No title

JOURNAL OF

SEMANTICS. VOL. I- 1982

1�1

JOURNAL OF SEMANTICS

(1982)

CONTENTS VOLUME I

page

Editorial statement

1 -2 Articles

Beesley, K.R.: Evaluative adjectives predicates in Montague gram mar

one-place

as

•

•

•

•

•

•

.

.

•

•

•

•

•

•

1 95-249

.

van Benthem, J. and van E ijck, J.: The dynamics of interpretation

•

•

.

•

.

•

.

.

.

.

.

.

•

•

•

Bree, D.S.: Counterfa ctuals and causality

.

•

•

.

•

.

•

.

•

.

.

.

•

•

.

.

•

•

.

•

•

.

.

.

.

.

.

•

•

.

3-20

1 47-185

.

van Eijck, J.: cf. van Benthem, J. Garrod, S.C. and Sanford, A.J.: The mental represen­ tation of discourse in a focuss ed memory system: Implications for the interpretation of anaphoric noun phrases •

•

•

•

•

•

•

•

.

•

•

.

•

•

•

•

•

•

•

•

•

•

•

•

•

•

. . . 21-41

•

Hintikka, J. and Kulas, J.: Russell vindicated:

Towards a general theory of definite descriptions

•

•

.

.

•

.

.

.

.

•

.

387-397

•

Kulas, J.: cf. H intikka, J. Kuno, S.: Principles of discourse deletion - case studies from English, Russian and Japanese

.

•

•

•

. . . 6 1 -93

Kuroda, S.-Y : Indexed predicate calculus . .

. . . 43-59

.

Marslen-Wilson, W . : cf. Tyler, L . K . Martin, J . N . : Negation, a mbiguity, and the identity test

.

.

.

Read, S.: Disjunction

.

•

.

.

.

.

.

.

•

•

•

•

.

.

.

.

•

.

•

•

•

.

.

.

•

•

.

25 1 -274

.

.

.

.

.

•

.

•

•

•

.

•

•

•

.

.

•

•

•

•

•

.

•

•

.

•

275-285

Reichgelt, H.: Mental models and discourse Richards, B.: Discourse and deixis

•

.

.

.

.

•

.

•

.

•

.

•

.

•

•

•

•

•

.

.

•

. .

.

.

•

.

.

.

•

•

37 1 -386

•

•

347-369

•

Rigter, B.: Intensional domains and the use of tense, perfect and modals in English

.

•

.

•

•

•

•

•

•

.

•

.

•

•

•

•

•

95- 1 4 5

Sanford, A.J.: c f . Garrod, S.C. Tasmowski-De Ryck, L. and Verluyten, S.P.: Linguistic control of pronouns

.

.

.

.

•

•

. . . . . . . . . .

.

323-346

Tyler , L .K. and Marslen-Wilson, W.: Processing utterances in discourse contexts: On-line resolution of anaphors .

•

.

.

.

.

.

.

•

•

.

•

•

•

•

•

.

.

.

.

.

.

.

•

•

•

297-3 1 4

Verluyten, S.P.: cf. Tasmowski-De Ryck, L. Yule, G.: Interpreting Anaphora without

. 3 1 5-322

identifying reference . . . . . . . . . . . . . . . . Special Issue

Editor 's preface

.

•

.

.

•

.

.

.

Anaphoro (vall, no.4)

on

.

.

..

.

.

•

.

.

.

.

.

.

.

.

293-397

.

.

.

•

Towards a genera l theory of definite descriptions

.

•

.

294-296

.

Hintikka, J. and Kulas, J.: Russell vindicated: Reichgelt , H.: Mental models and discourse Richards, B.: Discourse and dei:ris

.

.

.

.

.

.

.

.

.

.

.

• .

Tasmowsk1-De Ryck, L. and Verluyten, S.P.: Linguistic control of pronouns . . . . . . . .

.

.

.

.

•

.

.

.

•

.

•

.

•

•

.

. •

•

•

•

.

. 387-397

.

•

.

•

.

. . . . . .

•

.

3 7 1 -386

.

347-369

.

.

.

323-346

Tyler, L.K. and Marslen-Wilson, W.: Processing utterances in discourse contexts: On-line resolution of anaphors •

•

•

•

•

•

•

•

•

.

Yule, G.: Interpreting Anaphora without identifying reference . . . ... . ...

.

.

.

. . . . 297- 3 1 4

.

. . . . . . . . . . . . . . 3 1 5-322

.

Short Notice

Sterelny, K.: Against conversa t ional implicature

.

.

.

.

.

.

.

.

.

.

1 87- 1 94

.

Discussion

van der Auwera, J.: Against "Against conversa tional implicature

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

•

.

•

399-400

.

Book Review Ballmer, T. and Brennenstuhl, W.: Speech Act

Classification. A study in the lexical analysis of English speech activity verbs

(reviewed by Jens Allwood)

•

•

.

.

•

•

•

•

.

•

•

.

•

.

•

•

•

•

.

.

•

•

.

287-290

JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE

VOLUME 1,

no.

3/4, 1982

Reprinted with permission of the original publishers by

SWETS & ZEITLINGER BV LISSE

-

THE NETHERLANDS

-

1994

EVALUATIVE ADJECTIVES AS ONE-PLACE PREDICA TE5 IN MONT AGUE GRAMMAR•

Kenneth Reid Beesley

A bstract

•

I

Introduction

Adjectives form a very diverse class and present a challenge to any linguistic or logical analysis. To avoid confusion in a field which has become a terminological jungle, I feel it is important to present a few definitions and clarify a few usages at this point. I will speak of 'syntactic predicates', 'syntactic predicate position', or just 'predicate position' when referring to the appearance of adjectives, without accom­ I will panying nouns, after copulas in strings such as The sky is blue. speak of 'syntactic attributives' or 'attnbutive position' when referring to adjeCtives appear ing in strings together · with the noun they modify, as in blue sky, big {lea and good plumber. From the semantic point of view, a 'predicate' or 'one-place " p redicate' will be taken to be a iunction whose domain is the set of entities JOURNAL OF SE MANTICS, Vol.!, no.3, pp. 1 95-249

195

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

good, bad, In this paper I will argue that evaluative adjectives, such as clever and skilful, should be analysed as one-place predicates in, logical translation. This approach, which is basically the traditional logical trea tment of 'absolute ' adjectives, is to be contrasted with the approach in Montague (1974a) and Parsons (1972), wherein all adjectives are translated as two-place predicates, i.e. as semantic a ttributives. The move a way from the Montague-Parsons analysis is not new: Bartsch (1972, 1975), McConnell-Ginet (1973), Kamp (1975), Siegel (1976a, 1976b, 1979), Keenan &. Fultz (1978) and Klein (1980) have similarly advocated one-place predicate status, a t least for fairly stra ightforward qualities (e.g. red, carnivorous, stony) and even for degree adjectives (e.g. tall, short, heavy and old). Evaluatives, however, remain troublesome: Kamp concluded that their status was uncertain, and Siegel classified them as two-place predicates after much argument. My remarks are directed primarily against Siegel 's analysis; I intend to show that there are syntactic tests, some suggested by Siegel herself, which argue persuasively that evaluative adjectives should be interpreted as one­ place predica tes.

K ENNETH R. BEESLEY in a model and whose range is the set { 0, 1 } . More simply, a predicate like bam ' takes one argument x, written bam'(:r), and returns 1 (true) if x is a barn and 0 (false) if it is not. Alternately, bam' can be seen as representing the set of all barns in a model, and bam '(:r) will be true iff x e: barn'. I will speak of 'two-place predicates' or 'semantic attributives' when referring to analyses of the Montague-Parsons type; examples w ill follow shortly. The distinction made here is important, but that does not mean that there are not significant correlations between syntactic and semantic predicativity and attributiveness. Indeed, by arguing that evaluatives can and should be interpreted as one-place predicates,
,

Logics have traditionally treated adJectives as one-place predicates, and this sol\.ltion works nicely if one chooses his examples carefully. Red, for example, has usually, but not always, been considered an absolute adjective, and the corresponding one-place predicate is written red '. If we know that some entity x is a red barn, then the conjunction analysis in ( 1) supports ( I a) and ( l b) as logical consequences or entail­ ments. (1)

).. y

[red'(y) & barn '(y)] (x) ( I a) barn '(x) ( l b) red'(x)

That is, if we know that some entity is a red barn, we know intuitively that the entity is a barn and that the entity is red. Furthermore, a red barn is a red building and a red entity. The 'absolute' nature of red means that its meaning is not changed by the type of noun it mod ifies. It makes sense to speak of 'redness' or 'being red' pure and simple. Adjectives with this quality are called 'intersective ' , 'predicative' and 'absolute' by various authors. Some other, more (a) entailment, and a If w e know that an the ent ity is a flea, 1 96

troublesome, adjectives seem to allow only the few allow neither the (a) nor the (b) entailment. entity is a big flea, for instance, we know that but w e can object that the entity is not big. A JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The following names of adjective classes will suffice to get, the di scussion going: 'Absolute' (sometimes called 'predicative') adjectives are taken to include, for the time being, red white, blue, camivorous, metallic and, in their literal senses, pregnant, magnetic, virgin, dead and alive . 'Degree ' (sometimes called 'measure' or 'measuring') adjectives include big, small, tall, short, high, low, fat, thin, young, old, etc. ' Evalu­ ative ' (also called 'manner') adjectives include superb, excellent, skilful, beautiful, good, fair, poor, bad and awful. 'Relative' adjectives are taken to be a motley bunch including mere, chief, a lleged, ostensible, purported, fake and imita tion. These groupings w ill be challenged, defended, sub­ divided and collapsed as the analysis progresses.

EVALUATIVE A DJ ECTI V ES big flea can be a small animal. (2) x is a big flea (2a) x is a flea (2b) *x is big

(3) x is a good th1ef (3a) x is a thief (3b) *x is good The solution to this apparent failure of the conjunction analysis has generally been to treat degree adjectives like big and evaluative ad­ jectives like good as relative rather than absolute adjectives. Size is always relative to something, a comparison class, and it makes no sense to ascribe bigness or smallness except relative to such a class. A flea can be 'big . f or a flea' but 'small for an animal ' . Goodness and badness are also relative, but in a slightly different way. Evaluation is relative to some criterion or relevant function performed by the entity being evaluated. Thus, a woman can be 'good qua mother ' but 'bad qua wife'. It is here that the concept of a two-place predicate is sometimes used. Instead of saying simply that some entity is 'tal l ' , one says that that entity is 'a tall N ' , where N is the comparison class. Instead of saying that some entity is 'good ', one says that it is 'a good N ' . where N indicates the relevant role. I n logical syntax, both tall ' and good ' apply first to this N, yielding a one-place predicate which can then be predicated of the subject just like the one-place predicate red'. Sim­ plifying a great deal, we can characterise the traditional analyse� of the syntactically attributive adjectives we have met so far. (4) John is a blue-eyed jockey. (absolute adjective) (4a) A x [blue-eyed '(x) & jockey '(x)] (John) (4b) blue-eyed'(John) & jockey '(John) (de g ree adjective) (5) John is a tall jockey. (5a) >.x [( tall '(jockey •))(x)] (John) (5b) tall '(jockey ')(John) (evaluative adjective) (6) John is a good jockey. (6a) A x [(good '(jockey')J{ x)] (John) (6b) good'(jockey')(John) JS, voi . J , no.3

197

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

If big flea were analysed using the simple conjuction analysis used for red bam, then (2b) would follow automatically, and to hold that 'a big flea is small' would seem to involve a contradiction, that some­ thing can be big and small at the same time. Sim tlarly, If we know that someone is a good thief, then it follows that he is a thief, but not that he is absolutely good. A good thief can be, and usually is, a 'bad man. A woman can be a good· mother and a bad w ife simul­ taneously.

KENNETH R. BEESLEY In this analysis, whlie blue-eyed, tall and good can all be syntactically attributive, only ta ll and good correspond to semantic attributives, that is, to two-place rredicates. Blue-eyed' turns out to be semantically pre­ dicative even when the surface adjective appears as a syntactic attri­ butive. Conversely, these same adjectives can all appear in syntactic predicate posit10n, but sentences like (8) and (9) must be taken as ell iptical versions of the more �:omplete sentences (8a) and (9a).

Before moving on to a more detailed discussion of these analyses and claims, it is worth noting that some adjectives (the 'non-standard modi­ and fiers ' of Parsons 1972: 1 30) are even more troublesome than good tall. If we know that person x is a good thief, then we at least know that x is a thief. But if we know that x is an alleged thief, than we cannot even conclude that x is a thief . And if x is a pretend thief, then we must conclude that he is not a thief at all. Anxious to avoid any spurious logical consequences ansing from a conjunction analysis for such adjectives, Montague ( l 974a: 2 1 1- 1 3; henceforht EFL) goes to the extreme of denying conjunction analysis (and so one-place pred­ icate status) to all adJectives. The result is an analysis where all adjec­ tives correspond to semantic attributives much as in (5) and (6) above. Montague rejects the idea of defining multiple classes of adjectives, some of wh ich , the ' mtersectives' , would translate to allow (a) and (b) entailments as in ( 1), others of which, the 'subsectives ', would translate to allow the (a) entailments as in (2) and (3), and still others of which, like reputed and alleged, would allow neither; he felt such a solution would detract from the 'conceptual simplicity' of his untform treatment. As a result, all adjectives in EFL are treated as two-place predicates, ad-com mon nouns denoting functions from intensions of properties to properties. ( 1 0) and ( 1 1 ) show typical syntactic and seman­ tic trees. ( 1 0)

(I I)

big '('flea')

/ big' .

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(7) John is blue-eyed. (8) John IS tall. (8a) John is a tall N. (where N is a noun) (9) John is good. (9a) John is a good N. (where N is a noun)

"'

flea'

false '('friend ')

/

false '

� frtend'

.This blanket solut ion avoids invalid consequences; in fact it leaves_ Montague w ith no consequences at all. He cannot even show that a 198

JS, vol . l , no.J

EVALUATIVE ADJECTIVES red bam is red and a red bam is a bam are valid sentences. To patch up

this obvious deficiency, Montague mtroduces meaning postulates such as ( 1 2) and ( 13).

every or; is a r; I where 6 is an intersective or subsective adjective and r; is a common noun ( 13) I every 6z; is 6 I where 6 is an intersective adJective and c; is a common noun ( I 2)

I

In sp1te of Montague 's example, a number of writers in the general field of Montague Grammar, e.g. Dowty ( I 976: 2 1 0), continue to treat at least some adjectives as simple one-place pred1cates. Others like Kamp ( 1 97 5), Bartsch ( 1 973) and K lein ( 1 980) have argued that the Montague-Parsons analysis cannot be generalised to handle comparatives correctly; they claim that even degree adJeCtives should be translated as one-place pred1cates. Kamp ended his paper ( 1 975: 153-4) wondermg just how many other types of adject1ve could be g1ven predicate status; noting that a lleged seemed a hopeless case, he contmued: "The same can be said to be true, to an almost equal degree, Where p r ecisely we of adjectives such as fake, skilful, or good. should draw the boundaries of the class of adjectives [which are one-place predicates] I do not know. For example, does skilful belong to this class? Surely we must always ask 'skilful what? ' before we can answer the question whether a certain thing or person is indeed skilful; this suggests that the theory is not applicable to the word skilful . Yet there appears to be some plaus1billty in the view that having a good deal of skill does func­ tiOn as a predicate--be it a highly ambiguous one as there are so many skills." Siegel ( 1976a, 1 979) argued forcefully for treating degree adject1ves as a subset of the class of absolute adjectives, but she drew the line firmly at the evaluatives, which remained two-place predicates a!> in Montague ( 1 974a). I will proceed by f1rst examming the way in which degree adjectives have been elevated to absolute status. The challenge is to do this JS, vol. l , no.3

1 99

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

By way of example, red is an intersective adjective and tall 1s a subsec­ tive adjective in Montague 's term inology. Meaning postulate ( I 2) allows one to show that a red bam is a bam and a big flea is a flea a r e v a I i d but sentences. Sim ilarly, ( 13) supports the validity of a red bam is red not a big flea is big. These rules allow Montague to ass1gn all adjectives the same syntactic category and logical translation while still getting out JUSt the nght consequences. However, Montague is getting no free lurtch: he avoids adjective subclasses m the grammar, but the meanmg postulates have to refer to such subclasses m the end.

K ENN ETH R. BEESLEY without causing undue problems for logical consequence. I will then show that evaluative adjectives can be handled in much the same way. Some new tests involving adverbial modification of adjectives will lend support to my clai ms. In the end, it will be shown that extending one-place predicate status to evaluatives results in more consistent syntax and semantics, with a closer mapping between the two. 2 Degree Adjectives as One-Place Predicates 2.1

Traditional Analysis o f Degree Adjectives a s Two-Place Predicates

( 1 4) ( 1 4 ')

John is a tall man. Tall(John, >.x(x is a man))

&

John

£

>. x(x is a man)

The translation in ( 1 4') contains two clauses. In the first,tall is shown to translate as a two-place predicate holding between an entity and a set of entities. The set of entities, in this case the set of men, is the comparison class by which John's tallness is to be judged . This first clause is properly read as 'John is tall compared to men' rather than as ' John is tall for a man'. The second paraphrase presupposes that John is a man, and it is important to preserve a notation for representing a comparison between an entity and a set of which it is not a member. The second clause in ( 1 4 ') supports the intuition that if John is a tall man is true, then John is a man is also true, making the latter follow as a logical consequence. As we saw in section I, the less abstract Montague-Parsons analysis of tall man does not produce this logical consequence; meaning postulate ( 1 2) must be invoked. The relativity of tall is captured in the variability of the comparison class. John is a tall dwarf receives the interpretation in (15), which indicates that John's height is now to be considered relative to the set of dwarves. ( 1 5) Tall(John, >.x(x is a dwarf))

&

John

£

>. x(x is a dwarf)

Intuitively this accounts for the fact that the tallness of John can vary significantly when the comparison class changes from dwarves to men to basketball players or whatever. At first sight, (14') appears to be a simple and correct solution for characterising the relativity of degree adjectives. Any syntactically attributive construction A+N, where A is a degree adJective and N is a noun, will be translated with N representing the relevant comparison class. There are, however, two reasons why ( 1 4 ' ) is an incorrect transla200

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We may take Wheeler ( 1 972) as a typical analysis of degree adjectives as semantic attnbutives. Wheeler translates ( 1 4) as ( 1 4 ').

EVALUATIVE ADJECTIVES t ion of degree adjectives: First, I will show that the translation is too rigid to account for the data, and second, I will show that the relativity of degree adjectives can be better handled by appealing to context. 2.2 Deriving the Comparison Class 2.2. 1 Syntactic Attributive Constructions

( 1 6) Jones is a tall old man. These two readings, which involve, presumably, different bracketings of the noun phrase elements following tall are representable in Wheeler's system as ( 1 7) and ( 1 8). ( 1 7) Tall(John, AX(old(x , Ay(y is a man)) & x E: Ay(y is a man))) & John E: A x(old(x, A y(y is a man)) & x E: A y(y is a man)) ( 1 8) Tall(John, AX(x IS a man)) & John E: A x(x IS a man) & Old( John, A x(x is a man)) & John E: A x(x is a man) Sim ilar examples are discussed in Parsons ( 1 972: 1 32-3), Keenan & Faltz ( 1 978: 3 1 8- 1 9), and in most other works on the subject (Siegel 1 976: 1 30-4; 1 979: 247-8; McConnell-Ginet 1 973: 89, 1 33-4). The formalisms start getting unmanageable, and the intuitions start failing, however, as the noun phrases get more complicated. A sentence like John is a tall old fat man has one reading, for Wheeler, where the tallness is relative to old fat men, and "given the present order of its attributives, the sentence can be understood m at least two other ways", which again appear to involve different surface bracketmgs (Wheeler 1972: 32 1 ) . I will spare the reader the translations given for these readmgs (which are all, not unexpectedly, full of errors in the original paper anyway) and concentrate on the qualification by Wheeler that there are "at least" two other readings. It is com mon be under­ practice to Cite the simplest cases where "it will generally stood that the relevant class for grading is that specified by the noun" (McConnell-Ginet 1 973: 89; emphasis mine) and to avoid anything nastier. There is some uneasiness in the observations. If comparison classes are derived mechanically from syntactic structures, then it should JS, vol . 1 , no.3

20 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The problem with the Wheeler analysis is that it is not so easy to derive the relevant comparison class mechanically from the syntax. ( 1 4 ' ) shows the modified noun functioning automatically as the compari­ son class, but there arise difficulties with more complex noun phrases. Wheeler himsel f ( 1 972: 3 1 4) claims that ( 1 6) has two. readings, one ascribing tallness to Jones relative to old men and the other ascribing both tallness and age to him relative to men.

