Icelandic were, with the one exception discussed above, never assigned the Nom-Dat construction. This is not a coincidence. The statistical differences between the assignments of different case and argument structure constructions are highly significant. The differences in subject marking of all five verbs can occur by pure chance with a probability of less than one in a thousand (Pearson Chi-Square, p < 0.001; Cramer’s V = 0.565). The differences in object marking between the three transitive verbs can occur by pure chance with a probability of three or less in a thousand (Pearson Chi-Square, p < 0.003; Cramer’s V = 0.310). Both p-values are well below the general level of significance at 0.05, assumed within the social and behavioral sciences, and Cramer’s V is considerably high. If there were no correlation between the case marking of nonce verbs in this experiment and the case marking of their synonymous Icelandic priming verbs, such clear-cut statistics as these would not be expected. Instead, it would be expected that either all subjects were assigned nominative case and all objects were assigned accusative case, or that the distribution of case and argument structure constructions would be more evenly spread across the different nonce verbs. Table 4.3  Case and argument structure assignment to nonce verbs by adults and children Nom subj

Dat subj

Acc obj

Dat obj

N

%

N

%

N

%

N

%

9 20 20 10 20

40 100 100 48 100

13 – – 11 –

60 – – 52 –

– 20 17 – 20

– 100 85 – 100

– – 3 – –

– – 15 – –

14 20 20 15 20

64 100 100 68 100

8 – – 7 –

36 – – 32 –

– 20 16 – 19

– 100 80 – 95

– – 4 – 1

– – 20 – 5

Adults flokast ‘be bored’ slobba ‘barbeque’ kísa ‘kick’ tvíta ‘feel bad’ spofta ‘love’ Children flokast ‘be bored’ slobba ‘barbeque’ kísa ‘kick’ tvíta ‘feel bad’ spofta ‘love’



Chapter 4.  Nonce verbs 

According to the predictions of generative grammar and its structural vs. lexical case dichotomy (Zaenen, Maling and Thráinsson 1985, Yip, Maling and Jackendoff 1987, Holmberg 1986, Sigurðsson 1989, Allen 1995, Falk 1997, Jónsson 1997–98, 2003, Jónsson and Eythórsson 2005), only the Nom-Acc construction should be productive, and possibly the Dative subject construction, as dative case on subjects counts as “thematic” case, assigned on the basis of the meaning of the verb. However, the Nom-Dat construction should not be productive, as dative on objects counts as idiosyncratic case on the generative approach, non-productive by definition. Icelandic speakers’ assignment of the Nom-Acc construction to nonce verbs in this experiment is compatible with the generative approach. It is, of course, also compatible with the present cognitive-functional linguistic approach where case and argument structure is assigned either on the basis of a Nom-Acc verb-specific construction (to slobba ‘barbeque’ and spofta ‘love’) or on the basis of a high-level schematic construction with little or very general meaning (to flokast ‘be bored’, kísa ‘kick’ and tvíta ‘feel bad’). Icelandic speakers’ assignment of dative subjects to flokast ‘be bored’ and tvíta ‘feel bad’ is also compatible with the generative approach, given the assumption that this kind of case assignment is there regarded as being based on thematic roles. However, on the generative approach, it is unexpected that the speakers who assigned dative subjects to flokast ‘be bored’ and tvíta ‘feel bad’ would not assign a dative subject to spofta ‘love’, since spofta is also an emotion verb.2 If the assignment of the dative case to flokast and tvíta really were semantically based, it should also apply to spofta. This lack of dative case on the subject of spofta is, however, expected on the present approach where it is assumed that all three verbs are assigned case and argument structure on the basis of their low-level verb-specific synonymous priming constructions in Icelandic. Finally, the assignment of the Nom-Dat construction to kísa ‘kick’ is unexpected on the generative approach, but expected on the assumption that the verb-specific construction of the priming verb sparka ‘kick’ functions as a model for this case assignment. On several occasions in this chapter have I explicitly discussed the highest level of a high type frequency construction, like the Nom-Acc construction, as being devoid of semantic content. This may appear as contradicting the basic assumption within cognitive-functional CxG that a construction is a form–meaning correspondence. Observe, however, that the existence of a construction at a high schematic level does not preclude form–meaning mappings at a lower level. On the contrary, the existence of a construction at a high schematic level is based on low-level item-specific form–meaning mappings. Therefore, the highest schematic level of a construction only represents an abstraction across all the low-level verbspecific constructions, and does thus not, in and of itself, contradict the fundamental principle of cognitive-functional CxG that a construction is a form–function correspondence.

 Productivity

I remind the reader of the discussion in Section 2.5 where the principles of the lexicality–schematicity continuum are laid out. There I stress the point that the highest schematic level at which each construction exists in the minds of real speakers is not a model-theoretical issue but an empirical psycholinguistic issue that has to be established for each construction in each language. In other words, even though linguists can always abstract at higher and higher meta-levels, it is not a given that real speakers abstract at such high levels for all constructions. I also argue in Section 2.5 that the highest level at which a construction can be assumed to exist is a function of its type frequency and its semantics, in that low type frequency constructions only exist as verb-specific constructions, while very general high type frequency constructions may be assumed to exist at such a schematic level that their meaning at this high level becomes increasingly bleached. This, however, is strictly speaking a consequence of how constructions are organized into categories, and hence a consequence of our cognitive abilities to categorize, and does not invalidate the assumption within cognitive-functional CxG that a construction is a form–function correspondence. In fact, the present experiment suggests that only the Nom-Acc construction in Icelandic exists at a schematic level in the minds of speakers, a level which is so high on the lexicality–schematicity continuum that its meaning component at that level has become fairly vacuous. Only by assuming increased semantic vacuity at this highest level is it expected that the Nom-Acc construction be extended above its verb-specific constructions, as evident from the fact that nonce verbs not primed with the Nom-Acc construction were indeed assigned Nom-Acc in this experiment. The other constructions under discussion here, the Nom-Dat construction and the Dative subject construction, have not been extended above their verb-specific levels (with the one exception discussed above for the Nom-Dat construction, an exception which can equally well be due to priming effects). This shows that the highest level of schematicity of the Nom-Dat and the Dative subject constructions in Icelandic is not as high as that of the Nom-Acc construction. This is represented with thick solid lines at the highest level of the Nom-Acc construction in Figure 3.3 in Section 3.1.2.1 above, as opposed to thin solid lines at the highest level of the Nom-Dat construction in Figure 3.4 in Section 3.1.2.2, while the highest level of the Dative subject construction in Figure 4.6 in Section 4.3 below is represented with dashed lines. In conclusion, the fact that the highest level of schematicity of the Nom-Acc construction is more or less semantically vacuous does not contradict the fundamental assumption within cognitive-functional CxG that a construction is a formfunction correspondence. This is so because a high-level semantically-vacuous schematic construction is an abstraction in the minds of speakers across all the lower-level concrete lexically-filled item-specific constructions, and as such it does not exist independently of them. In other words, a high-level semantically-vacuous



Chapter 4.  Nonce verbs 

schematic construction is not a different construction than its concrete lexicallyfilled item-specific instances, it is only a generalization across its low-level instances and thus a representation, while the low-level instances are the real form–function correspondences here. However, the highest schematic level is of course not devoid of relational content. I have elsewhere in this book referred to it as grammatical semantic content as opposed to lexical semantic content. To summarize the content of this section, I have shown that the case and argument structure construction highest in type frequency in Icelandic, the Nom-Acc construction, is the only construction extended above its verb-specific constructions, suggesting that it exists at a highly schematic level in the minds of Icelandic speakers. This assumption is supported by the impression of lower response time for the participants who categorically assigned the Nom-Acc construction to all the nonce verbs in the experiment. Hence, Icelandic speakers assign case and argument structure constructions to unknown nonce verbs on the basis of: i. synonymous verb-specific constructions in Icelandic ii. high-level schematic constructions with little or no lexical semantic content The results of the experiment also contradict the predictions of generative grammar that the Nom-Acc construction and the Dative subject construction should be productive, as the Dative subject construction is not generally extended to nonce verbs construed as emotion verbs. Generative grammar also wrongly predicts that the Nom-Dat construction should be unproductive. Before finalizing this chapter, let us compare the Dative subject construction with the Nom-Dat construction, as these two constructions are considered to have approximately the same type frequency in Modern Icelandic. 4.3 The semantics of the Dative subject construction Recall from Section 3.1.1 above that the type frequency of the Dative subject construction is approximately 700 in Icelandic, which is very similar to the type frequency of the Nom-Dat construction, estimated as 750 by Maling (2002:  31). Therefore, if the type frequency of a construction were the only factor influencing its extensibility or productivity, as argued by for instance Bybee (1985: 132–34), Goldberg (1995: 134–36) and Michaelis and Ruppenhofer (2001: 93), these two constructions should be equally productive in Icelandic. However, the investigation in Chapter 3 of borrowed verbs in Icelandic only confirms the productivity of the Nom-Dat construction, not of the Dative subject construction, while the present psycholinguistic experiment yields a much higher degree of productivity for the Dative subject construction than for the Nom-Dat construction.

 Productivity

Observe that the figures discussed in Section 3.1.1 above are based on dictionary frequencies, i.e. counts of lexical items in Icelandic irrespective of their frequency of occurrence. As mentioned above, these counts yielded approximately 700 dative subject predicates and 750 dative object predicates in Icelandic. In contrast, the figures from my 2001 corpus of both written and spoken Icelandic are based on language use. The Nom-Dat predicates found in the 2001 corpus were 188, while it turns out that the dative subject predicates are only seventy-three. Of these seventy-three, twenty were passives of the Nom-Dat construction or the Ditransitive construction. This means that counts of type frequency based on language use may give different results than counts of type frequency based on dictionaries, and in this case the type frequency of the active Dative subject construction turns out to be 3.5 times lower than the type frequency of the Nom-Dat construction (188 vs. fifty-three). Obviously these differences in type frequency make contrasting predictions on the productivity of the Dative subject construction in comparison with the Nom-Dat construction, making it essential to distinguish between absolute type frequencies (dictionary frequencies) and relative type frequencies (corpusbased type frequencies) when discussing productivity. An analysis based on absolute type frequency only may predict a much higher degree of productivity than an analysis juxtaposing type frequencies found in real language use. Let us now compare the semantic range of the Dative subject construction, with the semantic range of the Nom-Dat construction, based on the occurrences in the 2001 corpus (see also Barðdal 2004 for a comparison of the semantics of the Dative subject construction in Icelandic with Faroese and German). Limiting the present analysis to the active Dative subject construction, the fifty-three exemplars of the Dative subject construction can be divided into the following eight narrowly-circumscribed semantic classes of verbs (for a comprehensive list of all the dative subject predicates occurring in the 2001 corpus, see Appendix A): 1. Verbs denoting (changes in) bodily states: renna kalt vatn á milli skinns og hörunds ‘be terrified’, verða kalt ‘be cold’, vera ómótt ‘feel nauseated’ … 2. Verbs of cognition: detta í hug ‘get an idea’, fljúga í hug ‘get an idea’, koma saman um ‘agree on’, skiljast ‘understand, gather’, vera hugleikið ‘be on sby’s mind’, verða hugsað ‘have a thought’, verða hugsað til ‘remember, think about sby’, verða ljóst ‘realize’, verða kunn ‘become to know’, vera ókunnur ‘not know’ … 3. Verbs of emotion: finnast ‘feel’, hitna í hamsi ‘become angry’, langa ‘want’, leiðast ‘be bored’, líða ‘feel’, líka ‘like’, lítast á ‘like’, nægja ‘suffice’, renna kalt vatn á milli skinns og hörunds ‘be terrified’, reynast erfitt/auðvelt ‘turn out to be difficult/easy for sby’, standa stuggur af ‘be scared of sth’, vera fjarri ‘feel remote’, vera heitt í hamsi ‘be angry’, vera leitt ‘be sad’, vera nóg ‘have enough’, vera nóg



Chapter 4.  Nonce verbs 

4.

5. 6. 7. 8.

boðið ‘be fed up’, verða rótt ‘become calm’, vera sama ‘not mind’, vera þægð í ‘be content with’, þykja ‘feel, think, seem’ … Verbs of perception: berast til eyrna ‘hear’, bregða fyrir ‘be briefly seen’, heyrast ‘hear, gather’, sýnast ‘seem, appear’, verða bilt við ‘be startled’, verða starsýnt á ‘stare’, virðast ‘seem, appear’ … Verbs of obligation: bera ‘oblige’, mega ‘may, have to’ … Verbs of ontological existence: festa saman ‘be fastened together’, koma við ‘be of sby’s business’, sæta ‘be the reason for’, vera bágt til bjargar ‘be difficult to save’ … Verbs of success and/or performance: ganga vel ‘be successful’, takast ‘manage, succeed’ … Miscellaneous verbs: fjölga ‘increase’, fylgja ‘accompany’, halla ‘decline’, ljósta ‘strike’ verða á mistök ‘make a mistake’, vera fært ‘be capable’ …

Notice that the third emotion verb in the list above, langa ‘want’, is an accusative subject verb in Icelandic, occurring, however, with a dative subject in the spoken part of the 2001 corpus, which is the reason why it is included in the current analysis. For a further discussion of this type of dative substitution, see Section 6.3 below.

Dative-(X)

COGNITION/EMOTION

[Existence] [Success] [Miscellaneous]

[Verb] ...

[Bodily states] [Cognition] [Emotion] [Perception] [Obligation]

[Verb] ...

Figure 4.6  The Dative subject construction in Icelandic

 Productivity

Consider the structure of the Dative subject construction, given in Figure 4.6. Observe that the vertical axis represents the lexicality–schematicity continuum. The lowest level at the bottom of the hierarchy is the semantically and lexically most concrete level, consisting of the fifty-three verb-specific constructions (labeled [Verb]). These fifty-three verb-specific constructions make up the next level above, which contains only eight verb-class-specific constructions. Five of these are subconstructions of the basic-event-type categories cognition/emotion while the remaining three do not make up a category of their own. This analysis of the Dative subject construction shows that its semantic range is considerably narrower than the semantic range of the Nom-Dat construction, concomitant with a considerably lower type frequency of the Dative subject construction than the Nom-Dat construction, as only fifty-three predicates instantiate the Dative subject construction while 188 predicates instantiate the Nom-Dat construction. This comparison of the two constructions is based on their occurrences in the same corpus of Icelandic texts and shows that the Dative subject construction is used proportionally less, more specifically 3.5 times less, by Icelandic speakers than the Nom-Dat construction (see Section 3.1.1 for a discussion of the representativity of the corpus) The semantic scope of the Dative subject construction is, in other words, considerably narrower than the semantic scope of the Nom-Dat construction, as the former construction subsumes eight narrowly circumscribed verb classes while the latter subsumes thirty-three such classes. This predicts that the Dative subject construction should be markedly less productive than the Nom-Dat construction, as its type frequency, semantic range, and thus its productivity domain is markedly smaller. This prediction is corroborated by the fact that none of the borrowed verbs in Section 3.2.1 above were assigned the Dative subject construction by Icelandic speakers while 36% were assigned the Nom-Dat construction. Observe also that only two of these 107 recently borrowed verbs, i.e. fíla ‘like’ and digga ‘dig, like’, have the right semantics to be assigned the Dative subject construction. They have, however, become conventionalized with the Nom-Acc construction in Icelandic, presumably either because they were borrowed into Icelandic together with their English verbspecific constructions, which are formally equivalent to the Icelandic Nom-Acc construction, or because their direct translational equivalents in Icelandic finna ‘feel’ and grafa ‘dig’ are Nom-Acc verbs. Observe that the meaning ‘like’ of fíla is not found with the English source verb feel, but is a derived meaning in Icelandic. How come, then, that the Dative subject construction shows such a high degree of productivity in the present nonce verb experiment? This, I believe, is a direct consequence of the fact that the experiment was designed in part to elicit the relevant Dative subject construction through the use of dative subject priming verbs. As such this experiment confirms the findings of Chapter 3 that the



Chapter 4.  Nonce verbs 

assignment of case and argument structure constructions to non-existing or nonconventionalized verbs is clearly based on low-level analogical extensions. For predicates with a meaning within the semantic range of the Dative subject construction, the chances that Icelandic speakers will assign the Dative subject construction to them are only 42–48%, as the Dative subject construction occupies the same semantic space as the Nominative subject construction in Icelandic. Hence, it is also expected, and at least equally likely, that Icelandic speakers assign the Nominative subject construction to verbs from this semantic field. For a further discussion of the productivity of the Dative subject construction in the history of Icelandic, see Section 6.3 below. The facts presented in this section of the type frequency and semantics of the Dative subject construction also show that the Dative subject construction cannot exist as a schematic construction in the minds of Icelandic speakers but only as a verb-specific or a verb-subclass-specific construction. This is compatible with the fact that only flokast ‘be bored’ and tvíta ‘feel bad’ were assigned dative subjects in the nonce verb experiment, while the Nom-Acc construction was the only construction extended above its verb-specific level to the other nonce verbs irrespective of their meaning. This experiment thus shows that the existence of the NomAcc construction at such a high schematic level, with little or no lexical meaning attached to it, further promotes its productivity above the productivity predicted by the inverse correlation between its type frequency and semantic coherence, as this highly schematic semantically-vacuous level gives the Nom-Acc construction a status similar to default status, which means that it can be freely assigned to new verbs irrespective of their meaning. That is, as the highest level of the Nom-Acc construction is so vacuous semantically, the Nom-Acc can be assigned to new verbs across all semantic fields. Therefore, on the present cognitive-functional CxG assumption that constructions are organized on a lexicality–schematicity continuum, default status of a construction can easily be modeled as a high schematic level of a high type frequency construction. This means, in other words, that default status effects are, on the present approach, a derivate of schematicity. Finally, this experiment has also shown that controlled elicitation tasks do not necessarily yield the same results on productivity as non-controlled methods such as investigating the productivity of specific constructions through material borrowed into a language. Obviously, almost all constructions can be extended in elicitation tasks where the participants are primed with item-specific constructions which are easily reproduced through the process of low-level analogical extensions, irrespective of whether such constructions are generally extended to new items in the language or not. In this particular case, the data on the Dative subject construction have shown that it can be extended to non-conventionalized verbs, given that Icelandic speakers are correctly primed, even though the Dative subject

 Productivity

construction does not show up with verbs borrowed into Icelandic (see, however, Section 6.3 below on the Dative subject construction attracting verbs from the Accusative subject construction). Whether or not elicitation tasks in general will yield the same results on productivity as non-controlled methods is of course entirely dependent on the linguistic situation in each language and the construction under elicitation. In sum, in this section I have shown (a) that the type frequency of the Dative subject construction is considerably lower than that of the Nom-Dat construction, and (b) that the semantic range of the former is also considerably narrower than the semantic range of the latter. As a consequence, the productivity domain of the Dative subject construction is also appreciably smaller than that of the Nom-Dat construction. This is unexpected as the absolute type frequency of the two constructions is very similar. However, their relative type frequency is not, as is evident from the fact that the Dative subject construction is much lower in type frequency in language use than what is expected by its absolute type frequency. This distinction between absolute and relative type frequency must therefore be borne in mind when establishing the type frequency of a construction and making predictions about its productivity. I have also shown that the Dative subject construction does not exist as a high-level semantically-vacuous schematic construction, but only as a low-level verb-specific or a verb-subclass-specific construction, as opposed to the Nom-Acc construction which clearly exists at a schematic semantically-vacuous level. This accounts for the default status effects of the Nom-Acc construction, evident from the fact that the Nom-Acc construction can be extended across all semantic classes of verbs. Finally, elicitation tasks need not yield the same results on productivity as uncontrolled methods, as elicitation tasks prime their participants to extend low-level verb-specific constructions which may or may not be productive in language use. 4.4 Summary In this chapter I have reported on a psycholinguistic experiment of nonce verbs and their case and argument structure assignment by Icelandic speakers. The experiment involved five different nonce verbs, two construed as agentive and three as emotion verbs. Three verbs were presented as synonymous with transitive verbs in Icelandic, two Nom-Acc verbs and one Nom-Dat verb, while two were presented as synonymous with dative subject verbs. The experiment was carried out with forty randomly selected monolingual Icelandic speakers, twenty adults and twenty children. The results of the experiment show that speakers assigned the Nom-Dat construction to the nonce verb presented as synonymous with an



Chapter 4.  Nonce verbs 

Icelandic Nom-Dat verb in 17.5% of the cases. They assigned the Dative subject construction in 42–48% of the cases, always to the two nonce verbs presented as synonymous with dative subject verbs in Icelandic. Finally, they always assigned the Nom-Acc construction to the two nonce verbs presented as synonymous with existing Nom-Acc verbs in Icelandic, except in one case where one of the nonce verbs was assigned the Nom-Dat construction. In all the cases, except the one just mentioned, where the case and argument structure assignment deviated from the case and argument structure assignment of the priming verbs, all five nonce verbs were assigned the Nom(-Acc) construction. The Dative subject construction, moreover, was never assigned to spofta ‘love’, the only nonce verb construed as an emotion verb, but presented as synonymous with a Nom-Acc verb. These findings show that Icelandic speakers assign case and argument structure constructions to unknown verbs on the basis of (a) their synonymous low-level verb-specific constructions in Icelandic, or (b) a high-level schematic semantically-vacuous NomAcc construction. These findings thus confirm the conclusions from Chapter 3 that syntactic productivity is, at least in part, a consequence of low-level analogical extensions, and that high type frequency is therefore not a necessary prerequisite for productivity, contra arguments in the literature. These findings also contradict the predictions of generative grammar that the Nom-Dat construction should be unproductive and that the Dative subject construction should be unrestrictedly extended to emotion verbs. A comparison of the type frequency and the semantics of the Dative subject construction and the Nom-Dat construction reveals that the Dative subject construction is indeed much lower in type frequency in language use than expected by its absolute type frequency. Hence, the semantic range of the Dative subject construction is also much narrower than the semantic range of the Nom-Dat construction. This predicts that the Dative subject construction should be much less productive than the Nom-Dat construction, a prediction that is borne out by the findings of Chapter 3 where only the Nom-Dat construction is extended to borrowed verbs while the Dative subject construction is not found to be extended at all. In the present psycholinguistic experiment, however, the Dative subject construction shows a greater degree of productivity than the Nom-Dat construction, a consequence of the fact that the present experiment aims at eliciting the Dative subject construction, again corroborating the findings of Chapter 3 that what appears as syntactic productivity based on high-level schematic constructions is an aftermath of low-level analogical extensions. The analysis of the semantics of the Dative subject construction also confirms that it only exists as a low-level verbspecific or a verb-subclass specific construction while the Nom-Acc construction clearly exists as a high-level schematic semantically-vacuous construction, at best exhibiting relational meaning. This highest level of the Nom-Acc construction

 Productivity

accounts for the default status effects found with the Nom-Acc construction, as it is the only construction in the experiment that can be extended to all the nonce verbs, irrespective of their meaning. On the present cognitive-functional CxG approach, default status effects are therefore clearly a derivative of schematicity. In this chapter I have shown that both low-level verb-specific constructions and high-level schematic constructions can be the source of productivity. In the next chapter, I report on a questionnaire survey of the acceptability of new verbs of communication in Icelandic which shows how low-level verb-specific constructions compete with verb-subclass-specific constructions as models for productivity. Notes 1 I tried to include five-year olds in the experiment but they refused to cooperate, and so did some six-year olds. The question arises as to whether they were too young to handle the experimental situation, or whether they were too young to know what to do with an unknown verb. I am inclined to think the latter. 2 See Jónsson (2003) for a different analysis of emotion verbs in Icelandic, where it is argued that verbs like elska ‘love’ are semantically different from líka ‘like’. This analysis is meant to capture the differences in subject case marking between the two verbs. However, since it is the difference in case marking which motivates the analysis, circularity is unavoidable.

chapter 5

New verbs of communication A questionnaire

New verbs of instrument of communication have been subject to some interest in the cognitive-functional linguistic literature (cf. Goldberg 1995, Croft, Barðdal, Hollmann Nielsen, Sotirova and Taoka 2001, Barðdal 2003). These are verbs like fax, email, text, etc., i.e. verbs denoting events of communication where the verb has often developed from the noun for the instrument or the technological product. These three verbs have emerged during the last two decades or so, concomitant with the relevant technology, and their development in the language can thus easily be traced. As such they are prime candidates for exploring the nature of productivity. In this chapter I focus on three new verbs of instrument of communication in Icelandic, faxa ‘fax’, (e)meila ‘email’ and smsa ‘text’. The verb faxa has developed from the noun fax ‘fax’, while emeila/í-meila or maila is confined to electronic messages, as snail-mail requires a different verb in Icelandic, senda ‘send’. The verb smsa ‘text’ has developed from the abbreviation sms which stands for Short Message System. This verb seems to have developed independently in, for instance, Icelandic and Swedish, and is found with two pronunciations in Icelandic, [esemesa] and [smesa], with subsequent orthographic differences. Icelandic is particularly well suited for this research, as in Iceland the use of cell phones and computers in the private sphere is one of the highest in the world; According to the National Statistical Institute of Iceland, 86% of the Icelandic population has access to the Internet, and 81% of the population use it regularly. This means that Internet use in Iceland is the highest in Europe. Also, 96% of the Icelandic population use cell phones. Hence, cell phone usage has become higher in Iceland than in Finland which has been at the top of the cell phone usage statistics in the world for years.1 In Section 5.1 I give an overview of three constructions that new verbs of instrument of communication may instantiate in Icelandic, i.e. the Transfer, the Caused-motion and the Ditransitive construction. The first two only differ as to the case marking of the object, the Transfer construction having accusative objects while the Caused-motion construction has dative objects. Also, traditionally, it is the Transfer construction and not the Caused-motion construction which alternates with the Ditransitive, not only in Icelandic, but in Germanic in general. Section 5.2 presents the questionnaire survey which was carried out with thirty-two

 Productivity

adolescents and adults in Iceland, age 14–85. Section 5.3 presents the findings which show that all three new verbs of instrument of communication are accepted in all three constructions, although to a varying degree. The Transfer construction is the most accepted construction with all three verbs, while younger speakers are more acceptant of the Caused-motion construction than older speakers. In Section 5.4 I discuss the competing factors underlying the productivity of the Transfer and the Caused-motion constructions, arguing that the extensibility of the Transfer construction is based on analogy with the verb senda ‘senda’, while it is in fact the Caused-motion construction that is the productive construction in Icelandic with verbs of caused motion. In Section 5.5 I summarize the findings and conclusions of this chapter. 5.1 The Transfer, the Caused-motion and the Ditransitive constructions New verbs of instrument of communication occur in at least three constructions in Icelandic, the Transfer constructiion, the Caused-motion construction and the Ditransitive construction. In 5.1.1 I describe the differences between the Transfer and the Caused-motion construction in Icelandic, and in 5.1.2 I give an outline of the semantics of the Ditransitive construction and its restrictions in Icelandic. 5.1.1 The Transfer and the Caused-motion constructions The Transfer and the Caused-motion constructions have already been discussed in Chapter 3 as Nom-Acc subconstruction 41 (verbs of transfer) and Nom-Dat subconstruction 21 (verbs of caused motion). I will refer to these as the Transfer construction and the Caused-motion construction in the remainder of this book. As is also discussed in Chapter 3, there is a substantial difference in the type frequency of the two constructions, with fifty predicates instantiating the Causedmotion construction, while only ten instantiate the Transfer construction (in the present material). These are listed again in (1) below for the sake of convenience. (1) a Caused-motion construction (50): banda frá ‘wave away’, beina ‘direct’, benda ‘direct, point’, berja (hrömmunum) ‘thrash (one’s paws)’, blanda ‘mix’, bylta (sér) ‘turn around’, dreifa ‘spread’, feykja ‘blow’, forða ‘get sth away’, fresta ‘postpone’, fylgja ‘follow, accompany’, fyllast ‘be(come) filled with’, henda ‘throw’, hella ‘pour’, hleypa ‘let, allow (sby into sth)’, hliðra ‘shirk’, kasta ‘throw’, klappa ‘pat’, koma (sér) ‘get off, go away’, mjaka ‘budge’, mæta ‘run into sby’, renna ‘glide’, ríða ‘ride’, róa ‘row’, safna ‘gather’, skella (í sig) ‘slam (in



Chapter 5.  New verbs of communication 

oneself), eat’, skila ‘return’, skjóta ‘shoot’, skjóta upp (kollinum) ‘pop up’, skutla ‘give a ride’, slengja ‘throw’, sleppa ‘drop, release’, smeygja (sér) ‘slide’, smjúga (fingrum) ‘slip’, snúa ‘turn’, sparka ‘kick’, spyrna ‘kick’, stilla upp ‘take a position’, stinga ‘insert’, sveipa ‘wrap’, svæla (í sig) ‘puff (in oneself), eat’, troða ‘squeeze (one’s way)’, tylla (sér) ‘take a seat’, varpa ‘throw’, velta ‘roll’, vefja ‘wrap’, víkja ‘make way’, ýta ‘push’, þrengja ‘push’, þrýsta ‘squeeze, thrust’ ... (1) b. Transfer construction (10): bera ‘carry’, draga ‘draw, pull’, flytja ‘transport’, flæma burt ‘drive away’, færa ‘move’, hreyfa ‘move’, senda ‘send’, taka út ‘withdraw’, toga ‘pull’, (út)breiða ‘spread, unfold’ ... Each Icelandic predicate is conventionally associated with only one of the two constructions. The evidence from the list of recently borrowed verbs in Icelandic, discussed in Section 3.2.1 above, shows that it is in fact the Caused-motion construction which is being activated when new verbs of caused-motion enter Icelandic. In addition to that, there is language internal evidence supporting the productivity of the Caused-motion construction. There are, for instance, certain verbs in Icelandic which alternate between the Nom-Acc and Nom-Dat constructions. These are not necessarily motion verbs, but, as shown in the (b) examples below, when used to denote movement they select for the Caused-motion construction and not for the Transfer construction (cf. Barðdal 1993, 2001a: Ch. 6, Maling 2002: 59–66): (2) a. Hann smyr brauð. he smears bread.acc ‘He butters the bread.’ (2) b. Hann smyr smjörinu á brauðið. he smears butter-the.dat on bread-the ‘He spreads the butter on the bread.’ (3) a. Hann hleður vegginn. he piles wall-the.acc ‘He piles the wall.’ (3) b. Hann hleður múrsteinunum upp. he piles bricks-the.dat up ‘He piles the bricks.’ (4) a. Hann mokar sand. he shovels sand.acc ‘He shovels sand.’