KENNETH R. BEESLEY always be possible to predict exactly how many readings are theoretical­ ly possible for any finite noun phrase construction involving degree adjectives. Siegel ( l 976a: 1 29-1 35; 1979: 247-8) goes farther than most in discuss­ ing the problems of interpretation of double degree adjectives as in ( 1 9) and horrors like (20). ( 1 9) Btlly is a heavy fast runner. (20) Btlly is a tall little red-headed basketball player.

"Presumably, it is the context of utterance that allows us to understand how much is included in the comparison class. It would appear that w ith prenominal measure [ 1.e. degree] adjectives, the comparison class may be taken as bemg either the set p1cked out by the basic common noun modified, or by another, more complex common noun to the right of the measure adject1ve." Going a step beyond Siegel, I claim that the comparison class, even in the simplest A+N constructions, may not be the set denoted by the noun at all. Consider the following conversation at a lmgu1stics convention. ( 2 1 ) Q: Which of the men over there is Quang? A: Quang is the short Vietnamese. Now there is a perfectly good, even preferable reading of short Vietnam­ ese in ( 2 1 ) which involves men' rather than Vietnamese 'as the comparison class for tall. That is, short Vietnamese can be paraphrased as 'short

compared to men and a Vietnamese ' . It is perfectly possible to hold that Quang is a short Vietnamese but that he is not short for a Vietnam­ ese. Wheeler 's analysis, with its mechanical, syntactic translation, is incapable of captunng such usages. 2.2.2 Syntactic Predicate Constructions Another problem for any Wheeler-like theory is to account for degree adjectives in syntactic predicate position. (22) John is tall. If such adjectives are · taken to be semantically attributive, then the 202

JS, vol. l , ;)u.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Noting that (20) seems to allow a reading where tallness is measured in comparison to little basketball players, skipping over red-headed alto­ gether, S iegel concludes, as did McConnel-Ginet, that context rather than syntax must p1ck out the comparison class.

EVALUATIVE ADJECTIVES noun (or predicate or set) used for comparison must be slipped into the translation by some fashion. There are basically two ways of doing this, both of which amount to the same thing for the purposes of this discussion.

For convenience, I will call such a process 'dummy noun deletion' regardless of the actual mechanism involved. Any actual attempt to interpret a degree adjective in predicate position would seem to involve something like (23B), which ulti mately relies on context. There have been claims, however, that the comparison class in such cases can be derived m an orderly way from semantic codings in the lexicon. Katz ( 1 967) presents the examples in (24), where the comparison classes are claimed to be named by the nouns m square brackets.

{

(24) The

skyscraper man flea United States tarantula

is big

[buildings] [humans] [insects] [countries] [spiders]

In general, "the class of entities with which something IS compared IS a category to which that thing belongs" (Katz 1 967: 1 86-1 87). Katz The sky­ even claims that "we cannot render The skyscraper is big as scraper is big for a physical object or in a more extreme case, The flea is big, as The flea is big for an animal". The choice of comparison class is taken to be li mited to the next highest concept in the great semantic chain of being. Sim ilar notions have appeared in Chafe ( 1 970: 1 95) and even in Bartsch ( 1 972), but the inadequacies of this approach are notorious (see McConneii-Ginet 1 97 3: 89; Damerau 197 5; Siegel 1 976a: 129; 1 979: 243 -244; Bierwisch 1 969: 1 6 5- 1 68). For instance, it is clearly possible, Katz ' claims to the contrary notwithstanding, to interpret The {lea is big as 'The flea is big for an animal', and such a reading can be useful and natural in the context of science fiction. The problem of determining the comparison class for derived lexical ite ms, which have no dictionary codings, will also be a problem. Chafe, Bierwisch JS, vo1. 1 , no.3

203

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(23) (A) An adjective like tall can be taken to apply to a noun as usual, yielding strings like John is a tall boy. Then a transformation can apply, deleting the noun (and deleting or suppressing the indefinite article) yielding John is tall. The predicate boy', or what­ ever, will remain in the semantic translation. (B) Alternatively, tall can be taken to apply to a dummy noun, call it b. , which never receives phonological form and so never has to be deleted. In the semantics, b. is instantiated with a nominal predicate by the context.

KENN ETH R. BEESLEY and Bartsch appreciate- that genericness and specificness can weight the choice, and Bartsch ( 1 972: 1 65) and Sampson ( 1 970: 257) point out further that proper name subjects like John give little clue as to what comparison class applies - so the context has the last say again. While attempts to assign comparison classes mechanically from lexical codings may have something to say about default assignments, it is clear that context can be an overriding factor.

2. 3

Appeals to Context

The solution of McConnell-Ginet ( 1 9 7 3: 1 1 5) is to employ 'delineations ', contextual variables which fix the extension of degree adjectives. The basic idea is that predicates apply with respect to a relevant delineation, and the formalism indicates a 'd', for delineation, as a subscript to any degree adjective. (25) John is tall. (25') talld(John) In the same spirit, Klein ( 1 980: 1 4- 1 6) indicates contextually specified comparison classes, written 'c ', as a subscript to formulas. (26) [ tall(John))c ' For reasons which will become clearer later, I prefer Klein's notation and w ill use it whenever the notion of comparison class needs to be made explicit. Along the same general lines are analyses proposed by Bartsch ( 1972: 1 65) and Keenan & Faltz ( 1 978: 1 78-1 79). These grammars interpret degree adjectives in such a way that comparison classes (or, in the case of Bartsch, averages over comparison classes) are built directly into them. The important point, for this discussion, is that all these schemes translate degree adjectives as one-place predicates.

20�

JS,

vo!. l,

no.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Given that only context can be the ultimate arbiter in the choice of a comparison class in the examples 1iscussed above, the question remains of how this comparison class 1s to be represented in logical translation. Wheeler's analysis ( 1 4 ') cements the class in as one argu ment of a two-place predicate. It is not wise, however, to treat a contextual factor on the application of a predicate as a syntactic argument . In addition, the fact that syntactically attributive A+N constructions may have companson classes distinct from N destroys any. syntactic justification, by analogy, for semantic attr ibutives.

EVALUATIVE A DJECTIVES 2.4 Syntactic Analogy Arguments 2.4. 1

Introduction

The common way of defending a solution like Klein 's (26) over on€­ like Wheeler 's ( 1 4 ' ) is to show that degree adjectives behave syntactical­ ly like absolute adjectives rater than like other adjecti ves thought to be irredeemably relat' i ve. Then by analogy, if absolute adjectives translate as one-place prPdi:::a tes, so should the degree class.

Stege! ( 1 976: 52-54 ; ' 1 979: 228) believes that there is, m most languages, a strong correlation between predicate position and one-place predicate status for adjectives. Degree adjectives are like absolute adjectives, and unlike many of her relative adjectives, in their ability to appear grammatically in predicate position. (27) Mary is pregnant. (28) Mary is tall. (29) *Mary is: •nere/alleged/ostensible.

(absolute) (degree) (relattve)

If ta ll is taken as a relative adjective so that (28) has to be generated by way of d4m my noun deletion, then something would have to be added , to the grammar to prevent the transformation from producing the strings in (29), which also involve relative adjectives. For Siegel, who has but one class of relative adjective, thts would involve the ad-hoc assignment of a feature [+dummy deletion] to all those relatives, like tall, which happen to appear in predicate position. If, however, tall is taken to be a subtype of absolute adjectives, then both (27) to absolute and (28) can be generated d i rectly by a rule which adds be adjectives to form syntactic predicates. The strings of (29) will never be generated by such a rule, and dummy noun deletion can be defenes­ trated. The net result is one less transformation and a closer translation between the syntax and the logtc. Si milar arguments for ' less abstract' analyses are found in Thomason ( 1976), K lein ( 1 980: 1 4- 1 5), Keenan & Faltz ( 1 978), and McConneli-Ginet ( 1 973). We shall see that Siegel's analysis of evaluative adjectives forces her to retai'n dummy noun deletion and adopt ad hoc lexical exceptton features. 2.4.3. The One Prefor m Argument Only certain adjectives seem to allow their arguments to be referred to by one as in (30) to (33). Aga11;, deg•ee adif'ctives behave 'like the absolutes.

;!05

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2.4.2 The Predicate Position Test

KENNETH R. BEESLEY (JO) This ( 3 1 ) This (32) *This (33)?*This 2.4.4

is is is is

a red box, a big box, an alleged an alleged

and that is a blue one. and that is a small one. thief, and that is a mere one. thief, and that is an ostensible one.

The Very Test

1 44, 1 50; Levi 1 978: 20; Keenan & Faltz 1 978: 1 64; Bartning 1 976: 9- l l , 58-59; Vendler 1 968: 109; Bolinger 1 972: 1 5). That is, pregnant, car­ nivoroWJ, virgin and other such all-or-nothing adjectives must take on secondary readings in very pregnant, very camivoroWJ and very virgin. might A woman either is pregnant or isn 't pregnant, but very pregnant refer to a woman's size, degree of morning sickness, craving for pickles, etc . Lees ( 1 963: 1 80- 1 8 1 ) sug g ests that all 'bona-fide' adjectives can be modified by very and Levi ( 1 973: 344) goes so far as to say that words like pregnant and virgin, by virtue of their resistance to very, are not adjectives at all but rather nouns. To add to the confusion, Wheeler ( 1 972: 3 3 1 ) holds that the possibility of modification by very is a sure sign of a two-place predicate.

It seems clear that if the present class of absolute adjectives is to be retained, then an internal distinction must be made between gradables and non-gradables. The gradables will allow modification by very, extremely, remarkably, quite and other such words. Semantically, the difference lies in the way an adjective div1des up the universe of a model. A predicate like pregnant ' sharply divides all entities into the pregnant and the non-pregnant, with nothing left over. The bounda­ ries of tall, on the other hand, will be a bit fuzzy, and between the group of tall entities and the group of short entities there may be a group of entities which are neither short nor tall. This last group lies in the 'extension gap'. (34) (35)

-pregnant

+pregnant

+short

+tall extension gap

If we postulate that very can modify any one-place predicate with a possible extension gap, then we can show once again that degree adjec­ tives act more like the absolutes than like the relatives. 206

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The use of very as a syntactic test for adjectives is very tricky. Because *very a lleged and *very mere are so grcssly ungrammatical while very red and very tall are so natural, one is tempted to advance modification by very as a test for one-place predicates. But some adjectives usually classified as absolutes par excellence do not allow modification by very w ithout taking on secondary meanings (Sapir 1 944; K iefer 1 978:

EVALUATIVE ADJ ECTIVES (36) (37) (38) (39)

The very red woman finally left the beach. The very tall woman got the job. *They sent me a very mere boy. *The very alleged thief was freed.

This test, if it is useful at all, can only provide posttJve evidence that an adjective is a one-place predicate. Failure to take very could in­ d icate either that the adjective lacks an extension gap or that it is not a one-place adjective at all. We shall see that Stegel's classification of evaluative adjectives forces her to deny even limited validity for the very test.

Yet another test proposed by Siegel ( 1 976a: 52, 1 2 3) is that only one­ place predicates can function as non-restrictive modifters of noun phrases. (40) (4 1 ) (42) (43) (44)

We all know naughty Nancy. We all know btg Bertha. *We all know mere Jonathan. *Ostensible Jonathan went home. *Say hello to alleged Bill.

(absolute) (degree) (relative) (relative) (relative)

2.4.6 The See-Catch-Find Test S tege! argues ( 1 979: 237-239; 1 976a: 76ft, 92) that the see-catch-find and there construction in (45) and (46) select unerringly for one-place predicates, and she uses these tests as a key tool in motivating her classification. Her examples again show degree adjectives behaving like absolutes. (45) (a) The warden saw the swimmers nude. (b) I've often seen the grass tall around that house. *(c) The warden saw the swimmers veteran.

(absolute) (degree) ( relative)

(46) (a) There is a bat asleep in the bathroom. (b) There were two pigs clean in the whole pen. *(c) There are several crt mes actual in the story.

(absolute) (degree) (relative)

In fact, Siegel gravely misunderstands these tests, which select not for one-place predicates as a whole, but rather for one-place predicates which are 'temporary ' in the carefully defmed technical sense of Bolinger ( 1 952: 42-43; 1 967: 9- 1 4), Bauer ( 1 975: 1 1 3, 1 1 9) and Dowty ( 1 975: 582-583). Dowty prefers to use 'temporary' and 'permanent' rather than Bolinger's terms 'accident ' and 'essence' (Bolinger 1 972: 38-39, 47-48; 1 973) because he thinks the latter terms to have inappropriate, JS, vol. l , no.3

207

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2.4.5 The Non-Restrictive Modifier Test

KENNETH R. BEESLEY m isleading connotations from their other uses in philosophy. Ironically, Bolinger adopted the terms 'accident' and 'essence' because of a similar concern about the connotations of 'temporary' and 'permanent'. In Siegel's case, Bolinger's fears are justified; she misinterprets 'temporary ' and 'permanent' in literal, everyday senses, a n d goes t o great lengths to demonstrate that some adjectives appearing in see-catch-find sentences are not temporary at all. This misunderstanding seriously flaws Siegel 's analysis, for English as well as for other languages she considers. 1 2.4.7 Nominalisation and Verbalisation Tests

(47) Adjective red tall big intelligent mere alleged ostensible

Noun redness tallness, height bigness, size intelligence *mereness, *merity *allegedness *ostensibi lity

The general intuiton is this: being red or tall involves partakmg of the qualities of redness or height respectively. But there is no way to make sense of a property of *mereness or *allegedness in the same way. The nouns allegation and ostentation are, of course, related to alleged and ostensible, but these are abviously not essences of a predicate like redness and size. Similarly, verbalisations, verbs with rough ' make Adjective' readings, and actual 'make Adjective' phrases appear to correspond only to one­ place predicates and never to relative adjectives.

(48� Adjective

red tall old big small dead pregnant mere alleged 208

Verb redden, make red, rubrify make tall make old, age make big, increase, augment make small, decrease kill, cause to become dead impregnate, make pregnant *make mere, *merify *make alleged, *allegify JS, vol . I, no.�

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Some people working on relative adjectives (e.g. Bartning 1 976; 1 0) have claimed that only those adjectives which are one-place predicates have meaningful nominalisations or corresponding nominal forms. Allow­ ing some slightly abstract, stilted nouns, the results of this test seem to support the others.

E. VALUATIVE

ADJECTIVES

but this verb is not reA llege, of course, is a verb related to alleged, lated to an adjective in the same way that impregnate and age are. 2.5 Rescuing the Conjunction Analysis

A major problem remains. The decision to treat degree as a kind of absolute adjective means that both will receive tion analysis. Yet it was the apparent failure of degree to support a conjunction analysis which origmally led to their tion as semantic attributives. Geach ( 1 956: 33) wrote

adjectives a conjunc­ adjectives classifica­

" 'Big ' and 'Small ' are attributive; 'x is a big flea' does not split up into 'x is a flea' and ' x is big ' , nor 'x is a small elephant' into 'x is an elephant' and 'x is small ' ; for if these analyses were legitimate, a simple argument would show. that a big flea is a big animal and a small elephant a small animal." In fact, a proper appreciation of context makes the problem disappear. Geach assumes that for a sentence like 'x is a big f lea ' , the comparison class must necessarily be the set of fleas. This is the same assumption which I showed to be inadequate in section 2.2. 1 . Even for simple A+N constructions like short Vietnamese and big flea, context can force comparison classes different from N altogether ( 2 1 ) . Also, Geach as­ sumes that 'x is big ' , where x is a flea, will necessarily have the com­ parison class of animals. This is an implicit statement of the Katz ( 1 967) method for automatic computation of comparison classes f rom lexical codings, which is similarly inadequate (see section 2.2.2). I have shown that context must choose the comparison class by which any sentence of the form 'x is big' will be evaluated. Let us assume that (49) is evaluated in a context where the compar­ ison class is c ' , where c' is the class of fleas. A conjunction analysis with contextual specification w ill look like ( 50). (49) x I S a big flea (50) [big '(x)]c• & flea'(x) Now let us assume that the pair of sentences in ( 5 1 ) is evaluated relative to the same context c ' . The translation will be (52). ·

JS, vol. l , no.3

2 09

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In summary, I have shown some of the history of degree adjective analysis, �hich shows an increasing trend toward interpreting them as one-place predicates. I have shown how syntactic computation of comparison classes is inadequate, and how an appeal to context is both necessary and natural. I have also outlined a number of arguments designed to show that degree adjectives behave syntactically like absolute adjectives, which lends support to a common translation.

KENN ETH R. BEESLEY ( 5 1 ) x is bi� and x is a f lea (52) [big'( x)]c ' & flea'(x) The equivalence of (50) and ( 52) shows that the conjunction analysis can indeed hold for degree adjectives. In fact, the conjunction analysis for degree adjectives holds whenever the context remains constant over the original clause and the constituent clauses of the conjunction.

(53)

{ [x is a big flea]c 1 } I= { [x is big]c 2 and [x is a flea] }

Given that context may assign different comparison classes for evalua­ ting a sentence of the form :r is big, the same type of contextual equiv­ ocation can make the evaluation of :r is big different from the evalua­ tion of another sentence w ith the same surface form. \54)

{ [x is big]c 1 }

I= { [X is big] c 2 }

And as I showed that contexts can force short Vietnamese to be evalua­ ted relative to the heights of Vietnamese, men, or any number of other classes, even (55) is possible. (55)

{ [short V ietnamese]c 1 }

I= { [short V1etnamese] c 2 }

As 1t IS hardly too much to require that context remain fixed throughout the various stages of an analysis, there is no barner to adopting a conjunction analysis for degree adJectives. 2.6

Hedges as Overt Specifiers of Context

Before leaving degree adject ives, I wish to add one final argument to support the treatment of compar ison classes as contextual subscripts rather than as syntactic arguments. So iar we have considered cases where context alone must select comparison classes. Tr.ere are, however, a variety of ways in which comparison classes can be spec ified syn­ tactically. ( 56) John is tall for Pygmy. For a Pygmy, John is tall. John is tall for i:l Pygmy but shor! f.;r 2!0

<1

Europe.m.

J�,

'loi.l, no.]

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The apparent failure of the conJunction analysis for a sentence like (49) is caused by the tendency for different comparison classes to be assigned to the structures in (49) and (5 1 ) . That is, it is possible to make the interpretation of :r is a big flea distinct from the interpreta­ tion of :r is big and :r is a flea, but to do so requires contextual equivoca­ tion. Using curly brackets around an expressiOn to indicate the value of that expression, this equivocation can be expressed as in (53), where c 1 is distinct from c 2 .

EVALUATIVE ADJECTIVES (57) John is tall As Pygmies (58) John is tall (59) John is tall ( 60) John is tall (61) John is tall (62) John is tall

as

Pygmies go/run. o, John is tall. compared with Pygmies. m comparison to a Pygmy in relation to among the Pygmies. when considered as a Pygmy. if while by Pygmy standards. from a Pygmy point of view.

1

{

i

(63) On the moon, John weigh s 40 pounds. (64) In John 's opinion, communists are diabolical. (65) In 20 years, everyone will own a computer. Just as the hedges in (63) to (65) indicate times, places and perspectives in relation to which their modified sentences must be evaluated, so the constructions in (56) to (62) specify comparison classes in relation to which their modified sentences must be evaluated. An y grammar treatmg compar ison classes as syntactic arguments in logical notation will run head-on into comparison class hedges. The easier and more attractive solution is to consider these hedges as syntactic realisations of c; this allows a close tic between the syntax and the semantics. (66) [taii(John)]c =for

8

Pygmy

Possible worlds and times are generally treated analogously to the hedge in (66). We might translate (67) as (68), where a superscript indicates a possible world in relation to which the sentence 's truth value is determined. (67) In John 's dream world, John is tall. Joh n '• dr e L m world (6g) [ta ll(John)]c=me n JS, vol. I , no. 3

2I I

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Sentences like those in (56) have often been used as illustrative paraphrases in adjective papers without ever getting much att � ntion as mteresting constructions m their own right (but see Bartsch & Venne­ mann 1 972: 89; Bartsch 1972: 1 64). Phrases l ike for a Pygmy, which can associate with individual adjectives or with a whole sentence, preposed or postposed, appear to act syntactically as some kind of sentence adverbial. They appear to be a type of 'hedge' (Lakoff 1 972), and related hedges for evaluative adjectives have been discussed briefly by Karnp ( 1 975: 149- 1 5 1 ). Semantically, these hedges act as context specifiers. Note their similarity to hedges specify ing possible worlds and belief spaces.

KENNETH R. B EESLEY It is because these overt specifications of context operate in the syntax as sentence modifiers that I have chosen K lein's notation over McConnell­ G inet 's (see (25') and 26)). 3 Evaluative Adjectives 3. 1

as

One-Place Predicates

Introduction

. 3.2 Preliminaries Both degree and evaluative adjectives are semantically 'subsective' (also called 'restrictive' (Keenan & Faltz 1 978: 68, 1 6, 225) or 'affir­ mative' (Kamp 1 975: 1 25)). This means simply that the set denoted by the result of applying one of these adjectives to a noun is always a subset of the set denoted by the noun alone. That is, A(N) c N. (69) { tall men } c { men } (70) { bad pluml.,e rs } c:= { plumbers } Another way of expressing this quality is to say that subsective adjec­ tives are compatible with the meaning postulate in ( 1 2), reprinted here for convenience. ( 1 2) revery ot is a t I where o is an intersective or subsective adjective and is a common noun

t

'Absolute' or ' intersect1ve' adjectives are also compatible with the stronger meaning postulate in ( 1 3). All absolute adjectives are sub­ sective.

212

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Given that degree and evaluative adjectives have traditionally been characterised together as relatives (Geach 1 956), a demonstration that one of these groups is really a subtype of absolute adjectives ' would tend to argue that the other is too. Nevertheless, there are differences between the classes, and I have already cited Kamp ( 1 97 5) on the doubts over extending one-place predicate status to adjectives like skilful. I maintain that evaluatives can indeed be analysed as one-place predicates, and that the alternative accounts in Montague ( 1 974a), Parsons ( 1 972) and Siegel ( 1 976a, 1 979) are inadequate. As Siegel addressed herself specifically to this question, concluding that evaluatives could not be interpreted as one-place predicates, my remarks will be addressed mainly to her work.

EVALUATIVE ADJECTIVES ( 1 3) revery 6 (,; is u -1 where 6 is an intersective adjective and

l,:

is a common noun

=

Although degree adjectives are usually assumed to be extensional modifiers, the matter has been debated somewhat (Kamp 1 975: 1 2 6- 1 27; Siegel· 1 976: 1 1 2- 1 1 6, 2 3 1 -232). Is it the case, for example, that in our world where all and only basketball players are trombone players { tall that it will always be the case that { tall basketball players } trombone players } ? Some people have the intuition that tallness for a basketball player might be assigned differently, probably more conser­ vatively, than tallness for a trombone player, even if the two classes happen to be extensionally identical. This effect is usually explained by appealing to possible worlds. It is generally the case in a world familiar to us that basketball players are significantly taller than { trombone trombone players. A world W where { basketball players } players } is a freak, so tallness might be assigned relative not to the actual extensional classes in W, but relative to an extensional class in a more normal world. =

=

Zwicky ( 1 969) illustrates these effects in what might be called 'disaster scenarios '. Assume a world much like ours in which basketball players are an exceptionally tall group. If John is a basketball player who is 6 ' 2" tall in this world, then he is tall for a man but not tall for a basketball player. Now if a sudden epidemic were to wipe out the entire population of basketball players over 6'2" tall and if no other basketball players are 6 ' 2-'' tall, then John would be the tallest basketball player in the world. We m ight still refuse, however, to say that he is tall for a basketball player. We seem able or obliged, m many cases, to use comparison classes from 'close' possible worlds (McConneii-Ginet 1 973: 90-9 1 , 1 33- 1 34). As an extreme illustration of the need to import a comparison class from another possible world, consider the following example: Adolphe Sax, the inventor of many musical instruments, suddenly conceived JS, vol.!, no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It has been claimed that degree adjectives are 'extensional' modifiers, whereas evaluative adjectives are ' intensional' modifiers (Wheeler: 1 972: 3 1 2; Kamp 1 975: 124- 1 25; Bartsch 1 975: 1 76; Siegel 1 976: 232-234; Keenan &. Faltz 1 978: 1 62-1 63, 2 1 9, 249). Formally, extensional functions are those whose values depend solely on the extensions of their argu­ ments. Informally, we may say that any extensional function, e.g. blue-eyed, will pick out the same members of the set denoted by its argument, say basketball players, no matter how that set is named. If the set of basketball players happens to be identical to the set of trombone players in a model. M, then it will be the case that { blue­ eyed basketball players } { blue-eyed trombone players } in M. Evaluative adjectives are obviously not extensional; m the same model M it may well be the case that { good basketbaU players } f. { good trombone play­ ers }.

KENN ETH R. BEESLEY the saxophone. Excited, he rushed to this workshop and soon produced a soprano model. Showing it to his assistants, he said, 'I've just invented the saxophone; this is a small one . ' The class of saxophones used for the comparison is only a twinkle in the eye of the inventor, but the class exists in his belief world.

3.3 History of Analyses The fact that evaluative adjectives are intensional modifiers was recog­ nised by Plato and Aristotle, who cited examples like good lyre player, and worned about the way that a good thief could be a bad man. Both concluded that good and bad modify relative to the · functions of the ir arguments, and this kind of analysis continues ltttle modified to the present day. Aristotle noted that if the function of a lyre player is to play the lyre, then the function of a good lyre player IS to play the lyre well. The very same rule, sprinkled with varying degrees of sanctifying formalism , turns up in even the most modern discussions of adjectives and adverbs (N!lson 1972: 90; Vendler 1 968: 2 1 -22; Keenan & Faltz 1 978: 1 34, 289). A problem for Plato, Aristotle, and anyone else try ing to explain all uses of evaluatives in this way is that the function of man, woman, person and other such 'empty ' nouns is far from obvious. W .D. Ross ( 1 930: 65- 74) proposed that there are two very separate readings of good. The first is the ' attr ibutive' type seen in the usual readmgs of good knife, good liar, good lyre player or anything else 'good of Its kind ' . The second is the 'predicati, v e ' (absolute) reading having the sense of ' moral excellence ' , as seen in one reading of good man and in constructions like courage is good. It soon becomes obvious that nouns indicating fairly obvious roles and functions will be most susceptible of Ross's 'attributive' good and that relatively 'empty' nouns like man w ill be most easily modified by the ' predicative' good. People mfluenced by Ross's theory include Katz ( 1 966: 289-290, 3 1 1 ) , Berman ( 1 973: 1 98- 1 99) and, as we shall see, Siegel. In spite of this, Geach ( 1 956: 30, 39) was to claim, however, that 214

JS, vol. I , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Such facts will cause difficulties for any theory which defines com­ parison classes to be extensional subsets of the universe of entities w ithin a given model (see, for instance, Klein 1 980: 1 3 - 1 4). To explain the intuition that John can be the tallest basketball player in the world without being tall for a basketball player, we must allow the predicate basketball player' to range over possible worlds, picking out a comparison class from one of them. Pragmatics wlil account for the fact that the default comparison class will normally be the extension class picked out by the predicate in the possible world of the sentence as a whole.

EVALUATIVE ADJECTIVES good and bad are always 'attributive', even when they appear in predica­ tive position. John is good, for instance, has to be interpreted as ' John is a good so-and-so' , implicitly involving dummy noun deletion. While ad­ mitting that it is hard to explain the meaning of good man in terms of the meaning of man, he rejects the idea that a separate kind of adjec­

tive is involved.

" 'Good ' often precedes words which are not funct1onal. ln such cases, in order to know what traits the thing in question would have to have in order to be called good, it is not suffici­ ent to know the meaning of the word. We have also to know what standard is to be adopted for judging the goodness of this sort of thing; and thls standard is not even partly (as in the case of functional words) revealed to us by the meaning of the word which follows 'good ' . I t must be emphasised that this difference between the behaviour of 'good ' when it precedes a functional word, and its behaviour when it precedes a non-functional word, is not due to any d1fference in the meaning of 'good' itself. We may say, roughly, that it means in both cases 'having the characteristic qualities (whatever they are) which are commendable in the kind of object in question'. The difference between the two cases is that the functional word does, and the non-functional word does not, give us clues as to what these qualities are." Hare 's comments are particularly astute, and we shall have occasion to refer back to his concept of 'standards' of evaluation. For now it is important to note how Hare challenged the notion that there is a self-evident standard for any noun. For example, even if we know the meaning and function of . a teacher, Hare claims quite correctly that there are still innumerable standards by wh1ch a good teacher might be judged. Even if we paraphrase good teacher as good qua teacher, we may still have to choose between standards of intell igence, clarity, elocution, legibility of handwritmg on the blackboard, discipl ine, etc. Context must oe the f ina! arbiter, and the standards suggested by the function of the modified noun may be ignored altogether. The standard of ' moral excellence' or 'ethical goodness' is only one possible standard among many. With a word like man, there are fewer natural suggestions for a standard, but context may force a good man to be evaluated as a soldier, as a lover, as a pol itician, as a moral being, etc. In conclusion, the standard of 'moral excellence ' , however JS, vol. l , no.3

215

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Hare ( 1 957: 1 07- 1 08), crit icising Geach, agrees that good and bad are 'attributives' but attacks the idea that the meaning of any A+N con­ struction, where A is an evaluative adjective, can be expla ined simply by knowing the meaning of N . Pointing out the distinction between he ' functional words ' like knife and 'non-functional words' like man, writes:

KENNETH R. BEESLEY special it m ight be to some philosophers, is nothing exceptional gram­ matically. Similar arguments have been offered by Keene ( 1 9 6 1 : 25 -26). The distinction between functional and non-functional words has been noted by a number of researchers. Vendler ( 1 968:9 1 -94) noted that his A3 class of adjectives, which correspond to our evaluatives, have a special affinity for nouns denoting 'certain functions'. Katz ( 1 966: 292-293) argued that only function words can be meaningfully modified by evaluatives. For each of Katz' s function words, such as knife there is an 'evaluation semantic marker ' built right into the lexical coding. ( 7 1 ) shows the coding for knife.

The syncategorematic function of good or bad, for Katz, is simply to set evaluation semantic markers such as ( 7 1 ) to plus ( + ) or mmus (-). The claim that evaluation standards are automatically derivable from lexical codings parallels Katz 's claims about the derivation of comparison classes, and it is open to all the same objections. Derived nouns, of course, have no lexical coding and so no evaluation semantic marker . .f'lon-function nouns, which also lack such a marker, can also be evaluated in many contexts. And, above all, there is simply no way to defend the notion that there is a simple self-evident criterion by which any­ thing. even the most functional noun, is evaluated. A knife whtch is to be given to a child, for instance, may be good only if the opposite of ( 7 1 ) applies (Sampson 1 970). Katz 's theory is perhaps the most explicit example of the kind of theory shown to be unworkable by Hare ( 1 9 57) and Keene ( 1 9 6 1 ) . Berman ( 1 973: 1 98- 1 99) presents a n analysis o f good which i s remarka­ bly close to that of Ross ( 1 930). That is, she takes good to be a doublet w ith semantically attributive and semantically predicative readings. The semantic attributives are called 'bound', indicating the claimed dependency of the meaning of good on the meaning of the noun it mod­ ifies, and the semantic predicate is called ' unbound ' , supposedly indi­ cating its absoluteness. The analysis of S iegel ( 1 976a, · 1 979) differs in no important ways from Berman 's, but it is formally elaborated and defended at length. Whereas Montague ( 1 974a) analyses all adjectives as semantic attribu­ tives and Kamp ( 1 97 5) argues for analysing adjectives as predicates, Siegel steers a middle course by adopting both solutions. Absolute adjectives, including degree adjectives, are interpreted as one-place predicates, and relative adjectives, including evaluative adjectives, are interpreted as semantic attributives. Relative adjectives are assigned the category CN/CN in the cate­ gorial grammar. This indicates simply that they are functions which take a com mon noun and yield another com mon noun as the result. 216

JS, voi . J , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 7 1 ) (Eva!: (ease of dividing substances softer than its blade))

EVALUAT IVE ADJECTIVES Relative adjectives are applied and ordered with rule ( 72). and B £ P e N ' then aB P e N , where F 6(a,B)

(72) If a £ P e N / e N F 6�,8)

=

Semantically, relative adjectives correspond to functions from intensions of one-place predicates into one-place predicates. Their type is therefore <<s,<<s,e>, t:»,« s,e>, t ». (73) If a £ P e N / e N and B £ P e N ' and a and B have translations a ' and B' respectively, then F 6� ,B ) translates as a\�8 ' ).

(74)

The mere boy died t

"'died / 1v I mere boy e N " boy / mere the mere boy T

e NIe N

eN

A one-place predicate within Montague grammar will be of type «s,e> , t >. There are two such predicate classes in Montague's ( 1 9 74b) PTQ:syntactic predicates and common nouns. To preserve the syntactic distinctions between them, syntactic predicates are assigned the category t/e and common nouns t/ /e. As Siegel proposes to treat some' adjectives as one-place pred1cates, and as their syntactic behaviour is different from both syntactic predicates and com mon nouns, she assigns them the category t/ I /e. Rule (7 5) forms a syntactic predicate from an adjectival predicate by adding a be. (75) If a £ Pt /// ' e

then F 2 0p ) £ P 1 y where F 2 o(l )

=

be a

For our purposes, the corresponding semantic interpretation rule may be taken to simply preserve the original interpretation. Rule (7 5) will allow Siegel to generate strings such as (76) and (77), but (78) will not be' generated by this rule because mere is not of cate­ gory t/I /e. ( 76) John is carnivorous. (77) John � nude . ( 78) *John is mere. To generate absolute adjectives in syntactic attributive position, Siegel resorts to an awkward version of the old relative clause reduction JS, vol. J , no.J

717

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(74) is a tree showing a derivation including mere, a typical relative ad­ jective.

KENN ETH R. BEESLEY transformation from transformational grammar. Such a solution, mtxmg TG and MG, is the sort of thing pursued by Partee ( 1 97 5, 1 976b). The reduction rule is shown in (79) and an example using it in (80). (79) If a e: P e N and a is of the form 6 such that hen be y where 6 e: P e N and y e: P t ///e , then F22 ( a ) e: P e N where F22 (a ) =y 6 The red barn collapsed t

(80)

/

�

/

barn e N

1s red t h eo'\. / '\_oe hea r

red 1 v

""' red t ///e

As ·we have seen, Siegel included degree adjectives among the absolutes. By using rules (75) and (79) she properly generates the strings in (8 1 ) and (82) without any recourse to dummy noun deletiOn. (8 1) The basketball player is tall. (82) The tall basketball player sat down. When taken to its full conclusion, this analysis makes some interesting predictions about possible readings for adjectives in attributive vs. predicate position. If there are adjective forms which have both a CNICN and a tl I le reading, then only the tl I le reading should be possible when the adjective is in predicate position. The adjective in attributive position should be ambiguous: absolute if derived by way of relative clause reduction and relative if derived directly. Consid­ er the case of moral in moral philosopher. A moral philosopher is either a practitioner of moral philosophy (Siegel's 'relative' reading), or he is a philosopher who is moral (Siegel ' s 'absolute' reading). But moral has only the absolute readmg when it appears in syntactic predicate position: The philosopher is moral. So far the predictions a re borne out. This test works for moral philosopher, plastic surgeon, abnormal psychologist,

fuzzy logician, legal secretary, deviant logician, Chinese restauranteur, mediaeval scholar and a number of similar examples discussed by Bolin­

ger ( 1 967). In fact, Siegel's absolutes and relatives correspond nicely with Bolinger 's 'referent modifiers' and 'reference modifiers' or with the traditional dist inction between 'inherent' and ' non-inherent' modifiers (Quirk et a!. 1 972: 259ft). Unfortunately 218

for Siegel's analysis, evaluatives refuse to act as JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the red bar n r collapsed1 v I red barn e N I barn such that it is red e N

EVALUATIVE A DJ ECTIVES her classification would predict. Good is taken to be a doublet, but unlike other doublets like moral, a good in predicate position can still have what Siegel calls a relative reading. (83) John is good. Since Siegel's relative readings come from semantic attributives, any relative reading of (83) must derive from John is a good 6 , where 6 is a common noun. This forces Siegel to retain dummy noun deletion. 3.4

Evaluatives as One-Place

Siegel has two related arguments, both of which I must counter, for handling such 'relative ' readings of evaluatives as semantic attribu­ tives. The first argument is based on the distinction between intensional and extensional modifiers which was discussed earlier. Siegel claims that to evaluate an expression like good car, which involves a 'relative' modifier, one must have a grasp of the intension of car - one must know what it means to be a car. But conversely, to evaluate expressions like red car or fast car, which involve absolute adjectives, one does not have to have a grasp of the intension of car; one needs simply to be able to identify car things and red-things or car things and things which move as fast as a car. Sim ilarly, to know what a bad library is, one has to have a grasp of the intension of library - a bad library IS poorly stocked, diff icult to use, badly designed in general for finding mformation. But a tall library, according to Siegel, can be understood with no reference to the characteristic functiOning of a library at all ( 1 979: 2 3 1 -232; 1 976a: 1 09ff). Relativ ity to the intension of a noun is taken to be explained by the form of the translations g•ven to so-called relative adjectives as in (84). (84) x is a bad library. (84 ') (bad '("library '))(x) That is, (84 ') shows bad mapping the intensiOn of library into a new predi­ cate which is predicated of x. · Extensionality of a modifier is taken to be explained by the structure of the conjunction analysis they receive. (85) x is a tall library (85') tall '(x) & library'(x) First, it must be pointed out that an analysis like (84 ) in itsel f explains nothing. Bad simply denotes a function mapping the intension of a CN library into a new CN bad libra r')'· There is no indication of how the '

JS, vol. l , no.3

219

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

On Arguments Against Translating Predicates

K ENN ETH R. BEESLEY meaning of the new whole is constructed from the meanings of its parts - one simply knows that a new predicate has been created. Signifi­ cantly, there is not even a way to show logically that the x in (84') is a l ibrary at all. Both Montague and Siegel have to resort to meaning postulate ( 1 2) to do this. As for the conjunction analysis in (85'), it is every bit as 'intensional' as the analysis in (84 '). All functions in PTQ, before simplification, apply to the intensions of their arguments. Siegel's argument that good in good car is relative to the intension of car while fast in fast car is not is simply confused. She points out that

In addition, we saw that the extensional status of degree adjectives is somewhat in doubt, so Siegel 's contention that their extensional status requires "them to be translated as one-place predicates cannot be accepted. In fact, the question of whether an adjective is an intension­ al or extensional mo difier simply does not have anything to do with their status as predicates in logical translation. Any theory like Siegel's which can appeal to context to guide the application of a predicate can simulate intensional effects with simple one-place predicates. Siegel's second argument for treating 'relative' readings of good and bad as semantic attributives centers on the possibilities of vague read­ ings in sentences such as (86) and (87).