 Productivity

(4) b. Hann mokar sandinum burt. he shovels sand-the.dat away ‘He shovels the sand away.’ (5) a. Hann barði hundinn. he hit dog-the.acc ‘He beat the dog.’ (5) b. Hann barði hrömmunum á hurðina. he hit paws-the.dat on door-the ‘He hit the door with his paws.’ Some of the accusative objects in the (a) examples could be classified as locatum or incremental themes. Others are ordinary affected objects. Observe that the dative objects in the (b) examples have all undergone a movement from one location to another. Therefore, the examples in (a) exemplify different subconstructions of the ordinary transitive construction, not necessarily conveying movement, while the sentences in (b) are examples of the Caused-motion construction, in which the subject referent has caused the object referent to move from one location to another. An argument in favor of analyzing the dative object as an indistinguishable part of the Caused-motion construction stems from the fact that when the locative/directional phrase is missing, all the (b) examples in (2–5) become somewhat ill-formed, while, as is also evident in the examples above, the same is not true for the corresponding (a) sentences with an accusative object: (6) a. ?Hann smyr smjörinu. he smears butter-the.dat (6) b. ?Hann hleður múrsteinunum. he piles bricks-the.dat (6) c. ?Hann mokar sandi. he shovels sand.dat (6) d. ?Hann barði hrömmunum. he hit paws-the.dat My explanation (Barðdal 1993) is that the sentences in (6) are ill-formed because they are incomplete; an obligatory part of the construction, the locative/directional phrase, is missing. In addition to the transitive verbs in (2–5) above, there are also numerous intransitive verbs in Icelandic, not necessarily motion verbs, which can be used to denote movement of an object, i.e. in the Caused-motion construction, and when



Chapter 5.  New verbs of communication 

used in this way they also occur with the Nom-Dat case frame and not with the Nom-Acc frame (also observed by Holland 1993 for Old Norse-Icelandic): (7) a. Hann kemur. he comes ‘He’ll come.’ (7) b. Hann kemur dótinu til þín. he comes stuff-the.dat to you ‘He’ll get the stuff to you.’ (8) a. Hann ruslaði út. he rubbished out ‘He made it look untidy.’ (8) b. Hann ruslaði þessu saman. he rubbished this.dat together ‘He got it together (in an untidy manner).’ (9) a. Barnið slefar. child-the drools ‘The child is drooling.’ (9) b. Hann slefaði þessu út úr sér. he drooled this.dat out of himself ‘He finally dragged it out (of himself).’ (10) a. Barnið sullar í vatninu. child-the splashes in water-the ‘The child is splashing in the water.’ (10) b. Hann sullaði matnum út fyrir diskinn. he splashed food-the.dat out of plate-the ‘He splashed the food out of the plate.’ All these examples are conventionalized expressions in Icelandic. However, examples of intransitive predicates, not conventionalized as occurring in the Causedmotion construction, can also be found in Icelandic with the moved object in the dative case: (11) Og sussaði þjóninum og stráknum í burtu. and hushed waiter-the.dat and boy-the.dat away ‘And hushed the waiter and the boy away.’ (from the novel Augu þín sáu mig by Sjón, p. 164)

 Productivity

There are, moreover, examples of motion verbs, which conventionally occur in the Transfer construction in Icelandic (as in 1b above), but which have begun to occur in the Caused-motion construction. I have registered the following examples: (12) a. Þú slærð þessu inn í tölvuna. you hit this.dat into computer-the ‘You type/log it into the computer.’ (email from staff at Landssími Íslands, 1999) (12) b. ... sagði Lína og bretti upp ermunum. said Pippi and rolled up sleeves-the.dat ‘... Pippi said and rolled up the sleeves.’ (from the novel Lína langsokkur [Pippi Longstocking], translated by Jakob Ó. Pétursson, p. 54) (12) c. Ég reytti af mér bröndurunum. I ripped off me jokes-the.dat ‘I ripped the jokes off.’ (overheard in 1992) The verbs slá ‘hit’, bretta ‘roll’ and reyta ‘rip’ all traditionally occur in the Nom-Acc construction and not in the Nom-Dat construction. The examples in (12), however, are attested examples from real language use. Hence, the occurrence of both intransitive and conventionalized transitive (Nom-Acc) verbs in the Caused-motion construction with a dative object clearly testifies to the productivity of the Caused-motion construction over the Transfer construction when translational and other types of motion are involved. Interestingly, however, new verbs of instrument of communication in Icelandic seem to be able to occur either in the Caused-motion construction or in the Transfer construction. The documented examples in (13–15), with (e)meila ‘email’, smsa ‘text’ and faxa ‘fax’, illustrate this: (13) a. Þú verður nú bara að meila þessu til mín ... you must now just to mail this.dat to me ‘You just have to email this to me ...’ (kiss.bloggar.is/blogg/127054) (13) b.

Má ég ekki bara meila þetta til þín? can I not just mail this.acc to you ‘Can’t I just mail this to you?’ (www.ismal.hi.is/Malfregnir_20_Gudrun_Kvaran_Malfar_i_ stjornsyslu.pdf)

(14) a.

Hann smessaði þessari til mín ... he texted this.dat to me ‘He texted this one to me ...’ (blog.central.is/g/gummiste/index.php?m=2&y=2007)



Chapter 5.  New verbs of communication 

(14) b.

Heiðmar smessaði þessa hringhendu til mín í morgun ... Heiðmar texted this.acc circumstanza.acc to me in morning ‘Heiðmar texted this circumstanza to me this morning ...’ (stikill.blogcentral.is/g/gummiste/index.php?m=3&y=2007)

(15) a. ... auk þess sem ég faxa þessu líka. as well as I fax this.dat too ‘... I’ll fax this too as well.’ (blog.central.is/valsson) (15) b. ... og þá bað hún mig að faxa þetta allt aftur ... and then asked she me to fax this.acc all again ‘... and then she asked me to fax the whole thing again ...’ (heidagella.blogspot.com/ 2007/10/rfl-rfl-rfl-og-sm-djamm.html) The fact that new verbs of instrument of communication can occur in the Causedmotion construction is unexpected since they are semantically closest to verbs of sending, which traditionally occur in the Transfer construction, and alternatively in the Ditransitive construction. Verbs of sending, however, are excluded from the Caused-motion construction in Icelandic by convention: (16) a. Ég sendi þetta til þín. I send this.acc to you ‘I’ll send it to you.’

Transfer construction

(16) b. Ég sendi þér þetta. I send you.dat this.acc ‘I’ll send it to you.’

Ditransitive construction

(16) c. *Ég sendi þessu til þín. I send this.dat to you

Caused-motion construction

This pattern is the original pattern found for verbs of sending in all the Germanic languages. The examples in (13a), (14a) and (15a), however, illustrate that new verbs of instrument of communication are accepted and used in the Caused-motion construction as well. Even though this is unexpected given the conventionalized distribution of verbs of sending across argument structure constructions, i.e. in the Transfer and the Ditransitive construction, this is in fact expected given my suggestion above that the Caused-motion construction is the productive construction in Icelandic for translational motion. To summarize, the examples in (13b), (14b) and (15b) show that the Transfer construction with an accusative object is productive, not generally with verbs of movement, but with novel verbs of instrument of communication, while the remaining data presented in this section and in 3.2.1 above suggest that the Causedmotion construction is generally a productive construction in Icelandic with verbs of movement, including new verbs of instrument of communication.

 Productivity

5.1.2 The Ditransitive construction Recent research on the Ditransitive construction has revealed at least seventeen semantic subclasses of Dat-Acc ditransitive predicates in Icelandic (cf. Barðdal 2007, Barðdal, Kristoffersen and Sveen 2007), as opposed to only nine in English (e.g. Goldberg 1995: 126): 1. Verbs denoting (prolonged) possession/owning: eiga sér e‑ð ‘have sth’ ... 2. Verbs inherently denoting giving or delivering: afhenda e‑m e‑ð ‘deliver sth to sby’, fá e‑m e‑ð ‘give sth to sby’ ... 3. Verbs of lending: lána e‑m e‑ð ‘lend sby sth’, leigja e‑m e‑ð ‘hire sth out to sby’, ljá e‑m e‑ð ‘lend sby sth’ ... 4. Verbs of paying: borga e‑m e‑ð ‘pay sby for sth’, bæta e‑m e‑ð ‘compensate sby for sth’, launa e‑m e‑ð ‘reward sby with sth’ ... 5. Verbs of sending: senda e-m e-ð ‘send sby sth’ ... 6. Verbs of bringing: bera e‑m e-ð ‘bring sby sth’, færa e‑m e‑ð ‘bring sby sth’ ... 7. Verbs of future transfer: bjóða e‑m e‑ð ‘offer sth to sby’, skulda e‑m e‑ð ‘owe sby sth’ ... 8. Verbs denoting transfer along a path: brjóta sér leið ‘break oneself a passage’, opna e‑m leið/dyr ‘open up a passage/door for sby’ ... 9. Verbs of enabling: auðvelda e‑m e‑ð ‘facilitate sth for sby’, heimila e‑m e‑ð ‘authorize sth to sby’, létta e‑m e‑ð ‘make sth easier for sby’ ... 10. Verbs of communicated message: boða e‑m trú ‘preach a belief to sby’, kenna e‑m e‑ð ‘teach sby sth’ ... 11. Verbs of instrument of communicated message: (e)meila e‑m e‑ð ‘(e)mail sby sth’, faxa e‑m e‑ð ‘fax sby sth’, smsa e‑m e‑ð ‘text sby sth’ ... 12. Verbs of creation: byggja sér e‑ð ‘build oneself sth’, elda sér e‑ð ‘cook sth for oneself¨, hekla sér e‑ð ‘crochet sth for oneself ’ ... 13. Verbs of obtaining: ávinna sér e‑ð ‘acquire sth for oneself ’, finna sér e‑ð ‘find sth for oneself ’, kaupa sér e‑ð ‘buy oneself sth’ ... 14. Verbs of utilizing: (hag)nýta sér e‑ð ‘make most of/use of sth’, nota sér eitthvað ‘use sth for oneself ’, notfæra sér e‑ð ‘take advantage of sth’ ... 15. Verbs of hindrance: banna e‑m e‑ð ‘forbid sby to do sth’, gera e‑m e‑ð ‘do sth (bad) to sby’, torvelda e‑m e‑ð ‘make sth difficult for sby’ ... 16. Verbs of constraining: setja sér e‑ð ‘determine to do sth’, setja e‑m úrslitakosti ‘give sby an ultimatum’ ... 17. Verbs denoting mental activity: fyrirgefa e‑m e‑ð ‘forgive sby sth’, gera sér e‑ð ljóst ‘realize sth’, hugsa sér e‑ð ‘think of sth’ ...



Chapter 5.  New verbs of communication 

There are, however, restrictions on verbs of obtaining and creation, for instance, such that they can only be used reflexively in Modern Icelandic (Holmberg and Platzack 1995: 190, 202, Maling 2002: 49–54, Barðdal 2007: 16–18): (17) a. Ég ætla að afhýða mér/*þér appelsínu. I intend to peal me/you.dat orange.acc ‘I intend to peal me/*you an orange.’ (17) b. Bakaðu *mér/þér köku! bake-you me.dat/yourself.dat cake.acc ‘Bake *me/yourself a cake!’ I have also pointed out elsewhere (Barðdal 2001a: 190, 2007: 16) that verbs of ballistic motion cannot be used ditransitively in Icelandic: (18) *Ég henti þér boltann/boltanum. I threw you.dat ball-the.acc/dat Maling (2002: 44) also observes that there is conflicting evidence for the occurrence of verbs of instrument of communication in the Ditransitive construction. She claims that faxa cannot be used in the Ditransitive construction in Icelandic, while (e)meila can: (19) a. %Geturðu faxað mér samninginn? can-you fax me.dat contract-the.acc

(Maling 2002: 44)

(19) b. Hún hefur e-meilað mér myndina. she has e-mailed me.dat picture-the.acc ‘She e-mailed me the picture.’

(Maling 2002: 44)

I have, however, encountered several examples of the ditransitive use of faxa, smsa and (e)meila: (20) a.

HEHE ertu að spá í að láta faxa þér stuðarana hehe are-you to think in to let fax you.dat bumpers-the.acc Vestur ... west ‘Hehe are you thinking about having the bumpers faxed to ou to the west coast ...’ (www.stjarna.is/forum/viewtopic.php?p=37458& sid=5f 36751839d314593dcea10859e79e8e)

(20) b. ... átti ég að fara að smsa honum söguþráðinn ... should I to go to text him.dat storythread-the.acc ‘... was I supposed to text him the thread of the story ...’

 Productivity

(20) c. ... e-maila þér uppskriftina á morgun ... e-mail you.dat recipe-the.acc in morning ‘... (I’ll) email you the recipe tomorrow ...’ (http://skralli.blog.is/blog/skralli/entry/338012) These facts show that further research is needed on the availability of the Ditransitive construction to new verbs of instrument of communication in Icelandic. This is the enterprise described in the next sections. I start with a questionnaire survey designed to throw light on the acceptability of these verbs in the Transfer, the Caused-motion and the Ditransitive constructions. 5.2 The questionnaire survey The questionnaire survey was conducted in Iceland with thirty-two randomly selected participants, at the age range 14–85. They were presented with the questionnaire (cf. Appendix C for an English version) which contained nine example sentences, three examples with each verb in one of the three constructions, the Transfer, the Caused-motion and the Ditransitive construction. As a control, the questionnaire also contained examples of senda ‘send’ in all three constructions. The sentences were randomly ordered and the participants were asked to rate each sentence on a 7-point acceptability scale, with 1 standing for “This is impossible in Icelandic”, 4 standing for “I could say this but normally I wouldn’t”, and 7 standing for “This is exactly how I would say it”. Hence, judgments 1–3 represent unacceptability while judgments 4–7 represent acceptability or felicitousness. This entails that the acceptability level is set at 4. 5.3 The findings Two participants rejected all three examples of (e)meila and two others rejected all three examples of smsa. In these four cases it is reasonable to assume that the participants were not rejecting the relevant verb in each and every one of the three constructions but rather that they were simply rejecting (e)meila and smsa as possible verbs in Icelandic. These twelve judgments are excluded from the statistics below. Table 5.1 gives the variance and mean score for all three verbs in all three constructions: a) gives the Caused-motion construction, b) the Transfer construction, and c) the Ditransitive construction. As a control, the same statistics is given for the conventionalized verb senda ‘send’, which can only occur in the Transfer and the Ditransitive constructions in Icelandic. This distribution is, in fact, clear from



Chapter 5.  New verbs of communication 

Table 5.1  Variance and mean score

(E)MEILA a. Ég (e)meila þessu til þín b. Ég (e)meila þetta til þín c. Ég (e)meila þér þetta

Smsa a. Ég smsa þessu til þín b. Ég smsa þetta til þín c. Ég smsa þér þetta

faxa a. Ég faxa þessu til þín b. Ég faxa þetta til þín c. Ég faxa þér þetta

Senda (for comparison) a. Ég sendi þessu til þín b. Ég sendi þetta til þín c. Ég sendi þér þetta

1

2

3

4

5

6

7

Mean

5 3 9

1 1 4

4 3 2

4 3 1

3 6 2

6 7 9

7 7 2

4.50 4.90 3.62

1

2

3

4

5

6

7

Mean

5 1 5

2 2 3

5 1 2

5 3 1

2 4 2

8 12 7

2 6 9

4.00 5.31 4.69

1

2

3

4

5

6

7

Mean

8 1 7

2 1 3

5 0 2

3 2 1

3 3 3

6 10 9

5 15 5

3.91 5.97 4.23

1

2

3

4

5

6

7

Mean

26

1

4 2

3 1

5 5

9 8

1 13 18

1.47 5.88 6.34

4

5

6

7

Mean

1

2

3

Table 5.1, as the mean score for senda in the Caused-motion construction is far below acceptability, i.e. 1.47 with the general acceptability level for this survey being at judgment 4. In contrast, the mean score of senda in the Transfer and the Ditransitive are 5.88 and 6.34 respectively, i.e. well above the acceptability level. This is also what one would expect given that senda is an existing conventionalized verb in Icelandic. As evident from the table, one of the participants judged senda as acceptable in the Caused-motion construction. As all the participants were native speakers of Icelandic, it seems reasonable to assume that this participant may have reversed the directionality of the judgment scale by mistake. Notice also that the construction which always obtains the highest mean score for all the three new verbs is the Transfer construction with an accusative object. These mean scores are by far the highest mean scores obtained in this study. For the other constructions, however, the mean score is either above acceptability or just below it. None of the mean scores are as low as the mean score for senda in the

 Productivity

Caused-motion construction with a dative object, which we know is ungrammatical in Icelandic. This suggests that all three verbs are accepted in all three constructions, although their use may not be fully conventionalized yet. In fact, it is only the distribution of judgments for faxa across constructions which is statistically significant (Pearson Chi-Square, p < 0.044; Cramer’s V = 0.338), while the distribution of judgments for (e)meila and smsa are not. To compare, the distribution of judgments for senda across all three constructions is also clearly statistically significant (Pearson Chi-Square, p < 0.001; Cramer’s V = 0.711). It thus seems reasonable to assume that the differences in the distribution of judgments across constructions for (e)meila and smsa are due to coincidence. This is in fact confirmed by the distribution of judgments across the same constructions for the two verbs, (e)meila and smsa, in an earlier study which was confined to (e)meila and smsa (cf. Barðdal 2003). The statistics for that study are given in Table 5.2, which shows that the mean score for (e)meila and smsa is always above the acceptability level of mean score 4. The variation in judgments between (e)meila and smsa in the 2003 study and the current study seems to indicate that the occurrence of precisely these two verbs across the three constructions is not fully conventionalized in Icelandic. This is confirmed by the fact that the differences between the statistics from 2003 and now are generally not statistically significant. The verb faxa, on the other hand, came into Icelandic long before (e)meila and smsa did, so one would expect that verb to have reached a higher degree of conventionalization than the other two, resulting in more clear-cut statistics for faxa and a higher level of statistical significance, which is also the case. Table 5.2  Variance and mean score (2003 study)

(E)MEILA a. Ég (e)meila þessu til þín b. Ég (e)meila þetta til þín c. Ég (e)meila þér þetta

Smsa a. Ég smsa þessu til þín b. Ég smsa þetta til þín c. Ég smsa þér þetta

1

2

3

4

5

6

7

Mean

5 4 1

1

2 1 3

2

2

3 4

4 4 3

7 5 6

4.63 4.53 4.80

1

2

3

4

5

6

7

Mean

3 2

1 2 2

5 1 2

3 1 1

2 1

3 3 3

6 5 6

4.94 4.67 4.76



Chapter 5.  New verbs of communication 

To conclude, the Transfer construction is judged the most acceptable with all three verbs, which is what one might expect given the constructional pattern found with the verb senda in Icelandic, i.e. in the Ditransitive and the Transfer constructions. It is surprising, however, given exactly this pattern of senda, that the three verbs are at all judged acceptable in the Caused-motion construction, but the mean score for the Caused-motion construction ranges from 3.91 to 4.50, which is either at the acceptability level or above it. I return to the productivity of the Causedmotion construction in the next section. It turns out, moreover, that there is a substantial difference in the mean age of the speakers who accept and reject the three new verbs of communication in the three relevant constructions, as is evident from Table 5.3. For the sake of clarity and convenience I have collapsed the three values for non-acceptability and the four values for acceptability into one value each. Observe that for the verb faxa there is little difference in the mean age of the participants who reject and accept faxa in the Transfer and the Ditransitive constructions. It is only for the occurrence of faxa in the Caused-motion construction that a difference emerges. Here the mean age for speakers who reject faxa is above 40 years, while the mean age for speakers who accept it is as low as 34 years. The same pattern is found for (e)meila: The mean age for the speakers who accept and reject (e)meila in the Transfer and the Ditransitive constructions is approximately the same, 33–35 years, while the mean age for the speakers who reject it in the Caused-motion construction is above 43 years. For smsa, the pattern is a little different, as the speakers who reject smsa in both the Caused-motion and the Ditransitive constructions are on average considerably older than the speakers who accept them, namely ca. 40 and 45 years vs. ca. 32 and 31 years. All this means that it is the older speakers who reject all three verbs in the Caused-motion construction, and who also reject smsa in the Ditransitive. The reason that they also reject smsa in the ditransitive may be that the text messaging technology is the most recent technology of the three. It is difficult to know at this point whether these age differences reflect the fact that older speakers are generally more conservative, or whether it means that older speakers have been less exposed to the technology and its accompanying language. It seems fairly reasonable to interpret these findings as showing that there is an ongoing change here, where younger speakers are leading the way in the increased productivity of the Caused-motion construction in Icelandic. To summarize, there is a clear correlation between judgments and the mean age of the participants. Younger speakers are generally more acceptant while the older ones are less acceptant. This correlation is strongest for the Caused-motion construction for all three verbs. A correlation between participant age, judgment and the age of the technology in the society can also be found; Faxing has existed longest and then e-mail, which has existed in Iceland for the last 15–17 years.

 Productivity

Table 5.3  Mean age across judgments

(E)MEILA a. Ég (e)meila þessu til þín b. Ég (e)meila þetta til þín c. Ég (e)meila þér þetta

SMsa a. Ég smsa þessu til þín b. Ég smsa þetta til þín c. Ég smsa þér þetta

Faxa a. Ég faxa þessu til þín b. Ég faxa þetta til þín c. Ég faxa þér þetta

1–3

4–7

43.40 34.57 35.53

30.55 34.91 33.43

1–3

4–7

40.75 36.50 45.10

32.71 35.96 31.26

1–3

4–7

40.80 38.50 36.92

34.12 37.17 35.50

Older speakers prefer the verbs faxa and (e)meila in the Transfer and the Ditransitive constructions, which are the two constructions senda may instantiate in Icelandic, and disprefer them in the Caused-motion construction. Older speakers also prefer smsa, which represents the most recent technology, only in the Transfer construction. Hence, the older the technology in the community, the more constructions are the verbs accepted in. This also suggests that there is a change going on in Icelandic in that the Caused-motion construction is gaining in productivity, perhaps at the cost of the Transfer construction (cf. Barðdal 2003). This situation entails that a new pattern has emerged for verbs of instrument of sending as they would seem to be the only verb class in Icelandic which can instantiate both the Transfer and the Caused-motion construction. The next section discusses the factors contributing to the productivity of the Transfer and the Caused-motion construction in Icelandic. 5.4 Analogy vs. high type frequency Let us start by reviewing the type frequency of Nom-Acc and Nom-Dat predicates in Icelandic. These numbers were presented in Table 3.5 above, which is repeated here as Table 5.4 for the sake of convenience.



Chapter 5.  New verbs of communication 

Table 5.4  The differences between Nom-Acc and Nom-Dat in various counts of Icelandic

Nom-Acc Nom-Dat Total

Dictionary Count

Corpus Count

Borrowed Verbs

N 1,381 738 2,119

N 303 188 491

N 68 39 107

f 65.2% 34.8% 100.0%

f 61.7% 38.3% 100.0%

f 63.6% 36.4% 100.0%

Generally, the numbers in Table 5.4 show that the Nom-Acc construction is two times as high in type frequency as the Nom-Dat construction. Confining this discussion to the intermediate column of the text corpus in Modern Icelandic gives 188 Nom-Dat verbs, as opposed to 303 Nom-Acc verbs (see Chapter 3). The issue, however, is not the total type frequency of the Nom-Acc vs. the Nom-Dat construction, but rather the type frequency of the relevant subconstructions, namely those of Transfer and Caused-motion. Recall that of the 303 Nom-Acc verbs discussed in Chapter 3, only ten occur in the Transfer construction, while fifty of the Nom-Dat verbs occur in the Caused-motion construction. As is evident from Table 5.5, this means that less than one percent of Nom-Acc verbs occur in the Transfer construction, while almost 27% of Nom-Dat verbs occur in the Caused-motion construction. Table 5.5  Proportion of motion verbs among Nom-Acc and Nom-Dat verbs

Transfer (Nom-Acc) Caused-motion (Nom-Dat)

N

f

10/303 50/188

0.33% 26.59%

Recall also from Chapter 3, in the discussion of borrowed verbs, that there were only five verbs of translational motion among the borrowed verbs which instantiated the Transfer construction, while twenty-two instantiated the Caused-motion construction. These are repeated in (21) below, for the sake of convenience: (21) a. Caused-motion construction (22): blasta ‘blast’, bomba ‘bomb’, bundla ‘bundle’, downloada ‘download’, dumpa ‘dump’, droppa ‘drop’, ejecta ‘eject’, expandera ‘expand’, exporta ‘export’, farta ‘fart’, innstalla/innstallera ‘install’, neimdroppa ‘name drop’, parkera ‘park’, peista ‘paste’, poppa ‘deliver when popping by’, pósta ‘post’, publisha ‘publish’, releasa ‘release’, slamma ‘slam’, streyma ‘stream’, uploada ‘upload’, untara ‘unzip’, unzippa ‘unzip’ ... (21) b. Transfer construction (5): browsa ‘browse’, hakka ‘hack’, logga ‘log’, ssh-a ‘ssh’, sörfa ‘surf ’ ...

 Productivity

Observe, moreover, that the five borrowed verbs in (21) b., instantiating the Transfer construction in Icelandic with an accusative object, all have a reflexive object. Therefore, these are not verbs of caused-motion but verbs of locomotion. All of this suggests that the Transfer construction with an accusative object is not a productive construction for verbs of caused-motion in Icelandic. The reason that new verbs of instrument of communication can occur in this construction, then, is because of the high degree of semantic overlap between them and the verb senda. Hence, the extensibility of the Transfer construction to new verbs of instrument of communication is in fact based on the lowest possible type frequency and the highest possible semantic coherence, or in other words on analogy. In contrast, the productivity of the Caused-motion construction in Icelandic is based on its much higher type frequency with verbs of caused-motion, and the high degree of semantic coherence found between the types. Figure 5.1 illustrates the observed productivity of the Caused-motion and the Transfer constructions in Icelandic. The extensibility of the Transfer construction is based on analogy with only one verb, senda, which also means that the productivity domain of the Transfer construction is fairly small, yielding only five new verbs within its domain (in the present material of borrowed verbs). Hence, the location of the Transfer construction on the productivity cline is at its bottom, rightmost side. The extensibility of the Caused-motion construction, in contrast, is based on a much higher type frequency, and hence the degree of semantic coherence between the predicates occurring in this construction and the new predicates the construction is being extended to is lower than the degree of semantic coherence found for the Transfer construction. This is evident in the fact that none of the existing caused-motion verbs in Icelandic are as semantically close to new verbs of instrument of communication as senda is. Therefore, the Caused-motion construction is located higher up on the productivity cline than the Transfer construction. The larger star in Figure 5.1 gives the estimated location of the Caused-motion construction and the smaller star gives the estimated location of the Transfer construction. The location of the Caused-motion construction higher on the cline yields a larger productivity domain, with a substantially better harvest than for the Transfer construction, as is evident from the number of borrowed verbs of causedmotion being attracted by this construction, discussed in Chapter 3 above.



Chapter 5.  New verbs of communication  Type Frequency High Full productivity

Different Degrees of Productivity

Analogy

Semantic Coherence Low

High

Figure 5.1  The productivity of the Transfer and the Caused-motion constructions in Icelandic

In the particular case discussed in this chapter, two different constructions, at different levels of schematicity, are being extended to new verbs of instrument of communication in Icelandic. On the one hand, a low-level lexically-filled concrete verb-specific construction is being extended, senda ‘send’, while on the other, a higher level verb-subclass-specific construction is being extended to the relevant set of verbs. This particular case illustrates again that the boundaries between “analogy” and “productivity” cannot be maintained, as both give rise to extensibility, to varying degrees. As such they are simply two sides of the same coin. 5.5 Summary In this chapter I have reported on the productivity of the Transfer construction, Caused-motion construction and the Ditransitive construction in Icelandic with three new verbs of instrument of communication: faxa ‘fax’, (e)meila ‘email’ and smsa ‘text’. These verbs have been documented as occurring in all three constructions in Icelandic, although their occurrence in the Caused-motion construction is unexpected, as for instance the verb senda ‘send’ in Icelandic can only instantiate

 Productivity

the Transfer and the Ditransitive constructions and not the Caused-motion construction. A questionnaire survey was carried out amongst native speakers, with participants ranging in age from 14 to 85, where example sentences with all three verbs in all three constructions were presented to the participants. They were asked to judge the acceptability of these sentences on a 7-point scale. The results show that all three verbs are accepted in all three constructions, although the Transfer construction obtained the highest mean score for all the verbs. This is expected on analogy with the verb senda which can only occur in the Transfer construction and not in the Caused-motion construction in Icelandic. The verb faxa also showed the highest mean score of all three verbs in the Transfer and the Ditransitive construction, which is expected given that this technology has existed longest of the three. A clear difference also emerged regarding acceptability and the age of the participants, in that the older speakers rejected the non-conventionalized occurrences of the three verbs in the Caused-motion construction while the younger speakers, in contrast, accepted the Caused-motion construction used with these verbs. Hence, there seems to be a change going on in Icelandic in that the Caused-motion construction is gaining in productivity, perhaps at the cost of the Transfer construction, for verbs of caused motion. This increased extensibility of the Causedmotion construction is clearly due to the fact that it is a verb-subclass-specific construction, existing at a considerably higher level of schematicity than the verbspecific construction of senda, and as such it attracts more verbs. On the one hand, there is analogy with only one verb, while on the other, there is analogy on the basis of a verb-subclass-specific construction much higher in type frequency, with the two located at different levels of the productivity cline, showing once again that the difference between so-called “analogy” and “productivity” is an artificial one. In this chapter I have shown how different low-level constructions, i.e. verbspecific constructions and verb-class-specific constructions, can compete with each other, yielding constructional variation with new verbs of instrument of communication. The last case study on the productivity of case and argument structure constructions in Icelandic, presented in the next chapter, pertains to historical productivity. Notes 1 For Internet use in Iceland, see http://www.utvefur.is/fjolmidlar/nr/2168, and for cell phone usage, see http://www.pta.is/upload/files/Tölfr_Farsímakerfi_til%2030.6.06.pdf.

chapter 6

Old and Modern Icelandic A frequency comparison

So far in this book, I have investigated the case marking of borrowed verbs in Icelandic (Chapter 3), case assignment to nonce verbs (Chapter 4), as well as case assignment of new verbs of instrument of communication (Chapter 5), in order to throw light on syntactic productivity. In this chapter I will discuss changes in syntactic productivity over time, more specifically the changes in the productivity of case and argument structure constructions in the history of Icelandic, including a general discussion on the development of case in the Germanic languages. I oppose two views on case, the generative view and the present cognitive-functional CxG view, aiming to demonstrate the superiority of the latter. In Section 6.1 I challenge the view that changes in case marking in Icelandic, in particular, and the Germanic languages, in general, are best described as a change from lexical to structural case, as has been extensively argued for in the generative literature. I show that there are changes in case marking in Icelandic which directly contradict this hypothesis. First, structurally case marked subjects have changed into lexically case marked subjects. Second, lexically case marked subjects have changed from one lexical case to another. Third, structural accusative objects have changed into lexical dative objects. Forth, lexical genitive objects have changed into structural nominative objects. Finally, in the history of English, Swedish and Faroese, structural nominative objects have changed into structural accusative objects. Several auxiliary mechanisms have been postulated to account for these changes, mechanisms which turn out to be derivatives of case marking facts in Germanic, and not predictive per se. In Section 6.2 I suggest a cognitive-functional CxG account of the development, arguing that its predictions hold not only for changes in case marking in Icelandic in general, but also for the development in all the Germanic languages. As the case and argument structure constructions in Germanic were partly synonymous, there were two logical ways for the case and alignment system to develop: (i) by merging the case and argument structure constructions, with subsequent loss of case distinctions and case morphology, or (ii) by eliminating the synonymous low type frequency constructions. The present cognitive-functional CxG view of productivity, based on type frequency, coherence, and an inverse

 Productivity

correlation between the two, predicts that high type frequency constructions will gain in type frequency over time, as they attract new and existing verbs, at the cost of low type frequency constructions. Rapid changes in the vocabulary are expected to speed up the development, as the proportion of new verbs in a language will be higher during periods of language contact than during other periods. As predicted, the development of case correlates with the amount of language contact found in the Germanic language area, as English has been exposed to the most language contact and earliest, with the loss of case morphology also taking place earlier in English than in the other languages. Swedish was also exposed to intense contact during the 13th century and later, which coincides in time with loss of case. German has been exposed to less foreign influence and has eliminated several of the Germanic low type frequency constructions, while Icelandic has been exposed to least foreign influence and maintained most of the Germanic case and argument structure constructions. In Section 6.3 I present a novel analysis of the factors behind the sudden onset of dative substitution in Icelandic during the latter part of the 19th century which has been an anomaly in the literature on dative substitution. I argue that even though the relative type frequency of the Dative subject construction has clearly gone down from Old Icelandic to Modern Icelandic, this has happened at the cost of the subconstruction of happenstance predicates, making the subconstruction of experience-based predicates more salient. This development has contributed to a semantically more coherent Dative subject construction in Modern Icelandic, which in turn explains why the Dative subject construction did not start attracting predicates from the Accusative subject construction until in modern times, evident in texts from after year 1850. Section 6.4 deals with the emergence of the “blended” construction in the history of the Germanic languages, in particular English, Swedish and Faroese. The blend has developed from Dat-Nom to Dat-Acc (on its way to Nom-Acc) and is thus formally a mixture of two case and argument structure constructions, i.e. the Dat-Nom and Nom-Acc constructions. The fact that the object changes its case form first, before the subject, is motivated by the differences in type frequency between dative subjects and nominative objects. Dative subjects are much higher in type frequency than nominative objects, as verbs selecting for nominative objects are only a subset of all dative subject predicates. As a consequence, dative subjects are much more entrenched in the grammatical system of these languages than nominative objects, and hence they resist the change longer. In Section 6.5 I summarize the results and conclusions of the current study for historical productivity, namely that the predictions of the present cognitive-functional CxG approach on productivity are not only borne out for synchronic structures but also for historical developments. The view that syntactic productivity is a



Chapter 6.  Old and Modern Icelandic 

function of type frequency, semantic coherence, and an inverse correlation between the two predicts that high type frequency constructions should increase their type frequency, that low type frequency constructions should decrease in their type frequency, and that only semantically coherent low type frequency constructions should show signs of productivity. These predictions are borne out in the history of Icelandic. Section 6.6 contains a summary of the content and conclusions of this chapter. 6.1

Structural vs. lexical case

It has been extensively argued in the literature that the breakdown of the case system in Germanic is a manifestation of structural case replacing lexical case (see Delsing 1991 and Falk 1997 for Swedish, Allen 1995 and Lightfoot 1999 for English, Askedal 2001 for Scandinavian, Eythórsson 2000a–b, 2002 for Icelandic, and Barnes 1986 and Jónsson and Eythórsson 2005 for Faroese). The predictions of this analysis should manifest itself in either of the following: (1) – Loss of morphological case – Structural case forms replace lexical case forms: Acc/Dat/Gen subjects > Nom subjects Dat/Gen objects > Acc objects The replacement of lexical case with structural case can involve a complete breakdown of the morphological case system, with case distinctions, at best, only present in pronouns. This has happened in English, Dutch and Mainland Scandinavian. It can also entail nominative becoming the subject case, accusative becoming the object case, dative becoming the case for indirect objects and genitive the case for nominal attributes, which is approximately what has happened in standard German. The predictions in (1) above are, however, not borne out for Icelandic. There are many examples in the history of Icelandic of accusative subjects changing not into nominative subjects but dative subjects, i.e. the so-called dative substitution (see Svavarsdóttir 1982, Halldórsson 1982, Rögnvaldsson 1983, Svavarsdóttir, Pálsson and Þórlindsson 1984, Smith 1994, Eythórsson 2000a–b, 2002, Barðdal 2001a:134–38, 2004, M. Smith 2001, Jónsson 2003, Jónsson and Eythórsson 2005): (2) Mig langar > Mér langar me.acc longs > me.dat longs There are also examples of nominative subjects changing into dative subjects:

 Productivity

(3) Ég hlakka til > Mér hlakkar til I.nom look forward > me.dat looks forward This last change can also be regarded as intermediated by an accusative subject (Eythórsson 2000b, 2002). Dative substitution is also well known from the history of English, German and Swedish (see Sections 6.2.1–6.2.3 below and the references there), and I will return to it in Section 6.3 below. The example in (2) illustrates that an alleged lexical accusative is being replaced with an alleged lexical dative with the verb langa ‘long for’. The example in (3) exemplifies alleged structural nominative being replaced with lexical dative with the verb hlakka til ‘look forward’. Therefore, with regard to subject case marking, the predictions of the lexical-to-structural-case hypothesis are far from being borne out in Icelandic, as lexical case is not being replaced with structural case but another lexical case in (2), and structural case is in fact being replaced with lexical case in (3). It is a well-known fact, however, that dative substitution only targets experience-based predicates, which in turn has given rise to yet another dichotomy within the generative tradition, namely the thematicity–idiosyncraticity dichotomy (Zaenen, Maling and Thráinsson 1985, Yip, Maling and Jackendoff 1987, Jónsson 2003, Jónsson and Eythórsson 2005, and for a critical discussion of both dichotomies, see Barðdal 2009a). Changes in case marking, as in (2–3), are assumed to take place on the basis of the semantics of these predicates, hence the term ‘thematic’. Other non-structural case marking and changes in case marking not based on semantic factors, are considered ‘idiosyncratic.’ Hence, in order to rescue the case assigning mechanism based on the structural–lexical dichotomy, additional auxiliary devices, like a dichotomy between thematic and idiosyncratic case, must be invented (cf. Barðdal 2009b). Turning to objects, there are verbs that could occur with either accusative or dative objects in Old Icelandic, but which can only occur with dative objects in Modern Icelandic (4); and conversely, verbs that could occur with either accusative or dative objects in Old Icelandic that can only occur with accusative objects in Modern Icelandic (5): (4) a. ... en fyrir því að ... glataði hann höfuð sitt ... Old Icelandic but for it that lost he head.acc his.acc ‘... but because of that ... he lost his head ...’ (Physiologus 1991: 46–48) b. Hann hafði glatað höfði sínu/*höfuð sitt. M odern Icelandic he had lost head.dat his.dat/head.acc his.acc (5) a. ... að enginn riddari stenst honum. Old Icelandic that no knight withstands him.dat ‘... that no knight is his equal.’ (Ívens saga 1979: 95–99)



Chapter 6.  Old and Modern Icelandic 

b. Enginn riddari stenst hann/*honum. no knight withstands him.acc/him.dat

Modern Icelandic

The example in (5) shows that a structural accusative has ousted lexical dative with the verb standast ‘withstand’, whereas (4) shows that lexical dative has ousted the structural accusative with the verb glata ‘lose’, which is unexpected on the lexicalto-structural-case account. In addition, dative objects should not be a productive category in Icelandic, as dative objects are lexically case marked (cf. Barðdal 2001a: 119–21), but as already discussed in Chapter 3 above, 36% of transitive verbs borrowed into Icelandic assign dative case to their objects. On the lexical-to-structural-case account, moreover, it is expected that genitive objects be replaced with accusative objects because genitive on objects is regarded as lexical whereas accusative on objects is regarded as structural.  Such cases exist; examples like those in (6) with the verb þurfa ‘need’ are well known from the history of Icelandic. (6) a. ... og þarf Hersteinn nú þinna heillaráða. Old Icelandic and needs Hersteinn now your.gen good-advice.gen ‘... and Hersteinn is now in need of your good advice.’ (Hænsna Þóris saga 1987: 1427) b. Ég þarf alla athyglina hjá pabba mínum líka. Modern Icelandic I need all.acc attention.acc at father mine too ‘I need all my father’s attention too.’ (barnaland.mbl.is/barn/19508/vefbok/8) However, it is not expected on the lexical-to-structural-case account that genitive objects change into nominative objects. Such a change is also found from Old Icelandic to Modern Icelandic (see also Berg-Olsen 2009 for the same change in Latvian). Consider the verb batna ‘recover from’ which selected for a genitive object in Old Icelandic (7a) but selects for a nominative object in Modern Icelandic (7b): (7) a. Þormóði batnaði þá skjótt augnaverkjarins ... Old Icelandic Thormod.dat got-better then swiftly eye-pain-the.gen ‘Thormod then swiftly recovered from the eye pain ...’ (Fóstbræðra saga 1987: 802) b. ... og Steinunni batnaði veikin. Modern Icelandic and Steinunn.dat got-better illness-the.nom ‘... and Steinunn recovered from the illness.’ (www.snerpa.is/net/thjod/fellsend.htm) On a lexical-to-structural-case account it is expected that a lexical genitive changes into a structural accusative (cf. Falk 1997:  77–78), but that is not the case in the

 Productivity

history of Icelandic with all genitive object verbs, as (7) shows. Hence, the predictions of the lexical-to-structural-case account are clearly not borne out in Icelandic. It must be pointed out that the existence of nominative objects has been discussed with within generative grammar (cf. Yip, Maling and Jackendoff 1987, Sigurðsson 1989 and subsequent work, Jónsson 1996). This is certainly needed as the case assigning mechanism originally postulated within this framework predicts that nominative objects should not exist, since nominative is supposed to be the structural case for subjects and accusative to be the structural case for objects. Yip, Maling and Jackendoff ’s modified account is based on the idea that structural case is assigned to the first argument in the clause which is not lexically case marked, in this case the object, since the subject is already case marked with a lexical dative. Thus, in order to account for the existence of nominative objects, the original concept of structural case being divided into nominative on subjects and accusative on objects has been abandoned. Yip, Maling and Jackendoff (1987: 224) make a note of this themselves. The problem, however, with their modified account is that it then predicts that predicates with lexically case marked subjects should assign structural nominative to their objects instead of the structural accusative case which is documented with the following predicates in Middle English, Old Swedish and Modern Faroese: (8) for þi ðat him areowe ow for that that him.obl pity you.obl ‘so that he would pity you’ (Allen 1995: 238)

Middle English

(9) Honom thykte sik wara j enom lystelikom stadh Old Swedish he.obl thought himself.obl be in a pleasing place ‘He felt as if he was in a pleasant place.’ (Falk 1997: 77) (10) Mær dámar væl hasa bókina. I.dat like well this book.acc ‘I like this book.’ (Barnes 1986: 33)

Modern Faroese

The verbs in (8–10) above, which instantiate the so-called “blended” construction, are all originally Dat-Nom verbs. Yet there has been a change from nominative objects to accusative objects, i.e. from one structural case to another.1 On Yip, Maling and Jackendoff ’s account, this should not happen, as structural nominative should be assigned here and not structural accusative, because of the lexical case marking of the subject. Hence, the original case assigning mechanism, that objects receive accusative case, must be invoked to account for this change. In other words, the original case assigning mechanism which was used to account for the changes in object case marking of þurfa in (6) makes wrong predictions about the object case marking of batna in (7), hence it needs to be modified. However, this modified



Chapter 6.  Old and Modern Icelandic 

case assigning mechanism makes wrong predictions about the “blended” construction in the history of Germanic (8–10), hence the original case assigning mechanism must be invoked again. Clearly, therefore, these two case assigning mechanisms are simply derivatives of case marking facts in Germanic rather than being predictive. I return to the case marking of the “blended” construction in Section 6.4 below where I argue that the change in case marking is motivated by differences in type frequency between nominative and accusative objects. To summarize the content of this section, I have shown that the predictions of the lexical-to-structural-case account are not borne out for Icelandic. Both structural and lexical case on subjects have been replaced by lexical case (dative substitution). This has given rise to a dichotomy of case assigning mechanism based on the notion of thematic vs. idiosyncratic case. There are also changes from accusative to dative on objects in the history of Icelandic, unexplained and unexpected, as structural object case must then have been replaced with lexical object case. Finally, genitive objects have changed into nominative objects, which is also unexpected on the lexical-to-structural-case account, but can be explained by a modified case mechanism which assumes that an object gets structural nominative if the subject is already lexically case marked. However, this modified case assignment mechanism does not explain the change from nominative objects to accusative objects in the history of English, Swedish and Faroese, although the original case assigning mechanism that accusative case is assigned to objects does. Hence, generative grammar must make use of different auxiliary mechanisms to account for case marking in Icelandic and changes in case marking in the history of Germanic, in addition to the original mechanism that nominative and accusative are assigned on the basis of their structure in the sentence. Therefore, these auxiliary mechanisms are simply derivatives of case marking facts in Germanic rather being predictive. As such they are of limited explanatory value. For a further critical discussion of this dichotomy and the dichotomy between thematic and idiosyncratic case, see Barðdal (2009a), where it is argued that all case marking in Icelandic is “lexical”, in the sense that it is bound to specific lexical entries. 6.2

The present cognitive-functional CxG approach

Recall the overview in Table 3.1 above of the case and argument structure constructions of one- and two-place predicates in the history of Icelandic, reproduced here as Table 6.1 for the sake of convenience. As no other case and argument structure constructions seem to have existed in any of the individual Germanic languages, Table 6.1 should accurately represent the case and argument structure constructions common for the Germanic language area before the breakdown of the case and alignment system in the individual languages.

 Productivity

Table 6.1  Case and argument structure constructions in Earlier Germanic Nom-subject

Acc-subject

Dat-subject

Gen-subject

Nom Nom-Acc Nom-Dat Nom-Gen

Acc Acc-Nom Acc-Acc Acc-Gen

Dat Dat-Nom Dat-Gen Dat-PP

Gen Gen-Nom Gen-PP Gen-S

Nom-PP Nom-S

Acc-PP Acc-S

Dat-S

With regard to the semantics of the various argument structure constructions, discussed in Chapters 3–4 above, recall that there is a substantial semantic overlap between the different argument structure constructions in Icelandic. The Nominative subject construction is the construction highest in type frequency and semantically the most open construction, as verbs from all semantic classes can have a nominative subject (see Chapter 3). The subconstructions of the Nominative subject construction, i.e. Nom-Acc, Nom-Dat and Nom-Gen, also overlap semantically with each other. For instance, all three subconstructions can be instantiated by verbs of cognition and emotion. Also, Nom-Dat verbs of connection are similar to Nom-Acc verbs of meeting and uniting, Nom-Dat verbs of ending and finishing share certain semantic features with Nom-Acc verbs of commencement, Nom-Dat verbs of obtaining and maintaining are similar to Nom-Acc verbs of (means) of gaining, Nom-Dat verbs of puttering partially occupy the same semantic field as Nom-Acc verbs of recuperation, verbs denoting destruction can be found among both Nom-Acc and Nom-Dat verbs, and so can verbs of motion. Moreover, the Dative subject construction also shares some of its semantics with the Nominative subject construction, as it is instantiated by verbs of cognition, emotion, perception, obligation, and ontological existence (see Chapter 4), and these verbs can also occur in the Nominative subject construction. The Genitive subject construction is mostly instantiated by predicates denoting ontological or perceived states (cf. Appendix A). Hence it is partly synonymous with both the Dative subject construction and the Nominative subject construction. The Accusative subject construction is instantiated by verbs of cognition, emotion, perception, and bodily states (see Appendix A), exactly like the Dative subject construction (cf. Barðdal 2001b, 2004 and Section 4.3 above), and in part the Nominative subject construction. To sum up, there is a substantial semantic overlap between the two subconstructions of the Oblique subject construction; and these two, in turn, overlap semantically with the Nominative subject construction, since the Nominative subject construction is the semantically most open construction. As the Accusative and Dative subject constructions were instantiated by verbs from



Chapter 6.  Old and Modern Icelandic 

the same semantic classes in the individual Germanic languages (cf. Barðdal 2004), it is clear that the four different subject constructions were partly synonymous, not only in Icelandic, but in the other Germanic languages as well. The question arises, of course, how and why a system with synonymous argument structure constructions would appear in the first place. This is, at least in part, a consequence of historical development where case markers have merged, either functionally or formally (cf. Luraghi 1987), yielding fewer argument structure constructions and more overlap between them. It has been assumed in cognitive-functional linguistics that languages have a tendency to avoid synonymous grammatical forms (see Goldberg 1995: 67, and the references cited there). The loss of morphological case in the individual Germanic languages can be regarded as a consequence of this, since the various case and argument structure constructions were partly synonymous with each other (cf. Barðdal and Kulikov 2008). Given this synonymy, there are logically two ways for languages to evolve: (11) – The morphological case distinctions disappear with a subsequent merging of the case and argument structure constructions – High type frequency constructions attract verbs from low type frequency constructions, thereby gradually causing low type frequency constructions to fall into disuse Mainland Scandinavian, English and Dutch seem to have evolved in the former way, as the case and argument structure constructions have disappeared in these languages. In contrast, German, Icelandic and Faroese have moved along the latter path, with the case and argument structure constructions highest in type frequency being generalized at the expense of the constructions lower in type frequency. According to the approach to productivity advocated here, syntactic productivity is a function of the type frequency of an argument structure construction and its semantic coherence, and an inverse correlation between the two (cf. Section 2.4 and 3.2 above). As high type frequency constructions are also semantically open and non-restricted, they are expected to attract new verbs entering the language, resulting in high type frequency constructions increasing in type frequency. Low type frequency constructions are usually more restricted semantically and are expected to gradually disappear unless they are high in token frequency, in which case they might be preserved as verb-specific lexically-filled constructions. If they attract new items, this should be on the basis of a high degree of similarity. High type frequency constructions can also attract verbs from low type frequency constructions, resulting in low type frequency constructions decreasing in type frequency. Moreover, if two low type frequency constructions

 Productivity

exchange items, it is expected that the construction lower in type frequency will lose items to the construction higher in type frequency. On this approach, it is predicted that a rapid change in the vocabulary may speed up the development, precisely because the bulk of new verbs will be attracted by the high type frequency constructions. This in turn should reduce the type frequency of low type frequency constructions, increasing the chances of them becoming extinct. It has been observed by various scholars that morphological case seems to have been lost at the same time as massive lexical borrowing is found, due to contact situations. This is discussed for Old English by Allen (1995), for Old Swedish by Wessén (1929, 1992), and for Scandinavian in general by Jahr (1995). Wessén argues that the case system was lost because the loan words could not easily adjust to the inflectional system. This has, however, been argued against by Norde (1994) who shows that the morphophonemic structure of the loan words cannot have constituted a problem for the noun inflection. It is nevertheless a fact that the breakdown of the case and alignment system in both English and Scandinavian coincides in time with massive lexical borrowings. It is also a fact that the correlation between loss of morphological case and the contact situation has so far not been appropriately included in the account of the development of case in Germanic. On the present account, it is expected that massive lexical borrowings will favor high type frequency constructions and disfavor the ones low in type frequency, and as will become evident below, this is borne out in Germanic. Therefore, the correlation between language contact and loss of morphological case is not an anomaly anymore but is satisfactorily accounted for on the present approach. To sum up, the predictions of the present cognitive-functional CxG approach of the development of case and argument structure constructions are the following: (12) – High type frequency constructions will attract new verbs and verbs from low type frequency constructions – Low type frequency constructions will decrease in their proportional type frequency since they do not attract new verbs, precisely because of their low type frequency and their restricted semantics – If a low type frequency construction attracts new items, it will be on the basis of a high degree of similarity – The construction lowest in type frequency will disappear first – Rapid changes in the vocabulary will speed up the development I will now consider the development of the case and alignment system in the Germanic languages in the light of the predictions of the present approach. I begin with Swedish (6.2.1), then I examine the development in English (6.2.2), then German (6.2.3) and subsequently I discuss the changes in frequency from Old Icelandic



Chapter 6.  Old and Modern Icelandic 

to Modern Icelandic (6.2.4). After that I proceed to show that the predictions of the present cognitive-functional CxG approach also hold for the onset of dative substitution in Icelandic (6.3) and the emergence of the “blended” construction in the Germanic languages (6.4). 6.2.1 Swedish In the Mainland Scandinavian languages the case and argument structure constructions have merged and case morphology disappeared. The Genitive subject construction, for instance, is not documented in Old Swedish at all. This is also the construction lowest in type frequency in Modern Icelandic (cf. its marginal status in Table 6.1 above, evident by the fact that it subsumes much fewer subconstructions than the other case and argument structure constructions). According to Delsing (1991), the first construction to disappear in Swedish was the infrequent Genitive object construction. This change had already happened before 1350. Next to disappear is the now lowest type frequency Accusative subject construction. That happens before 1400 (Falk 1997: 14–15). Around 1450 all case endings have been lost, which entails that the case distinction between nominative and dative subjects is lost for nouns, and the case distinction between accusative and dative objects as well. At this point in time, then, Swedish only exhibits two case forms on pronouns, i.e. nominative and non-nominative (or oblique). The only case frames that are now left in Swedish are the Nominative subject construction and the former Dative, now Oblique, subject construction, although this is only visible on pronouns. According to Falk (1997: 187–88), the number of different verbs occurring in the Oblique subject construction is as low as thirty-eight (compared to 700 dative subject predicates in Icelandic). These verbs either became associated with the Nominative subject construction or fell into disuse during the 16th and the 17th centuries. The last verb class to become associated with the Nominative subject construction is the class of ditransitive verbs occurring in the Oblique passive construction (where the indirect object of the active construction has been promoted to subject in the passive construction). That happens around 1800. It is interesting to note, however, that despite the low type frequency of the Oblique subject construction, there are still documented cases of it being mildly productive during medieval times. Falk (1997: 51) reports on such examples: (13) Än sidhan honom iäfwadhe tok han til at sionka but since him.obl doubted took he to sink ‘But since he doubted he started sinking’

(ca. 1420)

(14) ty ær thz sa som mik tænker therefore is that such as me.obl thinks ‘That is why it is as I think’

(ca. 1500)

 Productivity

The verbs iäfwa ‘doubt’ and tænka ‘think’ are conventionally associated with the Nominative subject construction in Old Swedish (Falk 1997). They have thus clearly been attracted by the salient cluster of oblique subject predicates, due to their lexical meaning which is compatible with the semantics of the Oblique subject construction as a whole. This is, in other words, an instance of an analogy based on a low-level verb-class-specific construction. To sum up, Swedish and Mainland Scandinavian in general have dealt with synonymous argument structure constructions by merging them, with a subsequent loss of the morphological case system. Moreover, according to the predictions of the present cognitive-functional CxG approach, the constructions lowest in type frequency should be the ones in most danger of disappearing. This prediction is borne out in Swedish, in that the most infrequent constructions disappeared first and the least infrequent constructions disappeared last. 6.2.2 English According to Allen (1995: 211–20) the first case and argument structure construction to disappear in English was the Genitive object construction. That happened in two stages: first the genitive of the impersonal Acc-Gen and Dat-Gen disappeared, then the genitive of the Nom-Gen construction. This is identical to the development in Icelandic (see 6.2.4 below) where the Dat-Gen construction has already disappeared while the Nom-Gen has decreased in frequency from Old to Modern Icelandic. Allen (1995:  218–19) makes a point of the fact that genitive objects disappear in two stages which on her generative approach can be interpreted as there being a structural difference between these two types of genitive objects (although the status of this structural difference is unclear). On the present approach, however, it is predicted that Acc-Gen and Dat-Gen disappear before Nom-Gen because of the differences in type frequency found between these constructions in Germanic. This prediction is also borne out. Second, the distinction between accusative and dative case is completely lost at the end of the 13th century for nouns. It is, however, maintained for pronouns. Third, the Oblique passive construction of ditransitives becomes infrequent in the late 14th century. Finally, the active Oblique subject construction remains in full use until the 14th century (as opposed to passives of ditransitives), starts declining in the 15th century and is completely lost in the 16th century. The productivity of the Oblique subject construction has been amply documented in Middle English (cf., for instance, Seefranz-Montag 1983, Allen 1995, inter alia) and it attracts both borrowed and already existing verbs. Allen (1995: 250) reports on a modal verb occurring in the oblique subject construction instead of in the conventionalized Nominative subject construction:



Chapter 6.  Old and Modern Icelandic 

(15) Wherefore us oghte ... have pacience. why us.obl should have patience ‘Why we should ... have patience.’