(86) That lutist is good. (87) That is a good lutist. Siegel claims that good in a sentence like ( 86) can be both ambiguous and vague. The claimed ambiguity is between being absolutely good (derived from a t/I /e) and being good as a something (derived from a CN/CN). In the case where (86) is derived from a CN/CN, the sen­ tence is vague; the something that the person is good as can be any noun in the lexicon. (86) might turn out, in the right context, to mean that ' That lutist is good as a chess player'. On the other hand, the good in (87) is claimed to be ambiguous but not vague ( 1 976: 55, 63-72). That is, the lutist m ight still be absolutely good or relatively good, but the relative goodness is limited to goodness as a lutist. It is impossible, 220

JS, vo!. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

a car can still be a car even if 1t does not run at all, and concludes from this that speed is not part of the intension of car - it is not a necessary attribute of car-hood. A fast car, therefore, can be charac­ terised without recourse to the intension of car. She claims that good car however, must be characterised relative to the intension of car. This ar­ gument simply does not hold. I may call a car good in the : 'relative' sense that it reliably transports me from point A to point B. But by Siegel's own argument, a car that does not run at all is still a car, so ability to transport me from place to place is not a necessary attribute of car-hood, and it therefore is not part of the intension of car any more than speed is. So, by Siegel's own argu ment, good is not an intensional modifier.

EVALUATIVE A DJECTIVES claims Siegel, for (87) to be read as 'Tnat is a good as a chess player lutist ' . Siegel believes that h e r analysis of (86) and (87) a s (86') and (87'), respectively, captures these observations about vagueness. (86'b) has. a variable slot ranging over common nouns. The same slot in (87 'b) is filled by lutist, blocking a similar kind of vagueness. (86') a. b. (87') a. b.

good '(that lutist) good '(A ll ) (that lutist) good '(x) & lutist '(x) good '(Aiutist ') (x)

absolute reading vague relative reading absolute reading non-vague relative reading

Siegel also flaws her arguments by using only functional words in her examples showing the possibilities of vagueness. In practice, there is strong pragmatic pressure to evaluate goodness relative to lutist-like qualities in any construction like good lutist, just as there is strong pragmatic pressure to evaluate shortness relative to Vietnamese in a construction like short Vietnamese. It is easy to be fooled into think­ ing that the usual default readings for such constructions are the only readings available. Since Plato, however, it has been recognised that examples involving more general words like man, woman and person are much harder to explain (Bolinger ( 1 967: 26-30, 1 4 , 20; Vendler 1 968: 2 1 -22, 90-94; Hare 1 957: · 1 07- 1 09; Katz 1 966: 292-293; Berman 1 973: 1 99). (88) John is a good man. It is much easier to get a range of 'vagu e ' readings for (88) than it is for Siegel's example in (78). The sentence ·might be utte red, for example, by John's sergeant, his drinking buddy, his wife, the president of his chess club or even his religious leader, and each utterance could JS, vol. 1 , no.3

221

·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Siegel, however, is wrong. Semantically, it is indeed possible, despite her objections, to read the good in sentence (87) as 'good as a chess player'. Hare ( 1 957), Sampson ( 1 970) and Keene ( 1 9 6 1 ) have shown the futility of trying to derive standards of evaluation from the meaning of modified nouns, and they have shown how context can override even the strongest hints of functional nouns. Consider the hypothetical case of a chess school which specialises in teach!ng musicians. When asked how lutists, as opposed to oboists, take to chess, an instructor might say, ' V(e get some good lutis�s and some bad lutists' . In this context, the goodness will be relative not to lute playing but to chess playing. (Syntactically, Siegel errs in presenting the reading she wants to reJect in a way which is ungrammatical in surface English. Adjectives with explicit complements cannot be preposed, so good as a chess play­ er lutist is simply ungrammatical. The reading becomes clear and acceptable when the phrase is postposed: 'That is a lutist good as a chess player '.)

K ENNETH R. BEESLEY imply goodness according to a different standard. Yet for Siegel, any reading other than that of absolute moral goodness must be derived from a semantic attributive. Her translation in (89), with man plugging up the vagueness slot, is completely inadequate. (89) *good '( "man') (John) 3.5 Syntactic Tests and Evaluative Adject ives 3.5.1

Introduction

3.5.2 The Predicate Position Test It is quite obvious that evaluatives appear freely in predicate position just like absolute and degree adjectives. (90) Mary is good/tall/pregnant : To defend her analysis, Siegel has to argue that free occurrence in predicate position is not a reliable test for t/ I /e adjectives in English. The problems Siegel runs mto in the attempt to. make her analysis work suggest strongly that something is basically wrong. After including evaluative adjectives among the CN/CN 's, Siegel is forced to retain an otherwise unnecessary dummy noun deletion transformation ( 1 976: 54; 1 979: 230). Given the modern standards for attractiveness of analyses, which place a premium on avoiding transformations and abstract analyses, this is a Bad Thing. The dummy noun deletion transformation causes its own difficulties, for it threatens to generate ungrammatical strings such as *John is mere and * Bill is utter. To compensate for this new problem, Siegel is forced to introduce an ad hoc feature [+dum my deletion] for every word in the CN/CN class ( 1 976: 93). In the end, Siegel lists a large class of CN/CN adjec­ tives with the features [+dummy deletion], [ +preposing] and [ +subsective] (in the sense of Montague, see the meaning postulate ( 1 2)). This class, barrin g a few mistaken entries, is in fact the class of evaluative adjec­ tives ( 1 976: 1 8 1).

222

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We have seen that Siegel and others have used a number of syntactic tests to identify absolute adjectives. Siegel uses these to defend her classification of degree adjectives with the absolutes. I believe that her tests and her conclusions, as regards degree adjectives, are generally valid. I claim, moreover, that evaluative adjectives are also a subtype of the absolute adjectives and that the same syntactic tests w ill support my previous arguments.

EVALUATIVE ADJECTIVES 3.5.3 The One Proform Test Here again, evaluative adjectives act like absolutes and unlike real semantic attributives. (9 1 ) (92) (93) (94)

That 's a red box, and that's a blue one. That 's a tall man, and that's a short one. That 's a good boxer, and that 's a bad one. *That's an utter fool, and that 's a good/tall/fat/prize/consummate one. (95) *That's an abnormal psychologist, and that 's a normal one. (on CN/CN reading for abnormal)

Because evaluative adjectives occur freely with very and other extent adverbs, Siegel has to argue that very is not a reliable positive test for t///e status ( 1 976: 1 46; 1 979: 256ft). Gradability, she claims, is a feature ranging through all the categories of grammar, and she cites (96) as an example of gradability on a CN/CN. (96) I will follow Peter, because he is the more rightful king. This specific objection is a relic of the fact that the CN/CN class is poorly understood and analysed. R ightful properly interprets as some­ rightful thing like A PA x [rightfui'(1P { x } ]], so the interpretation of king turns out to be AX [nghtful '( 1king'(x)])]. We may read this as 'all x such that it is rightful (i.e. proper, legal, etc.) that x is a king'. R ightful turns out to correspond to a one-place predicate over formulas, which denote facts or states of affairs. If we believe that modification by very is characteristic of predicates of various sorts, then (96) follows quite naturally. CN/CN 's which do not translate as predicates of any kind, such as mere and utter, will be expected to resist modification by very. In conclusion, it is important to note that even if this syntactic test is totally unreliable, it still cannot rule out the classification of evaluative adjectives which I have proposed. 3.5.5 The Non-Restrictive Modifier Test In yet another "test, Siegel ( 1 976: 52, 1 23) argues that non-restrictive adjective modifiers can only be of category t/ I/e. Evaluatives work here too. (97) The crowd cheered for Good K ing John. (98) The king, good at warfare, suspected the abilities of his general.

JS, vol . l , no.3

223

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

3.5.4 The Very Test

KENN ETH R. BEESLEY 3.5.6 The See-Catch-Find Test

.)

The see-catch-find and there constructions are claimed by Siegel to be positive test for t///e status ( 1 9 7 6 : 7 6 f f ;92; 1 979:237-239). This is quite correct, but it means, of course, that failure of an adjective to pass this test is no guarantee of CN/CN status. Siegel cites examples like (99) to show that tall is necessarily a t/ I /e. Then she has to explain in away the perfectly good and perfectly devastating example of good ( I OO) by calling it an ' idiomatic' usage. (99) I ' ve often seen the grass tall around that house. ( I OO) I ' ve seen the children good, but not very often.

( 1 0 1 ) Nureyev was good last night; it's been a long time since we've seen him (so) good. Even degree adjectives, interestingly, can fail the see-catch-find test; Siegel has simply chosen her examples well. ( I 02)?*1've often seen Bill tall. In our world, grass is a type of thing which is constantly changing in height--it gets cut short and grows right back again. The height of grass, therefore, is a ' temporary ' or 'accident' state. (I use these words in their technical sense.) Humans, however, although they grow and change over time, have a characteristic 'permanent ' or 'essence' height. We don ' t expect people to be 6' tall one day, 5 ' 6 " tall the next, 6 ' 2 " tall the next, etc. In a world like Wonderland, however, where Alice constantly grows and shrinks by eating and drinking magic cakes, mushrooms, liquids, the Mad Hatter could utter a perfectly acceptable sentence parallel to ( I 02). ( 1 03) I 've often seen Alice tall. In conclusion, the see-catch-find test supports my classification over Siegel 's. The good in ( I OO) must be a t/ I/e. but is has a 'relative ' reading in Siegel ' s sense; i .e. the goodness is probably to be evaluated relative to proper be haviour for children. 3.5.7 The Nominalisation and Verbalisation Tests If one-place predicates have corresponding nouns representing their nominalised qualities and corresponding verbs w ith the reading ' make Adjective ' , then evaluatives also seem to be one-place predicates.

224

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Other devastating examples are easy to construct.

EVALUATIVE A DJ ECTIVES ( 1 04) Adjective

Verbalisation

red big good bad

redden, make red make big (bigger), increase, augment make good (better), improve, reform make bad (worse), harm , ruin, corrupt

mere

*make mere

Nominalisation redness size goodness_ badness *mereness

3.5.8 AU-or-Part Relativity Test

( 1 05) Bill is a tall little red-headed basketball player. ( 1 06) Bill is a former little red-headed basketball player. Quite rightly, Siegel points out that the comparison class for tallness in ( l 05) may well be the set of little basketball players, with red -headed m issed out altogether. Quite rightly again, Siegel points out that no such skipping is allowed in ( 1 06), which has the CN/CN former. But in a sin of om ission, Siegel never tries using evaluatives in the same con­ structions. ( 1 07) Billy is a good little red-headed basketball player. In ( 1 07) also, the goodness may very well be evaluated relative to some standard for little basketball players, with red-headed skipped over. Indeed, I have shown that standards of evaluation may have nothing to do with the following noun phrase, just as the comparison class for a degree adjective can be distinct from the modified com mon noun, even if that common noun is a single word. I conclude that Siege l 's own -syntactic tests show that her analysis is untenable. In all these cases evaluative adjectives behave syntactically like absolute and degree adjectives. 3.6 Rescuing the Conjunction Analysis Again The appeal to contextual standards, j ust like the appeal to contextual JS, voi . J , no.3

225

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Siegel argues that CN/CN modifiers have the quality of applying ne­ cessarily to the ·entire common noun following them, and ' that therefore such adjectives are always relative to that whole common noun. Degree adjectives, on the other hand, are not similarly relative, for their comparison classes are not necessarily the class denoted by the full CN following them. The key examples are ( 1 05) and ( 1 06) ( 1 976: 1 32- 1 34 , 7 9 ; 1 979: 248).

KENN ETH R. BEESLEY comparison classes, allows evaluative adjectives to receive a conjunction analysis. As long as context remains co n stan t , the truth conditions of John is a good thief and John is good and John is a thief will be the same. The argument for this runs completely parallel to the one for degree adjectives in section 2.5. 4

An Adverb Test for Adjective Classification

( 1 08)

The red barn collapsed t the re

�

�

I

apsed 1 v

rn T

red barn c N

I

barn such that it is red c N ""' h eo ts red t barn c N

/

�T

'be red 1 v "red t

///e

Relative adjectives, on the other hand, apply very straightforwardly to their arguments, with no intermediate clause. ( 1 09)

A mere boy arrivedt / arrt":e d I V a me e boy r

T

""- ·

mere boy c N

/ mere

c NI c N

"

boy c N

The presence of the embedded t-node in ( 1 08) predicts implicitly that is, as well as the t-node of the whole sentence, should be susceptible of modification by sentence adverbs. Conversely, the absence of such an extra t-node in ( 1 09) would predict that only the sentence as a whole can take sentence adverbs. ( 1 1 0) ( I l l) 226

Obviously, the red barn collapsed. The obviously red barn collapsed. JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Siegel's ( l 976a, l 976b, 1979) analysis of adjectives in Montague Gram mar provides grossly different analyses for absolute and relative adjectives. Absolute adjectives in attributive position arrive there only after starting out in the predicate position of an embedded sentence which later undergoes relative clause reduction.

EVALUAT I VE A DJECTIVE�

( 1 1 2) Obviously, a mere boy arrived. ( 1 1 3) *An obviously mere boy arrived.

The tree corresponding to ( I I 0) is the same as (1 08) but w i th applying to the highest t-node.

( 1 1 4) obviousl

obvious!)

Obviously, the red barn collapsed t

{;

t

�

the red ba

/

the red barn T

I

"

collapsed 1 v

I

barn such that it is red C N

/

barn c N

"'1s red t / "' be red1 v heo T I red heo

t ///e

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

red barn c N

collapsed t

( 1 I 1) is not only grammatical and intelligible, but its truth conditions are different from those of ( 1 1 0). I n Obviously, the red bam collapsed, it is the fact that a paricular red barn collapsed which is obvious. In

The obviously red bam collapsed, it 1s the redness of the barn which is

obvious. Although Siegel's relative clause reduction rule provides no way to carry along a sentence adverb from a reduced clause, the tree in ( 1 1 5) would seem to be the analysis for ( I I I ). The obviously red barn collapsed t

( 1 1 5)

"

/

collapsed 1 v

the obviously red barn T obvious!

)

I

red barn c N

barn such that obviously it is red C N

/

barn c N

obviously

obv

tu

h�

sly t / t

is

> is

he

/

heo T

red t

red t

"'be

red 1 v

I red

t /// e

The analysis of ( 1 1 2) requires only a straightforward addition· to 0 09), but ( 1 1 3) 1s ungrammatical. Obviously mere s e e m s u n i n t erpretable, and this intui tion fits in nicely with the observation that in the grammar JS, vol. l , no.3

227

KENN ETH R. BEESLEY there is no obvious way that a t/t modifier can apply to an argument of type CN/CN . There is simply no node in ( 1 09) for a sentence adverb to attach to other than the top node of the whole tree. All these ob­ servations are consistent with, and even explained by Siege l ' s dual handling of adjectives. Sentence adverb be very productive. ( 1 1 6) ( 1 1 7)

modification

of

absolute adjectives

turns out

turned

to

The close association of Wl(ortunately to mad in ( 1 1 6) is i n tuitively quite undeniable. A n 'unfortunately mad patient' is a patient such that ' un­ fortunately he is mad. If 1D1(ortW10tely were a modifier of the full sentence, it would clash with or cancel out w ith fortW10tely, w hich it ob­ viously doesn ' t . ( 1 1 7) shows a sentence modifier associated with either a single adjective handsome or with the conjoined pair handsom e and dis ­ tinguished--an ambiguity of the old men and women-type. Not only do all of Siege l ' s absolute adjectives allow sentence adverbs, but most of her exclusively relative adjectives seem to reject them. ( 1 1 8) *That is the obviously main reason. ( 1 19) *He is an obviously alleged cr iminal. ( 1 20) *Bill is an obviously blithering idiot. In cases where adjectives are ambiguous between relative and absolute readings (as in moral philosopher, plastic surgeon, fuzzy logician, ab­ normal psychologist and criminal lawyer), only the absolute reading is possible with sentence adverbs. ( 1 2 1) ( 1 22 ) ( 1 23) ( 1 24)

The obviously moral philosopher lost his job. The previously fuzzy logician got a shave. The occasionally abnormal psychologist needed help. John is an obviously criminal lawyer.

As a test for absolute adjectives, sentence adverbs are even better tests than very. It has long been known that very can disambiguate moral

philosophers. ( 1 25) ( 1 26)

Jane is a very moral philosopher. The very abnormal psychologist studied himself.

ex­ But some adjectives like pregnant, virgin and carnivorous and also treme ends of scales like perfect and fla wless resist very. Very pregnant, very virgin, very flawless, etc. take on secondary readings. But sentence adverbs can apply to the words with their literal senses. 228

JS, vol. l , no.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Fortunately, the unfortunately mad patient was cured. The unmistakably handsome and distinguished actor politics.

to

EVALU ATIVE A D J ECTIVES ( 1 27) ( 1 28)

The obviously pregnant princess continued her engagements. My previously carnivorous friend became a vegetarian.

Siegel's work contains the claim that degree adjectives (tall, short, high, are a subtype of absolute adjectives, and that they should be generated in attributive position by way of relative clause reduction. Given the long history of degree adjectives being considered semantic attributives, this is a bold claim. If the sentence adverb test is valid and Siege l ' s claim is true, degree adjectives should appear freely with sentence adverbs; and indeed they do.

low, fat, thin, etc.)

The obviously tall ballerina was rejected. The previously fat capitalist lost 20 pounds. The presently low wages were once respectable.

For evaluative adjectives, however, which Siegel claims are ambiguous between absolute and relative readings, the adverb tests indicate other­ wise. The doublet moral, as was seen, can have only its absolute reading i n the presence of a sentence adverb. In parallel constructions, however, good is perfectly capable of retaining what Siegel calls 'relative' readings. ( 1 32) ( 1 33) ( 1 34)

He is an obviously moral philosopher. (only the absolute reading is possible) She is an obviously beautiful dancer. ( 'relative' readings are still possible) John is an obviously bad monk. (the 'relative' reading is even preferred)

Though it is impossible to read the moral in ( 1 32) as being 'relative' to the meaning of philosopher (i.e. in the sense of moral-philosophy), the beautiful in ( 1 33) can still be 'relative' to dancer. An obviously beautiful dancer can still be a 'dancer who dances beautifull y ' . Similarly, an obviously bad monk need only be relatively 'bad qua m o n k ' r a t h e r than absolutely bad. Siegel would represent her two claimed readings of bad monk w i th the following trees. ( 1 35)

John is a bad monk t

/

John T

be

ba

JS, vol . 1 , no.3

�

(relative reading)

ad monk 1 v

I

bad monk c N

(.

/C N

�onk C N 229

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 29) ( 1 30) ( 1 3 1)

KENN ETH R. BEESLEY (absolute reading)

John is a bad monk t

( 1 36)

/

Jotin T

" be a bad I

monk

/ monk c

monk 1 v

bad monk c N

J

ch that he is bad c N

N

he o

h

Only ( 1 36) provides a natural and consistent way to account for the presence of the sentence adverb in ( 1 34). Any attempt to allow sentence adverbs to apply to CN/CN words like bad m ( 1 35) will overgenerate producing *obvioU8ly mere, *allegedly moral (with its CN /CN reading) and *previOU8ly ostensible. This would also destroy the explanatory insights that the theory seems to support so far. In this as well as in other syntactic tests, evaluative adjectives consistently act like absolute adjectives. A gross syntactic and logical ambiguity in good is neither tenable nor necessary. The standard of moral uprightness which supposedly characterises Siegel's 'absolute' reading is simply one standard among the many possible. These results point to an analysis parallel to that suggested in Hare ( 1 957), where only one good is postulated, and it is a fuzzy one-place predicate. 5 The Grammar 5.1

The internal S tructure of Adjective Phrases

Absolute, degree and evaluative adjectives are all of category t// /e, reflecting their status as one-place predicates. Any grammar wishing to characterise their syntactic behaviour w ill have to deal with the following facts: ( A) Gradable t// /e's, those which can apply fuzzily over a universe of entities leaving an extension gap, can be modified by very. extremely, quite, a wfully and other such extent adverbs: very tall, extrem ely good,

quite red.