Middle English

The internal order of distinctions being lost is the same in English as in Swedish, which is consistent with an overall assumption that the type frequency of the different verb classes may have been relatively similar in the different Germanic languages, with some minor deviations to be expected. There is one difference between the development of case in Swedish and English, namely the internal order of the loss of the Oblique subject construction, as opposed to the Oblique passive construction of ditransitives. In Swedish the loss of the Oblique subject construction precedes the loss of the Oblique passive construction while the order is reversed in English (this reverse order of events has also been reported by Knudsen 1956: 36–41 for Danish). On a usage-based account this is expected to be a manifestation of differences in type frequency and semantic coherence between oblique subject predicates and ditransitives in the two languages, in that the Oblique passive construction had higher type frequency than the Oblique subject construction in Swedish as opposed to English, and vice versa. This might be because the vocabulary may not have been renewed at the same rate in the two languages. Obviously, extensive borrowing will increase the type frequency of the most productive construction, and hence reduce the type frequency of other less productive constructions. Another explanation for differences in type frequency of constructions in different languages is that verbs may be borrowed or coined in a particular language for one semantic field at a higher rate than for another semantic field, which in turn may result in differences in type frequency between equivalent constructions in related languages. This, in turn, can contribute to increased semantic coherence of a low-type frequency construction in one language, as opposed to the other language, which again strengthens the chances of that construction. However, as neither Falk nor Allen gives any numbers for the ditransitives they investigate, the matter cannot be determined at present. Table 6.2  Case and argument structure constructions in Modern English and Mainland Scandinavian Nom Nom Nom-Acc Nom-PP Nom-S

 Productivity

Compare now the case and argument structure constructions in earlier Germanic, given in Table 6.1 above, to the situation in both Swedish and English, after the loss of the morphological case, shown in Table 6.2. The latter table shows that only the Nominative subject construction, i.e. the construction highest in type frequency in Germanic, still exists. All the low type frequency case and argument structure constructions have disappeared from the two languages. 6.2.3 German German differs from Swedish and English in that it has maintained its morphological case, like Icelandic and Faroese. However, German has clearly developed in the direction that the constructions lowest in type frequency have disappeared, and the remaining low type frequency constructions have become even lower in type frequency. The Nom-Acc, Nom-Dat and Nom-Gen constructions still exist in Modern German, although the Nom-Dat construction is only instantiated by approximately 100 verbs, while the corresponding figure for Modern Icelandic is 750 verbs (Maling 2002: 31).2 The Nom-Gen construction is instantiated by approximately five predicates in Modern High German language use. According to Seefranz-Montag (1983: 171–89) the genitive object of the AccGen and Dat-Gen constructions, as well as the genitive subject of genitive subject predicates, merged with nom/acc forms in the 13th century. Furthermore, most of the verbs occurring in the Accusative and Dative subject constructions have either fallen into disuse in German or been attracted by the Nominative subject construction. However, during the Middle High German period there was considerable variation between the different constructions, in that impersonal verbs could readily occur in the Accusative, Dative and the Nominative subject construction. The Dative subject construction attracted many verbs from the Accusative subject construction, and the Accusative subject construction has, likewise, attracted (somewhat fewer) verbs from the Dative subject construction (1983: 162–63). The fact that the verbs were so easily exchanged between the Accusative and Dative subject constructions is a consequence of the close semantic similarity of these constructions. Finally, German has maintained the Dative passive construction. This is parallel to the Swedish situation in that the Oblique subject construction disappears before the Oblique passive construction of ditransitives. Compare now the case and argument structure constructions in earlier Germanic, given in Table 6.1 above, to the remaining case and argument structure constructions in Modern German, given in Table 6.3. The cross in this table signals that the relevant case and argument structure construction is so low in type frequency that it is on the boarder of becoming extinct.



Chapter 6.  Old and Modern Icelandic 

Table 6.3  Case and argument structure constructions in Modern German Nom

Acc

Dat

Nom Nom-Acc Nom-Dat

Acc Acc-Nom Acc-PP

Dat Dat-Nom Dat-PP

† Nom-Gen Nom-PP Nom-S

Acc-S

Dat-S

The situation in Modern High German is such that the Accusative/Dative subject construction has more or less fallen into disuse, except with some Dat-Nom predicates, a few Acc-Nom predicates, and intransitive adjectival predicates of the type mir ist kalt (literally ‘me.dat is cold’). Thus, most subjects in German are in the nominative case, most objects are in the accusative case, indirect objects are in the dative case and nominal attributes are in the genitive case. Hence, all the high type frequency case and argument structure constructions have been maintained in Modern High German, at the expense of the low type frequency constructions which have either disappeared or are only instantiated by a few predicates in the modern language. 6.2.4 Icelandic The construction lowest in type frequency in Old Icelandic is presumably the DatGen construction, as I have only found it instantiated by two verbs, batna ‘get better’ and létta ‘abate (of illness)’.3 This construction is also the only case construction that has disappeared in Icelandic. Not surprisingly, batna and létta were subsumed by the more common Dat-Nom construction which was much higher in type frequency (a count of alternating Dat-Nom/Nom-Dat predicates in Modern Icelandic reveals 111 predicates, cf. Barðdal 2001b, and there are also several non-alternating Dat-Nom preedicates in Modern Icelandic which raises the type frequency even further). This is a natural development as the two case and argument structure constructions are similar in both form and meaning. That is, both constructions are two-place predicate constructions with a dative subject, and the predicates shared by them are experience-based. Therefore, the Dat-Gen construction can be regarded as a proper subconstruction of the more general Dative subject construction, as shown in Figure 6.1:

 Productivity

[Dat-V-X]

[Dat-V]

[Dat-V-Nom]

[Dat-V-Gen]

[Dat-V-PP]

vera kalt

líka

batna

geðjast

[Dat-V-S] virðast

Figure 6.1  The Dative subject construction and its subconstructions

Examples of the verb-specific constructions in Figure 6.1 are given in (16) below: (16) a. Það var daginn sem öllum var kalt ... it was day-the which everybody.dat was cold ‘It was the day everyone was cold ...’ (privat.bloggar.is/blogg/77623) b.

Mér líkar þessi vinna rosalega vel ... me.dat likes this work.nom exceedingly well ‘I really like this job ...’ (torleifur.blogspot.com/ 2005_08_01_torleifur_archive.html)

c. Þormóði batnaði þá skjótt augnaverkjarins og ... Thormod.dat got-better then swiftly eye-pain-the.gen ‘Thormod then swiftly recovered from the eye pain ...’ (Fóstbræðra saga 1987: 802) d. Mér geðjaðist að hugmyndinni um hugleiðslu. me.dat liked at idea-the of meditation ‘I liked the idea of meditating.’ (www.al-anon.is/hlekkurinn.asp?Frettir_ID=9501) e. Mér virðist sem að fæstir þeirra hafi tíma til að hlusta ... me.dat seems as if fewest them have time for to listen ‘It seems to me that the fewest of them have time to listen ...’ (www.doktor.is/grein/efni/grein.asp?id_grein=3366&flokkur=16) In contemporary Icelandic, the Dat-Gen construction does not exist; Therefore a figure like Figure 6.1 for Modern Icelandic would not contain the Dat-Gen construction. Both batna and létta are now Dat-Nom predicates. It is also a fact that verbs which occur in one of the subconstructions in Figure 6.1 readily occur in some of the other. The verb líka ‘like’ is one example; it can occur in three out of four subconstructions of the Dative subject construction:



Chapter 6.  Old and Modern Icelandic 

(17) a.

Mér líkar þessi vinna rosalega vel ... Dat-Nom me.dat likes this work.nom exceedingly well ‘I really like this job ...’ (torleifur.blogspot.com/ 2005_08_01_torleifur_archive.html)

b. Mér líkar við þig en ... Dat-PP me.dat likes with you.acc but ‘I like you but ...’ (frontpage.simnet.is/united/TE_stelpur_segja.htm) c. ... mér líkar að hann sé með gras í kjaftinum. Dat-S me.dat likes that he is with grass in mouth-the ‘... I like that it has grass in its mouth.’ (www.ljosmyndakeppni.is/resultimage.php?imageid=1314&challen geid=70) The present analysis, based on higher type frequency of the Dat-Nom construction as opposed to the Dat-Gen construction, and their similarity in form and meaning, makes the prediction that an alleged lexical genitive object will change into an alleged structural nominative object in the history of Icelandic, a change which is unexpected on the lexical-to-structural-case account, as on that account one would expect the object to change into accusative and not nominative (discussed in Section 6.1 above). Since the Dat-Nom construction is higher in type frequency than the Dat-Gen construction (at least 111 predicates vs. 2–5), it attracts verbs from the Dat-Gen construction, causing it to fall into disuse. The change from genitive objects to nominative objects is thus not an incongruity on the current approach. The same factors are responsible for the change from genitive objects to accusative objects with verbs like þurfa ‘need’ in (6) above. As mentioned in Section 6.1, þurfa was originally a Nom-Gen verb, but the Nom-Gen construction was and is a low type frequency construction. The Nom-Acc, in contrast, is the highest in type frequency of all transitive constructions in Icelandic. Therefore, it attracts verbs from the Nom-Gen construction. Let us now compare frequency figures for subjects and objects in Old Icelandic and Modern Icelandic texts (reported on in Section 3.1 above). In this comparison, only half of the Modern Icelandic text material was used. That is, only the four Modern Icelandic genres that can be regarded as having a correspondence in the Old Icelandic material were used, i.e. Icelandic fiction, translated fiction, biographies and memoirs, and non-fiction. The corresponding Old Icelandic texts are Icelandic family sagas and tales, romantic sagas (translated), contemporary sagas (of kings and bishops) and non-fiction (see Barðdal 2001a for further information on the two corpora).

 Productivity

Table 6.4  Subject frequency in Old Icelandic and Modern Icelandic Old Icelandic

Nom Acc Dat Gen

Modern Icelandic

N

%

N

%

299 13 72 7

76.3 3.5 18.4 1.8

395 14 48 7

85.0 3.2 10.3 1.5

391

100

464

100

Table 6.4 gives the number of subjects in different case form in both language stages. However, it only specifies nominative subject case for two-place predicates, i.e. Nom-Acc, Nom-Dat and Nom-Gen, leaving out intransitive and monotransitive verbs with nominative subjects. Most importantly, the table shows that the type frequency of nominative subjects has gone up on a per/word basis of ca. 9%. It also shows that the type frequency of dative subjects has gone down from seventy-two to forty-eight types in the corpus, a reduction of approximately 8%. Accusative and genitive subjects, however, have remained stable in type frequency between the two periods of Icelandic. A closer look at the Accusative and Genitive subject constructions reveals that some of the predicates instantiating them in Old Icelandic have fallen into disuse and some are shared across the two language stages (see Appendix A). This last fact suggests that the constructions have been maintained in Icelandic because the few lexical items instantiating them have not fallen into disuse. However, it is possible that the Accusative and Genitive subject constructions may have gone down in relative type frequency also, although the present corpus is presumably not large enough to capture this. Consider now Table 6.5 on object frequency which illustrates that the NomAcc construction is proportionally higher in type frequency in the Modern Icelandic corpus than in the Old Icelandic corpus. The difference is around 6%, i.e. from 52.1 to 58.4%. It is interesting in this context that the Nom-Dat construction has not only remained stable from Old Icelandic to Modern Icelandic but also that there has been a slight increase in its type frequency, namely of ca. 3%, from 31.6 to 34.7%. The increase in the type frequency of Nom-Acc and Nom-Dat has taken place at the cost of verbs selecting for nominative and genitive objects, since their type frequency is drastically reduced.



Chapter 6.  Old and Modern Icelandic 

Table 6.5  Object frequency in Old Icelandic and Modern Icelandic Old Icelandic Dat-Nom Nom-Acc Nom-Dat Nom-Gen

Modern Icelandic

N

%

N

%

33 173 105 21

10.0 52.1 31.6 6.3

11 237 141 17

2.7 58.4 34.7 4.2

332

100

406

100

Observe that the two constructions highest in type frequency, Nom-Acc and NomDat, which exhibited approximately 52 and 32% of the types, respectively, are the ones which have increased their share, namely of approximately 6 and 3%, respectively. The fact that the Nom-Dat construction has increased in type frequency from Old to Modern Icelandic of ca. 3% may suggest that the minimum type frequency needed for a category to be stable is around 30% of the types. Clearly more research is needed to establish that, but it is in accordance with the predictions of the present approach to productivity, based on type frequency and coherence, that such large categories will increase their type frequency in course of time. The fact that a category containing approximately 30% of the types is also productive sustains the validity of this approach. Tables 6.4–6.5 show that the constructions highest in type frequency have indeed increased their type frequency, i.e. Nom-Acc and Nom-Dat, and the ones lowest in type frequency have gone down in frequency, i.e. Dat-Nom and NomGen. Two constructions are of the same size, Accusative and Genitive subject constructions, although they may also have gone down in relative type frequency, not detectable here because of the limited size of the corpora. Table 6.6 (which is the same table as Table 6.1 above) thus presents the case and argument structure constructions in Modern Icelandic. The case and argument structure constructions marked with † have either gone drastically down in type frequency in Icelandic or disappeared (either because they have fallen in disuse or been attracted by other case and argument structure constructions higher in type frequency). These are also the case and argument structure constructions which were lowest in type frequency in Old Icelandic. Therefore, of the four languages discussed here, Icelandic has changed the least from Proto-Germanic and maintained most of the case and argument structure constructions common to the Germanic language area.

 Productivity

Table 6.6  Case and argument structure constructions in Icelandic Nom-subject

Acc-subject

Dat-subject

Gen-subject

Nom Nom-Acc Nom-Dat Nom-Gen Nom-PP Nom-S

Acc † Acc-Nom Acc-Acc † Acc-Gen Acc-PP Acc-S

Dat Dat-Nom † Dat-Gen Dat-PP Dat-S

Gen † Gen-Nom Gen-PP Gen-S

In sum, the predictions of the present cognitive-functional CxG approach to changes in case and argument structure constructions are borne out for Icelandic, as the high type frequency constructions have gained in type frequency while the low type frequency constructions have lost in type frequency. I return to dative substitution in Icelandic, i.e. the change from accusative subjects to dative subjects, in Section 6.3 below. 6.2.5 Vocabulary, language contact, type frequency and loss of case A large degree of the vocabulary has been maintained from Old Icelandic to Modern Icelandic (cf. Kvaran 1996, Rögnvaldsson 1997). In that sense, Icelandic differs from the three other Germanic languages discussed above. A comparison of the predicates occurring in the Old and Modern Icelandic corpora with a list of the 100 most frequent predicates in Modern Icelandic (Pind 1991) reveals that of the ninety-one most frequently occurring transitive predicates in Modern Icelandic language use, eighty-three predicates occur in the Modern Icelandic corpus and eighty-two in the Old Icelandic corpus. This suggests more than a 90% overlap in the verbal vocabulary between the two language stages. This is furthermore in accordance with my hypothesis that there is a correlation between the rate of the vocabulary replacement and the development of case in Germanic. Table 6.7 shows that there is an implicational hierarchy between the time span and the amount of contact found for the four languages, in that Icelandic has maintained most of the inherited vocabulary, then German, then Swedish and then English. The implicational hierarchy in Table 6.7 is the same implicational hierarchy as that for the time span and the loss of morphological case, as English lost its case marking first, Swedish lost its case marking later, German has maintained more case and argument structure constructions than Swedish and English, and Icelandic has maintained most of the original Germanic case and argument structure constructions.



Chapter 6.  Old and Modern Icelandic 

Table 6.7  The correlation between language contact and loss of case Vocabulary replacement:

English < Swedish < German < Icelandic

Development of case:

English < Swedish < German < Icelandic

This correlation is predicted on the present cognitive-functional CxG approach which takes type frequency as its point of departure. That is, English leads the development with the most extensive borrowings of all the languages considered here, beginning in the 11th century (Allen 1995), evident by the enormous amount of borrowed vocabulary. Swedish has also been involved in much language contact, with massive Low German influence beginning in the late 13th century, as is evident from the fact that several affixes, both prefixes and suffixes, as well as auxiliaries have been borrowed into the language (Wessén 1929, 1992). German has not had the extensive replacement of the vocabulary found in both English and Swedish, but it has nevertheless been more influenced by other languages than Icelandic has, for instance by Latin and French during the middle ages (Polenz 1991). Icelandic is the least influenced language of the four, with most of the Old Icelandic vocabulary being maintained (cf. Kvaran 1996, Rögnvaldsson 1997). Obviously, the faster the vocabulary is renewed, the sooner the high type frequency constructions will increase in type frequency, and the sooner the low type frequency constructions will decrease in their type frequency, as the renewal of the vocabulary favors the construction highest in type frequency and disfavors the ones lower in type frequency. Therefore, on the present cognitive-functional CxG approach it is predicted that the language that has been subject to most foreign influence will lead the development, and that the least influenced language will lag behind. This prediction is borne out for Germanic. Furthermore, this language contact and the changes in the vocabulary coincide in time with the disappearance and the breakdown of the case and alignment system. The chronology of the development is described in Sections 6.2.1–6.2.4 above, and summarized in Table 6.8 for English, Swedish and German. Table 6.8 clearly illustrates that the internal order of the development is more or less the same, as the constructions lowest in type frequency disappeared first, then the constructions second-lowest in type frequency, and so on. The dates in Table 6.8 refer to the disappearance of the case and argument structure constructions, unless specified otherwise.

 Productivity

Table 6.8  The chronology of the breakdown of the case and alignment system in English, Swedish and German

Gen-subject Acc/Dat-Gen Nom-Gen Acc-subject Nom-Dat Dat-subject Ditr. passive

English

Swedish

German

non-attested 1100–1150 1150–1200 1200 (acc/dat distinction lost) 1200 (acc/dat distinction lost) 1300–1500 >1300

non-attested non-attested >1350 >1400 1450 (all case marking lost) 1500–1700 1800

1200–1300 1200–1300 1200–1300 reduced reduced reduced maintained

In sum, the breakdown of the case and alignment system has gone farthest in English, then in Swedish, then in German, and least far in Icelandic. The development correlates with language contact, as rapid changes in the vocabulary will speed up the process. Indeed, the breakdown of the morphological case system in the individual Germanic languages correlates, not only with the density of the contact, but also with the time span of the contact, and with the loss of the individual case and argument structure constructions. 6.3

Dative substitution in the history of Icelandic

Another change in case marking in Icelandic, subject to considerable attention in the international literature, is dative substitution, sometimes also referred to as “dative sickness” (see Svavarsdóttir 1982, Halldórsson 1982, Rögnvaldsson 1983, Svavarsdóttir, Pálsson and Þórlindsson 1984, M. Smith 1994, Eythórsson 2000a–b, 2002, Barðdal 2001a:134–38, 2004, M. Smith 2001, Jónsson 2003, Jónsson and Eythórsson 2005). This change entails that verbs conventionally occurring in the Accusative subject construction have started occurring more and more in the Dative subject construction. This was shown in (2) above, repeated here for convenience: (2) Mig langar > Mér langar me.acc longs > me.dat longs This change has also been documented in the history of German (see Section 6.2.3 above), while in Icelandic it started in the late 19th century (Halldórsson 1982). At this point, two questions arise: a) Why have accusative subject predicates changed into dative subject predicates and not dative subject predicates into accusative ones?; and b) Why does this change not happen until the late 19th century? Starting with the first question, the reason that there is a transition from one construction to the other is due to the fact that the two constructions (partly) share the



Chapter 6.  Old and Modern Icelandic 

same semantics. Both constructions are instantiated at least by verbs of emotion, cognition, perception, and bodily states. It is also expected that accusative experience-based predicates be attracted by the Dative subject construction and not the Nominative subject construction. This is because the Dative subject construction is much more restricted in its semantics and semantically closer to the Accusative subject construction than the Nominative subject construction which is a semantically open construction, instantiated by verbs from all semantic fields. Moreover, on the present cognitive-functional CxG approach to productivity, it is expected that the Accusative subject construction changes into dative, and not vice versa, because of the differences in type frequency between the two. The accusative subject construction is much lower in type frequency than the Dative subject construction. Table 6.4 above gives the relative type frequency of fifteen accusative subject types vs. forty-eight dative subject types from a subset of the Modern Icelandic corpus, and Table 3.3 gives the absolute type frequency of ca. 200 accusative subject types vs. ca. 700 dative subject types in Modern Icelandic. Table 6.4 also reveals that the differences between the two constructions were even more prominent in Old Icelandic. Given the assumption that the Dative subject construction attracts predicates from its semantically similar Accusative subject construction because of the higher type frequency of the former, let us turn to the second question, i.e. why this process begins as late as in the latter part of the 19th century. This is a legitimate question because if the structural prerequisites for this change are found already in Old Icelandic, given that the differences in type frequency between the constructions are common for the Germanic linguistic area, why does the onset of this change in Icelandic not take place earlier than in the 19th century? In order to answer this question, let us compare the predicates instantiating the Dative subject construction in the Old and Modern Icelandic corpora. As already discussed in Section 6.2.4 above, there are seventy-two types occurring in the Dative subject construction in the Old Icelandic corpus, but only forty-eight types in the corresponding Modern Icelandic corpus. Excluding passives from the comparison, the Old Icelandic corpus contains sixty-six active dative subject predicates, while the Modern Icelandic corpus contains thirty-three active dative subject predicates. This entails in fact a 50% reduction in the use of the active subconstruction of the Dative subject construction on a per/word basis (see Table 6.9 below). An earlier analysis of all the ca. 700 documented types occurring in the Dative subject construction in Icelandic (Barðdal 2004) revealed two major semantic

 Productivity

classes of verbs, namely Experience-based predicates and Happenstance predicates, which in turn can be divided into seven subclasses each. These subclasses are: (18) a. Experience-based predicates: Verbs of Emotions, Attitudes, Cognition, Perception, Bodily states, Changes in bodily states, Obligation b. Happenstance predicates: Verbs of Decline, Failing/mistaking, Success/performance, Ontological states, Social interaction, Gain, Personal properties An examination of the sixty-six types found in the Dative subject construction in the Old Icelandic corpus reveals that they divide evenly on Experience-based and Happenstance predicates, with thirty-three predicates in each of the two superordinate classes: Experience-based predicates: 1 Verbs of emotion: bjóða hugur ‘want’, finnast ‘think, feel, seem’, geta(st) að ‘like’, líka ‘like’, lítast á ‘like’, mislíka ‘dislike’, vera að harmi ‘be grief-stricken’, vera ást á ‘love’, vera eftirsjá ‘regret’, vera ekki þökk á ‘dislike’, vera heyrilegt ‘be pleasurable’, vera í skapi ‘feel like’, vera ekki um e‑n ‘dislike’, þykja ‘feel’ ... 2 Verbs of attitudes: fara vel ‘suit well’, sýnast ‘appear, seem’, vandast málið ‘become difficult’, vera best ‘be best for sby’, vera dul ‘be impossible’, vera nauðsyn ‘be necessary/important for sby’, vera vant ‘be difficult for sby’, vera vant ‘be in need of ’, verða vei ‘be woe to’, verða hamingja að ‘be of happiness for sby’, vera þungt ‘be difficult for sby’ ... 3 Verbs of cognition: ganga til ‘have the intention’, koma í hug ‘get an idea’, segja hugur um ‘have an instinct’, vera forvitni á ‘be curious’, vera grunur ‘suspect’, verða kunnugt ‘find out’ ... 4 Verbs of bodily states: vera sár fótur ‘have pain in the leg’ ... 5 Verbs of obligation: bera ‘be obliged’ ... Happenstance predicates: 6 Verbs of failure/mistake: leiða illt ‘get into trouble’, misfarast ‘be unsuccessful’, standa mein að ‘have problems with’, verða mein að ‘be harmful to sby’, verða sein förin ‘be late’, verða seint ‘be late’, vera til dauða ‘cause to die’ ... 7 Verbs of success/performance: endast ‘last, manage, be able to’, greiðast ‘go well’, takast ‘succeed’, vaxa afl ‘become strong’, verða gott til fjár ‘become rich’ ... 8 Verbs of ontological states: fara fjarri ‘be absurd’, fylgja ‘accompany’ halda ‘stay in the same way’, vera borgið ‘be safe’, vera farið ‘be in a certain way’, vera maklegt ‘be deserved’, vera ofurefli ‘be inferior to’, vera síður ‘be long’ ...



Chapter 6.  Old and Modern Icelandic 

9 Verbs of gain: berast ‘receive’, byrja ‘get wind’, dæmast ‘receive by ruling’, gagna ‘be of use’, vera hollur ‘be faithful’ ... 10 Verbs of personal properties: vaxa fjaðrir ‘grow feathers’ ... 11 Other happenstance verbs: fara ‘become of sby’, ganga út ‘become of sby’, ljúka ‘come to an end’, mega ‘happen to, become of ’, snúa ‘get turned’, verða að munni ‘accidentally speak’, verða vísa á munni ‘happen to speak out a poem’ ... The structure of the active Dative subject construction in Old Icelandic can thus be represented as in Figure 6.2. It is clear that the construction consists of two equally entrenched subconstructions, i.e. one denoting happenstance events and the other denoting events which I here label events of cognition/emotion. Given these semantic restrictions, the highest level of schematicity of the active Dative subject construction in Old Icelandic is fairly low, i.e. the construction cannot be assumed to exist at a higher level than the intermediate level of the basic human event type categories happenstance and cognition/emotion. The Old Icelandic Dative-(X)

HAPPENSTANCE

COGNITION/EMOTION

[Failure/mistake]

[Emotion]

[Success]

[Attitudes]

[Ontological states]

[Cognition]

[Gain]

[Bodily states]

[Personal properties]

[Obligation]

[Happenstance]

[Verb] ...

[Verb] ...

Figure 6.2  The active Dative subject construction in Old Icelandic

 Productivity

data do not motivate the postulation of a schematic level higher than this intermediate level of the lexicality–schematicity hierarchy. This intermediate level is, in other words, the highest level of the construction which can be assumed to be psychologically real in the minds of speakers. This fact is indicated with the dashed lines above the intermediate level in the hierarchy in Figure 6.2. Let us now compare the status of the active Dative subject construction in Old Icelandic with the status of the construction in the Modern Icelandic corpus. The thirty-three predicates only instantiate seven of the ten classes found in the Old Icelandic corpus: Experience-based predicates: 1. Verbs of emotion: finnast ‘feel’, hitna í hamsi ‘become angry’, líða ‘feel’, líka ‘like’, renna kalt vatn á milli skinns og hörunds ‘be terrified’, standa stuggur af ‘be scared of sth’, vera (ekki) sama ‘(don’t) mind’, vera leitt ‘be sad’, verða bilt við ‘be startled’, vera þægð í ‘be content with’, þykja ‘feel, think, seem’ ... 2. Verbs of attitudes: vera bágt til bjargar ‘be difficult to save’, vera nóg ‘be enough for sby’ ... 3. Verbs of cognition: detta í hug ‘get an idea’, fljúga í hug ‘get an idea’, sýnast ‘seem, appear’, vera hugleikið ‘be on sby’s mind’, verða hugsað til ‘remember, think about sby’, verða kunn ‘be known’, verða ljóst ‘realize’, vera ókunnur ‘be unknown’, virðast ‘seem, appear’ ... 4. Verbs of bodily states: verða kalt ‘be cold’, vera ómótt ‘feel nauseated’ ... 5. Verbs of perception: heyrast ‘hear’ ... Happenstance predicates: 6. Verbs of success/performance: takast ‘manage, succeed’ ... 7. Verbs of ontological states: fara fjarri ‘be not the case’, festa saman ‘be fastened together’, fylgja ‘accompany’, reynast ‘turn out to be’ ... 8. Miscellaneous: halla ‘decline’, ljúka ‘come to an end’, verða starsýnt á ‘stare’ ... Observe that of these thirty-three predicates, twenty-five are experience-based and only eight are happenstance predicates. Therefore, an analysis of the structure of the active Dative subject construction in the Modern Icelandic corpus, shown in Figure 6.3, reveals that the subconstruction denoting happenstance events does not make up a significant part of the construction in Modern Icelandic anymore. This means that the highest level of schematicity of the active Dative subject construction in Modern Icelandic is at the intermediate level of the basic human event type category cognition/emotion. In other words, only the subconstruction of cognition/emotion can be assumed to be psychologically real in the minds of speakers and not the happenstance construction.



Chapter 6.  Old and Modern Icelandic 

Dative-(X)

COGNITION/EMOTION

[Success]

[Emotion]

[Ontological states]

[Attitudes]

[Miscellaneous]

[Cognition] [Bodily states] [Perception]

[Verb] ...

[Verb] ...

Figure 6.3  The active Dative subject construction in Modern Icelandic

Table 6.9 summarizes the statistics for both language stages. The observed differences in distribution can occur by pure chance with a probability of less than twenty-five in a thousand (Pearson Chi-Square, p < 0.0141; Yates Chi-Square, p < 0.0253; Cramer’s V = 0.2466). As Cramér’s V shows that the association between the distribution of dative subject verbs across subconstructions in the two periods of Icelandic is as strong as 25%, this change in the structure of the Dative subject construction cannot be ignored. Table 6.9  Distribution of dative subject predicates across subconstructions in Old Icelandic and Modern Icelandic

Experience Happenstance

Old Icelandic

Modern Icelandic

33 33

25 8

66

33

 Productivity

As a consequence of the fact that the use of happenstance predicates is drastically reduced from Old Icelandic to Modern Icelandic, the cognition/emotion subconstruction has become the only dominant subconstruction of the construction. The active Dative subject construction has thus shifted from having two equally entrenched subconstructions, one denoting events of cognition/emotion and another denoting happenstance events, to only having the cognition/emotion subconstruction as central. This of course makes the Dative subject construction considerably more coherent semantically in Modern Icelandic than it was in Old Icelandic, and much more focused on experience-based predicates. And, indeed, the class of accusative subject predicates which undergoes dative substitution in Icelandic belongs to the experience-based class. While more research is needed to determine if this change was gradual in the history of Icelandic, and to determine the pace at which this change took place, it is still clear that the cognition/emotion subconstruction has become the most salient subconstruction of the active Dative subject construction in Modern Icelandic. This means that although the relative type frequency of the Dative subject construction has gone down from Old to Modern Icelandic, the construction has still become considerably more coherent semantically during modern times. On the present approach to syntactic productivity, as a function of type frequency and semantic coherence, it is expected that semantically more coherent constructions are more extendable than constructions that are semantically less coherent, given the same type frequency. The present approach thus explains why dative substitution only got started as late as in the latter part of the 19th century and not already in Old Icelandic, as it was not until that late that the semantic structure of the active Dative subject construction had shifted towards having the cognition/emotion subconstruction as central. 6.4

The “blended” construction in English, Swedish and Faroese

In the process of the breakdown of the case system, many Germanic languages have developed the so-called “blended” construction. This means that the nominative object of the former Dat-Nom construction, realized as Obl-Nom at this point in the development, turns up in the accusative, and the construction is hence realized as Obl-Obl. Examples (8–10) above from Middle English, Old Swedish and Modern Faroese illustrate this (repeated here for convenience): (8) for þi ðat him areowe ow for that that him.obl pity you.obl ‘so that he would pity you’ (Allen 1995: 238)

Middle English



Chapter 6.  Old and Modern Icelandic 

(9) Honom thykte sik wara j enom lystelikom stadh Old Swedish he.obl thought himself.obl be in a pleasing place ‘He felt as if he was in a pleasant place.’ (Falk 1997: 77) (10) Mær dámar væl hasa bókina. I.dat like well this book.acc ‘I like this book.’ (Barnes 1986: 33)

Modern Faroese

Allen reports that traditionally this blend has been regarded as an accident in the prevalent language material. She argues, however, and quite convincingly so in my opinion, that the blend deserves a better explanation than that. Given that the blend seems to arise independently in the Germanic languages, it certainly does not seem like an accident, but requires a systematic explanation. Allen herself argues that the blend emerges when her postulated case-marking hierarchy disappears, which on her account is a part of the change from lexical to structural case in English. However, she also argues that the case-marking hierarchy disappeared on the basis of the existence of the blend, thus her account is not independently motivated. Falk (1997: 77–78) also argues for Old Swedish that the blend represents structural case marking on objects. The problem with these analyses is that they do not address the question of why only the object should become “structurally” case marked and not the subject. In other words, why should oblique subjects retain their non-canonical case marking longer than nominative objects in a system which is otherwise in the process of falling apart? Since the case of the subject and the object does not change at the same time, why does not a Nom-Nom construction emerge instead of an Obl-Obl? This is a legitimate question as it is not a priori given that the object must change its case form first and the subject later. It is equally plausible and equally logical that the subject changes its case form before the object, but that does not happen in Middle English, Old Swedish and Modern Faroese. On the cognitive-functional CxG account laid out here, it is predicted that subjects will become nominative and objects accusative because of the high type frequency of the Nominative subject construction and the Accusative object construction, and that the case form higher in type frequency will resist the change longer because it is more entrenched in the system. As I have already outlined for Icelandic above, the nominative object is most prevalent in the Dat-Nom construction whereas dative subjects can select for different types of complements. The Dative subject construction comprises around 700 predicates (see the discussion around Table 6.1 above) while the Dat-Nom subconstruction comprises ca. 111 predicates in Modern Icelandic (Barðdal 2001b: 54–55). Hence, the Dative subject construction is much higher in type frequency than the Nominative object construction and is thereby predicted to maintain its

 Productivity

case form longer. That prediction is borne out for the “blended” construction in the history of the Germanic languages, which in turn sustains the validity of the present cognitive-functional CxG model. No explanation for this anomaly, however, has been suggested in the generative literature. 6.5

Historical productivity

The present cognitive-functional view of productivity, that productivity is a function of type frequency, coherence, and an inverse correlation between the two, predicts that the constructions highest in type frequency should gain in type frequency and that the constructions low in type frequency should lose in type frequency. The study presented in this chapter confirms this for Icelandic, German, Swedish and English, in that the case and argument structure constructions high in type frequency have gone up in Icelandic and German, and the low type frequency constructions have gone down in the two languages. The low type frequency constructions were also the ones to disappear first in the history of German, Swedish and English, partly in parallel with the breakdown of the case system, with the constructions lowest in type frequency disappearing first, then the one next-lowest in type frequency, and so forth. At the same time, it is also expected that low type frequency constructions may be maintained on the basis of a high degree of token frequency, a high degree of semantic coherence, or both. This is confirmed by the mild productivity of the Dative subject construction in the history of English, Swedish, German and Icelandic. As reported in 6.2 above, it is documented in the literature that the Dative subject construction has attracted experience-based predicates from other constructions in English and Swedish (Seefranz-Montag 1983, Allen 1995, Falk 1997. It is also documented in the literature that the Accusative and Dative subject constructions exchanged items in the history of German (Seefranz-Montag 1983), and that the Dative subject construction started attracting experience-based predicates from the Accusative subject construction in the latter part of the 19th century in Icelandic (Halldórsson 1982). These are clear cases of analogy on the basis of low-level verb-specific or verb-subclass-specific constructions. That is, changes in case marking of individual verbs or verb classes, like the change from genitive objects to either nominative objects or accusative objects, and the change from accusative subjects to dative subjects, are motivated not only by the type frequency of the relevant constructions, but also by the similarities in form and meaning between the relevant case and argument structure constructions. The study of Dative substitution in Icelandic has furthermore revealed that the distinction between absolute and relative type frequency, suggested in Section 4.3, is needed in order to unearth changes in the semantics of case and argument



Chapter 6.  Old and Modern Icelandic 

structure constructions that are caused by changes in language use. It is therefore not sufficient to elaborate only with absolute type frequencies, as these may be quite different from type frequencies found in real language use. The present study also suggests that the threshold for maintaining an intermediate-sized construction may lie around approximately 30% of the types for case and argument structure constructions. This is evident by the fact that the Nom-Dat construction, which is of this size, has been maintained in the history of Icelandic. All the other case and argument structure constructions in Icelandic, which are lower than that in type frequency, have gone down in their type frequency from Old to Modern Icelandic (except for the ones lowest in type frequency which have remained stable because the predicates instantiating them have not fallen into disuse). This, however, does not necessarily entail that the low type frequency constructions do not show signs of productivity. Whether or not a low type frequency construction is productive depends entirely on the semantic coherence found between the types occurring in the relevant construction. In the history of Icelandic, the Dative subject construction became productive in the latter part of the 19th century, due to its increase in semantic coherence, while the semantically low-coherent Nom-Gen construction, for instance, has not shown signs of productivity in the history of Icelandic. To conclude, the predictions of the present cognitive-functional CxG view of syntactic productivity, that syntactic productivity is a function of type frequency, semantic coherence, and an inverse correlation between the two, is not only borne out for synchronic structures but also for diachronic developments. 6.6

Summary

The present examination has revealed that the lexical-to-structural-case account of loss of case is not borne out for Icelandic, as there are documented changes in case assignment of verbs in both directions, i.e. lexical case seems to replace structural case and structural case seems to replace lexical case in the history of Icelandic. Rather, on the present cognitive-functional CxG approach to loss of morphological case, the development in four Germanic languages, Icelandic, Swedish, English and German, can be accounted for. The analysis is based on the fact that the original case and argument structure constructions were partly synonymous, and that logically there are two ways for languages to eliminate such synonymy: i. by high frequency constructions attracting verbs conventionally occurring in the less frequent constructions, thereby causing less frequent constructions to fall into disuse ii. by merging two synonymous case and argument structure constructions into one, with a subsequent loss of morphological case

 Productivity

English, Dutch and Mainland Scandinavian have taken the latter alternative, while Icelandic, Faroese and German have gone the former way. Also, German has developed much farther in this direction than Icelandic. I have also pointed out that there is a clear correlation between the rate of the vocabulary renewal and the development of case, which is predicted on the present cognitive-functional CxG approach, relying on the type frequency of the constructions in question. English leads the development with the most extensive borrowings of all the languages being considered here, beginning in the 11th century. Swedish has also been involved in much language contact, with massive Low German influence beginning in the late 13th century. German has not had the extensive replacement of the vocabulary found in both English and Swedish, but it has nevertheless been more influenced than Icelandic, which is the least influenced language of the four. Clearly, the faster the vocabulary is renewed, the sooner the high type frequency constructions increase in type frequency, and the sooner the low type frequency constructions decrease in their type frequency. Thus, on the present cognitive-functional CxG approach it is predicted that the language subject to most foreign influence will lead the development, and the language that has been least influenced will lag behind. That prediction is borne out for Germanic. A cognitive-functional CxG approach to dative substitution, the well-known change from accusative to dative case on subjects in Modern Icelandic, has also been suggested. It is generally assumed in the literature that accusative subjects have changed to dative subjects because of the semantic overlap between the two constructions. I have, moreover, argued that the order of this change, from Accusative to Dative, as opposed to Dative to Accusative is motivated by the fact that the Dative subject construction is much higher in type frequency than the Accusative subject construction. Given that, the question arises why this change takes place as late as in the latter part of the 19th century and not already in Old Icelandic. A systematic comparison of the predicates instantiating the construction across these two periods of Icelandic reveals that there were two salient subconstructions of the active Dative subject construction in Old Icelandic, i.e. one denoting events of cognition/emotion and the other denoting happenstance events. In Modern Icelandic, in contrast, only one of these, the one denoting events of cogntion/emotion, exists. This shift in the semantics of the active Dative subject construction from Old Icelandic to Modern Icelandic, due to a reduction in Icelandic speakers’ use of dative subject happenstance predicates, and hence the increased salience of the cognition/emotion subconstruction, explains why the onset of dative substitution takes place so late in the history of Icelandic. I have also discussed the emergence of the “blended” construction in the history of the Germanic languages, a construction which is formally a mixture of the Dat-Nom and Nom-Acc constructions. Hitherto, the explanations offered in the



Chapter 6.  Old and Modern Icelandic 

literature for the blend have either assumed that it is an accident in the prevalent language material, or that it represents a change from lexical to structural case marking. These analyses, however, have not offered any systematic explanation as to why the case of the object should be replaced first instead of the subject case. I have proposed a cognitive-functional CxG analysis which predicts that the more entrenched argument, i.e. the one highest in type frequency, will resist the change longer. This prediction is borne out for the blend in the history of Germanic, as the dative is more entrenched as a subject case than the nominative is as an object case which in turn sustains the validity of the present cognitive-functional CxG approach. This study of historical productivity has furthermore revealed that the predictions of the present cognitive-functional CxG approach to syntactic productivity, as a function of type frequency, semantic coherence, and an inverse correlation between the two, is borne out, not only for synchronic structures but also for historical developments. The case and argument structure constructions highest in type frequency are expected to increase in type frequency, as they are the semantically most open constructions, and hence most extendible to new verbs entering the language. This is what has happened to the Nom-Acc construction in the history of Icelandic. In the same vein, a construction of intermediate size can be expected to maintain its status as a construction of intermediate type frequency, with existing verbs falling into disuse and new being attracted to the construction, given an intermediate degree of schematicity of the construction. This is exemplified by the development of the Nom-Dat construction in the history of Icelandic. Finally, case and argument structure constructions low in type frequency are expected to go down in type frequency, because of their restricted semantics. These low type frequency constructions, however can still be extended to new or existing verbs, through analogy with either verb-specific or verb-class-specific constructions, given a high enough degree of semantic similarity between the relevant items. This is what has happened in the case of dative substitution in Icelandic, during the latter part of the 19th century. Notes 1 All existing Old Swedish examples of thykia ‘feel, seem’ together with a small clause are ambiguous between a nominative and accusative form of the pronoun of the lower argument, thus it is not self evident that thykia was a Dat-Nom verb when selecting for small clauses in Old Swedish. However, since the cognate of thykia in the earliest period of the other Germanic languages was a Dat-Nom verb it is reasonable to believe that this is a common Germanic inheritance, which has already been lost, or is in the process of being lost, at the time of the oldest Swedish examples.

 Productivity 2 Maling counts 140 dative object predicates in German. However, around 40 of these are DatNom predicates where the dative has traditionally been regarded as an object (for arguments against an object analysis of the dative in Dat-Nom constructions, cf. Eythórsson and Barðdal 2005 and Barðdal 2006a). I exclude these 40 from the present number of dative object predicates as I am mainly concerned with Nom-Dat predicates here and not Dat-Nom predicates. 3 Jóhannes G. Jónsson (p.c.) reports that he has encountered three more verbs in the DatGen construction in Old Icelandic: bætast ‘get better’, fá ‘get’ and ljá ‘lend’.

chapter 7

Synthesis An investigation of the different senses and usages of the terms productive and productivity in the linguistic literature has uncovered a wide-ranged nonconformity in scholars’ use of this terminology, including formal definitions, as well as lax, informal uses of the term. A systematic analysis of a non-exhaustive list of nineteen different senses shows that the metaconcept of productivity is made up of three subconcepts, the concepts of generality, regularity and extensibility. Figure 7.1 (Section 2.2) shows the interrelations between the three subconcepts. Generality can be regarded as entailing extensibility, while regularity and extensibility do not necessarily entail each other but are concomitant in the largest, most productive categories in the world’s languages. This, I believe, is the reason that the concept of productivity is structured as it is, because of local situations where factual entailments and contextually invited inferences have contributed to the beginning of meaning change on a semantic change continuum, ultimately resulting in semantic polysemy. The nineteen senses found for productive and productivity are related to the subconcepts in various ways: Some of them are elaborations or further specifications of the subconcepts, some are entailed, some are co-occurrence relations, some are metaphorical relations, while yet others have become synonymous with productive and productivity through antonymy relations. The unraveling of the metaconcept of productivity, and the identification of the subconcepts, is a prerequisite for a fruitful discussion of productivity in linguistics. Generality Ent

Co

Regularity

Extensibility Co

Figure 7.1  The three subconcepts of productivity

 Productivity

However, not only are there several different notions of productivity found in the morphological literature, there are also at least two notions of syntactic productivity around. One stems from the generative literature, i.e. speakers’ ability to understand and generate new sentences that they have never encountered before (Chomsky 1965). The other refers to the ability of an argument structure construction to attract new or existing verbs. The first notion of syntactic productivity is an elaboration of the regularity concept, as it captures regular performance, while the second is an elaboration of the extensibility concept. The subconcept of productivity investigated in this work is the extensibility concept, as the research questions posed here relate to the ability of case and argument structure constructions to attract new or existing verbs. The view of productivity advocated here is that syntactic productivity is a function of type frequency, semantic coherence, and the inverse correlation between these two factors. By coherence I mean the internal consistency found between the relevant items. This view of productivity can be illustrated as in Figure 7.2 (Section 2.4) where the vertical axis represents type frequency and the horizontal axis represents semantic coherence. The upper left-most sphere of Figure 7.2 represents high type frequency and thus a high degree of schematicity, while the lower right-most sphere represents low type frequency and a high degree of specificity. The inverse correlation between these two factors is depicted as a cline of productivity, located in between these two extreme points in Figure 7.2. Type Frequency High Regularity–Generality–Open Schema

Different Degrees of Productivity

Analogy

Semantic Coherence Low

Figure 7.2  The productivity cline

High



Chapter 7.  Synthesis 

The most productive constructions are located at the upper left-most corner of the productivity cline. These are the constructions which are highest in type frequency and show the least semantic coherence between their items. Or, stated differently, they are high in type frequency and high in schematicity. In contrast, the constructions lowest in type frequency can range from exhibiting a low degree of semantic coherence to exhibiting a high degree of semantic coherence. If they show low degree of semantic coherence, they are located on the left side at the bottom of the figure, while if they show a high degree of semantic coherence, they are located on the right side at the bottom of the figure, at or closer to the productivity cline. In other words, constructions high in type frequency are always high in schematicity, while constructions low in type frequency can vary in their semantic coherence. This in turn means that constructions high in type frequency are always productive, i.e. extendable, while the ones low in type frequency vary in their productivity depending on their degree of semantic coherence. At the bottom of the productivity cline, we get the lowest possible type frequency, i.e. only one, and the highest possible semantic coherence, i.e. full synonymy. Extensibility at this level is well known in the literature, where it goes under the label analogy. Hence, on the present cognitive-functional CxG view of productivity, full productivity and analogy are two sides of the same coin, not different in ontological status, only different in degree. The present view of productivity also entails that productivity is gradient. At the top of the productivity cline, we get full productivity, at the bottom of the cline, we get low productivity, and at the intermediate points on the cline, we get different degrees of productivity. All the categories and constructions located at the cline are productive, although their productivity domains vary according to their location on the cline. The categories and constructions at the upper end of the cline exhibit a larger productivity domain than the categories and constructions located at the lower end of the cline. In contrast, the categories and constructions below the cline, i.e. in the left lower sphere of the figure, are non-productive because they are not high enough in type frequency and not coherent enough semantically. And the categories and constructions which should be located at the right higher sphere of the figure, above the productivity cline, are the ones high in type frequency, exhibiting a low degree of schematicity (i.e. a high degree of specificity). The only problem is that such constructions do not exist, as high type frequency takes a high degree of schematicity as its corollary, and not a low degree of schematicity (cf. Section 2.5). Observe also that the constructions located at the top of the productivity cline, i.e. the most productive constructions, are also the constructions that are most general and most regular, as already mentioned above. This is a consequence of the fact that a high degree of extensibility is concomitant with a high degree of

 Productivity

generality and regularity, as the constructions highest in type frequency are usually also the constructions which are the most general ones and the most regular ones in the world’s languages. Hence, the generality and the regularity concepts can in fact be derived from the extensibility concept. This means that the extensibility concept is of primary theoretical status, while the generality and the regularity concepts are secondary. Regularity, however, is also found in constructions that do not exist at a high schematic level, like with strong verbs in English (cf. Section 2.2). Moreover, a description of language and grammar in terms of type frequency and semantic coherence, i.e. a description of language which assumes different levels of schematicity, predicts the degree of productivity for each construction in the language. As the highest level of schematicity specifies the productivity domain of each construction, specific theories of productivity are not needed. Productivity can, as stated, be derived from the highest level of schematicity of each syntactic construction. I have also argued against the view that it is the most entrenched level of a construction that is the best predictor of productivity, as is claimed in the traditional cognitive-functional linguistics literature. A construction’s most entrenched level is derived from the interaction between its type and token frequency, and it disregards the degree of semantic coherence found for that construction. In contrast, a construction’s highest level of schematicity is derived from its type frequency and semantic coherence, and this highest level varies from construction to construction, depending on the interaction between these two factors. The identification of the highest level of each construction must thus be done on the basis of empirical facts, like the verb classes occurring in an argument structure construction, in order for this level to properly represent a psycholinguistically real level in the minds of speakers. An investigation of the three subconstructions of the Nominative subject construction in Modern Icelandic, i.e. the Nom-Acc, Nom-Dat and Nom-Gen constructions, of both the type frequency and the semantics of the subconstructions, reveals that the highest level of schematicity varies for the three constructions (Chapter 3). The analysis is based on occurrences in a corpus of Modern Icelandic texts, of the size of 40,000 running words, composed of six different genres, five written and one spoken genre. The corpus contains 303 Nom-Acc predicates, 188 Nom-Dat predicates and twenty-four Nom-Gen predicates. The proportions between the three subconstructions in the text corpus mirrors the proportions found between them in a dictionary count which contains four times as many predicates as the corpus. The 303 Nom-Acc predicates can be divided into forty-six different narrowly-circumscribed semantic verb classes, the 188 Nom-Dat predicates into thirty-two verb classes and the twenty-four Nom-Gen predicates into four verb classes. The highest level of schematicity also varies for the three subconstructions,



Chapter 7.  Synthesis 

in that the Nom-Acc construction exists at an abstract, schematic level, the NomDat construction exists at a lower, basic-event-type level, while the Nom-Gen construction only exists at a verb-specific and verb-subclass-specific level. These different levels of schematicity of the three subconstructions of the Nominative subject construction also make predictions about the degree of productivity of these constructions. The Nom-Acc construction is expected to be most productive, as it is an open schema, the Nom-Dat construction is expected to be less productive than the Nom-Acc construction, as it does not exist above the level of basic human event types, and the Nom-Gen constructions is not expected to be productive at all. These predictions are borne out, as is evident from the behavior of recently borrowed verbs in Icelandic, mostly from the area of information technology. Of 107 borrowed verbs, 64% are assigned the Nom-Acc construction and 36% are assigned the Nom-Dat construction by Icelandic speakers. No borrowed verb is assigned the Nom-Gen construction. These frequencies are almost exact reflections of the proportions between existing Nom-Acc and NomDat verbs in Modern Icelandic. A closer inspection of the syntactic behavior of the individual borrowed verbs reveals that they are assigned case and argument structure constructions, by Icelandic speakers, on the basis of low-level verb-specific or verb-subclass-specific constructions. This is evident from the fact that the distribution of case and argument structure constructions of these borrowings across the available constructions in Icelandic overlaps with the case and argument structure constructions of existing Nom-Acc and Nom-Dat verbs. If the assignment of the Nom-Acc construction to borrowed verbs had taken place independently of existing Nom-Acc verbs, perhaps as default assignment as argued in the generative tradition, one would expect the Nom-Acc construction to be assigned evenly to verbs and verb classes irrespective of their semantics. This, however, is not the case. The borrowed verbs that are assigned the Nom-Acc construction fall into the narrowly-circumscribed semantic classes of Nom-Acc predicates. It also turns out that a majority of the borrowed verbs are assigned the same case and argument structure construction as their synonymous translational equivalents in Icelandic. The remaining borrowed predicates are clearly attracted by small clusters of verbs with the same or a similar meaning. This means that even though the highest level of schematicity of each construction accurately predicts its degree of productivity and its productivity domain, the actual extensions are in fact based on low-level analogical extensions. One argument against the assumption that it is a construction’s most entrenched level that is the best predictor of productivity stems from the productivity of the Nom-Dat construction in Icelandic. It is clear from the data that it is the Caused-motion construction that represents the most entrenched level of the Nom-Dat construction, meaning that new Nom-Dat verbs should be confined to

 Productivity

caused-motion verbs or verbs used in the Caused-motion construction. This, however, is not borne out for borrowed verbs, as the borrowed verbs assigned the Nom-Dat case and argument structure construction are spread across the semantic domain of the Nom-Dat construction, not restricted to caused-motion. It has also been argued in the literature that it is type frequency, not token frequency, which contributes to productivity, as entrenched, or lexicalized, tokens are not regarded as contributing to the entrenchment of their higher level constructions. What is more, it is even argued that high token frequency detracts from productivity. However, if full productivity and analogy are two sides of the productivity coin, then clearly token frequency should be relevant for analogical extensions. My hypothesis is that the importance of token frequency stands in an inverse correlation with the importance of type frequency. This can be graphically represented as in Figure 7.3 (Section 3.2.2) which suggests that high token frequency is more important for lower-level verb-specific analogical extensions than for the productivity of high type frequency constructions, and the lower down on the productivity cline, the more important it is, as that is where the low-level analogical extensions are located on the cline. Type frequency is an indicator of the highest level of schematicity each construction exists at, and hence an indicator of the semantic scope of the construction and its productivity domain, but token frequency is an important psycholinguistic factor singling out model items for speakers when they extend low-level constructions. Token frequency is thus of utmost important for productivity, contra what has hitherto been assumed in the traditional cognitive-functional linguistic literature. The results of a nonce-probe experiment on the case and argument structure assignment of nonce verbs by Icelandic speakers, reported in Chapter 4, are in partial agreement with the results of the investigation on borrowed verbs. The experiment was carried out with forty Icelandic speakers, twenty children and twenty adults, and it involved five nonce verbs, which were presented to the participants as synonymous to already existing Icelandic verbs. Two of these were Nom-Acc verbs, one was a Nom-Dat verb and two were dative subject verbs. Two of the verbs were agentive and three were emotion verbs. The nonce verbs presented as synonymous to the Nom-Acc verbs were generally assigned the NomAcc construction. The nonce verb presented as synonymous to the Nom-Dat verb was assigned the Nom-Dat construction by 17.5% of the participants and was assigned the Nom-Acc construction by the remaining participants. The verbs presented as synonymous to the dative subject verbs were assigned dative subjects by 42–48% of the participants and assigned nominative subjects by the others.



Chapter 7.  Synthesis 

Type Frequency High

Token Frequency Low

High

Figure 7.3  The inverse correlation between type and token frequency for productivity

The findings of this nonce-probe experiment show that Icelandic speakers assign case and argument structure constructions to unknown verbs either on the basis of a synonymous low-level verb-specific construction, as with the dative subject verbs, or on the basis of a high-level abstract schematic construction which is semantically unrestricted, as with the Nom-Acc verbs. This is suggested by the fact that several participants who assigned Nom-Acc categorically to all the nonce verbs simply rattled off their responses in no time, while the other participants who assigned dative subjects to the nonce verbs that were synonymous with Icelandic dative subject verbs took an unusually long time to think through their answers. These differences in case assignment stand in a stark contrast to the assignment of case and argument structure of the borrowed verbs, which were clearly not assigned case on the basis of a high-level abstract schematic construction, but only on the basis of the lower-level verb-specific or verb-subclass-specific constructions. Another important issue which also emerged from this experiment is that the notion of type frequency needs to be elucidated and properly defined. That is, type frequency can be defined in terms of absolute type frequency, i.e. the type frequency obtained when all the items occurring in a construction are counted in a dictionary. Or it can be defined in terms of relative type frequency, i.e. the type frequency obtained when all the items occurring in a construction are counted in a representative corpus. These two different notions of type frequency will

 Productivity

not necessarily make the same predictions about productivity. The Nom-Dat construction and the Dative subject construction, for instance, display similar absolute type frequency, i.e. ca. 700 in Icelandic. However, corpus counts reveal that the Nom-Dat construction is used 3.5 times more often than the Dative subject construction. An analysis of the semantics of the two constructions, based on the corpus occurrences, shows that the Dative subject construction is much more restricted semantically than the Nom-Dat construction. The Dative subject construction cannot be assumed to exist at a higher level than a verb-class-specific level, while the Nom-Dat construction exists at a basic-event-type level, much higher in the lexicality–schematicity hierarchy. The productivity of the two constructions is also predicted to vary accordingly. Such a prediction is borne out by the fact that only the Nom-Dat construction, and not the Dative subject construction, was found amongst the borrowed verbs, reported on above. The nonce-probe experiment, reported on in Chapter 4, showed that speakers assign case and argument structure constructions either on the basis of low-level analogy or on the basis of a high-level abstract schematic construction. In contrast, a questionnaire survey of three new verbs of instrument of communication, faxa ‘fax’, (e)meila ‘email’ and smsa ‘text’, suggests a competition between extensions based on low-level analogy and extensions based on a verb-subclass specific construction. This questionnaire survey was carried out in Iceland, with thirtytwo adolescents and adults at the age 14–85 years, and it involved the occurrence of these three new verbs in the Transfer construction with an accusative object, the Caused-motion construction with a dative object, and the Ditransitive construction. The participants were presented with the three verbs in the three constructions, randomly ordered, and asked to judge them on a 7-point scale. All three verbs were most accepted in the Transfer construction, although their occurrence in both the Caused-motion and the Ditransitive constructions were around or above the acceptability threshold at mean score 4. The occurrence of the three new verbs of instrument of communication in the Transfer construction is expected on analogy with the verb senda ‘send’, which conventionally occurs in the Ditransitive and the Transfer construction with an accusative object. The occurrence of these verbs in the Caused-motion construction, however, is unexpected. It can only be explained by assuming that speakers assigned case and argument structure construction to these new verbs on the basis of a verb-subclass-specific construction of caused-motion verbs with a dative object. The Caused-motion construction is much higher in type frequency than the Transfer construction, and its degree of semantic coherence is high. This is shown with the smaller star in Figure 7.4 (Section 5.4) for the verb senda ‘send’ of the Transfer construction which is located at the right-most bottom level of the pro-



Chapter 7.  Synthesis 

ductivity cline, and with the larger star for the Caused-motion construction which is located higher up on the cline, as it is higher in type frequency. The questionnaire survey thus reveals a competition between one verb as a model verb and a cluster of verbs as a model for the extension of the Transfer vs. the Caused-motion construction, respectively. The two constructions differ regarding their levels of schematicity and their productivity also varies accordingly. The productivity of the Transfer construction is based on the low-level verb-specific construction of senda ‘senda’, while the Caused-motion construction exists as a verb-subclass-specific construction much higher in type frequency and semantic coherence. This case study thus shows, once again, that the boundary between analogy and productivity cannot be maintained, as both give rise to extensibility, only to varying degrees. As such, “analogical extension” and “full productivity” may be regarded as two sides of the same coin. The questionnaire survey also uncovered a difference in the distribution of the acceptability of the Transfer and the Caused-motion constructions with verbs of instrument of communication, depending on age: It turns out that younger speakers accept these verbs in both the Transfer and the Caused-motion construction while older speakers prefer them in the Transfer construction. This suggests that the Caused-motion construction, with a dative object, may be gaining in productivity in Icelandic. Type Frequency High Full productivity

Different Degrees of Productivity

Analogy

Semantic Coherence Low

High

Figure 7.4  The productivity of a verb-specific construction vs. a verb-subclass-specific construction