( B) Some adjectives of the t/ I /e class, whether modi fied by extent adverbs or not, can be further modified by certain manner adverbs: ostentatioU8ly (very) rich, quietly (extremely) assertive, discretely 230

JS, vol. 1 , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

�

" is

E V A L U ATIVE A D JECTIVES

(quite) powerful, wwssumingly intelligent, generously wealthy, graciously hospitable, lavishly generous, coolly defiant, etc. These constructions can themselves be preceded by very, etc., but the bracketing will always be thus: [very ostentatiously] rich. [very discretely] [quite powerful], [extremely stubbornly] defiant. It is clear that the extent modifiers of (A) must apply to a t/ I /e before the manner adverbials of (B): *very [os­ tentatiously rich]. An excursus here into manner adverbtals is i mpossible, and only the gross syntactic possibilities will be considered.

(C) Adject ives of class t/ I /e, whether or not mod i f ted by extent adverbs and by manner adverbs, are further modifiable by sentence adverbials:

The analysis must, therefore, accommodate constructions at least as complex as the tree in ( 1 37), ensuring that the hierarchy of modifiers remains fixed. (sent-adv (man-adv (ext-adv ( t///e))))

( 1 37) sent

�v

� v (t/I/e)))

(man-adv (ext

/ man-adv

"

(ext-adv (t///e))

e

xCadv

""'- t/I /e

The way to ensure that modifters apply in the correct order ts to distinguish the different levels of the tree by distmct categortes. As the semantic category at each level is always a one-place predtcate, we need only extend our analysis with a few extra slash categories. Mamtammg t/ I /e as the basic category of absolute adjectives, succeeding levels can be distinguished by adding an extra slash.

/

o

( 1 38)

�

� r ich t ///// e

ously o tentatiously v

obv iously t ( ///// e ) /( t///// e )

/

�

tentatiou

osten tat i ously t//// / e ) /( t//// e ) (

very rich t ///// e

/very � \.1///e

very t// ( 1 / e) /( t /// e )

rtch t

///e

The unusual abiltty of very to apply recursively to its own output (very very very rich) will be tgnored here. Other extent adverbs block

any possibil ity of other extent adverbs applying; this is reflected in their proposed category (t/ I I /e)/(t/ I /e). Even the troublesome ve·ry blocks further modification by extent adverbs other than very: *quite very rich. (See Backlund 1 973: 1 58; N ilsen 1 972: 1 1 3ff.) It does not seem possible JS, vol. l , no.3

231

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

obviously (ostentatiously) (very) rich, previously openly defiant, presently very peaceful. The modifiers of (A) and (B), if any, must apply before those of (C): *very [obviously defiant], *modestly [previously humble]. Some sentence adverbs may themselves be mod i fied: [very obviously] rich.

K E N N ETH R. BEESLEY to define hard rules about which lexical items can be graded, so the matter of constraining extent adverbs is left open. 2 Many adverb forms normally thought o f as manner adverbs seem also to have a reading as extent adverbs: impressively lucid, irritatingly bad, a wfully good, horribly contagious, terrifically talented (Bolinger 1 972: 22-23; Backlund 1 973: 1 2, 1 65ft, 277; Zwicky 1 970). Nevertheless, i t cannot be concluded that all premodifiers of adjectives can be reduced to extent adjectives. The examples in (B) have typical manner force.

The resulting structure of an adjective phrase bears a striking, hardly accidental, resemblance to the structure of a sentence. The analysis of absolute adjectives begins to resemble, i n overall respects, the adjective analysis proposed in Lytle ( 1 9 7 1 ; 1 97/f) and even that in Anderson & Bower ( 1 9 7/f), where all adjectives appear in sentence­ like structures. It is important to note that such sentence-like structures as ( 1 38) are not relative clauses which are reduced by some syntactic transformation; they are hierarchical semantic representations from which adjective phrase strings are computed directly. (For a related approach see Bartsch 1972: 1 66.) We can now look at the rules assumed so far. ( 1 39)

( l ifO )

If a

e:

� t////e)/ ( t///e )

then F 6( a , 8 )

e:

where F 6 (a , 8 ) If a

e:

=

and 8

P t//// e

P t///e

e:

aS

P (t////e)/(t///e )

and 8

Pt/// e

e:

and a and 8 translate as a� and 8 � respectively, then F 6(a , 8 ) translates as a � ( 8 �). A

These rules and translations will meet the present purposes. The seman­ tics of extent modifiers are surprisingly complex and have been described elsewhere (Kamp 1 975: l lf6 - l lf7; Keenan & Faltz 1 978: 1 71f, 228, 207- 2 1 0; Klein 1 980: 2/f-26).

p ( t//1// e )/( t//// e ) and F 6 ( a , 8 ) e: P t/////e .

( ! If ! ) If a e: then

( l lf2)

232

If a e:

P(t/////e)/(t//// e )

8

and 8

e:

£

p t//// e

P t ////e

and a and 8 translate as a � and 6� respectively, then F6( a , s ) translates as a' ( "ll ' ).

JS, voi . J , no. 3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Manner adverbs on adjectives seem to block modification by any other manner adverbs; they too bump the category one slash higher. Sentence adverbs, however, allow recursive application: obviously pre­ viously rich, previously obviously rich. Their category is (t//1//e)/(t// (//e))

EVALUATIVE A DJ ECTIVES

This rule treats manner adverbs modifying adjectives as functions mapping intensions of << s,e> , t > predicates into < < s,e> , t > p r e d icates, just as in PTQ and in Thomason & Stalnaker ( 1 97 3). The whole question of manner adverb analysis is vexed - Parsons ( 1 972) and R ichards ( 1 976) have even argued that they are formula modifiers. I have no desire or space to pursue the issue; these rules will suffice for the present purposes.

( 1 43)

( 144)

If a E If a E

;

Pt t Pt 1 t

then F s o(a ) E

and a translates as a '

( 1 46)

A P A y( a ' ( 1

then F s o ( a ) translates as

The application of is straightforward. ( 1 45)

P( t///// e )/{t/////e)'

sentence adverbs,

If a E

P (t/////e)/( t///// e)

a

y } ] )).

of class (t/ I II /e)/(t/ I I I /e),

E

P t/////

e

and 6 E

P t/////

e

P (t/////e)/( t///// e ) and 6 F 6( a ,6 ) E P t/////e .

If a E

then

now

P{

where F s o(a )

and a and il translate as a ' and 6 ' respectively, then F 6( a , 6 ) translates as a' ( � 6 ' ) .

5.2 Adjectives in Predicate Position To generate a sentence with an absolute adjective in predicate positiOn, rule ( 1 47) is used. Unlike the Montague-Parsons analysis, this rule generates adjectives in predicate position directly, with no recourse to a dummy noun deletion transformation. The class IV /(t/ II I /e) contains the copula verbs - for simplicity we will limit our example5 to be . ( 1 47)

If a E

P IV/(t///// e )

and 6 E

P t/////e ,

then F 6(a , 6 )

E

P1v.

In the interests of simplicity, I assume that the semantic rule correspond­ ing to ( 1 47) merely preserves the translation of the argument intact. Also, I assume that I V ' s map their subjects into truth values rather than vice-versa as in PTQ (Thomason 1 976: 79; Keenan & Faltz 1 978: 23). JS, vc-1. 1 , no. 3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The proposed class of (t/ I I I /e)/(t/ I I I /e) adverbs appears to be coexten­ sive w ith the class of t/t sentence adverbs. E ither ( l ) a class of (t//1//e)/ (t/1///e) adverbs can be defined in the lexicon, w ith the stipulation that i t is coextensive with the t/t class, or (2) a syntactic rule can be defined to bump t/t adverbs into (t/1///e)/(t/////e)'s, with a corre­ sponding change in the semantics. As there are complex t/t's like very obviously, � he second option offers some economy.

KENNETH R. BEESLEY The iota operator is imported for definite expressions; no other quanti­ fied expressions appear in the examples. Tense and the contribution of the copula be will be ignored in the semantic trees. Using rule ( 1 47), the analysis tree for John is obviously ostentatiously very Mch is ( 1 48). The corresponding semantic tree is ( 1 49). John is obviously ostentatiously very rich t

( 1 48)

/

'\.

Johnr be obviously ostentatiously very rich 1 v

be

OV(

11

obYi "'( y

obYiou•ly o•ten

HI/III o l/(
obv10us ly t t l

�ou•ly my cich

ol

/ ti� enta

y '"'Y

ostentatiously t ( lllll e ) l( t llll e) ver � t

( 1 49)

< II111,

.:::..

h


/ very

�

lIIl e) l( t l I/ e )

ch t llll

e

rich t l

I le

).. y [obviously'< l ( lostentatiously ' <1 very ' ( 'rich')] )] ) { y } ] )] ( 'John)

/

"'

John

l. y[obviousl y ' ( l< lostentatiously'( lvery' (rich')] )] ) { y } ] )]

/

).. P ).. y[obviously '(lP obv

(c;'usl y '

{y

} ] )]

"'

ostentatiously' ( 'very' ( 'rich '))

ostenta

!ously'

�ery ' ( 'rich ') / """ .

very '

nch'

Because words l ike obviously create opaque contexts, the lambda expres­ sion in formulas like ( 1 49) cannot always be reduced. Before moving on, we must deal with the fact that not all tl Ile adjectives get bumped up by mod ifiers into tlllle 's and then tllllle 's. Rule ( 1 47) is defined only in terms of tl I I I le 's, so the rules so far shown could not generate the simpler string John is rich. We cannot sim­ ply create classes of basic tlllle ' s and tlll//e 's in the lexicon, stipula­ ting that they are coextensive with the class of tl 1/e's. Rule ( 1 47) must also be able to map derived t// //e's l ike timidly assertive into tie's. Rule ( 1 47) could be redefined to map tllle ' s or tlllle's or tlll//e 's into tie 's; or, which is equivalent, one can define empty rules which simply bump tl I le's and t/ I I le's into the higher classes (preserving is the translation throughout). Under the first alternative John is good derived from ( 1 50), under the second, from ( 1 5 1 ) .

( 1 50)

234

John is good t

/

John T

'

be good 1 v

/ ."

be

good t

l1l e

JS, vol . l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(, / l

EVALU ATIVE ADJECTIV ES (151)

John i s good 1

/

John T

'

be good 1 v

/

be

'\.

r

go d t ///// e

g

�

goo d

t ////e t ///e

The former solution requires a more complicated rule ( 1 47), in effect abbreviating three rules into one. The second solution requtres rule ( 1 47) as i t stands plus two bumping rules. Nothing hinges on the choice.

( 1 52)

If a E P I V ( t e ) and fl E P � //e / // then F6(a , fl ) E P 1 v where F6(a , fl ) = a a fl

•

A parallel rule will place adjectival prepositional phrases in predicate position. All these rules mapping various one-place predicates into t/e's should eventually be merged. ( ! 5 3)

John is a boy 1 { Johr be aboy 1 v '\: boy

� - l V /( t // e J

( 1 54)

/

John

t//e

boy ' ( 'John)

'\.

). x[boy' (x)]

/

be

"

boy '

Space precludes the obvious extension of these rules to allow sentence and manner adverb modifications for nominal predicates, adjectival preposittonal phrases, and participles, both present and past. ( 1 55)

John is obviously

JS, vol. l , no.3

tall a man out of breath speaking digging a ditch persecuted

235

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Rule ( 1 52) is l ike ( 1 47) except that IV's are formed from common noun predicates (t//e) rather than from adjective predicates (t/////e). English requires an indefinite article to be inserted before the predicate nominal. The class IV/(t//e) is l t m i ted here to be.

K E N N ETH R. BEESLEY ( 1 56)

The man, obviously

( 1 57)

John is ostentatiously rich a member of Mensa in favour with the K ing

( 1 58)

The Pope, obviously wishing for an end to the war, prayed for peace. The man, obviously asked to wait, was growing i m patient.

For the t i me being, adjectives which cannot appear in predicate position will be analysed as CN/CN ' s. This is, in fact, a very diverse class requiring special treatment. Because no rules are defined to convert CN/CN 's l ike mere, utter, main, present, past and future into t/////e 's, they are never generated in predicate position. ( 1 60) (161) ( 1 62) 5.3 5.3. 1

*The boy is mere. *The king is future. *The idea is main. Adjectives in Attributive Position Relative Adjectives

Just as the syntactic predicate-ness of the t/ I /e classes is reflected in the semantic predicate-ness of their translations, so the syntactic attributive-ness of the CN/CN class is reflected in their semantic a ttr ibutiveness. ( 1 63) ( 1 64) ( 1 65)

If C1 £ P C N / C N and 6 £

P eN

Pe N'

I f a £ P c N / C N and 6 £ P c N and a and 6 translate as a � and 6 � respectively, then F 6(1l , 6 ) translates as 11 � ( ' 6 � ) . John is a mere boyt

/ John T

"'

be a mere bo y 1 v

/

be I V ( t // e )

"'

�

/mere

merec NIc N

236

then F 6( a ,6 ) £

yC N

boy N c

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 59)

lost , asked for a map. a soldier in a hurry needing help confused by the terrain

EVALUATIVE A DJ ECTIVES ( 1 66)

mere'('boy') nohn)

(

�

Joh

e ' ( "boy')

m

/ ""- boy' mere'

Cases l ike the normal readings of moral philosopher, abnormal psycholo­ which are candidates for this gist, criminal lawyer and plastic surgeon, rule, are probably better treated as

5.3.2

CN compounds in the lexicon.

Absolute Adjectives

W ithin M G, the trend, following Thomason ( 1 976), is to pursue as straightforward a syntax as possible, complicating the semantics as necessary. Instead of putting absolute adjectives in predicate position by relative clause reduction, the following rules can change a tl I I I le into a CNICN . (Such a rule appears in Bartsch 1 97 5: 1 7 9 , and the solution was also suggested to me independently by Theo Janssen.) ( 1 67) ( 1 68)

If a

P

e:

t /////e

then Fo ( a ) If a

Pt

e:

e:

/////e

P

and a does not have a complement where Fo (a )

c N/C N

and a translates as

=

a

a'

then F o ( a ) translates a s ). P ;�. x [ a ' (x) & P { x } ].

The rules ( 1 63) and ( 1 64) serve to order and interpret these new CNICN '-s.· The requirement in ( 1 67) that no tlll//e with a complement can be promoted to a CN/CN reflects the fact that no adjectives w i th comple­ ments can appear in attributive position in English. Although there are no adjective complements in the present fragment, the condition must be considered in any more complete account. ( 1 69)

John is a bad actor t

/

John T

JS, vol. l , no. 3

"

be a bad actor1 v

"

/

Dei V ( t //e )

ba

c(

" CN

bad act N/C N

actor c N 237

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

H istorically, even some linguists who recognised the i mpossibility of generating all syntactically attributive adjectives by way of relative clause reduction retained the transformation for those cases where i t did work (Smith 1 96 1 : 346; Winter 1 965: 487; Berman 1 973). The analysis of Siegel ( 1 976a, 1 979) is in this tradition. There are indepen­ dent reasons for avoiding all syntactic transformations, however, and analyses in linguistics and logic have become increasingly surfacy in recent years.

K E N N E TH R. BEESLEY ( 1 70)

bad'(�John) & actor '( �John)

/ John

"'

h[bad '(x) & actor'(x) ]

/

"' actor '

>. P >. x[bad'(x) & P{ x}]

J

d'

It is now possible to show the semantic and syntactic differences be­ tween ( 1 7 1 ) and ( 1 72) w i th no recourse to relative clause reduction. Obviously, the red barn collapsed. The obviously red barn collapsed. Obviously the red barn collapsed t

/

"'

the red barn collapsed t

obviously t/ t

/

the red barn T

I

"collapsed

red bar n c N

/

red c N I c N

I

red t

( l 7 l b)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 7 1) ( 1 72) ( 1 7 l a)

" barn

1v

cN

/////e

obviously'(1collapsed ' ( � t x[red '(x) & barn'(x)] )] )

/

obviously' t

"

collapsed ' ( � t x[red'(x) & barn '(x)] )

/

"'

collapsed '

x[red'(x) & barn'(x)]

I

>. x[red'(x) & barn'(x)]

/

>. P >.x[red '(x) & P{ x }]

"

I

barn'

red' ( l 7 2a)

The obviously red barn collapsed t

/

the obviously red barn T I

"'

collapsed 1 v

obviously red barn c N '\ y red c N / C N barn c N obvio

;. I

ob

�

�

ously red t

/////e

obviou ly t / ( III/ e ) /( t /I II/ e ) obviously t / t 238

""' red t /

II Ile

JS, vol. l , no.J

EVALUATIVE ADJECTIVES collapsed '( " l x(obviously '( 1red '(x) ] ) & barn '(x)] )

(7 2b)

/

1 x[obviously ' < 1red'(x)l)

I

& barn '(x)]

"

collapsed '

>.x[obviously' < 1red'(x)] ) & barn'(x)]

/

>. PA x[obviously'< 1red'(x)] ) & P{x} ] I

"

bar n '

>. y[obviously'( 1red'(y)] )]

/

AP>.y[obviously'(1P{ y }] )]

red '

obviously ' 5.4

Postposed Adjectivals

Certain 'tempo rary ' or with complements, will they modify. ( 1 73)

( 1 7 4)

If a £ P t

lllll e

'accident ' adJectives, as well as adjectives require a rule postposing them to the noun

where

a

denotes an 'acciden t ' state

or where a is an adjectival with a syntactic complement then F o(a) £ P c N II c N. (Same translation as in rule ( 1 68).) If a £ p C N II c N

and a £ p C N

then F7 (all ) £ P c N

where F 7 (,. , a) aa (Same translation as in rule ( 1 64).) =

If adjectival prepositions are viewed as trans1t1ve adjectives with their objects representing a kind of complement, then rule ( 1 7 3) also serves to order and interpret such prepositional phrases. Note that the tentative rule for preposing and postposing attributive adjectives in EFL, which states ·that all multi-word adjectivals must be postposed, is totally madequate; it would even postpose very good (Hamblin 1 976: 249; Thoma­ son 1 976: 8 1 -82). Bartsch 's ( 1 973: 57) slightly better rule, which preposes those adjectivals which end in an adjective, will still incorrectly order examples such as ( 1 7 5). ( 1 7 5) *The

+

eager to help you become famous

+

man

Present participles are derived from I V ' s (both trans1t1ve and in­ transitive), by rules parallel to ( 1 67) and ( 1 68). Lees ( 1 963: 33, 97) JS, vol. l , no.3

239

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I

"

K ENN ETH R. BEESLEY notes that only those -ing forms which are genuine i n t ransitives are likely to be preposed to a noun. A transitive present participle with its object may never be preposed. ( 1 76) *the stealing/eating boy ( 1 7 7) the shivering/walking dog ( 1 78) *the walking the dog man

(pseudo-intransitive) (true intransitive) ( transitive)

If we view the object of a transitive participle as a type of complement, analogous to seeing the object of a preposition as a complement, then a generalised rule ( 1 73) will handle participles as well. the man walking the ,dog

Intransitive post pose .

present

participles

with

other

types

of complement

also

( llSO) the man shivering with cold. 5.5

HedgP-s as Overt Context

Context figures decisively i n the analysis of both degree and evaluative adjectives. For degree adjectives, we can represent the relevant com ­ parison class as a ' c ' subscript to a formula. For evaluative adjectives, we can represent the standard or criterion involved as an 's' subscript. Syntactically, context can be made explicit in 'hedges' which behave like sentence adverbials. The class of hedges is highly variable and hard to characterise syntacticall y . While I don ' t claim to offer a gram­ mar of hedges in the form of derivational and interpretive rules, the pairings of sentences and interpretations in ( 1 3 1 ) to ( 1 89) are consistent with m y claims and intuitively satisfying. ( 1 8 1)

John is tall. [tall •nohn)]c

( 1 82)

John is good. [good'nohn)]8

( 1 83)

Quang is a short Vietnamese. [short'(AQuang)]c &: Vietnamese' ( AQuang)

( 1 84)

Quang is a good com munist. [good'( �uang)], &: communist '(AQuang)

( 1 85)

For a Vietnamese, Quang i s tal l. [tall ' ( AQuang)] c

240

=

VIetnam ese

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 79)

EVALUATIVE ADJF.CT:VES ( 1 86) John is short compared to a Watusi. [short'("John)] = w a t u s l c ( 1 87) John is tall for a Pygmy and short for a Watusi. A x[ [tall '(x) c : P y g y & [short '(x)]c = w a t l ] ( "John) us m ( 1 88) In a sense, John is OK. 3s [ [OK '("John)]5 ] ( 1 89)

My car is g ood aesthetically but bad mechanically. A X [ [g oo d '(x)]s = a e s t h e t i c a l l y & [bad'(x)Js = m e c h a n i c a l l y ] (\my car)) ·

Examples ( 1 87) and ( 1 89) are extremely interesting in the light of the analysis proposed for other sentence adverbials. Hedges, just like sentence adverbs, must be applicable to individual adjectives within larger sentence structures.