 Productivity

The present approach to syntactic productivity also makes certain predictions about historical productivity, i.e. the productivity of argument structure constructions over time. It predicts that the argument structure constructions highest in type frequency will attract the most new verbs, while the argument structure constructions lowest in type frequency will not attract new verbs. If, however, low type frequency constructions attract new verbs, this will be on the basis of a high degree of semantic overlap, i.e. on the basis of analogy. It is also hypothesized here that an alignment system, like that in Germanic where the case and argument structure constructions are partly synonymous, will strive towards the elimination of this synonymy. This can either happen through a merger of case and argument structure constructions, and subsequent loss of morphological case, or through one case and argument structure construction taking over the functions (in this case, verbs), of another. This latter alternative entails the preservation of high type frequency constructions at the cost of low type frequency constructions. Both these developments are documented in the history of the Germanic languages, dividing them into two groups: those that have lost morphological case, like English and Swedish, and those that have maintained morphological case, like Icelandic and German. Both Icelandic and German show a reduction in the alignment system in that the Germanic case and argument structure constructions lowest in type frequency have been lost, while the Germanic case and argument structure constructions highest in type frequency have been preserved. In fact, a comparison between the developments in these four Germanic languages reveals the same internal order of events in all languages, namely that the construction lowest in type frequency disappears first, then the one next lowest in type frequency, etc., while the constructions highest in type frequency are preserved and strengthened. This development also coincides in time with massive language contact in Europe. The languages being exposed to most contact have led the development while the languages exposed to least contact have lagged behind. This is unsurprising, as the languages being exposed to most contact have also borrowed most vocabulary, which favors high type frequency constructions, thus contributing to their productivity, gradually causing low type frequency constructions to fall into disuse. This is the reason why situations of massive language contact speed up the trimming of alignment systems where the case and argument structure constructions are (partly) synonymous. The present usage-based cognitive-functional CxG approach also explains two additional anomalies in the literature on case marking and argument structure constructions: (a) the emergence of the “blended” construction in English, Swedish and Faroese, and (b) the late onset of dative substitution in Icelandic. The blended Dat-Acc construction, documented in the history of English, Swedish and Faroese, is not original in the Germanic languages. It has developed from the Dat-Nom



Chapter 7.  Synthesis 

construction but acquired the Acc from the Nom-Acc construction. The anomaly, not previously addressed in the literature, relates to the fact that the object argument changes its case marking first, while the subject argument does not change its case marking until much later. This raises the question of why the object would change its case marking first, and not the subject? On the present approach it is expected that the more entrenched argument will be preserved longer than the less entrenched argument. As nominative objects occur more or less only together with dative subjects, they are low in type frequency, while the type frequency of dative subjects is much higher, as they can occur with all kinds of complements. That is, Dat-Nom predicates are only a small subset of all dative subject predicates. The predictions of the present usage-based cognitive-functional CxG approach are thus borne out. No explanation of this fact has been suggested in the literature before. Dative substitution, i.e. the change from accusative subjects to dative subjects with a small class of so-called impersonal predicates, has received abundant attention in the literature, and the consensus is that the change is motivated by the semantics of these predicates. That is, the verbs that have gone from accusative to dative subjects are experience-based predicates, in conformity with the fact that the prototypical dative subject verb is experience based. Two questions arise at this point: (a) Why do accusative subject verbs change to dative subject verbs, and not vice versa, since both are experience-based?, and (b) Why does the change not start until the late 19th century? The answer to the first question is predicted by the present usage-based cognitive-functional CxG approach, namely that the type frequency of the Dative subject constructions is much higher than the type frequency of the Accusative subject construction. Hence, it is the Dative subject construction that attracts members from the Accusative subject construction, and not vice versa. The answer to the latter question relates to changes in the semantic structure of the Dative subject construction from Old Icelandic to Modern Icelandic. In Old Icelandic two clearly distinct equally entrenched subconstructions of the Dative subject construction existed, one denoting events of emotion/cognition and another denoting happenstance events. In Modern Icelandic, however, only the first subconstruction can be argued to exist as a higher-level construction, while the few happenstance predicates that still remain in use exist only as low-level verb-specific constructions. This change in language use and thus in the semantic structure of the Dative subject construction has therefore contributed to increased semantic coherence of the construction in Modern Icelandic, which in turn explains why the onset of dative substitution is as late as the late 19th century; an increase in the semantic coherence of the construction will contribute to its increased productivity. The approach to productivity presented and argued for in the previous chapters of this book, i.e. as a function of type frequency, coherence, and the inverse

 Productivity

correlation between the two, not only accounts for different degrees of productivity, it also gives a unified account of the notions of full productivity and analogy. The two are here regarded as two sides of the same coin, full productivity being observed in categories that are high in type frequency and low in semantic coherence, while analogy is observed in categories that are low in type frequency and high in semantic coherence. This follows from a definition of productivity based on the extensibility concept, not the regularity or the generality concepts, which in turn constitutes an argument for the theoretical primacy of the extensibility concept. Syntactic productivity can, furthermore, be derived from the assumptions of the present cognitive-functional CxG approach to language and grammar, namely that constructions exist at different levels of schematicity, depending on their type frequency and semantic coherence. Given these assumptions, specific hypotheses on productivity can simply be dispensed with from linguistic theories. The present approach to language and grammar, and hence to syntactic productivity, has here been successfully applied to case and argument structure constructions in Old and Modern Icelandic, demonstrating beyond doubt its superiority over generative analyses which rely on the dichotomy between lexical and structural case and the distinction of lexical case into thematic and idiosyncratic case.

References Allen, Cynthia L. 1995. Case Marking and Reanalysis: Grammatical Relations from Old to Early Modern English. Oxford: Clarendon Press. Anshen, Frank and Mark Aronoff. 1989. “Morphological productivity, word frequency and the Oxford English Dictionary”. In Language Change and Variation, R.W. Fasold and D. Schiffrin (eds), 197–202. Amsterdam & Philadelphia: John Benjamins. Aronoff, Mark. 1976. Word Formation in Generative Grammar. Cambridge, MA: MIT Press. Aronoff, Mark. 1983. “Potential words, actual words, productivity and frequency”. Proceedings of the 13th International Congress of Linguists. Oslo: Oslo University Press. Askedal, John Ole. 2001. “‘Oblique subjects,’ structural and lexical case marking: Some thoughts on case assignment in North Germanic and German”. Grammatical Relations and Change, J. T. Faarlund (ed), 67–97. Amsterdam & Philadelphia: John Benjamins. Baayen, Harald. 1992. “Quantitative aspects of morphological productivity”. In Yearbook of Morphology 1991, G. Booij and J. van Marle (eds), 109–49. Dordrecht: Kluwer. Baayen, Harald. 1993. “On frequency, transparency and productivity”. In Yearbook of Morphology 1992, G. Booij and J. van Marle (eds), 181–208. Dordrecht: Kluwer. Baayen, R. Harald. 2001. Word Frequency Distributions [Text, Speech and Language Technology 18]. Dordrecht: Kluwer. Baayen, Harald and Rochelle Lieber. 1991. “Productivity and English derivation: A corpusbased study”. Linguistics 29: 801–44. Barðdal, Jóhanna. 1992. Um fallmörkun sagna í íslensku: Þolfall eða þágufall [On case assignment of verbs in Icelandic: Accusative or dative]. Unpublished manuscript, University of Iceland. Barðdal, Jóhanna. 1993. Accusative and dative case of objects of some transitive verbs in Icelandic and the semantic distinction between them. In Flyktförsök: Kalasbok till Christer Platzack på femtioårsdagen 18 november 1993, från doktorander och dylika, 1–13. Lund. Barðdal, Jóhanna. 1998. “Argument structure, syntactic structure and morphological case of the impersonal construction in the history of Scandinavian”. Scripta Islandica 49: 21–33. Barðdal, Jóhanna. 1999a. “Case in Icelandic — A construction grammar approach”. TijdSchrift voor Skandinavistiek 20(2): 65–100. Barðdal, Jóhanna. 1999b. “Case and argument structure of some loan verbs in 15th century Icelandic”. In Alla tiders språk: En vänskrift till Gertrud Pettersson november 1999 [Lunda­ studier i Nordisk Språkvetenskap A 55], I. Haskå and C. Sandqvist (eds), 9–23. Lund: Institutionen för nordiska språk. Barðdal, Jóhanna. 2000. “Case assignment of nonce verbs in Icelandic”. SKY Journal of Linguistics 13: 7–27. Barðdal, Jóhanna. 2001a. Case in Icelandic — A Synchronic, Diachronic and Comparative Approach [Lundastudier i Nordisk Språkvetenskap A 57]. Lund: Department of Scandinavian Languages.

 Productivity Barðdal, Jóhanna. 2001b. “The perplexity of Dat-Nom verbs in Icelandic”. Nordic Journal of Linguistics 24: 47–70. Barðdal, Jóhanna. 2001c. “The role of thematic roles in constructions? Evidence from the Icelandic inchoative.” In Proceedings of the 18th Scandinavian Conference of Linguistics 2000, A. Holmer, J-O. Svantesson and Å. Viberg (eds), 127–37. Lund: Department of Linguistics. Barðdal, Jóhanna. 2003. “Case and argument structure of novel verbs of communication in Icelandic”. In Grammar in Focus: Festschrift for Christer Platzack 18 November 2003, Vol. II, L.-O. Delsing, C. Falk, G. Josefsson and H.Á. Sigurðsson (eds), 25–35. Lund: Department of Scandinavian Languages. Barðdal, Jóhanna. 2004. “The semantics of the impersonal construction in Icelandic, German and Faroese: Beyond thematic roles”. In Focus on Germanic Typology [Studia Typologica 6], W. Abraham (ed), 101–30. Berlin: Akademie Verlag. Barðdal, Jóhanna. 2006a. “Construction-specific properties of syntactic subjects in Icelandic and German”. Cognitive Linguistics 17(1): 39–106. Barðdal, Jóhanna. 2006b. ”Predicting the productivity of argument structure constructions”. To appear in Berkeley Linguistics Society 32. Barðdal, Jóhanna. 2007. “The semantic and lexical range of the ditransitive construction in the history of (North) Germanic”. Functions of Language 14(1): 9–30. Barðdal, Jóhanna. 2009a. “Lexical vs. structural case: A false dichotomy”. To appear in Empirical Approaches to Morphological Case, C. Donohue and J. Barðdal (eds). A special issue of Morphology. Barðdal, Jóhanna. 2009b. “The development of case in Germanic”. In The Role of Semantic, Pragmatic and Discourse Factors in the Development of Case, J. Barðdal and S. Chelliah (eds) 123–160. Amsterdam & Philadelphia: John Benjamins. Barðdal, Jóhanna and Thórhallur Eythórsson. 2006. “Control infinitives and case in Germanic: ‘Performance error’ or marginally acceptable constructions”. In Case, Valency and Transitivity, L. Kulikov, A. Malchukov and P. de Swart (eds), 147–77. Amsterdam: John Benjamins. Barðdal, Jóhanna, Kristian E. Kristoffersen, and Andreas Sveen. 2007. “The ditransitive in WestScandinavian: A family of constructions”. Presented at 10th International Cognitive Linguistics Conference in Krakow, Poland. Barðdal, Jóhanna and Leonid Kulikov. 2008. “Case in decline”. In The Oxford Handbook of Case, A. Malchukov and A. Spencer (eds), 470–79. Oxford: Oxford University Press. Barðdal, Jóhanna and Valéria Molnár. 2003. “The passive in Icelandic – Compared to Mainland Scandinavian”. In Structures of Focus and Grammatical Relations, J. Hetland and V. Molnár (eds), 231–60. Tübingen: Niemeyer. Barnes, Michael P. 1986. “Subject, nominative and oblique case in Faroese”. Scripta Islandica 37: 13–46. Basbøll, Hans. 2003. “Prosody, productivity and word structure: The stød pattern of Modern Danish”. Nordic Journal of Linguistics 26(1): 5–44. Bauer, Laurie. 2001. Morphological Productivity [Cambridge Studies in Linguistics 95]. Cambridge: Cambridge University Press. Berg-Olsen, Sturla. 2009. “Lacking in Latvian: Case variation from a cognitive and constructional perspective”. In The Role of Semantic, Pragmatic and Discourse Factors in the Development of Case, J. Barðdal and S. Chelliah (eds) 181–202. Amsterdam & Philadelphia: John Benjamins.



References  Bernódusson, Helgi. 1982. Ópersónulegar setningar [Impersonal sentences]. Master’s thesis, University of Iceland. Boas, Hans C. 2003. A Constructional Approach to Resultatives. Stanford: CSLI Publications. Boas, Hans C. 2005. “Determining the productivity of resultative constructions: A reply to Goldberg and Jackendoff ”. Language 81(2): 448–64. Bock, Kathryn and Zenzi M. Griffin. 2000. “The persistence of structural priming: Transient activation or implicit learning?” Journal of Experimental Psychology: General 129: 177–92. Böðvarsson, Árni (ed). 1988. Íslensk orðabók [Icelandic Dictionary]. 2nd Edition. Reykjavík: Bókaútgáfa Menningarsjóðs. Braine, Martin D. S., Ruth E. Brody, M. Fish Shalom, Mara J. Weisberger, and Monica Blum. 1990. “Can children use a verb without exposure to its argument structure?” Journal of Child Language 17: 313–42. Bybee, Joan L. 1985. Morphology: A Study of the Relation between Meaning and Form [Typological Studies in Language 9]. Amsterdam & Philadelphia: John Benjamins. Bybee, Joan. 1995. “Regular morphology and the lexicon”. Language and Cognitive Processes 10(5): 425–55. Bybee, Joan. 1998. “The emergent lexicon”. Chicago Linguistic Society 34: 421–35. Bybee, Joan. 2001. Phonology and Language Use [Cambridge Studies in Linguistics 94]. Cambridge: Cambridge University Press. Bybee, Joan L. and Carol Lynn Moder. 1983. “Morphological classes as natural categories”. Language 59: 251–70. Bybee, Joan L. and Dan I. Slobin. 1982. “Rules and schemas in the development and use of English past tense”. Language 58: 265–89. Bybee, Joan and Sandra Thompson. 1997. “Three frequency effects in syntax”. Berkeley Linguistic Society 23: 378–88. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Chomsky, Noam and Howard Lasnik. 1995. “The theory of principles and parameters”. In The Minimalist Program, 13–127. Cambridge: MIT Press. Clausner, Timothy C. and William Croft. 1997. “Productivity and schematicity in metaphors”. Cognitive Science 21(3): 247–82. Clausner, Timothy C. 2002. “How conceptual metaphors are productive of spatial-graphical expressions”. In Proceedings of the 24th Annual Conference of the Cognitive Science Society, W.D. Gray and C.D. Shunn (eds), 208–13. Mahwah, NJ: Erlbaum. Croft, William. 2000. Explaining Language Change: An Evolutionary Approach. Harlow: Longman. Croft, William. 2001. Radical Construction Grammar: Syntactic Theory in Typological Perspective. Oxford: Oxford University Press. Croft, William. 2003. “Lexical rules vs. constructions: A false dichotomy”. In Motivation in Language: Studies in Honour of Günter Radden, H.B. Cuyckens, Th. Berg, R. Dirven, and Kl.-U. Panther (eds), 49–68. Amsterdam & Philadelphia: John Benjamins. Croft, William, Jóhanna Barðdal, Willem Hollmann, Maike Nielsen, Violeta Sotirova, and Chiaki Taoka. 2001. “Discriminating verb meaning: The case of transfer verbs”. Presented at the LAGB Autumn meeting in Reading, U.K. Croft, William and Alan D. Cruse. 2004. Cognitive Linguistics. Cambridge: Cambridge University Press.

 Productivity Delsing, Lars-Olof. 1991. “Om genitivens utveckling i fornsvenskan [On the Development of the Genitive in Old Swedish]”. In Studier i svensk språkhistoria 2, S.-G. Malmgren and B. Ralph (eds), 12–30. Acta Universitatis Gothoburgensis, Göteborg, Delsing, Lars-Olof. 2002. “Svenskt foge-s” [Linking ‑s in Swedish]. In Studier i Svensk språkhistoria 6, G. Harling-Krack and H. Lehti-Eklund (eds), 67–78. Helsingfors: Föreningen. Demuth, Katherine. 2000. “Bantu noun class systems: Loanword and acquisition evidence of semantic productivity”. In Systems of Nominal Classification, G. Senft (ed), 270–92. Cambridge: Cambridge University Press. Durie, Mark. 1997. “Grammatical structures in verb serialization”. In Complex Predicates, A. Alsina, J. Bresnan and P. Sells (eds), 289–354. Stanford: CSLI Publications. Eckhoff, Hanne M. 2009. “A usage-based approach to change: Old Russian possessive constructions”. In The Role of Semantic, Pragmatic and Discourse Factors in the Development of Case, J. Barðdal and S. Chelliah, (eds) 161–180. Amsterdam & Philadelphia: John Benjamins. Enfield, Nick J. 2002. “Cultural logic and syntactic productivity: Associated posture constructions in Lao”. In Ethnosyntax: Explorations in Culture and Grammar, N.J. Enfield (ed), 231–58. Oxford: Oxford University Press. Eythórsson, Thórhallur. 2000a. “Dative vs. nominative: Changes in quirky subjects in Icelandic”. Leeds Working Papers in Linguistics 8: 27–44. Eythórsson, Thórhallur. 2000b. “Fall á fallanda fæti [Unstable case marking].” Íslenskt mál og almenn málfræði 22: 185–204. Eythórsson, Thórhallur. 2002. “Changes in subject case marking in Icelandic”. In Syntactic Effects of Morphological Change, D. Lightfoot (ed), 196–212. Oxford: Oxford University Press. Eythórsson, Thórhallur and Jóhanna Barðdal. 2005. “Oblique subjects: A common Germanic inheritance”. Language 81(4): 824–81. Falk, Cecilia. 1997. Fornsvenska upplevarverb [Experiencer Verbs in Old Swedish]. Lund: Lund University Press. Fillmore, Charles J. 2002. “Idiomaticity”. Lecture notes from the spring 2002. Available at http:// www.icsi.berkeley.edu/~kay/bcg/lec02.html. Fillmore, Charles J., Paul Kay, and Mary Kay O’Connor. 1988. “Regularity and idiomaticity in grammatical constructions: The case of Let alone.” Language 64: 501–38. Fleischer, Wolfgang. 1982. Wortbildung der deutschen Gegenwartssprache. 5th Edition. Tübingen: Niemeyer. Fóstbræðra saga. 1987. In Íslendinga sögur og þættir II, B. Halldórsson, J. Torfason, S. Tómasson and Ö. Thorsson (eds), 775–851. Reykjavík: Svart á hvítu. Fromkin, Victoria and Robert Rodman. 1995. An Introduction to Language. 5th Edition. Fort Worth: Harcourt Brace College Publishers. Gibbs, Raymond W. Jr. 1995. “Idiomaticity and human cognition”. In Idioms: Structural and psychological perspectives, M. Everaert, E.-J. van der Linden, A. Schenk and R. Schreuder (eds), 97–116. Hillsdale, NJ: Erlbaum. Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago & London: University of Chicago Press. Goldberg, Adele E. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Goldberg, Adele E. and Ray Jackendoff. 2004. “The resultative as a family of constructions”. Language 80: 532–68.



References  Gropen, Jess, Steven Pinker, Michelle Hollander, Richard Goldberg, and Ronald Wilson. 1989. “The learnability and acquisition of the dative alternation in English”. Language 65: 203–57. Haegeman, Liliane. 1991. Introduction to Government and Binding Theory. Oxford: Basil Blackwell. Hallan, Naomi. 2001. “Paths to prepositions? A corpus-based study of the acquisition of a lexico-grammatical category”. In Frequency and the Emergence of Linguistic Structure, J. Bybee and P. Hopper (eds), 91–120. Amsterdam & Philadelphia: John Benjamins. Halldórsson, Halldór. 1982. “Um méranir: Drög að samtímalegri og sögulegri athugun [About dativizings: Preliminaries of a synchronic and diachronic investigation]”. Íslenskt mál og almenn málfræði 4: 159–89. Hare, May and Jeffrey L. Elman. 1995. “Learning and morphological change”. Cognition 56(1): 61–98. Haspelmath, Martin. 2002. Understanding Morphology. London: Arnold. Hay, Jennifer and Harald Baayen. 2002. “Parsing and productivity”. In Yearbook of Morphology 2001, G. Booij (ed), 203–35. Dordrecht: Kluwer. Holland, Gary. 1993. ”Transitivity, causativity, and surface case in Old Norse”. Arkiv för nordisk filologi 108: 19–37. Hólmarsson, Sverrir, Christopher Sanders, and John Tucker. 1989. Íslensk–ensk orðabók: Concise Icelandic–English Dictionary. Reykjavík: Iðunn. Holmberg, Anders. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Doctoral Dissertation, Stockholm University. Holmberg, Anders and Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax. New York & Oxford: Oxford University Press. Hornby, A.S., A.P. Cowie, and A.C. Gimson. 1974. Oxford Advanced Learner’s Dictionary of Current English. 3rd Edition revised and updated. London: Oxford University Press. Hænsna-Þóris saga. 1987. In Íslendinga sögur og þættir II, B. Halldórsson, J. Torfason, S. Tómasson and Ö. Thorsson (eds), 1417–36. Reykjavík: Svart á hvítu. Israel, Michael. 1996. “The way constructions grow”. In Conceptual Structure, Discourse and Language, A.E. Goldberg (ed), 217–30. Stanford: CSLI Publications. Itkonen, Esa. 2005. Analogy as Structure and Process: Approaches in Linguistics, Cognitive Psychology and Philosophy of Science. Amsterdam & Philadelphia: John Benjamins. Ívens saga. 1979. F. W. Blaisdell (ed). Copenhagen: C.A. Reitzels forlag [Editiones Arnamagnæanæ Series B, vol. 18]. Iwata, Seizi. 2008. Locative Alternation: A Lexical-Constructional Approach [Constructional Approaches to Language 6]. Amsterdam & Philadelphia: John Benjamins. Jahr, Ernst Håkon. 1995. ”Nedertysk og nordisk: språksamfunn og språkkontakt i Hansa-tida [Middle low German and Scandinavian: Language society and language contact during the Hansa period]”. In Nordisk og nedertysk: Språkkontakt og språkutvikling i seinmellomalderens. E.H. Jahr (ed), 9–28. Oslo: Novus. Jónsson, Jóhannes Gísli. 1996. Clausal architecture and case in Icelandic. Doctoral dissertation, University of Massachusetts Amherst. Jónsson, Jóhannes Gísli. 1997–98. “Sagnir með aukafallsfrumlagi [Verbs selecting for oblique subjects]”. Íslenskt mál og almenn málfræði 19–20: 11–43. Jónsson, Jóhannes G. 1998. “A list of predicates that take a quirky subject in Icelandic”. Unpublished manuscript, University of Iceland.

 Productivity Jónsson, Jóhannes G. 2003. “Not so quirky: On subject case in Icelandic”. In New Perspectives on Case and Case Theory, E. Brandner and H. Zinsmeister (eds), 127–64. Stanford: CSLI Publications. Jónsson, Jóhannes G. 2005. “Merkingarhlutverk, rökliðir og fallmörkun [Thematic roles, argument structure and case marking]”. In Setningar: Handbók um setningafræði, H. Thráinsson (ed), 350–409. Reykjavík: Almenna bókafélagið. Jónsson, Jóhannes G. and Thórhallur Eythórsson. 2005. “Variation in subject case marking in Insular Scandinavian”. Nordic Journal of Linguistics 28(2): 223–45. Josefsson, Gunlög. 1997. On the Principles of Word Formation in Swedish [Lundastudier i Nordisk Språkvetenskap A 51]. Lund: Lund University Press [Also published in 1998 by John Benjamins under the title Minimal Words in a Minimal Syntax]. Kay, Paul. 2002. “Patterns of Coining”. Presented at ICCG2 in Helsinki, Finland. Available at http://www.icsi.berkeley.edu/~kay/coining.pdf. Kay, Paul and Charles Fillmore. 1999. “Grammatical constructions and linguistic generalizations: The What’s X Doing Y? construction”. Language 75: 1–34. Knudsen, Tryggve. 1956. Kasuslære I [Case Studies 1]. Oslo: Universitetsforlaget. Kvaran, Guðrún. 1996. “Þættir úr sögu orðaforðans [On the History of the Icelandic Vocabulary]”. In Erindi um íslenskt mál, 35–48. Reykjavík: Íslenska málfræðifélagið. Langacker, Ronald W. 1988. “A usage-based model”. In Topics in Cognitive Linguistics, B. Rudzka-Ostyn (ed), 127–61. Amsterdam & Philadelphia: John Benjamins. Langacker, Ronald W. 2000. “A dynamic usage-based model”. In Usage-Based Models of Language, M. Barlow and S. Kemmer (eds), 1–63. Stanford: CSLI Publications. Leonard, Laurence B. 2000. “Understanding grammatical deficits in children with specific language impairment: The evaluation of productivity”. In Methods for Studying Language Production, L. Menn and N.B. Ratner (eds), 333–52. Mahwah, NJ: Erlbaum. Lightfoot, David. 1999. The Development of Language: Acquisition, Change, and Evolution. Malden, MA: Blackwell. Lieber, Rochelle. 1981. On the Organization of the Lexicon. Bloomington, IN: Indiana University Linguistics Club. Luraghi, Sylvia. 1987. “Patterns of case syncretism in Indo-European languages”. In Papers from the VIIth International Conference on Historical Linguistics. A.G. Ramat, O. Carruba, and G. Bernini (eds), 355–71. Amsterdam & Philadelphia: John Benjamins. Magnússon, Einar Björn. 2004. Slangurorðabók [Slang Dictionary]. Available at http://www. hauksi.com/slangur. Maling, Joan. 2002. “Það rignir þágufalli á Íslandi [It is raining dative in Iceland]: Verbs with dative objects in Icelandic”. Íslenskt mál 24: 31–105 [Also published in a slightly different English version in Working Papers in Scandinavian Syntax 70: 1–60, 2002]. Mayerthaler, Willi. 1980. Morphologische Natürlichkeit. [Linguistische Forschungen 28]. Wiesbaden: Athenaion. McGlone, Matthew S., Sam Glucksberg, and Cristina Cacciari. 1994. “Semantic productivity and idiom comprehension”. Discourse Processes 17: 167–90. Michaelis, Laura A. and Josef Ruppenhofer. 2001. Beyond Alternations: A Constructional Model of the German Applicative Pattern. Stanford: CSLI Publications. Moder, Carol Lynn. 1992. Productivity and Categorization in Morphological Classes. Doctoral Dissertation, State University of New York at Buffalo, NY. Norde, Muriel. 1994



References  Norde, Muriel. 1997. The History of the Genitive in Swedish: A Case Study in Grammaticalization. Amsterdam: University of Amsterdam. Norde, Muriel. 2006. “Demarcating degrammaticalization: The Swedish s‑genitive revisited.” Nordic Journal of Linguistics 29(2): 201–38. O’Grady, William, John Archibald, Mark Aronoff and Janie Rees-Miller. 2001. Contemporary Linguistics: An Introduction. 4th Edition. Boston & New York: Bedford/St. Martin’s. Physiologus. 1991. In Heimskringla III, B.S. Kristjánsdóttir, B. Halldórsson, J. Torfason and Ö. Thorsson (eds), 43–55. Reykjavík: Mál og menning. Pind, Jörgen (ed). 1991. Íslensk orðtíðnibók [Icelandic Word Frequency book]. Reykjavík: Orðabók Háskólans. Pinker, Steven. 1989. Learnability and Cognition: The Acquisition of Argument Structure. Cambridge MA: MIT Press. Pinker, Steven. 1999. Words and Rules: The Ingredients of Language. London: Phoenix. Plag, Ingo. 1999. Morphological Productivity: Structural Constraints in English Derivations [Topics in English Linguistics 28]. Berlin and New York: Mouton de Gruyter. Polenz, Peter von. 1991. Deutsche Sprachgeschichte vom Spätmittelalter bus zur Gegenwart, Vol. I. Berlin: Walter de Gruyter. Poplack, Shana. 1992. “The inherent variability of the French subjunctive”. In Theoretical Analyses in Romance Linguistics, C. Laeufer and T.A. Morgan (eds), 235–63. Amsterdam & Philadelphia: John Benjamins. Poplack, Shana. 1995. “The sociolinguistic dynamics of apparent convergence”. In Towards a Social Science of Language: Papers in Honor of William Labov, G. Guy, J. Baugh and D. Schiffrin (eds), 285–309. Amsterdam & Philadelphia: John Benjamins. Ragnarsdóttir, Hrafnhildur, Hanne Gram Simonsen, and Kim Plunkett. 1999. “The acquisition of past tense morphology in Icelandic and Norwegian children: An experimental study”. Journal of Child Language 26: 577–618. Rainer, Franz. 1987. “Produktivitetsbegriffe in der Wortbildungstheorie”. In Grammatik und Wortbildung romanischer Sprachen: Beiträge zum Deutschen Romanistentag in Sigen, H.-M. Gauger and H. Geckeler (eds), 187–202. Tübingen: Narr. Ritz, Marie-Eve A. and Dulcie M. Engel. 2003. “The use of the present perfect in Australian English: Examples of semantic productivity”. Abstract for the DGfS-workshop “Explaining Productivity” in Munich, Germany. Rögnvaldsson, Eiríkur. 1983. “Þágufallssýkin og fallakerfi í íslensku [Dative sickness and the case system in Icelandic]”. Skíma 16: 3–6. Rögnvaldsson, Eiríkur. 1997. “Orðafar Íslendinga sagna [Wordings in the Icelandic Sagas]”. In Milli himins og jarðar. A. Agnarsdóttir, P. Pétursson and T. H. Tulinius, 271–86. Reykjavík: Háskólaútgáfan. Sandøy, Helge. 1997. “Genuset åt substantiv på ‑ning i nynorsk [The gender of ‑ning nouns in New Norwegian]”. In Språket er målet: Festskrift til Egil Pettersen, J. Bondevik, G. Kristoffersen, O. Nes and H. Sandøy (eds), 160–77. Bergen: Alma Mater. Seefranz-Montag, Ariane v. 1983. Syntaktische Funktionen und Wortstellungsveränderung. Die Entwicklung ‘subjektloser’ Konstruktionen in einigen Sprachen. München: Wilhelm Fink Verlag. Sigurðsson, Halldór Ármann. 1989. Verbal Syntax and Case in Icelandic. Doctoral dissertation, Lund University. Smith, Henry. 1994. “‘Dative Sickness’ in Germanic”. Natural Language and Linguistic Theory 12: 675–736.

 Productivity Smith, Michael B. 2001. “Why quirky case really isn’t quirky (Or how to treat dative sickness in Icelandic)”. In Polysemy in Cognitive Linguistics, H. Cuyckens, B. Zawada, and D. Tuggy (eds), 115–59. Amsterdam & Philadelphia:  John Benjamins. Smith, K. Aaron. 2001. “The role of frequency in the specialization of the English anterior”. In Frequency and the Emergence of Linguistic Structure, J. Bybee and P. Hopper (eds), 361–82. Amsterdam & Philadelphia: John Benjamins. Smyk, Dorota. 2003. Why do “unproductive” processes yield new forms? Presented at the DGfSworkshop “Explaining Productivity” in Munich, Germany. Svavarsdóttir, Ásta. 1982. “‘Þágufallssýki” [‘Dative Sickness’]”. Íslenskt mál og almenn málfræði 4: 19–62. Svavarsdóttir, Ásta, Gísli Pálsson, and Þórólfur Þórlindsson. 1984. “Fall er fararheill: Um fallnotkun með ópersónulegum sögnum [Fall is a sign of luck: On case use with impersonal verbs]”. Íslenskt mál og almenn málfræði 6: 33–55. Tomasello, Michael. 1998. “Cognitive linguistics”. In A Companion to Cognitive Science. W. Bechtel and G. Graham (eds), 477–87. Massachusetts & Oxford: Blackwell Publisher. Tomasello, Michael. 2000. “Do young children have adult syntactic competence?” Cognition 74(3): 209–53. Traugott, Elizabeth Closs and Richard B. Dasher. 2002. Regularity in Semantic Change [Cambridge Studies in Linguistics 96]. Cambridge: Cambridge University Press. Wessén, Elias. 1929. “Om det tyska inflytandet på svenskt språk under medeltiden [On the German influence on the Swedish language during the Middle Ages]”. Nordisk Tidsskrift 5: 265–80. Wessén, Elias. 1992. Svensk språkhistoria 1 [Swedish language history 1]. Stockholm: Institutionen för nordiska språk. Yip, Moira, Joan Maling, and Ray Jackendoff. 1987. “Case in tiers.” Language 63: 217–50. Zimmer, Karl E. 1964. “Affixal negation in English and other languages: An investigation of restricted productivity”. In Supplement to Word, Vol. 20(2) [Monograph 5]. New York: Linguistic Circle of New York. Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson. 1985. Case and Grammatical Functions: The Icelandic Passive. Natural Language and Linguistic Theory 3: 441–83. Zlatev, Jordan. 1997. Situated Embodiment: Studies in the Emergence of Spatial Meaning. Stockholm: Gotab.

Appendix A Case and argument structure constructions in the text corpora and the predicates which instantiate them

Modern Icelandic Dat-Nom (11 types): finnast ‘feel’, fylgja ‘accompany’, heyrast ‘seem to hear’, mega ‘may, have to’, renna (kalt vatn) milli skinns og hörunds ‘feel terrified’, standa (stuggur) ‘fear’, sýnast ‘feel’, vera ljóst ‘be clear’, verða kunn ‘be known’, vera þægð í ‘be happy’, þykja ‘feel’. Nom-Acc (303 types): afflytja ‘lure’, afgreiða ‘expedite’, afvopna ‘disarm’, aka (Þrengslin) ‘drive’, ala (manninn) ‘be for some time’, angra ‘bother’, annast ‘care for’, athuga ‘check’, auglýsa ‘advertise’, auka ‘increase’, ákveða ‘decide’, átta (sig) ‘realize’, ávarpa ‘address’, baða ‘bathe’, barna ‘make pregnant’, beita ‘graze’, bera ‘carry’, bera fram ‘ask’, bera (kennsl) ‘recognize’, bera (sig) að ‘go about sth’, berja ‘beat’, beygja ‘stoop’, biðja ‘ask’, binda ‘tie’, bíta ‘bite’, bjóða ‘offer’, blekkja ‘deceive’, blessa ‘bless’, boða ‘proclaim, preach’, bora ‘drill’, borða ‘eat’, borga ‘pay’, brenna ‘burn’, brjóta ‘break’, brýna ‘emphasize’, búa undir ‘prepare’, búa til ‘make’, dikta ‘compose’, drekka ‘drink’, dylja ‘hide’, draga ‘draw, pull’, drekka ‘drink’, drepa ‘kill’, drífa ‘hurry’, efla ‘strengthen’, eggja ‘spur’, eiga ‘own, have’, eiga til ‘tend to’, eignast ‘come into possession’, (endur)gera ‘(re)do, make’, endurheimta ‘regain’, endurnýja ‘renew’, endurreisa ‘rebuild’, endurtaka ‘repeat’, elska ‘love’, erfa ‘inherit’, éta ‘eat’, fara (ferð) ‘take on a journey’, fatta ‘understand’, fá ‘get’, fela ‘hide’, fela í sér ‘entail’, fella niður ‘cancel, dismiss’, festa (ást) ‘fall in love’, finna ‘feel, find’, flytja ‘transport’, flytja (erindi) ‘give a speech’, flækja ‘complicate’, flæma burt ‘drive away’, framleiða ‘produce’, frelsa ‘free’, fræða ‘enlighten’, fullvissa ‘assure’, fullyrða ‘claim’, færa ‘move’, gefa ‘give’, gefa (sig) á tal ‘start conversation’, gefa (sig) út ‘make appearance’, geifla (sig) ‘grimace’, geyma ‘preserve’, gifta ‘get married’, gista ‘stay the night’, gleypa ‘swallow’, greiða ‘pay’, greina ‘distinguish’, grípa ‘catch’, græða ‘grow’, hafa ‘have’, hafa ekki (augun) af ‘not take one’s eye of ’,

 Productivity

hafa (hug) ‘intend’, hafa hugfast ‘keep in mind’, hafa (spurnir) ‘hear’, hafa upp úr krafsinu ‘gain’, halda ‘believe’, halda (erindi) ‘give a speech’, handtaka ‘arrest’, hefja ‘begin’, heilla ‘charm’, heimsækja ‘visit’, heyra ‘hear’, hitta ‘meet’, hirða ‘pick up from the street’, hlaupa (hring) ‘run’, hljóta ‘get’, hluta ‘cut up’, hneppa ‘get’, hnipra (sig) ‘curl up’, hremma ‘snatch’, hreyfa ‘move’, hrifsa ‘snatch’, hrista ‘shake’, hrífa upp ‘wake up’, hræða ‘scare’, hugleiða ‘ponder’, húkka ‘hitchhike’, hylla ‘enchant, hvessa (augun) ‘give (a sharp look)’, hvíla ‘rest’, hvísla ‘whisper’, hvítta ‘whiten’, höggva ‘hew’, iðka ‘practice’, innbyrða ‘consume’, kalla ‘call’, kaupa ‘buy’, kenna ‘recognize’, keyra ‘push’, kjafta til ‘convince’, kjósa ‘choose, vote’, klippa ‘cut’, klípa ‘pinch’, klæða ‘dress’, knýja ‘run (machines)’, kosta ‘finance’, kreppa (fæturna) ‘clench (one’s feet)’, kunna ‘know how’, kveðja ‘say goodbye’, kveikja ‘light’, kynna ‘present’, kyssa ‘kiss’, kæra ‘accuse’, laga ‘make’, laga að ‘adjust’, lagfæra ‘mend’, láta ‘let’, leggja ‘lay’, leggja (áherslu) ‘stress’, leggja (land undir fót) ‘travel’, leggja (sig) ‘take a nap’, leggja (stund á) ‘practice’, leiða ‘lead’, lengja ‘lengthen’, lesa ‘read’, leyfa ‘allow’, lita ‘color’, líða (skort) ‘suffer (privation)’, líta (dagsins ljós) ‘appear’, losa ‘loosen’, lýsa ‘lighten up’, læra ‘learn’, læsa ‘lock’, margfalda ‘multiply’, marka ‘demarcate’, merkja ‘mark, mean’, missa ‘lose’, muna ‘remember’, mynda ‘form’, myrða ‘murder’, mæla ‘measure’, mæla ‘speak’, naga ‘gnaw’, naglhreinsa ‘refine wood’, nálgast ‘approach’, nema ‘learn’, nema (land) ‘settle’, nefna (nöfn) ‘name’, nota ‘use’, (ný)skapa ‘(re)create’, nýta ‘utilize’, opna ‘open’, orða ‘put into words’, óhreinka ‘dirty’, óttast ‘fear’, pikka upp ‘pick up’, pína ‘force’, plata ‘con’, prófa ‘try’, prýða ‘decorate’, reka í ‘bump into sth’, reka upp ‘cry out’, reka úr sér tunguna ‘stick one’s tongue out’, rekja ‘trace’, reyna ‘try’, reyra saman ‘tie together’, rétta ‘hand’, rita ‘write’, rífa ‘pull, tear’, rífa kjaft ‘use foul language’, rukka ‘collect (bills)’, rýja ‘shear’, ræða ‘discuss’, rægja ‘slander’, ræskja (sig) ‘clear one’s throat’, rökstyðja ‘give arguments for’, sameina ‘unite’, selja ‘sell’, semja ‘compose’, senda ‘send’, setja ‘put’, setja upp ‘put on scene’, sérhæfa ‘specialize’, sigra ‘conquer’, sjá ‘see’, sjúga ‘suck’, skemma ‘damage’, skera ‘cut’, skilja ‘understand’, skilja eftir ‘leave behind’, skipa (sess) ‘occupy a place’, skjóta ‘shoot’, skoða ‘observe’, skrá ‘register’, skrifa ‘write’, skynja ‘perceive’, skýra ‘explain’, slá ‘hit’, slíðra ‘sheath’, slíta ‘rip’, smíða ‘construct’, smjúga ‘penetrate’, snapa ‘beg’, snerta ‘touch’, sofa (nóttina) ‘sleep (through the night), spara ‘save’, spenna ‘tighten, hitch’, spyrja ‘ask’, standa að e-u ‘catch redhanded’, standa e-ð af sér ‘wait for sth to stop’, standa vörð ‘stand guard’, stilla (sig) ‘calm down’, stofna ‘found’, stoppa ‘stop’, strjúka ‘touch’, styðja ‘support’, styrkja ‘strengthen’, stöðva ‘stop’, sveipa ‘wrap’, sverta ‘disparage’, svíða ‘singe’, syngja ‘sing’, sýna ‘show’, sækja ‘fetch’, sækja (skóla) ‘attend’, taka ‘take’, taka út ‘withdraw’, takast (á herðar) ‘shoulder’, taka (ákvörðun) ‘decide’, telja ‘consider’, tendra ‘lighten’, teygja ‘stretch’, tilkynna ‘declare’, tína upp ‘pick up’, toga ‘pull’, toppa ‘top’, trufla ‘disturb’, tryggja ‘secure’, tvíhenda ‘grasp with both hands’, tygja sig ‘prepare for departure’, tyggja ‘chew’, tæta upp ‘tear up’, umlykja ‘surround’, umskera ‘circumcise’, umsteypa ‘reshape’, undirrita ‘sign’, undirstrika ‘emphasize’, (út)breiða ‘spread’,



Appendix A 

útiloka ‘exclude’, útvega ‘provide’, uppgötva ‘discover’, vara við ‘warn’, vega ‘weigh’, veiða ‘fish’, vekja ‘awake’, velja ‘choose’, venja ‘acquire a habit’, verja ‘defend, protect’, vilja ‘want’, vinna (þrekvirki) ‘do great feat’, virða ‘view, respect’, vita ‘know’, víla fyrir sér ‘hesitate’, vísa (veginn) ‘show the way’, volgra ‘heat’, yfirbuga ‘overpower’, yrkja ‘compose’, þekkja ‘know’, þétta ‘make tight’, þrýsta ‘squeeze’, þurfa ‘need’, þýða ‘translate, mean’, æsa upp ‘incite’, öðlast ‘obtain’, ösla (snjóinn) ‘wade (snow), örva ‘arouse’. Nom-Dat (188) banda frá ‘wave away’, beina ‘direct’, beita ‘apply’, beita sér ‘do one’s utmost’, benda ‘direct’, berja (hrömmunum) ‘thrash one’s paws’, bindast ‘tie with’, bjóða ‘invite’, blanda ‘mix’, bregða ‘startle’, (um)breyta ‘change’, bylta (sér) ‘turn around’, búa (búi) ‘farm’, deila ‘share’, depla (auga) ‘blink’, dreifa ‘spread’, duga ‘last’, dunda (sér) ‘putter’, eigna ‘attribute’, einbeita ‘concentrate’, etja ‘egg’, eyða ‘destroy’, fara (höndum um) ‘treat’, fá ‘take’, fá e-u framgengt ‘get one’s will through’, feykja ‘blow’, fjölga ‘increase’, fletta upp ‘flick’, flýta ‘hurry’, forða ‘get sth out of the way’, fórna ‘sacrifice’, fresta ‘postpone’, fylgja ‘follow, accompany’, fyllast ‘be filled with’, gangast á hönd ‘surrender’, gefa ‘give’, gegna ‘obey’, geispa golunni ‘die’, gera ljóst ‘make clear’, gera (í hugarlund) ‘imagine’, gera (viðvart) ‘tell, warn’, geta til ‘guess’, giftast ‘marry’, glata ‘lose’, gleyma ‘forget’, gnísta (tönnum) ‘grind one’s teeth’, haga ‘arrange, behave’, halda ‘keep, hold’, halda áfram ‘continue’, halda uppi (vörnum) ‘defend’, halla ‘lean’, hampa ‘hold up’, hegða ‘organize, behave’, heilsa ‘greet’, heita (nöfnum) ‘carry names’, henda ‘throw’, herma ‘tell’, hella ‘pour’, hita ‘warm’, hjálpa ‘help’, hleypa ‘let, allow’, hliðra ‘shirk’, hlynna ‘care’, hlýða yfir ‘recite’, horfa (bænaraugum) ‘beg with the eyes’, hugsa ‘think’, jafngilda ‘be equal to’, kasta ‘throw’, kenna ‘teach’, kenna um ‘accuse’, kinka (kolli) ‘nod’, klappa ‘pat’, koma á óvart ‘surprise’, koma á fót ‘establish’, koma fyrir ‘arrange’, koma sér ‘get off, go away’, koma (e-u í lag) ‘get sth in order’, kunna vel ‘like’, kynnast ‘get to know’, kvíða ‘be anxious’, kvænast ‘marry’, leika ‘play’, leyfa ‘allow’, líkja saman ‘compare’, líkjast ‘resemble’, líta augum ‘regard’, ljúka ‘finish’, loka ‘close’, lúta ‘obey’, lyfta ‘lift’, lýsa ‘describe’, lýsa yfir ‘declare’, mjaka ‘budge’, mótmæla ‘object’, múta ‘bribe’, mæta ‘run into sby’, ná ‘obtain’, ráða ‘control’, ráða bana ‘kill’, renna ‘glide’, ríða ‘ride’, róa ‘row’, ræna ‘kidnap’, safna ‘gather’, segja ‘tell’, segja (skrækri röddu) ‘say with a squeaky voice’, sinna ‘attend to’, skemmta ‘entertain’, skella (e-u í sig) ‘slam (in oneself), eat’, skila ‘return’, skila (árangri) ‘lead to result’, skipa ‘order, organize’, skipta ‘divide’, skipta (máli) ‘matter’, skjóta ‘shoot’, skjóta (rótum), ‘settle’, skjóta upp (kollinum) ‘pop up’, skrifa ‘write’, skutla ‘give a ride’, slengja ‘throw’, sleppa ‘drop, release’, slíta ‘end’, smeygja (sér) ‘slide’, smjúga (fingrum) ‘slide’, snúa ‘turn’, sparka ‘kick’, spá ‘predict’, spyrna upp ‘kick’, stela ‘steal’, stilla upp ‘take a position’, stinga ‘insert’, stjórna ‘control’, styðja (olnboganum) ‘support’, stýra ‘gov-

 Productivity

ern’, svara ‘answer’, sveia ‘curse’, sveipa ‘wrap’, svæla (í sig) ‘puff (in oneself), eat’, sæta (undrum) ‘be of surprise’, taka ‘greet, receive’, tengjast ‘connect’, tilheyra ‘belong’, tilkynna ‘declare’, troða ‘stuff, squeeze’, treysta ‘trust’, trúa ‘believe’, tylla (sér) ‘take a seat’, una (sér) ‘be content’, valda ‘cause’, varpa ‘throw’, velta ‘roll’, vera (aufúsugestur) ‘be welcome’, vera áþekkur ‘resemble’, vera búinn ‘be equipped’, vera eignaðar ‘be attributed to’, vera jafnhár ‘be as tall’, vera hlaðinn ‘be loaded’, vera hliðstæð ‘be equivalent’, vera hulið ‘be hidden’, vera hættulegt ‘be dangerous’, vera klæddur ‘be dressed’, vera (ó)kunnugur ‘be (un)known’, vera lagt í munn ‘be put in sby’s mouth’, vera nálægt ‘be near’, vera nákominn ‘be close’, vera nær ‘be close’, vera (orðum) aukið ‘be exaggerated’, vera tengdur ‘be related’, vera troðinn ‘be stuffed with’, vera samgróinn ‘be rooted’, vera sammála ‘agree’, vera vafinn ‘be wrapped’, vera (ó)vanur ‘(not) be used to’, vera vaxinn ‘be grown’, vera viðbúinn ‘be ready’, verða til ama ‘bother’, vera til sóma ‘makes by proud’, vefja ‘wrap’,venjast ‘get used to’, verjast ‘defend’, víkja ‘retreat’, voga ‘dare’, ýta ‘push’, þakka ‘thank’, þjóna ‘serve’, þrengja ‘push one’s way’, þrýsta ‘squeeze’, þvo ‘wash’. Nom-Gen (24 types): afla ‘obtain’, biðjast (hjálpar) ‘ask (for help)’, bíða ‘await’, ganga duldir ‘hide’, gæta ‘make sure’, hefjast (handa) ‘begin’, krefjast ‘demand’, leita (aðstoðar) ‘seek (help)’, mega (sín) ‘be influential’, nema (staðar) ‘stop’, njóta ‘enjoy’, njóta (hylli) ‘be popular’, óska ‘wish’, sakna ‘miss’, skammast (sín) ‘be ashamed (of oneself)’, spyrja (spurninga) ‘ask (questions)’, vekja (máls) ‘mention’, vera fullviss ‘be sure’, verða áskynja ‘realize’, verða vísari ‘find out’, verða vart ‘be(come) aware of ’, vitja ‘visit’, vænta ‘expect’, þurfa ‘need’. Dative subject predicates (73 types): bera ‘be obliged’, berast til eyrna ‘hear’, bregða fyrir ‘be briefly seen’, detta í hug ‘get an idea’, festa saman ‘be fastened together’, finnast ‘feel’, fjölga ‘increase’, fljúga í hug ‘get an idea’, fylgja ‘accompany’, ganga vel ‘be successful’, halla ‘incline’, heyrast ‘hear, gather’, hitna í hamsi ‘become angry’, koma saman um ‘agree on’, koma við ‘be of sby’s business’, langa ‘want’, leiðast ‘be bored’, líða ‘feel’, líka ‘like’, lítast á ‘like’, ljósta ‘strike’, mega ‘may, have to’, nægja ‘suffice’, renna kalt vatn á milli skinns og hörunds ‘be terrified’, reynast ‘turn out to be’, skiljast ‘understand, gather’, standa stuggur af ‘be scared of sth’, sýnast ‘seem, appear’, sæta ‘be the reason for’, takast ‘manage, succeed’, verða á mistök ‘make a mistake’, vera bágt til bjargar ‘be difficult to save’, vera beint ‘be directed’, verða bilt við ‘be startled’, vera fjarri ‘be remote’, vera fært ‘be capable’, vera haldið ‘be kept’, vera haldið fram ‘be argued’, vera heitið ‘be promised’, vera heitt í hamsi ‘be angry’, vera hugleikið ‘be on sby’s mind’, verða hugsað ‘have a



Appendix A 

thought’, verða hugsað til ‘remember, think about sby’, vera hætt ‘be ended’, verða kalt ‘be cold’, verða kunn ‘be known’, vera leitt ‘be sad’, verða ljóst ‘realize’, vera lokið ‘be finished’, vera lyft ‘be raised’, vera lýst ‘be described’, vera meinað ‘be prohibited’, vera mútað ‘be bribed’, vera nóg ‘have enough’, vera nóg boðið ‘be fed up’, vera ókunnur ‘be unknown, vera ómótt ‘feel nauseated’, vera ósvarað ‘be unanswered’, verða rótt ‘become calm’, vera rænt ‘be kidnapped’, vera sagt ‘be told’, vera safnað ‘be collected’, vera sama ‘don’t mind’, vera sinnt ‘be attended to’, vera slegið föstu ‘be decided’, verða starsýnt á ‘stare’, vera stjórnað ‘be controlled’, vera stolið ‘be stolen’, vera takmörk sett ‘be limited’, vera tekið ‘be received’, vera þægð í ‘be content with’, virðast ‘seem, appear’, þykja ‘feel, think, seem’. Accusative subject predicates (14 types): bresta kjark ‘lack courage’, dreyma ‘dream’, finna ‘find’, fýsa ‘desire’, gruna ‘suspect’, langa ‘want’, leggja ‘waft’, lysta ‘want’, mega heyra ‘may be heard’, reka í rogastans ‘become surprised’, skorta ‘lack’, taka ‘take’, undra ‘be surprised’, vanta ‘need’. Genitive Subject Predicates (7 types): gerast þörf ‘is needed’, gæta ‘be perceptible’, vera að geta ‘be mentioned’, vera dæmi ‘examples exist’, vera getið til ‘be guessed’, vera gætt ‘be considered’, verða vart ‘be perceptible’. Old Icelandic Dat-Nom (33 types): bjóða (hugur) ‘want’, draga til dauða ‘cause to die’, dæmast ‘judge’, endast ‘last’, fara ‘suit’, fylgja ‘accompany’, greiðast vel ‘go well’, koma í hug ‘get an idea’, mega ‘have to’, lítast ‘seem’, mislíka ‘dislike’, segja (hugur) um ‘have a feeling’, standa mein að ‘have problems with’, sýnast ‘feel’, vandast ‘have problems’, vaxa (afl) ‘grow stronger’, vera (ást) ‘love’, vera (dul) ‘hide’, vera (eftirsjá) ‘regret’, vera (forvitni) á ‘be curious’, vera gerðar (aðfarir) ‘be attacked’, vera (grunur á) ‘suspect’, vera (kostur) ‘be a possibility’, vera maklegt ‘be deserved’, vera (mein) ‘be harmful’, vera (nauðsyn) ‘be necessary’, vera sár (fóturinn) ‘be hurt’, vera (þökk) ‘be grateful’, verða (hamingja) ‘be of happiness’, verða kunnigur ‘be known’, verða (sein förin) ‘be late’, verða (vísa á munni) ‘accidentally speak’, þykja ‘feel’. Nom-Acc (173 types): auka ‘increase’, bannsetja ‘excommunicate’, benda ‘bow’, bera ‘carry’, berja ‘beat’, biðja ‘ask’, bíða ‘await’, binda ‘tie’, birta ‘show’, bjóða ‘offer’, blekkja ‘deceive’, brjóta

 Productivity

‘break’, búa ‘make ready’, byggja ‘inhabit’, byrja ‘put in a house’, byrja upp ‘start’, brenna ‘burn down’, draga ‘draw’, draga (grun) ‘suspect’, drekka ‘drink’, drepa ‘kill’, dubba ‘knight’, láta (sig) dvelja ‘dwell, be delayed’ dýrka ‘adore’, dæma ‘judge’, döggva ‘wet’, eggja ‘spur’, eiga ‘own, have’, eiga (yfir höfði sér) ‘be threatened’, elska ‘love’, elta ‘follow’, erfa ‘inherit’, fá ‘give’, fella ‘fell’, finna ‘feel’, fletta ‘peel’, flytja ‘transport’, flytja út ‘recite’, forðast ‘avoid’, forvitnast ‘be curious’, fregna ‘hear (news)’, frelsa ‘free’, fremja ‘do’, frétta ‘hear news’, fyrirláta ‘leave’, fæða ‘feed, raise’, færa ‘move’, gefa ‘give’, gefa (í vald) ‘surrender’, gefa gaum ‘attend to’, gera ‘do, make’, geta ‘conceive’, gifta ‘give, marry’, gjalda ‘pay’, glata ‘lose’, grípa ‘catch’, græða ‘heal’, gyrða ‘fasten’, gæða ‘endow’, hafa ‘have’, hafa burt ‘kidnap’, hafa fram ‘get sth through’, hafa upp ‘lift up’, hafa úti ‘finish’, halda ‘put up’, halda (vörð) ‘stand guard’, heyja orrustu ‘fight a battle’, hefja (sig fram á veginn) ‘make it’, heimta ‘fetch’, hengja ‘hang’, herða ‘tie’, heyra ‘hear’, hitta ‘meet’, hrista ‘shake’, huga ‘think’, hugsa ‘think up’, íhuga ‘think about’, kalla ‘say’, kalla ‘call’, kanna ‘explore’, kaupa ‘buy’, kenna ‘recognize’, kjósa ‘elect’, klæða ‘dress’, kunna ‘know how to’, kyssa ‘kiss’, kveða ‘say’, kveða ‘recite’, kveina ‘complain’, kæra ‘complain’, lasta ‘defame’, lauga ‘bathe’, láta ‘let’, leggja ‘lay’, leggja af ‘give up’, leggja til ‘suggest’, leiða ‘take’, leika ‘treat’, lima ‘dismember’ lífga ‘give life to’, líta ‘see’, lofa ‘praise’, merkja ‘mean’, mikla ‘boost’, muna ‘remember’, mæla ‘say’, mæta ‘value’, nálgast ‘approach’, nefna ‘name’, nema ‘settle’, nema ‘learn’, næra ‘nourish’, pína ‘torture’, pússa saman ‘marry’, reisa ‘build’, reka (erindi) ‘run errands’, reyna ‘try’, reyna ‘experience’, reka flóttan ‘chase fleeing men’, rétta ‘stretch’, rifta ‘break (contract)’, rjúfa ‘break, róa (vík) ‘force to submission’, segja ‘say’, senda ‘send’, setja ‘put’, setja (fund) ‘declare a meeting open’, sigra ‘conquer’, sjá ‘see’, skapa ‘create’, skera ‘cut, make’, skilja ‘understand’, skilja ‘divide’, slá manngarð ‘form a ring of men’, spyrja ‘ask’, strjúka ‘hit’, styrkja ‘strengthen’, syngja ‘sing’, svíkja ‘betray’, sýna ‘show’, sækja ‘attack’, sæma ‘award’, sæta ‘wait’, söðla ‘sadle’, taka ‘take’, taka (ógleði) ‘become sick’, taka (af lífi) ‘kill’, taka (veður) ‘get weather’, telja ‘assume’, torkenna ‘disguise’, tortryggja ‘suspect’, tryggja ‘secure’, undra ‘wonder’, vara við ‘warn’, varða ‘protect’, varðveita ‘maintain’, vega ‘kill’, veiða ‘hunt’, vekja ‘awake’, vilja ‘want’, vinda (segl) ‘wind’, vinna ‘do, work’, virða ‘respect’, virða ‘view’, vita ‘know’, vígja ‘consecrate’, yrkja ‘write poetry’, þakka ‘thank’, þiggja ‘accept’, þola ‘endure’, þora ‘dare’, æpa ‘scream’, ætla ‘intend’. Nom-Dat (105 types): bergja ‘taste’, beita (vélum) ‘con’, bjóða ‘invite’, bjóða ‘order’, borga ‘pay’, bölva ‘curse’, eftirfrétta ‘ask’, eindaga ‘give a final due date’, etja (kappi) ‘fight’, eyða ‘destroy’, fagna ‘greet’, fara ráðum ‘follow (advice)’, fá ‘give’, fleygja ‘throw’, fylgja ‘be accompanied with’, færa ‘bring’, gegna ‘be useful’, gera ‘do’, grípa hendinni ‘catch’, halda ‘keep, hold’, heita ‘promise’, heyra til ‘belong’, hlífa ‘not attack’, hlýða ‘listen’, hlíta ‘obey’, hnjóða ‘hit’, (af)hyggja ‘give up’, hæfa ‘suit’, hætta ‘stop’, jafnast ‘be equal’, játa ‘ad-