Appendix The fact that sentence adverbs can modify attributive adjectives in isolation from the nouns those adjectives modify provides a strong argument against Montague's EFL treatment of adjectives. Adverbs like formerly and adjectives like former correspond to formula modif iers in logic. There exist syntactic constructions in which the argument of formerly is a formula wh1ch appears, in EFL, only as a product of the meaning postulates. In other words, in an EFL-like grammar, th(:�<:: are some formula modifiers in logical structure which must wait an.•"nd for the meaning postulates to apply in order to f ind their argum':nts. I find this result quite absurd. L�t us consider a practical example. Using the EFL approach, the string John is a bald pre."!ident would be translated as ( 1 ). Because JS, vol- 1 , no.3

241

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Sentences with bare degree and evaluative adjectives like ( 1 8 1 ) and ( 1 82) rely completely on context for the choice of a suitable cpmparison class or standard. In attributive constructions ( 1 83) arid ( 1 84) there is a strong tendency, but only a tendency, for the comparison class or standard to be 'relative ' , in Siegel's sense, to the modified noun. The more the modified noun is connected with a characteristic role or purpose, the stronger the hints of standards become. To refer to some entity as an adjective + noiDl will usually mean than the entity's no!Dl-ness is somehow salient in that context. We may thus save the intuition that That is a good lutist will almost certainly be evaluated relative to some kind of lutist standard, just as John is a tall Pyg my wtll usually be evaluated relative to the height of Pygmies. This is a pragmatic effect, however, which can, with varying degrees of ease, be overridden by context.

" K ENN ETH R. BEES L EY bald is an intersective adjective, the meaning postulates are able to justify the conclusions that 'John is bald' and that ' John is a president'. M y approach is to treat ba ld' as a one-place predicate from the be­ ginning. Grammar rule ( 1 67) (in the main !text) maps bald into a CN/CN, and the semantic interpretation becomes A I'>. x [bald'(x) &. P{ x} ]. Applying this to president ' , we arrive eventually at the formula in (2), which generates 'John is bald' and ' John is a president' without any recourse to meaning postulates.

( 1) (bald'(Apresident ')) nohn) (2) A x[bald'(x) &. president '(x)] (A John)

(3) (4) (5) (6)

formerly bald president former bald president bald former president formerly bald former president

Taking (3) first, one knows intuitively that a person who is a formerly bald president is a president who, formerly, was bald. The formerness applies only to the baldness, not to the president-hood . In (4), however, the formerness applies over both the baldness and the president-hood. These readings are extremely easy to represent if ba ld ' is treated as a predicate forming the core of a formula separate from the president ' for­ mula. (3) and (4) should be translated as (7) and (8) respectively. (7) A x[formerly'( 1bald'(x)] ) & president'(x)] (8) >. x[for merly'( 1bald'(x) &. president '(x)] )] In the grammar, former is assigned the category CN/CN and ts translated as A PA x[formerly'(1p{ x J] )]. Formerly is a t/t, and it can be bumped to a (t//1//e)/(t/////e) by rule ( 1 4 3 (see main text)). The deriva­ tion trees for (3), (4), (7) and (8) can now be shown. (3')

formerly bald president "' c N / prestdent former fy bald cN cN1c N "

r/

forme y t ( /III/ e )/( t/III/ e ) formeriy t / t

i

b ld

r

t/III / e

b dt 1111 e bald t

242

/1/ e

JS, vol. I , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The problems arise in constructions like (3) to (6).

EVALUATIVE AD,JECTIVES

(4')

former bald president

;" � �

/ former

c

aid presi

c N /C N

nt

cN

president

bald l c N1c N b ld t///// e

j

cN

ba ld t //// e l bald t /// e AX[formerly'( 1bald'(x)] ) / A PAx[formerly'( 1bald'(x)] ) &. P{ x }] I :l.x[former ly'(1bald'(x) ] )] / " A P :l.x[formerly'( 1P{ x } ] )] bald '

&.

president'(x)]

" president'

I

formerly' (8')

A x[formerly '( 1bald '(x) &. president'(x)J )] / " APAx[formerly'(1P{ x} ] )] A x[bald'(x) &. president'(x)] / "'president' A PA x[bald'(x) &. P{x} ]

I

bald' In an EFL treatment, (4) could be translated as (7), but (3) is not proper­ ly representable. The formerly in (3) applies to a formula which appears only as the output of a Montagovian meaning postulate. (9) A x[former '(�bald'( �president')(x)] ( I 0) formerly bald president formerly - bald'(x)

l

by EFL meaning postulate

bald'( �president ')(x) On the same pattern, (5) but not (6) is manageable in an EFL-Iike grammar. School of Epistemics University of Edinburgh

JS, vol. l , no.3

243

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(7')

K ENNETH R. BEESLEY Notes

244

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

* A primitive ancestor of this paper was read to the Department of General Linguistics at the University of Groningen in November, 1 9 8 1 . I wish to thank H an Reichgelt, Theo Janssen, Crit Cremers, Pieter Seuren and my advisors Keith Brown and Barry Richards for their criticism and comments. The usual disclaimers apply. 1 Siegel's thesis appears to be an attempt to formalise the excellent analysis of Bolinger ( 1 967), who shows that the relative clause reduc­ tion transformation proposed by Chomsky ( 1 957: 72, 1 1 4) and elaborated by other transformational grammarians (Smith 1 96 1 ; Burt 1 9 7 1 ; G ivon 1 970: 828; Vendler 1 968: 2 f , 85ff) simply will not work for many exam­ ples. Bolinger describes two independe nt, separately sufficient, conditions for relative clause reduction failure: , (A) Adjectives which are 'reference' (i.e. relative) rather than 'ref­ erent' (i.e. absolute) modifiers cannot appear grammatically in predicate position. As the sentences in ( l ) to (3) are uninterpretable and ungram­ matical, it is unattractive and uninsightful to use them as a deeper representation of attributive structures. the mere boy ( l ) *The boy ts mere. (2) *The stranger is total. the total stranger (3) *The reason is main. the main reason (B) Adjectives which describe 'temporary' (i.e. accident) states resist placement m prenominal attributive position. So while the adjectives in (4) to (6) are grammatical in predicate position, relative clause reduction will produce ungrammatical attributives. *the asleep boy (4) The boy is asleep. •the ready man (5) The man is ready. •the faint girl (6) The girl is faint. The (A) condition is accounted for by Siegel with her distinction between t/1 /e and CN/CN adjectives. The (B) condition, however, is the essence­ accident distinction which she mi sunderstands and tries to explain away. Bolinger uses the essence-accident distinction to explain the adjective effects in constructions like (7) and (8). (7) (a) The only navigable river to the south is the Thames. (b) The only river navigable to the south is the Thames. (8) (a) The visible stars were Aldebaran and Sirius. (b) The stars visible were Aldebaran and Sirius. The preposed adjectives in the (a) sentences show characteristic prop­ erties. For instance, the Thames is characteristically thought of as a navigable r iver; any list of 'navigable rivers' in Britain will include it. Where adjectives can be postposed as in the (b) examples, the same adjectives describe accident states. A river which is not characteristical­ ly navigable may happen to be temporarily navigable because of heavy rains and resulting high water. Conversely, a characteristically navigable river like the Thames could conceivably become temporarily unnavigable and yet be in no danger of losing its status as a 'navigable river '. Predicate position is weighted for accident readings, and attributive position is weighted for essence readings, as is seen in (9) and ( I 0).

EVALUATIVE ADJECTIVES (9) I have a red house. ( I O) The house was red in the sunset. Some adjective forms have exploited this distinction, splitting into two related but distinct meanings, one typically [+essence] and the other typically [+accident]. ( I 1) (a) (b) ( 1 2) (a) (b)

Is this a handy tool? Are your tools handy? John is a responsible man. John is the man responsible.

(e ssence-'useful') (accident - 'at hand') (essence - 'trustworthy ') (accident - 'in charge' or 'guilt y ' or 'accountable')

Siegel, having abandoned the essence-accident distinction, tries to explain the effects in terms of a contrast between CN/CN ' s and tl//e's, with disastrous results. Example ( I 3a), for instance, is claimed to contain an example of red which is necessarily a t/ I /e, while the red in ( 1 3b) can be understood as either a t///e or a CN/CN ( 1 976: 78). ( I 3) (a) There were a lot of faces red in the room that night. (b) There were a lot of red faces in the room that night. ( 1 3a), she claims, excludes the 'relative' reading of ' red for a human face' because it cannot refer to American Indian faces as ( 1 3b) can; the red in ( 1 3a) must be absolute. This is clearly wrong: the redness of faces in ( 1 3a) may be no more than a blushing p ink, i.e. only ' red' in relation to faces. Handy, navigable and responsible get similar mis­ treatment ( 1 976: 1 22, 1 46). Siegel fails to spot an essence-accident distinction in some of her Russian data ( I 976: 1 9), and her Ngamambo examples, which show a classic essence-accident distinction for the adjectives and verbs in that language, are also forced into an incorrect CN/CN v. t///e explanation ( 1 976: 1 50- 1 70). 2 Very perfect or very flawless can make sense only i f an arbitrary standard applies; a ' flawless' diamond, for instance, is one which has no flaws humanly visible under 1 0-power magnification, by definition. A god, however, is either perfect or imperfect; there is no room {no extension gap) for very perfect to make sense. In everyday usage, as Sapir ( 1 944) points out, perfect is often used in the general sense of 'approaching closely to perfection ' , which is a predicate fuzzy enough to allow extent modification. 3 Zwicky ( 1 970) notes a striking distinction between pairs of positive and negative marked sentence adverbs: the positive members retain a sentence adverb reading before adjectives while the negative members have an extent adverb reading. JS, vol. 1 , no.3

245

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

navigable, visible, It is i mportant to appreciate that both readings of these examples involve one-place predi­ cates. The issue in essence-accident distinctions is not connected with gross syntactic category.

red, handy and responsible in

KENNETH R. BEESLEY The children are John is Mary is The

{

{ {

{

usually noisy. unusually

typically quiet atypically characteristically uncharacteristically

studious

ordinarily outgoing man became depressed. extraordinarily

References

Anderson, J.R. & Bower, G.H., 1 974: Human associative m emory. Hemi­ sphere, Washington D.C. Backlund, U., 1 973: The collocation of adverbs of degree in English. Doctoral dissertation, Uppsala University. Bartning, I., 1 976: Remarques sur la synta.re et la semantique des pseudo-adjectifs denominau.r en francais. Doctoral disserta­ tion, Institut d ' Etudes Romanes de I ' Universite de Stockholm . Bartsch, R., 1 972: Relative adjectives and comparison in a Montague grammar. In: Bartsch, R. & Venneman, T.; 1 57- 1 86. Bartsch, R., 1 973: The semantics and syntax of number and numbers. In: Kimball, J.P. (ed.); 5 1 -93. Bartsch, R . , 1 9 7 5: Subcategorization of adnomina1 and adverbial modi­ f iers. In: Keenan, E.L. (ed.); 1 75 - 1 87. Bartsch, R . & Vennemann, T ., 1 972: Semantic structures: a study in the Athenaeum Verlag, relation between semantics and S)'lta.r. Frankfurt am Main. Bauer, L.J., 1 97 5 : Nominal compounds in Danish, English and French. U npublished doctoral dissertation, Edinburgh University. Berman, A . , 1 973: Adjectives and adjective complement constructions in English. Doctoral dissertation, Harvard University. Bierwisch, M., 1 969: On certain problems of semantic representations. Foundations of Language 5; 1 5 3- 1 84. Bolinger, D.L., 1 952: Linear modification. PMLA 6 7. 1 1 1 7- 1 1 44. Page references to reprint in Householder, F. W. (ed.), Syntactic Theory 1: Structuralist. Penguin, Harmondsworth, England. Bolinger, D.L., 1 967: Adjectives in English: attribution and predication. Lingua 1 8; 1 -34. Bolinger, D.L., 1 972: Degree words. Mouton, The Hague . Bolinger, D.L., 1 973: Essence and accident: English analogs of H ispanic ser/estar. In: Kachru, B.B. et a!. (eds.); 58-69. 246

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Whatever the origin of this pattern, the task of grammar is to assign each of these modifiers to its correct class.

EVALUATIVE ADJECTIVES Burt, M . K . , 1 97 1 : From deep to surface strocture: an introduction to transformational syntax. Harper & Row, New York & London. Chafe, W.L., 1 970: M eaning and the strocture of language. U n i v e r s i t y of Chicago Press, Chicago & London. Chomsky, N., 1 9 57: Syntactic stroctures. ( Janua Linguarum 4) Mouton, The Hague. Damerau, F .J., 197 5: On 'fuzzy' adjectives. IBM Research Report RC 5340, Thomas J. Watson Research Center, Yorktown Heights, N .Y . Davidson, D. & Harman, G . (eds.), 1 972: Semantics o f natural language. D. Reidel, Dordrecht, Holland. Davis, S. & Mithun, M. (eds.), 1 979: Linguistics, philosophy, and Montague gram mar. University of Texas Press, Austin & London. Dowty, D.R., 197 5: The stative in the progressive and other essence/ accident contrasts. Linguistic Inquiry 6; 579-588. Dowty, D.R., 1 976: Montague grammar and the lexical decomposition of causitive verbs. In: Partee, B. (ed.); 201 -246. Geach, P., 1 956: Good and evil. Analysis 1 7; 33-42. Givon, T., 1 970: Notes on the semantic structure of English adjectives. Language 46; 8 1 6-837. Hamblin, C.L., 1 976: Questions in Montague English. In: Partee, B. (ed.). Hare, R.M., 1 957: Geach: Good and evil. Analysis 1 7; 1 03- 1 1 1 . Kachru, B.B., Lees, R.B., Malkiel, Y ., Pietrangeh, A. & Saporta, S. (eds.), 1 973: Issues in L inguistics: Papers in Honor of Henry and Renee Kahane. University of Illinois Press, Urbana. Kamp, J.A. W ., 197 5: Two theories about adjectives. In: Keenan, E.L. (ed.); 1 23- 1 55. Harper & Row, New York Katz, J.J., 1 966: The philosophy of language. & London. Katz, J.J., 1 967: Recent issues in semantic theory. Foundations of Language 3; 1 24- 1 94. Keenan, E.L. (ed .), 197 5: Formal semantics of natural language. Cambridge University Press, Cambridge. Keenan, E.L. & Faltz, L.M., 1 978: Logical types for natural language. (UCLA Occasional Papers in Linguistics 3) UCLA Lingu istics Dept ., Los Angeles. Keene, G.B., 1 96 1 : Language and reasoning. D. van Nostrand, London. Kiefer, F. 1 978: Adjectives and presuppositions. Theoretical Linguistics 5; 1 35- 1 73. Kimball, J.P. (ed .), 1 973: Syntax and semantics. vol. 2. Seminar Press, New York & London. Klein, E.H., 1 980: A semantics for positive and comparative adjectives. Linguistics and Philosophy 4; 1-45. Lakoff, G., 1 972: Hedges: a study in meaning criteria and the logic of fuzzy concepts. CLS 8; 1 83-228. Lees, R.B., 1 963: The gra m mar of English nominalizations. Indiana Unisity Press, Bloomington, Ind. Levi, J.N., 1 973: Where do all those other adjectives come from? CLS 9; 332-34 5. =

247

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

JS, vol. l , no.3

KENN ETH R. BEESLEY

248

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Levi, J.N., 1 978: The syntax and semantics of complex nominals. Academ­ ic Press, San Francisco &: London. Lytle, E.G., 1 97 1 : Structural derivation in Russian. Unpublished doctoral dissertation, U n iversity of Illinois. Lytle, E.G . , 1 974: A gra m mar of subordinate structures in English. Mouton, The H ague. McConneii-Ginet, S., 1973: Compara tive constructions in English: a syntactic and semantic analysis. Unpublished doctoral disserta­ tion, University of Rochester (New York). Montague, R., 1 974a: English as a formal language. ln : Thomason, R.H. (ed.); 1 88-22 1 . Montague, R., 1 9711b: The proper treatment of quantification i n ordinary English. In: Thomason, R.H. (ed.); 247-270. N ilsen, D.L.F. (ed .), 1 972: English adverbials. Mouton, The Hague. Parsons, T ., 1 972: Some problems concerning the logic of gram matical modif iers. In: Davidson, D. &: Harman, G. (eds.); 1 27- 1 4 1 . Partee, B.H., 1 975: Montague gram mar and transformational grammar. Linguistic Inquiry 6; 203-300. Partee, B.H. (ed.), 1976a: Montague gram mar. Academic Press, New York. Parteoe, B.H., l 976b: Some t ransformational extensions of Montague gratnmar. In: Partee, B.H. (ed.). Quirk , R., Greenbaum, S., Leech, G. &: Svartvik, J., 1 972: A gra m mar of contemporary English. Longman, London. R ichards, B., 1 976: Adverbs: from a logical point of view. Synthese 32; 329-372. Ross, W.O., 1 930: The right and the good. Clarendon Press, Oxford. Sapir, E., 1 944: Grading: a study in semantics. Philosophy of Science 1 1 ; 93- 1 1 6. Page references te reprint in: Mandelbaum , D.C. (ed.), 1 949: Selected writings of Edward Sapir in language, culture and personality. University of California Press, Berkeley. Sampson, G., 1 970: Good. Linguistic Inquiry 1; 257-260. Siegel , M . E.A., 1 97 6a: Capturing the adjective. Unpublished doctoral dis­ sertation, University of Massachusetts (Amherst). Siegel, M . E.A., 1 976b: Capturing the Russian adjective. In: Partee, B.H. (ed.); 293:-309. Siegel, M.E.A., 1 979: Measure . adjectives in Montague grammar. Ir� Davis, S. &: M ithun, M . (eds.); 223-262. Smith, C.S., 1 96 1 : A class of complex modifiers in English. Language 3 7; 342-365. Yale University Press, Thomason, R.H. (ed.), 1974: Formal philosophy. New Haven &: London. Thomason, R.H., 1 976: Some extensiOns of Montague grammar. In: Partee, B.H. (ed.); 7 7 - 1 1 7 . Thomason, R.H . &: Stalnaker, R.C., 1 973: A semantic theory o f adverbs. Linguistic Inquiry 4; 1 95-220. Mouton, The Hague. Vendler , z . 1 968: Adjectives and nominaliza tions. Wheeler, S. 1 972: Attributives and their modifiers. Nous 6; 3 1 0-334. Wint�r, W., 1 965: Tr ansforms without kernels. Language 41; 484-489.