Appendix A 

mit’, kasta ‘throw’, kenna ‘accuse’, kippa ‘pull’, klæðast ‘dress’, koma fram ‘get through’, koma í málið ‘involve’, leggja ‘park’, lenda ‘park’, leyna ‘hide’, líkjast ‘resemble’, líta (augum) ‘like’, ljósta ‘throw’, lýsa ‘declare’, mæla (hárri röddu) ‘speak’, ná ‘catch’, nema ‘equal’, neita ‘deny’, nenna ‘want’, ráða ‘steer’, ráða ‘advice’, renna ‘draw’, ræna ‘steal’, safna ‘collect’, samþykkjast ‘agree, obey’, segja ‘tell’, sigla ‘sail’, sinna ‘care for’, sitja nær ‘sit near’, skipa ‘decide’, skipa ‘order’, skipta ‘divide’, skipta (málum) ‘associate’, skjóta ‘slide’, skrifta ‘accept confession’, slá til ‘throw a party’, snúa ‘turn’, snúa ‘translate’, spá ‘predict’, standa ‘stand’, stefna ‘supine’, svara ‘answer’, sveifla ‘swing’, standast ‘resist’, svæfast (svefni) ‘fall asleep’, sæta ‘endure’, taka ‘greet, receive’, taka (fæti) ‘take’, upplyfta ‘lift up’, una (ráði) ‘be happy’, varpa ‘throw’, veifa ‘wafe’, veita ‘give’, vera barinn ‘be beaten’, vera gefinn ‘be given’, vera girtur ‘be enclosed’, vera hlaðinn ‘be loaded’, vera hollur ‘be faithful’, vera hulinn ‘be covered’, vera klædd ‘be dressed’, vera lagður (til skaða) ‘be dangerous’, vera (ó)líkur ‘be like’, vera sagt ‘be told’, vera saumuð ‘be sewed with’, vera síður ‘be long’, vera stuttur ‘be short’, vera trúr ‘be faithful’, verða afhuga ‘lose interest’, verða feginn ‘feel relief ’, verða að skömm ‘be ashamed of oneself ’, verða til bana ‘cause to die’, verja (fé) ‘spend (money)’, þakka ‘thank’, þjóna ‘serve’. Nom-Gen (21 types): biðja ‘propose’, biðja (griða) ‘ask’, bíða ‘await’, fá ‘get’, gera ‘do’, gera (sakar) ‘commit crime’, gjalda ‘pay’, gæta ‘look after’, gæta ‘check’, hefna ‘revenge’, kenna ‘feel’, krefja ‘demand’, leita ‘look for’, leita (ráðs) ‘seek advice’, ljá ‘give’, missa ‘lose’, njóta ‘benefit’, neyta ‘use’, neyta (matar) ‘enjoy’, vera vís ‘find out’, vilja (ráða sinna) ‘want’. Dative subject predicates (72 types): bera ‘be obliged’, berast ‘receive’, bjóða hugur ‘want’, byrja ‘get wind’, dæmast ‘be judged’, endast ‘last’, fara ‘be successful’, fara vel ‘suit well’, finnast ‘think, feel, seem’, fylgja ‘accompany’, gagna ‘be of use’, ganga til ‘have the intention’, ganga (vel) ‘be (un)successful’, getast að ‘like’, greiðast ‘go well’, halda ‘hold’, koma í hug ‘get an idea’, leiða illt ‘get into trouble’, líka ‘like’, lítast á ‘like’, ljúka ‘end’, mega ‘must’, misfarast ‘be unsuccessful’, mislíka ‘dislike’, segja hugur um ‘have a feeling’, snúa ‘turn’, standa mein að ‘have problems with’, sýnast ‘appear, seem’, takast ‘succeed’, vandast málið ‘become difficult’, vaxa afl ‘become strong’, vera að harmi ‘be griefstrucken’, verða að munni ‘accidentally speak’, vera áheyrilegt ‘like’, vera ást á ‘love’, vera best ‘be best for sby’, vera borgið ‘be safe’, vera fjarri ‘be remote’, vera forvitni á ‘be curious’, vera eftirsjá ‘regret’, vera farið ‘be in a certain way’, vera gerðar aðfarir ‘be attacked’, vera gefið ‘be given’, vera goldið ‘be payed back’, verða gott til fjár ‘become rich’, vera grunur ‘suspect’, verða hamingja að ‘be of happiness for sby’, verða kunnugt ‘become known’, vera í skapi ‘feel like’, vera játað ‘be confessed’, vera maklegt ‘be deserved’, vera veitt ‘be given’, verða mein að ‘be harmful’, vera nauðsyn ‘be necessary’,

 Productivity

vera ofurefli ‘be inferior to’, vera sagt ‘be told’, vera sár fótur ‘have pain in the leg’, verða sein förin ‘be late’, verða seint ‘be late’, vera síður ‘be long’, vera siglt ‘be sailed’, vera skipað ‘be given a place to sit’, vera slitið ‘be ended’, vera tekið far ‘arrange sby a lift’, vera til dauða ‘cause to die’, vera ekki um e-n ‘dislike’, vera vant ‘be difficult for sby’, verða vei ‘be woe to’, vera þungt ‘be difficult for sby’, vera þökk ‘deserve thankfulness’, votta ‘testify’, þykja ‘feel’. Accusative subject predicates (13 types): bera ‘be obliged’, bera undan ‘float away’, bresta ‘lack’, forvitnar ‘be curious’, fýsa ‘want’, gefa til ‘get favourable weather’, gera fúsan ‘become eager’, greina á ‘disagree’, hungra ‘hunger’, reka ‘drift’, saka ‘harm’, sjá ‘see’, skorta ‘lack’. Gen subject predicates (7 types): mega vera ‘may be’, vera auðið ‘have the possibility, vera freistað ‘be tempted’, vera getið ‘be mentioned’, vera kostur ‘be a possibility’, vera tíðinda ‘be new’, vera von ‘be expected’.

Appendix B Recent borrowings in Icelandic

Nom-Acc: archive-a ‘archive’, battla ‘battle’, biddsslappa ‘bitchslap’, blasta ‘blast’, bojkotta ‘boycut’, builda ‘build’, bomba ‘bomb’, branda ‘brand’, browsa ‘browse’, bösta ‘bust’, compilera ‘compile’, copy-a ‘copy’, digga ‘dig, like’, döbba ‘dub’, editera ‘edit’, erasa ‘erase’, fíla ‘like‘, fixa ‘fix’, flexa ‘flex‘, formatta ‘format’, fótósjoppa ‘photoshop’, gúggla ‘google’, hakka ‘hack’, hössla ‘hussle’, kidda ‘kid‘, krakka ‘crack’, logga ‘log’, massa ‘finish with style’, meisa ‘spray with tear gas’, modda ‘modify’, mounta ‘mount’, muffa ‘bang’, mönnsa ‘munch’, offa ‘off ’, óna ‘own’, paira ‘pair’, partiona ‘partition’, patcha ‘patch’, peista ‘paste’, pinga ‘ping’, plögga ‘plug’, pródúsera ‘produce’, prógrammera ‘program’, publisha ‘publish’, r[e]nta ‘rent’, releasa ‘release’, rendera ‘render’, resetta ‘reset’, resolva ‘resolve’, restora ‘restore’, rippa ‘rip’, rokka ‘rock’, skratsa ‘scratch’, skvassa ‘squash, break’, slamma ‘slam’, ssh-a ‘ssh’, stúdera ‘study’, supporta ‘support’, sörfa ‘surf ’, synca ‘synchronize’, synkrónisera ‘synchronize’, tagga ‘tag’, testa ‘test’, tóka ‘smoke hash’, trimma ‘trim’, updata ‘update’, upgreida ‘upgrade’, verifya ‘verify’. Nom-Dat adda ‘add’, blasta ‘blast’, bomba ‘bomb’, bundla ‘bundle’, convertera ‘convert’, downloada ‘download’, deleta ‘delete’, de-multiplexa ‘demultiplex’, dumpa ‘dump’, droppa ‘drop’, ejecta ‘eject’, expandera ‘expand’, exporta ‘export’, farta ‘fart’, innstalla/innstallera ‘install’, krassa ‘crash’, mnsa ‘msn’, mökka ‘contaminate’, neimdroppa ‘namedrop’, offa ‘off ’, parkera ‘park’, peista ‘paste’, poppa ‘deliver when popping by’, pósta ‘post’, publisha ‘publish’, releasa ‘release’, resolva ‘resolve’, restarta ‘restart’, rippa ‘steal’, sjera ‘share’, slamma ‘slam’, starta ‘start’, statta ‘stat, dublicate’, streyma ‘stream’, tilta ‘tilt’, umba ‘be agent for’, unzippa ‘unzip’, uploada ‘upload’, untara ‘unzip’.

Appendix C The questionnaire

Age _____________________ F 

M

Below are some examples of how people talk about email and cell phones. I am interested in finding out what YOU say when you talk about these things. Read the examples below and choose between the seven possible judgments. Choose the judgment that you feel is the most appropriate, and draw a ring around the corresponding number. 1 2 3 4 5 6 7

This is impossible in Icelandic This is hardly possible in Icelandic It might be possible to say this but I would never do it I could say this but normally I wouldn’t I might perhaps say this I could very well use this formulation This is exactly how I would say it

Finally here are the examples: Ég (e)meila þessu til þín Ég sendi þér þetta í pósti Ég faxa þetta til þín Ég smsa þér þetta Ég sendi þetta til þín í pósti Ég (e)meila þér þetta Ég smsa þessu til þín Ég (e)meila þetta til þín Ég faxa þér þetta Ég smsa þetta til þín Ég faxa þessu til þín Ég sendi þessu til þín

1 1 1 1 1 1 1 1 1 1 1 1

2 2 2 2 2 2 2 2 2 2 2 2

3 3 3 3 3 3 3 3 3 3 3 3

4 4 4 4 4 4 4 4 4 4 4 4

5 5 5 5 5 5 5 5 5 5 5 5

6 6 6 6 6 6 6 6 6 6 6 6

7 7 7 7 7 7 7 7 7 7 7 7

Name index A Allen, C.L.  79, 109, 139, 142, 146, 148–149, 157, 164–166 Anshen, F.  26, 28 Archibald, J.  12 Aronoff, M.  11, 12, 17, 25–26, 28, 49 Askedal, J.O.  139 B Baayen, H.  10, 11, 26, 28, 36 Barðdal, J.  xi, 4, 5, 14, 28, 33–35, 45, 46, 55, 57–59, 63, 83, 84, 87, 98, 99, 103, 106, 112, 119, 121, 122, 126, 127, 130, 132, 139, 140, 141, 143–145, 151, 153, 158, 159, 165, 170 Barnes, M.P.  139, 142, 165 Basbøll, H.  13 Bauer, L.  xii, 1, 3, 11, 25, 28, 36, 44, 90, 91 Berg-Olsen, S.  141 Bernódusson, H.  58 Bjarnadóttir, K.  59, 106 Blum, M.  101 Boas, H.C.  14, 44, 69 Bock, K.  105 Böðvarsson, Á.  59 Braine, M.D.S.  101 Brody, R.E.  101 Bybee, J.L.  xii, 2, 3, 14, 16, 18, 19, 23, 25, 28, 36, 43, 45, 49, 50, 52, 92, 95, 96, 107, 111 C Cacciari, C.  15 Chomsky, N.  1, 29, 79, 80, 172 Clausner, T.C.  xii, 14, 28, 44, 45, 48, 49, 85, 92, 96 Cowie, A.P.  50, 51 Croft, W.  xii, 4, 5, 14, 28, 31, 32, 35, 44–49, 63, 85, 92, 96, 119 Cruse, A.D.  4, 5, 28, 63, 96 D Dasher, R.B.  11, 19

Delsing, L.-O.  12, 33, 139, 147 Demuth, K.  13, 28 Durie, M.  13, 14 E Eckhoff, H.M.  14 Elman, J.L.  17 Enfield, N.J.  13 Engel, D.M.  16 Eythórsson, T.  57, 58, 79, 109, 139, 140, 158, 170 F Falk, C.  33, 79, 109, 139, 141, 142, 147–149, 165, 166 Fillmore, C.J.  14, 15, 36, 37 Fleischer, W.  11 Fromkin, V.  13 G Gibbs, R.W.Jr.  15 Gimson, A.C.  50, 51 Glucksberg, S.  15 Goldberg, A.E.  xii, 3, 5, 14, 27, 28, 35, 66, 111, 119, 126, 145 Goldberg, R.  27, Griffin, Z.M.  105 Gropen, J.  27 H Haegeman, L.  80 Hallan, N.  10 Halldórsson, H.  139, 158, 166 Hare, M.  17 Haspelmath, M.  17, 33, 44, 45 Hay, J.  26, 28 Holland, G.  123 Hollander, M.  27 Hollmann, W.  119 Hólmarsson, S.  58 Holmberg, A.  79, 109, 127 Hornby, A.S.  50, 51 I Israel, M.  31 Itkonen, E.  2, 90

Iwata, S.  5, 14, 46 J Jackendoff, R.  3, 14, 79, 109, 140, 142 Jahr, E.H.  146 Jónsson, J.G.  58, 79, 99, 109, 118, 139, 140, 142, 158, 170 Josefsson, G.  30, 80 K Kay, P.  14, 15, 36, 37 Knudsen, T.  149 Kristoffersen, K.E.  46, 126 Kulikov, L.  145 Kvaran, G.  124, 156, 157 L Langacker, R.W.  xii, 14, 29, 45, 47 Lasnik, H.  79, 80 Leonard, L.B.  12 Lieber, R.  26, 183 Lightfoot, D.  139, 186 Luraghi, S.  145 M Magnússon, E.B.  78 Maling, J.  3, 58, 59, 79, 99, 109, 111, 121, 127, 140, 142, 150, 170 Mayerthaler, W.  11 McGlone, M.S.  15 Michaelis, L.A.  16, 111 Moder, C.L.  50, 107 Molnár, V.  63 N Nielsen, M.  119 Norde, M.  33, 146 O O’Grady, W.  12, 13 P Pálsson, G.  139, 158 Pind, J.  59, 156 Pinker, S.  3, 12, 27, 35, 80

 Productivity Plag, I.  11, 25, 26, 28, 36, 44 Platzack, C.  127 Plunkett, K.  40–42, 85, 107 Polenz, P.v.  157 Poplack, S.  16 R Ragnarsdóttir, H.  40–42, 85, 107 Rainer, F.  25 Rees-Miller, J.  12 Ritz, M.-E.A.  16 Rodman, R.  13 Rögnvaldsson, E.  139, 156–158 Ruppenhofer, J.  16, 111 S Sanders, C.  58 Sandøy, H.  16, 17 Seefranz-Montag, A.v.  148,

150, 166 Shalom, M.F.  101 Sigurðsson, H.Á.  79, 109, 142 Simonsen, H.G.  40–42, 85, 107 Slobin, D.I.  23, 50, 107 Smith, H.  139, 158 Smith, K.A.  16 Smith, M.B.  139, 158 Smyk, D.  31 Sotirova, V.  119 Svavarsdóttir, Á.  139, 158 Sveen, A.  46, 126 T Thompson, S.  16, 25, 28, 49, 92 Taoka, C.  119 Thráinsson, H.  3, 79, 109, 140 Tomasello, M.  4, 10, 18, 19, 63 Traugott, E.C.  11, 19

Tucker, J.  58 Þ Þórlindsson, Þ.  139, 158 W Weisberger, M.J.  101 Wessén, E.  146, 157 Wilson, R.  27 Y Yip, M.  3, 79, 109, 140, 142 Z Zaenen, M.  3, 79, 109, 140 Zimmer, K.E.  11, 28, 29 Zlatev, J.  30

Subject index A affectedness  65, 67, 69, 82, 96 alternating verbs  58 analogical leveling  32 analogy  2–4, 6, 10, 34, 37, 43–44, 50, 52, 54, 56, 76, 78, 85, 88–94, 96–98, 101, 105, 115, 117, 120, 134–136, 148, 166, 169, 173, 175–176, 178–180, 182 B Bantu  13, 28 basic human event types  66–67, 71–72, 76, 96, 175 borrowings  3, 6–7, 28, 54, 56, 77–89, 92–94, 97, 101, 111, 114–117, 121, 133–134, 137, 141, 146, 148–149, 157, 168, 175–178, 180 C Canadian French  16 category-conditioned degree of productivity  26 cognitive-functional linguistics  1–2, 4–7, 10, 28, 34, 44–45, 47, 52, 85, 89, 91, 96–98, 101, 109–110, 115, 118–119, 137–138, 143, 145–148, 156–157, 159, 165–169, 173–174, 176, 180–182 coherence  2, 4, 6, 9–10, 27–28, 33–34, 38–39, 41, 44–45, 52–53, 72, 76, 89, 91, 134, 137–139, 149, 155, 164, 166–167, 172–174, 181–182 semantic coherence  6, 9, 24, 28, 34–35, 38–39, 43–44, 52–56, 62, 78, 89–92, 98, 115, 134, 139, 145, 149, 164, 166–167, 169, 172–174, 178–179, 181–182 phonological coherence  41, 50 coining  36–37

compositionality  22 conventionalization  18–19, 48, 53, 82, 114–115, 123–125, 128–130, 136, 148 core  80 creativity  3, 37 D Danish  13, 149 dative substitution  7, 113, 138–140, 143, 147, 156, 158, 164, 168–169, 180–181 default status  13, 18–19, 115–116, 118 degrees of productivity  3, 4, 6, 10, 28, 34, 47, 54, 62, 89, 96, 111–112, 114, 117, 173–175, 182 derivation  5, 30, 183 Dutch  139, 145, 168 E English  7, 13–14, 16–18, 23, 26–28, 31, 33, 35–36, 40–41, 48, 50–51, 58, 66, 83, 86, 92–94, 114, 126, 128, 137–140, 143, 145–146, 148–150, 156–158, 164–168, 174, 180 Australian English  16 Middle English  142, 148–149, 164–165 Old English  146 entrenchment  34, 45, 47–49, 52, 54, 61–62, 85, 89, 94, 97–98, 138, 161, 164–165, 169, 174–176, 181 extensibility concept  1–4, 20, 22–25, 27, 29–33, 37–38, 44, 45, 50, 52–53, 88, 92, 95–96, 98, 111, 120, 134–136, 171–174, 179, 182 F Faroese  7, 112, 137–139, 142–143, 145, 150, 164–165, 168, 180

frequency absolute type frequency  112, 116–117, 159, 177–178 document frequency  94 relative type frequency  116, 138, 154–155, 159, 164, 166, 177 text frequency  27 token frequency  6, 16, 25–27, 30, 34, 49–52, 54, 56, 78, 89–96, 98, 145, 166, 174, 176–177 type frequency  2–7, 9, 18–19, 24, 26–30, 33–36, 38–41, 43–45, 47–50, 52–58, 60–62, 74, 76–78, 86–87, 89–92, 94–97, 98, 101–102, 106–107, 109–112, 114–117, 120, 132–134, 136–139, 143–151, 153–157, 159–169, 172–174, 176–182 frozenness  11, 15–16, 20, 57 full productivity  2, 10, 13, 34, 39, 44, 52, 78, 89–92, 96–97, 173, 176, 179, 182 G generality concept  2–4, 20, 22–24, 34, 36–38, 45, 52–53, 171, 174, 182 generative linguistics  1, 3, 29, 36, 56, 79–80, 87–89, 97, 99, 109, 111, 117, 137, 140, 142, 143, 148, 166, 172, 175, 182 genitive -s  12, 32, German  7, 11, 33, 112, 138–139, 140, 145–146, 150–151, 156–158, 166–168, 170, 180 Germanic  7, 119, 125, 137–139, 143–150, 155–159, 164, 165–169, 180 North Germanic  16 Proto-Germanic  57, 155 global productivity  26

 Productivity gradience  2, 6, 9, 36–37, 44, 48–50, 52–53, 92, 173 H hapax  26–27 hapax-conditioned degree of productivity  26 happenstance  138, 161–162, 164, 168, 181 I idioms  15, 48, 92 idiosyncratic case  3, 79, 80, 87–89, 99, 109, 140, 143, 182 idiosyncratic set phrases/expressions  56, 63, 77, 92 incremental theme  122 inflection  5, 30, 146 inverse correlation  6, 9, 24, 28, 34–35, 44, 52–53, 89–91, 94–95, 98, 115, 139, 145, 166–167, 169, 172, 176–177 irregular verbs, see also strong verbs  3, 23, 48, 79, 80 irregularity  12, 80 K Kivunju  80 L language contact  7, 138, 146, 156–158, 168, 180 levels of schematicity  4, 6, 10, 34, 36, 45–47, 49–50, 52, 54, 67, 69, 72, 74, 76, 78, 91, 95, 98, 110, 135–136, 161–162, 174–176, 179 lexical case  3, 56, 79–80, 87, 89, 97, 109, 137, 139–143, 153, 165, 167, 169, 182 lexicalization  25, 27, 49, 76, 92, 176 lexicality–schematicity hierarchy  4–5, 10, 46–47, 69, 74, 110, 114–115, 162, 178 linking element  12 locatum  122 loss of case  137–138, 145–146, 156–157, 167, 180 M measurements of productivity  24–27

metaphors  22, 44–45, 48 motion locomotion  84, 134 non-translational motion  7, 65, 67, 71–72, 82 translational motion  65, 67, 71, 74, 82, 84, 125, 133 N nonce verbs  3, 6–7, 18, 98, 101–118, 137, 176–177 Norwegian  16, 17 O open schema  36, 40, 43, 175 P past-tense formation  23, 40–44, 48, 79, 104 performance  30, 33, 113, 160, 162, 172 periphery  80, 89 prefixes  11, 157 haupt-  11 productivity cline  2, 10, 19, 34, 36, 38–41, 43–44, 52, 89, 90, 92, 96, 134, 136, 172–173, 176, 179 productivity domain  6, 34, 39–40, 47, 74, 89, 95, 98, 102, 114, 116, 134, 173–176 R regular verbs, see also weak verbs  3, 23 regularity concept  2–4, 12, 18–20, 22–24, 29, 31–34, 37–38, 52–53, 171–172, 174, 182 regularization  32 relics  16, 11, 20 rules  2–3, 10, 12–14, 17–18, 22, 25–26, 29, 33, 37, 56, 79, 80, 88, 90–92, 96 S Scandinavian Modern Mainland Scandinavian  33 Old Scandinavian  32, 33 schematicity  2, 4, 6, 9–10, 14–15, 18, 34, 36, 38, 43–49, 52–55, 69, 74, 77, 91, 98, 110, 114–115, 118, 169, 172–174

semantic range  36, 44, 85, 101, 102, 112, 114–117 sense relations  9, 20, 24 antonym-to-antonym –> synonym relation  20–21, 24, 53, 171 co-occurrence relation  20–24, 171 elaboration relation  20–22, 24, 53, 171–172 entailment relation  20–22, 24, 171 metaphorical relation  20–21, 24, 53, 171 Sesotho  28 similarity  41, 43, 47, 50, 54, 89, 91, 145–146, 150, 153, 169 token similarity  10, 50, 93, 98 strong verbs  17, 23, 40, 41, 43, 50, 174 strung verbs  23, 51–52, 93 swept verbs  23, 51–52, 93 structural case  3, 56, 79, 80, 87, 89, 109, 137, 139–143, 153, 165, 167, 169, 182 suffixes  13, 16–17, 23, 25–26, 31, 41, 48, 157 -er  13 -ibility  17 -ibleness  17 -ing  17 -ity  25 -ivity  25–26 -iveness  25–26 -ning  17 -ness  25–26 -th  26, 27, 33 suppletion  48, 49 Swedish  7, 12, 119, 137–140, 143, 146–150, 156–158, 164, 166–169, 180 Old Swedish  142, 146–148, 164, 165, 169, 186 synonymy  7, 20, 43, 78, 85, 87–88, 90, 92, 97, 101, 104–106, 108–109, 111, 116–117, 137, 144–145, 148, 167, 171, 173, 175–177, 180 syntactic creativity  1, 29 T thematic case  3, 79, 80, 99, 109, 140, 143, 182



Subject index  translational equivalents  56, 78, 85, 86, 87, 88, 90, 93–94, 114, 175 transparency  10, 13, 21–22, 48, Turkish  80

U unconditioned degree of productivity  26 V verbs of instrument of communication  7, 98, 118, 119–136, 137, 178, 179

vocabulary replacement  156 W weak verbs  17, 23, 40, 41 word formation  25, 27, 30–32 wug-experiment  104

Constructions index A Accusative object construction, see also Nom-Acc, AccAcc  165 Accusative subject construction  57–58, 60, 116, 138, 144, 147, 150, 158–159, 166, 168, 181 Acc-Gen  57, 148, 150 B Basic event type constructions  6, 66–69, 72–74, 114, 175, 178 “Blended” construction  7, 138, 142–143, 147, 164–166, 168, 180 C Caused-motion construction  7, 36, 37, 71, 83–87, 97, 119–125, 128–136, 175, 176, 178, 179 D Dative subject construction  57–58, 60, 62, 101–102, 106, 108–117, 138, 144, 150–152, 158–168, 178, 181 Dat-Gen  58, 148, 150–153, 170 Dat-Nom  58, 138, 142, 151–153, 155, 164–165, 168–170, 180–181 Dat-only  105–106 Dat-PP  153 Dat-S  153 Ditransitive construction  7, 112, 119–120, 125–129, 131–132, 135–136, 178 G General constructions  14 Genitive object construction  147–148 Genitive subject construction  6, 57, 60, 62, 144, 147, 154–155 Gen-Nom  57 Gen-only  57

Gen-PP  57 Gen-S  57 H High level constructions Schematic constructions  5–6, 10, 37, 46, 54, 56, 78, 88, 91–92, 95–98, 101, 106–107, 109–111, 115–118, 177–178 I Inchoative construction  98–99 Incredulity construction  18 Intermediate level constructions Verb-class-specific constructions  5–6, 46, 55, 74, 107, 115, 136, 148, 169, 178 Verb-subclass-specific constructions  5–7, 46, 55–56, 87–88, 91, 93, 97–99, 115–116, 118, 135–136, 166, 175, 177–179 L Low-level constructions Item-specific islands/constructions  10, 48, 52, 95, 109–111, 115 Verb-specific constructions  5–7, 10, 42, 46, 52, 54–56, 69, 74, 76–78, 83, 85, 87–88, 91–94, 96–98, 104, 106–107, 109–111, 114–116–118, 135, 145, 152, 166, 169, 175–177, 179, 181 N Nominative object construction, see also Dat-Nom  165 Nominative subject construction  6, 55–56, 58, 60–62, 75–77, 88, 96, 115, 144, 147, 148, 150, 159, 165, 174–175 Nom-Nom  165

Nom-Acc  6, 33, 55–56, 58, 60, 62–63, 66–69, 71, 74–89, 91, 93–94, 96–97, 101, 104–107, 109–111, 114–118, 120–121, 123–124, 132–133, 138, 144, 150, 153–155, 168–169, 174–177, 181 Nom-Dat  6, 33, 55–56, 58, 60, 62, 69–89, 91, 93–94, 96, 97, 99, 101–102, 105–106, 108–112, 114, 116–117, 120–121, 123–124, 132–133, 144, 150–151, 154–155, 167, 169, 170, 174–176, 178 Nom-Gen  6, 33, 55–56, 58, 60, 62, 75–79, 96, 144, 148, 150, 153–155, 167, 174, 175 O Oblique passive construction  147–150 Oblique subject construction, see also Accusative subject construction, Dative subject construction, Genitive subject construction  6, 61–62, 144, 147–150 Obl-Obl  164, 165 S Semantically compositional constructions, see general constructions Semantically non-compositional constructions, see specific constructions Sentence-type constructions  5 Specific constructions  14 T Transfer construction  7, 65, 67, 105, 119–121, 124–126, 128–129, 131–136, 178–179 W Way-construction  31–32

In the series Constructional Approaches to Language the following titles have been published thus far or are scheduled for publication: 8 7 6 5 4 3 2 1

Barðdal, Jóhanna: Productivity. Evidence from Case and Argument Structure in Icelandic. 2008. xiii, 209 pp. Hilpert, Martin: Germanic Future Constructions. A usage-based approach to language change. 2008. ix, 205 pp. Iwata, Seizi: Locative Alternation. A lexical-constructional approach. 2008. xiv, 239 pp. Leino, Jaakko (ed.): Constructional Reorganization. 2008. vi, 155 pp. Fried, Mirjam and Hans C. Boas (eds.): Grammatical Constructions. Back to the roots. 2005. viii, 246 pp. Östman, Jan-Ola and Mirjam Fried (eds.): Construction Grammars. Cognitive grounding and theoretical extensions. 2005. viii, 325 pp. Fried, Mirjam and Jan-Ola Östman (eds.): Construction Grammar in a Cross-Language Perspective. 2004. vi, 209 pp. Kuno, Susumu and Ken-ichi Takami: Functional Constraints in Grammar. On the unergative– unaccusative distinction. 2004. ix, 242 pp.