EVALU ATIVE ADJECTIVES Zwicky, A.M., 1 969: A note on becoming. CLS 5; 293-294 . Zwicky, A . M . , 1 970: Usually and unusually. Linguistic Inquiry 1 ; 1 45.

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

JS, vo1. 1 , no.3

249

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

NEGATION, AMBIGUITY, AND THE IDENTITY TEST

John N . Martin

Abstract

In recent years the idea that presupposition should be explained in semantic terms has been attackP.d from various directions. It has been claimed that alleged cases do not require non-classical truth-values, that they are cancellable and hence pragmatic, that purely pragmatic JOU RNAL OF SEMANTICS, Vol . 1 , no.3, pp. 2 5 1 -274

251

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Negation has been closely tied to semantic presupposition since the concept was first discussed. In most accounts there is a definition or a theorem to the effect that A presupposes B if and only if A and the negation of A , in one sense of negation, both entail B. The multiple senses of negation assumed by such principles have been criticized and along with it the concept of presupposition. Indeed one of the most interesting arguments against semantic presupposition is the joint claim that many-valued semantics for presupposition require a mbiguous negation and that negation as found in English is not a mbigu­ ous. In this essay I propose to discuss quite generally the idea of ambigu­ ity and the role of negation in presupposition theory. Along the way I shall argue that it is quite difficult to explain precisely how the usual identity test for ambiguity employed by linguists should apply to a logical connective like negation, and tha t most versions of the test when clarified do not yield the result that negation in English is ambiguous. I argue for these conclusions by a ttempting to clarify what the theoretical properties of language would have to be if this critique of semantic presupposition were right. The kind of �tax and formal semantics needed to support the identity test when combined with the relevant data about natural language usage does not yield the result that negation is ambiguous. The argument is based on details that are of some interest in themselves. An effort is made to formulate precisely what the identity test is, and in particular what the conditions are tha t must be m e t before a meaningful conjunctive a bbreviation is permissible. Two different sorts of conditions are distinguished which really amount to two quite different versions of the test. Only one of U1ese is really relevant to the issue of negation. This variety is also of interest because failure in this sense a mounts to what philoso­ phers have called zeugma. Both sorts are distinguished from a third version of the test, probably the most common, in which it establishes �tactic but not semantic a mbiguity.

JOHN N . M ARTIN explanations in terms of implicature and other such notions are possible, and that those cases that are not pragmatic are really varieties of classical i mplications requiring no special semantic account. But Jay David Atlas and to a lesser extent Ruth Kempson have developed their own critique questioning in a way others have not the assumption they f ind in semantic accounts that natural language negation is ambigu­ ous. 1

Briefly, the argument runs like this. One criterion for ambiguity recognized by linguists is the identity test. According to this procedure transformations that abbreviate conjunctions of similar structure and overlapping content to shorter phrases that omit redundant material sometimes signal the presence of an ambiguity by rejectin g some read­ ings because they are equivocal. For example, ( 1 ) and (2) are each open to two readings, one verbal and the other nominal, but their abbreviation ( 3) admits readings in which both must be verbal or both nominal; so-called cross-readings are rejected. (1) ( 2) ( 3)

I sa w her duck. I saw her swallow. I saw her duck and swallow.

The rejection of cross-readings is thus taken as a mark of ambigui ty. 252

JS,

vol . l , no.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The postulation of ambiguity arises as follows. If we suppose that A semantically presupposes B, then we know by basic principles shared by such theories that whenever A or its negation not-A are true, so is B. Then it should not be possible to find a case of B failing while not-A is true. Yet there are uses of negation in natural language, often called ' metalinguistic' and translated 'it is not true that ' , that are true when the contained sentence is anything other than true. Such a not-A will be true so long as A has any property that keeps it from being true, whether this be falsity, incoherence, absurdity, undefinedness, or even ungrammaticalness. Thus if A 's presuppositions fail, A will not be true, and thus not-A will be true, contradicting the previous claim that when not-A is true, so are A's presuppositions. Presuppositionalists, however, have always been aware of this use of negation and usually maintain that a different sense, sometimes called internal or choice negation, or negation in secondary occurrence, is operative when A presupposes B. Such severe critics of the semantic notion as Wilson, and Boer and Lycan have not called into question the postulation of this ambiguity on the part of negation. 2 But what they and traditional theorists overlook, it is alleged, is the fact that there is really just one natural language negation and the so-called metalin­ guistic uses are just additional proof that presuppositions are cancellable and hence not universal in the way required in the semantic account. Atlas' strategy is to show in detail that natural language negation fails the ambiguity tests recently adumbrated by Zwicky and Sadock, and W ilson 's argument is similar.

N EGATION, AM � IGU ITY , AND THE IDENTITY TEST Atlas makes use of this criterion by observing that abbreviations making use of not are open to cross-readings in which not is simultaneously used to abbreviate expressions that logicians have taken to represent different 'senses' of negation. The use of not in (4), for example, is called exter­ nal, and that in (5) internal, but both sentences yield an acceptable reduction (6). (4) (5) (6)

The king of France is not wise·. The queen of England is not wise. The king of France is not wise and the same thing goes for the queen of England.

The argument is · d ifficult and interesting for several reasons. First of all , it raises the question of exactly what the tests are for ambiguity and how seriously they should be taken. The accepted view seems to be that they offer mere 'criteria' for ambiguity, and that criteria in this sense are just marks or symptoms of a phenomenon. Exactly what this cautious linking amounts to is difficult to say. I think part of what is meant is that the link is not very deeply understood. The mechanism el'plaining the tie is unknown, and it is even left open how inextricably they go together. Essentially what the tests record is a rather rough generalization from examples. Many, many cases of conjunctive abbreviation that reject cross-readings harbour intuitive ambiguities. So at the very least, rejection of cross-readings seems to be a kind of prima facie ground for suspecting an ambiguity. But it remains an open question what the theory would be l ike that explains why this is the case. One of the tasks I'm interested in pursuing here is to sketch what such a theory would be like. In particular I ' m inter­ ested in the mathematical and formal features of syntax and semantics that would yield the result that failure of cross-readings marks ambigu­ ity. A second interesting feature of Atlas' argument is that i t seems to presuppose a very tight tie between failure of cross-readings and ambiguity. Indeed, for the argument to work the failure must constitute not only sufficient but also necessary conditions for ambiguity. On the usual account finding a failure of cross-readings is taken to mark an ambiguity, to be suffici�nt at least prima facie for concluding that there is an ambiguity. But Atlas' argument works in reverse. He observes the acceptability of cross-readings in a few examples and generalizes that it is not the case that some cross-reading is ever rejected in conjunctive reductions involving not. He also assumes that failure of cross-readings is a necessary condition for ambiguity - that if there is ambiguity, then there is some failure of cross-readings. It follows then by modus tollens that uses of negation are not ambiguous. Thus Atlas goes somewhat further than is customary in tying criterion and phenomenon. In doing so he helps to sharpen the inquiry into what JS, vol. l , no.J

253

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Thus, Atlas concludes that negation in English is not ambiguous.

JOHN N. MARTIN sort of metatheory could explain all these claims. It would have to be one that yields as a theorem an inevitable, one to one lmking of criterion to what it marks.

The purpose of the paper may now be set forth more clearly. It is a kind of rational reconstruction of the formal metatheory that would be necessary for language to be as Atlas says it is. The interest of such reconstruction is to see whether language so viewed is plausible. I shall argue that it is not. The reconstruction has several limitations to serve as guides. First of all there are the quasi -theoretical claims of Atlas himself. An expression must be semantically ambiguous if and only if some of its conjuntive reductions have failures of cross­ reading. Secondly, there is the constraint of the standard sort of linguis­ tic theory that seems to be presupposed by users of the identity test. These assumptions w ill require us to give definitions to a string of standard concepts like conjunctive reduction, reading, cross-reading, syntactic ambiguity, semantic ambiguity, and the dist inction often made between general and ambiguous expression. For this standard material I will assume for the syntax a stripped-down version of trans­ formational grammar. For the semantics I will sketch a minimal possible world semantics and explain its motivation as we go along. In both it will be my goal to assume the minimu m , only as much structure as is necessary to get the desired results. Another constraint of the enterprise and measure of its success will be how well the predictions of the theory match the facts of usage. For the most part, however, I will not be arguing w ith Atlas' data. The question is rather how this data, accepted as genuine, is to be explained. When we get down to serious detail, will a metatheory of the sort he envisages hold together? Let us begin with !>yntax. The basic building blocks of a syntax are assumed to be the following.

254

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A third interesting feature of the argument is that it must be taken as claim ing that the ambiguity in question is semantic rather than syntactic. As we shall see in the discussion, many of the ambigu ities marked by the identity test can be straightforwardly explained as syntactic. Indeed this explanation is available for the examples ( I ) t o (3). But the issue i n metalogic that Atlas i s addressing concerns a claim about the semantic ambiguity of negation in natural language. No logician has made any claim about the natural language syntax of not. What is at issue is rather its semantic analysis, whether its various natural language occurrences shift from context to context, meamng sometimes external and sometimes internal negation. Thus the theory that would be necessary to explain Atlas' link between negation, ambiguity, and cross-reading can be sharpened even more. It must be a theory which is capable of formulating the distinction known as semantic ambiguity, and it must yield the result that such ambiguity goes hand in hand with failure of cross-readings.

NEGATION, AMBIGUITY, AND THE IDEN TITY TEST

It will also prove useful to define the notion of a syntactic type or JS, vol. l , no.3

255

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The elements of syntax consist of sets of basic or lexical expressions, a set of formation operations for 'deep structure ', and a set of single­ valued transformations yielding 'surface structure '. For brevity we shall refer to both sorts of functions as grammatical operations, and we im­ pose various structural conditions on them to insure that every gram­ matical expression, defined as any argument or value of a syntactic operation, can be generated in a finite manner. I shall list here just enough of these conditions to develop a workable identity test. In order to ensure that surface structure is transformed from a prior deep structure, we require that formation operations are never defined for the values of both formation operations and transformations. Though transformations are all one-place, taking one expression at a time as arguments, formation operations are typically defined over a series of argument expressions, each operation always taking a set numer of input expressions as an argument series and pairing with it a unique expression as value. Formation functions, but not necessarily transforma­ tiOns, are required to be syntactically unambiguow in the sense that they are uniquely decomposable: no two formation operations and no single operation assign the same value to more than one argument series. Two different transformations or even the same transformation, are, however, allowed to pair the same value with different arguments. Finitary construction of expressions, without loops, is assured by requiring that every expression have at least one gra mmatical tree. These are finite trees with the expression in question occupying the position of maximal element, lexical expressions those of minimal elements, and such that the expression occurring at any node is obtained from those at its immediate predecessors by the application to these expres­ sions of one of the grammatical operations. Such a tree is required to be annotated in the sense that along with it (in the form, say, of a function on its nodes) comes certain information about the con­ struction. Specifically, it must be stipulated for each expression at a non-minimal node which funct ion generated tt and the order of its immediate predecessors used. This additional information serves to distinguish the various trees of an expression producible in more than one way. Trees also allow us to distinguish the various occurrences of a single expression within a longer expression, and within two differ­ ent expressions. We merely identify an occurrence of an expression rela­ tive to a particular grammatical tree with one of the nodes it occupies. When there is no possibility of confusion, we shall speak of a node and the occurrence of the expression 'occupying' it as the same. Strictly speaking, a single expression could have two or more isomorphic trees with the same lexical expressions occupying minimal nodes, the same expressions assigned to corresponding non-minimal nodes and generated from its immediate predecessors by the same functions, applied to its arguments in the same order, and the same annotation assigned to corresponding nodes. But these need not be distinguished for most purposes.

JOHN N. M A RTIN part of speech. By a sub-domain of a function let us mean the set of entities such that for some i, these entities are all those occupying the i-th place in some argument series for which the function is defined. By a function' s range we shall mean as usual the set of all its values. We may then identify a syntactic type with any sub-domain of any range of any formation function, and we require that if a formation is defined for some member of a syntactic type, it is defined for all.

An operation corresponding to an n-place formation function is to be an n-place function on semantic values. It is allowed that two of these functions, unlike the formation rules they correspond to, may assign the same value to different arguments. Operations corre­ sponding to transformations are more complex. In order t o determine the value of the transformed expression f rom that of its unsimplified prototype, two things need to be known: first, whether the syntactic and semantic history of the expression allows that the transformation be meaningful, and second , what the value of the untransformed expres­ sion is. If the transformation is meaningful for the expression, then the transformed version has exactly the same meaning as its untrans­ formed original. Transformations preserve meaning, if they are defined at all. But it is allowed that if an expression violates certain grammati­ cal and semantic constraints, the transformation may be grammatical but semantically meani!")gless because the corresponding semantic operation is undefined for it. Exactly this situation arises in the trans­ formations that are used in the identity test. Exactly what these con­ straints are is a long story that we will take up later. For now we shall just set up the framework so that transformations will be sensitive to such information. Accordingly, we require that the semantic correlate of a transformation be a function from what we shall call the s e m a n t i c history o f an expression, b y which w e shall mean one of its grammatical trees together with an assignment of semantic values to all nodes but the top one. It shoul d be remarked that in formal semantics, it is not all that unusual to f ind semantic rules that need more information than the reference of the immediate part in order to determine the reference of the whole. Semantic rules for the operators of modal logic or the connectives of supervaluations are good examples. But 256

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The necessary semantic theory postulates a series of elements corre­ sponding to the elements of the syntax which I shall call their correlates. Corresponding to each set . of lexical expressions there is a set of appropriate semantic values, and to each syntactic operation there is a unique semantic operat.ion. By a language let us mean a specification of the elements of syntax (lexical classes and syntactic operations) and corresponding elements of semantics (types corresponding to the syntactic operations). The motivation for the semantic structure parallel­ ing syntax is that if the right structure is imposed on it, a projection of meaning to all expressions is determined from an assignment of meaning to just the lexical items. Let me now list a minimum set of such structural conditions sufficient for stating the identity test.

NEGATION, AMBIGUITY, AND THE IDENTITY TEST

What is perhaps the only novel formal feature of the notion of interpretation we shall use is that in conformity with practice in linguis­ tics, it allows for different occurrences of the same expression to have d ifferent meanings in the same context of use. People do as a matter of fact equivocate in this way , and the possibility of such equivocation is assumed in the identity test. We capture it here by defining interpretations relative to grammatical trees in such a way that the various occurrences of an expression may have different meanings, but that the meaning of any occurrence is determined by applying the corresponding semantic rule to the meanings o f its imme­ diate predecessors. Any interpretation for an expression · that i s obtained this way is legitimate. More precisely, let us first define the notion of an interpretat ion of a gra m ma tical tree as any partial function on the nodes of the tree such that ( 1 ) every m inimal element is assigned something in the corresponding set of appropriate semantic values, (2) any node generated by a formation function is assigned that value determined by applying the correlate of the function to the previously defined interpretations of the node' s immed iate predecessors, in the annotated order, and (3) any node generated by a transformation is assigned that value, if there is one, determined by applying the correlate of the transformation to the node 's semantic history. We may now as any define the simpler notion of an interpretation for the language . partial function on expressions that assign a value to an expression only if some interpretation of one of its trees assigns that value to JS, vol . l , no.3

257

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

the ancillary information needed by these more familiar cases consists really of facts about the semantic interpretation of the immediate parts throughout other possible semantic interpretations. In the case of transformations like the ones we shall discuss, the supplementary infor mation is of a different sort, concerning gram matical details of the deeper parts of the sentence itself and of their semantic values. Nevertheless semantic correlates determine a general interpretation in much the standard way. Information about the parts of a sentence will determine an interpretation for the whole. A second feature of the · correlates to transformations that we m ust make explicit is that if they are defined, �hey assign to a transformed expression exactly the same value that is assigned to its immediate part as determined in its semantic history. That is, they 'preserve sense' . If they are defined at all for a semantic history, they assign to it exactly the same semantic value that its maximal expression has in that history. A last condition, already alluded to, is that the semantic correlates of transformations may be partial functions. It is not the case that they are defined for all semantic histories. Thus it is possible that some expression of surface structure has no value, and that it is meaning­ less in this sense. To make a projection of meaning well-defined, we also require that if the i-th sub-domain of a syntactic operation f is included in the range of another syntactic operation g, then the i-th sub-domain of the correlate of f is included in the range of the correlate of g.

JOHN N. MARTIN its maximal node, and it is undefined for an expression only if some interpretation for one of its trees is undefined for the maYimal node.

It should be acknowledged that from the perspective of formal semantics the theory is a bit complicated and somewhat inelegant. Transformations are interpreted by rather baroque functions and interpre­ tations are assigned not to individual expressions primarily but to grammatical trees. But some such complications seem required by the project of capturing standard linguistic assumptions. The various versions of the identity test are formulated in terms of a family of transformations that, in Chomsky's words, "permit or require the deletion of repeated elements, in whole or in part, under well-defined conditions." He gives as examples that (7) may be transformed to (8), and (9) to ( 1 0).4 ( 7) ( 8) ( 9) ( I O)

I don 't like John 's cooking any more than Bill's cooking. I don ' t like John's cooking any more than Bill 's. I know a taller man than Bill, and John knows a taller man than Bill. I know a taller man than Bill, and so does John.

Lakoff discusses other examples, as do Zwicky and Sadock in their summary of identity tests, and Atlas constructs similar cases employing negations in his application of the test to presupposition theory. 5 Lakoff discusses and so does constructions. Zwicky and Sadock cite the straightforward deletion of repetition without proform as in the deletion from ( I J ) yielding ( 1 2). (11) ( 12)

I saw her duck and I saw her swallow. I saw her duck and swallow.

The way such examples figure in evidence for ambiguity can be informal­ ly sketched. Each of the various abbreviated forms is open to different John 's readings, some of which are closed to its transformation. Thus cooking and Bil l 's cooking in (7) may refer either to single acts or to a 258

JS, vol. J , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The reason why logicians avoid equivocation on occurrences of the same expression is that it tends to undermine the purely formal nature of logical truth and validity. When we can mean different things at different occurrences of P, then ' if P then P' is no longer a logical truth. Thus, for the purposes of logical theory, special importance attaches to the subset of interpretations defined above in which all occurrences of each expressiOn have the same value. But since we are Jess interested here in logic than in making sense of the identity test, it is reasonable to allow the relevant equivocation. An expression may then be said to be (semantically) ambiguow if there are interpreta­ tions of the language that assign it different values. J

NEGATION, AMBIGUITY, AND THE IDENTITY TEST product. Moreover, one may refer to an act and the other to a product. But such so-called cross-readings are excluded in (8). Therefore, what is intended by cooking must either be a single act for both John and Bill or a product, but not one of each. L ikewise in (9) each occurrence of know a man taller than Bill may be read as know a man taller than Bill is or know a man taller than Bill does, but ( 1 0) can abbreviate only cases in which one or the other is used in both occurrences. In ( 1 1 ) duck may be a verb or a noun, and likewise for swallow, but ( 1 2) abbrevi­ ates only uses in which both are verbs or both nouns. When such cross­ readings are eliminated, the sentence abbreviated is claimed to be ambiguous.

( 1 3) ( 1 4)

Harry kicked Sam and Pete kicked Sam. Harry kicked Sam and so did Pete.

He explains that Harry may have kicked Sam with his left foot but Pete may have done so in a different way, for example with his right foot, and yet both sentences may be used to describe this situation. Cases that admit cross-readings are contrasted with genuinely ambiguous ones and are variously called vague (Lakoff), general (Zw icky and Sadock), and non-specific (Atlas). But explainmg in detail why such a test marks ambiguity, if it does, is no easy matter. My previous remarks proceeded by giving examples of the relevant transformations and then suggesting that in certain conditions they mark ambiguity. The mathematical task is to find a general characterization of the relevant class of transforma­ tiOns and a general statement of the conditions in which they mark ambiguity. Details must conform with the background theory already laid down. Syntactically the family of abbreviating transformations act by simplifying conjunctions. Both conjuncts typically have the same overall structure, but within this structure they differ at one place. Where the first has one phrase or expression the second has another, though these are of the same part of speech. The simplification then consists in disregarding the long conjunction in favour of a shorter expression with the same structure as each of the conjuncts with the position of the variable phrase or expression taken by some combination of the disparate parts. This very vague procedure can be made precise by first laying down some technical terms. By a conjun c t i \;e formation rule we shall mean a formation operation that maps pairs of sentences onto their conjunction. Let N be an n-tuple of arguments in the domain of a syntactic operation f , and let N-M be the n-m-tuple of arguments obtained by deleting m arguments from N. Let M be the m-tuple of JS, vol. 1 , no. 3

259

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

These cases are contrasted to others in which cross-readings of the reduced form are acceptable. Thus, Lakoff says that no reading appropriate to ( 1 3) is inappropriate for ( 1 4):

JOHN N. M ARTIN

·

=

(i)

(a) (b)

T is a transformation defined over some subset of the range of a conjunctive formation operation (which we shall call f); there is some iterated relativized syntactic operation (which we shall call g) such that if T is defined for an expression, that expression is obtained by applying f to conjuncts obtained from g, i.e., it is f(g(A) , g(B)), for some expressions A and B.

Let us say the common content of g( A) and g(B) in the conjunction f(g(A), g(B)) consists pf the occurrences of expressions occupying nodes in the trees of both g(A) and g(B), but not in the trees of just A and B. Thus, any expression C in the common content of a conjunction has two occurrences, which we shall call parallel, one of which is part of the tree headed by g(A) and the other of which is part of the tree headed by g(B), and these occurrrences are such that the sub-trees headed by them are isomorphic. Then condition (i) may be summarized as requ1rmg that the relevant transformations are defined only for conjunctions both parts of which have the same structure and a common content. 6 It is not the syntactic but the semantic conditions on the transfor­ mations that are the most interesting. There are in fact two rather different semantic constraints. On the whole they have not been clearly d istinguished in the literature, and their difference proves important in applications of the identity test to presupposition theory. The first condition is fairly non-controversional and is George Lakoff ' s requirement that t h e common content o f the conjunction must have the 'same meaning' if the tran5formation is itself to be- meaningful. In addition to (i), we thus require: 260

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

arguments deleted. Then a syttactic operation relativized to f and M is any function g that assigns to N-M what f assigns to N. For example, if f is a 2-place formation function defined for expression A , then g(X) :: f(A,X) would define a syntactic operation relative to f and A. An operation on expressions is said to be iterated and to be relative to the set 0 of formation operations and the set E of expressions if it is a one-place function defined by applying to a given expression a f inite series of relativized syntactic operations such that 0 is the set of all formation functions these operations are relativized to, and E is the set of expressions they are relativized to. For example, if f is a syntactic operation relative to formation function g and expres­ sion A, and h is another such operation but relative to formation func­ tion j and expression B, then k(X) h(g(X)) would define an iterated syntactic operation relative to 4,j > and . We can now state the first condition characteristic of the kind of transformations used in the identity test. This condition is purely syntactic. For a transformation T to be used in the identity test we would require as a minimum:

NEGATION, A MBIGUITY, AND THE IDENTITY TEST (ii)

the semantic correlate of T is defined for the grammatical history of a conjunction f(g(A), g(B)) if and only if the parallel occurrences of every expression in its com mon content have the same semantic value in that history.

The condition says in essence that we cannot equivocate over a deleted element. The requirement is natural enough and lies behind claims like Lakoff's that because ( 1 5) does not admit cross-readings, it is ambiguous. ( 1 5)

Selma likes visiting relatives and so does Sam.

I have chosen to state the condition semantically, in terms of a restriction governing cases in which the semantic correlate of the transformation is defined. From many of the examples given in the linguistic literature it is not clear that the relevant condition need be interpreted this way. In a sense, cross-readings of ( 1 5) and of other examples like Lakoff 's ( 1 6) and ( 1 7) are eliminated by the syntactic conditions (i) without mentioning semantic structure at all. ( 1 6) H arry was disturbed by the shooting of the hunters and so was A I . ( 1 7) The chickens are ready to eat and so are the children. As (i) is formulated, the syntactic structures of g(A) and g(B) fed into the transformation T in the form of the conjunction f(g(A), g(B)) must be structurally isomorphic except that where a tree headed by A appears in the first a tree headed by B appears in the second. Both conjuncts must be of the same part of speech, and it is not implausible to think that the various senses displayed in Lakoff 's examples ( 1 5)-( 1 7) represent different parts of speech. Many ordinary uses of the test by linguists can be viewed as uncovering in this way what are really syntactic ambiguities. But the kind of ambiguity that interests Atlas and that we are trying to explain in metatheory is semantic, and we can construct other examples using more traditional sorts of lexical ambiguity that cannot plausibly be traced to switches in syntactic type. ( 1 8) Tony Benn is a radical and so is the square root of 2. ( 1 9) Ink goes in pens and so do pigs. It I S implausible to think that radical and pens here a r e syntactically ambiguous, falling into different parts of speech. Rather what is wrong is that there has been an equivocation. Literally the same syntactic entity has been used in two different senses. It is, moreover, a semantic constraint as formulated in (ii) that JS, vol. l , no.3

261

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Her visiting relatives may mean the act of going to VISit relatives or the relatives who visit, but not both. Though the condition is plausible enough, a number of important points need to be made.

JOHN N . MARTIN

·

is appealed to in applications of the ambiguity test to negation in presupposit ion theory. Atlas suggests that in some examples negation may be part of the common content of a reduction, yet because cross­ readings of the reduced form are acceptable, the reduced sentences are not ambiguous. His example is (20), but I think ( 2 1 ) would do as well. (20) (2 1 )

The king of F rance is not w ise and the same (thing) goes for the queen of England. The king of France and the, queen of England are not wise.

It is important to see that just as in ( 1 8) and ( 1 9) the acceptability of (20) and ( 2 1 ) is not a syntactic matter resolvable by appeal to (i) alone. No one has questioned that the formation rules of English seem to use exactly the same syntax for the various sorts of negation. Rather the issue facing Atlas and ourselves, as we try to state a metatheory compatible w ith Atlas' prescriptions, is semantic. Given a single for­ mation rule for negation, should some uses be interpreted semantically by the semantic operation for exclusion negation and some by one for chotce negation? Believers of the ambiguity thesis say yes, doubters no. Thus to make sense of the debate over the ambiguity of negation and the use of the identity test to settle it, we must assume the seman­ tic condition ( ii). Now let us see exactly how (ii) bears on ambiguity in cases for which cross-readings are excluded. Ideally what we want to do is first define a conjunction reduction transformation as any transformation meeting conditions (i), (ii) and perhaps other conditions as well. We then need to define the notion of a reading, and prove some theorem like the following: If T is a conjunctive reduction transformation, then f(g(A), g(B)) is ambiguous if and only if there are some readings for f(g(A), g(B)) that are not readings for T(f(g(A), g(B))). As will emerge the readings open to the conjunction but closed to its reduction are what we have been calling informally the excluded 'cross-readings '· The theorem as stated is quite strong. It requires that the failure of some cross-readings is both a necessary and sufficient condition for ambiguity. Strictly spt:aking the so-called identity test for ambiguity requires only that it be a sufficient condition. What we do is hunt 262

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

One glosses the examples by explaining that the king of F rance is not w ise because there is no such person, and the Queen of England is not w ise because though existing she lacks the relevant properties. It does seem true that such a context can be convincingly described in which it is fair to say one could summarize the situation by either (20) or ( 2 1 ).

NEGATION , AMBIGU ITY, AND TH E IDENTITY TEST around for an example of a reduction that does have some excluded cross-readings and then conclude that the transformed sentence harbours an ambiguity. But Atlas' application of the test to presupposition theory requires the converse also. He generalizes from cases like (20) to the thesis that all cross-readings of the conjuctions involving negation also apply to their reductions. Then by appeal to the converse he can conclude that the conjunction is not ambiguous. Part of our task then is to see whether the properties of the transformations will support both directions of the principle.

(22) (23) (24) (25)

I saw a dog and a man. The search uncovered only a bug and a bat. A man's aggressive sexual behavior is correlated to testosteron levels and so IS a dog 's. The wings of the bat are covered by a thin memQrane and so are those of the bug.

But a technical and , from our viewpoint, a more interesting prelimi­ nary to evaluating either direction of the identity test is finding the right analysis of 'reading ' . This 1dea as 1t is used in linguistiC discussions is a bit slippery. On the one hand, genuinely ambiguous expressions are said to be so because they have more than one reading. In this usage, which I shall call intensiOnal, reading seems to mean something like meaning, sense, semantic representatiOn, or synonymous paraphrase. It IS what is represented in our foregoing theory by the idea of an expressions's semantic interpretation. Thus Zwicky and Sadock speak of (26) as having multiple 'unders.tandings ' . (26)

They saw her duck.

On the other hand, what makes some sentences vague, general, and non-specific 1s also that they have many readings. They differ from ambiguous expressions in that all of their readings also apply to their conjunctive reductions. Usmg essentially th1s idea of reading, Lakoff allows that Harry kicked Sam may be consistent with Harry ' s kicking Sam with the left foot or the right foot, and likewise Zwicky and Sadock explain that (27) may have as d1stinct understandings as both (28) and (29). 7

JS, vol. l , no.3

263

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Before proceedmg further into theory, it is relevant to point out that the converse viewed as a generalizatiOn about language is not very accurate. It has some counter-examples. Both (22) and (23) admit cross-readings. In the f irst contained expressions are ambiguous between a wider class and its male subset, and in the second they are ambiguous between disjoint sets. But even by the identity test itself, the terms in question are ambiguous: there are cases, (24) and (25), m which cross-readings fail. In (24) both terms are li mited to the male reading, and m (25) both stand for animals.

JOHN N . M ARTIN (27) (28) (29)

M y sister is a prominent composer. M y sister is the composer of "Concerto for Bassoon and Tympani". My sister published a concerto last week.

· Now this sense of reading differs from the first in several important ways. First, this sort of reading is clearly not synonymous with the expression it interprets and is something quite different from the tradi­ tional notions of an expression 's sense or intension. Harry ' s kicking Sam is not equivalent to his kicking with the left foot or to his kicking with the right. L ikewise (27) is not synonymous with either (28) or (29), and none of these entail any of the others.

Now the advantages and disadvantages of using possible worlds in semantic theory are well-known. Their attraction here i s both general and specific. They can be used first of all to give concrete examples of languages and semantic interpretations as these notions have been defined in our general background theory. Indeed, i t would tell against these notions if they could not be seen as embracing the ordinary possible world semantics of logical theory. In these accounts semantic values or 'intensions' are set-theoretic constructions made up out of a postulated set of possible worlds. The sense of a sentence, for example, is a function that pairs a possible world with the truth-value of the sentence in that wqrld. How to define semantic operations on such intensions so as to generate well-defined semantic interpretations for simple formal languages is now well-known, and extensions of these methods to richer languages closer to natural speech is now also commonplace. It would be a strength of any account of the identity test to explain how it fits with these standard ideas. More specifically, the idea of possible world also provides a straight­ forward analysis of the second usage of reading. In this sense, which I shall call referential, a reading of a sentence is any possible world in which it is true. A sentence would then have more than one reading, and various other sentences would be partial specifications of it. Thus ( 1 6) and ( 1 7) help to specify a reading of ( 1 5) in that they could each be true in some world in which ( 1 5) is trut. All three might even be true together. It must be admitted at once that this notion of reading is rather trivial. Every sentence but a contradiction would have various 264

JS, vol. l , no. 3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It m ight be possible to explai_n this sense of 'understanding ', as Zwicky and Sadock suggest, as any state of affairs describable by the sentence. But in introducing states of affairs to semantic theory we would be adding a new theoretical concept that would itself need to be explained. Some recent formal work on facts or events might be used to this end. But there is a more obvious approach that does not introduce any more semantic entities than are regularly appealed to in intensional logic. We interpret state of affairs in the full-blooded sense of possible world. A possible world is, if you like, a total and complete state of affairs.

NEGATION, AMBIGUITY , AND THE IDENTITY TEST readings, and saying of a sentence that it is general, vague, or non­ specific because is has various readings is then to say not very much of interest. Most sentences, ambiguous or otherwise, would be general in this sense. It is not a very interesting idea. To be fair to those who use the notion, we should say that it is not given much weight; it is used for little more than as a means of contrasting the genuinely ambiguous from the merely general. But what is very interesting about the referential account is its ability to provide a simple and useful sense of 'reading'. Given the rather weak notion of referential reading, we can explain how the identity test establishes ambiguity. But before we do that let us backtrack to the intensional sense of ' reading' and investigate to what extent it is reconcilable with the identity test.

llleoce m

Proof

-In any language in which T is a conjunctive reduction and in which there are some intensional readings of f(g(A), g(B)) that are not also readings : of T(f(g(A), g(B))), the former expression contains an ambiguity. -Let T be as specified and let there be some reading for the conjunction that is not one for its transformation under T. The only reason a reading for the conjunction would not be the same as that of its T -transformation is that the semantic correlate of T is undefined for the semantic history of f(g(A), g(B)). But by (ii) it is undefined only if some expressions shared by g(A) and g(B) are ambiguous. B

This result also holds for the referential use of reading. Let a lan­ guage be said to have a possible world semantics if it assigns as a seman­ tic value to each sentence a function from possible worlds to truth­ values. We make no assumptions about the number of truth-values beyond the classical two values T and F, but require merely that the semantic operation corresponding to conjunction be normal in the many-valued sense: if both parts of a conjunction are assigned classical truth-values T or F, then the semantic operation ass1gns that value dictated by the classical truth-table for conjunction. We can now prove that lack of cross-readings in the referential sense marks ambiguity. Since th1s is essentially the use of 'reading' employed by linguists, the result is really a statement of the identity test. Theorem

Proof

(Tile Identity Test) - In any language with a possible world semantics, and in which T is a conjunctive reduction, and in which there is some referential reading of f(g(A), g(B)) that is not also a reading of T(f(g(A), g(B))), the former expression contains an ambiguity. -Assume the conditions. There is some world in which the conjunction is true but its T -reduction isn 't. Since the seman-

JS, vol. l , no. J

265

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

For the time being let us call a conjwlCtive reduction any transforma­ tion meeting conditions (i) and (ii). Something like the identity test follows directly from the def initions.

JOHN N. MARTIN tic correlate of T, if defined, is an identity mapping, the correlate of T must be undefined, and therefore, as in the last proof , g(A) and g(B) share ambiguous expressions. It is important to see, however, that the relevant converses of both theorems fail. Theor-em

Tileor"em

To the extent that these results fail to justify in our metatheory the converse of the identity test, they tend to undermine Atlas' argument. But they are in a sense too strong. What is established is just that acceptable cross-readings are consis­ tent with ambiguity when the ambiguity referred to is in those parts of the conjuncts which differ. All Atlas needs for his argument, however, is that acceptable cross-readings are inconsistent with ambiguity oc­ curring in those parts of the conjuncts which are the same. It is negation he wishes to argue is not ambiguous, and in all his examples negation is in the common content of the reduction. Within the contraints of our reconstructed metatheory, aren 't acceptable cross-readings formallv 266

JS, vol . J , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Proof

-There are languages in which T is a conjunctive reduction and the conjunction contains some ambiguous expressions, yet every intensional reading of the conjunction is also a reading of the reduction. -There are languages with possible world semantics in which T is a conjunctive reduction and the conjunction contains some ambiguous expressions, yet every referential reading of the conjunctions is also a reading of the reduction. -To see why the firts result holds it suffices to note that the semantic correlate of T is always defined if and only if the intensional readings of the conjunction are always readings of the reduction. Moreover, according to (ii) it is always defmed if and only if all the common content of g(A) and g(B) is univocal. Languages are easily constructed that give the same interpretation to the com mon content but different interpretations to parts of A and B, by, for example, giving lexical items in the common content the same value, but ite m s in A and B different values. In such languages the semantic correlate is always defined and hence all cross-readings are acceptable. Yet the expession contains ambiguities. Indeed with the right choice of f and g the conjunction itsel f can be made ambiguous. For the referential case observe first that every referential reading may hold of both the conjunction and the reduction, yet the semantic correlate of T might still be undefined for some values of the conjunction. We can nevertheless construct a language in which all cross-readings apply to both and the correlate of T is always defined. We may then proceed as In the intensional case.

N EGATION, AMBIGUITY, AND THE IDENTITY TEST

Theorem -In any lan guage in which T is a conjunctive reduction and

in which some expression in the common content of g(A) and g(B) is ambtguous, there is some intensional reading of f(g(A), g(B)) that is not a readmg of T(f(g( A), g(B))). Theorem -There is some language with posstble world semantics in which T is a conjunctive reduction and in which :;orne expres­ sion in the common content of g(A) and g (B) is ambiguous, yet every referential reading of f(g(A), g(B)) is also a readmg of T(f(g(A), g(B))).

Discussion could stop at this point if it were not for the fact that we may have under-represented conJunction reduction in an important way. It is common in the literature to impose additional condttions in the form of category constraints on the disparate parts, and it might seem posstble to obtain Atlas' converse if these additional assump­ tions are incorporated into the semantic theory. As I shall argue, I think such an approach fails, but it is interesting, especially in its need to clarify what sort of metatheory these category constraints presuppose. The condition in question requires that the disparate parts of the conjunction can be yoked in a reduced form only if they are originally of the same type. "Roughly", say Zwicky and Sadock, "to be eligible for reduction two conjoined clauses must be of the forms X--A--Y and X --B--Y, where A and B are constituents of the same type." Likewise, Chomsky posits "some general condition of the appltcability of deletion operations such as the one that gives ( 3 1 ) from (3Q), a rather abstract condition that takes into account not only the structure to which the operation applies but also the history of derivation of this structure." 9

JS, vol. l , no.3

267

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

inconsistent with ambiguity in the common content? The answer depends on the sense of reading. It is true that under the intensional notion all cross-readings apply if and only if the semantic correlate of the transformation is always defined. Hence by (ii) ambiguity in the common content entails that the correlate is undefined for some arguments and, therefore, that not all cross-readings apply. But under the referen­ tial sense of readmg this inference does not follow. It is compatible with that notion to have a language in which every world in which the conjunction is true is also a world in which the reduction is true, i .e., for all cross-readings to apply in this sense, yet the transformation 's correlate might also be undefined for some values. A trivial example can be constructed by defining a correlate that assigns to any proposition (function from worlds to truth-values) itself except in those cases in which the proposition is contradictory (assigns F to every wor ld) , and for these cases we let the correlate be undefmed. Then any world that satisfies the conjunction will also satisfy its reduction. Other more complex examples are also possible. We have in effect estabilshed two st mple results.

JOHN N . M A RTIN (30) (31)

I don ' t like John's cooking any more than Bill's cooking. I don ' t l ike John's cooking any more than Bill 's.

These conditions would allow the reduction only if cooking i n both con­ juncts was of the same type, both a general practice or an isolated act .

But the syntactic interpretation is not the most interesting, and the intentions of linguists in stating the condition are not always clear. Chomsky, like many others, posits a parallel between syntactic and semantic structures. Zwicky and Sadock are also typical in shifting from syntactic to semantic vocabulary in a manner justified by postu­ lating a vague correspondence. Moreover, it is clear that the syntactic interpretation will not help Atlas' argument. For the syntactic restric­ tion is already captured in the notion of a conjunctive reduction, and that idea does not y ield the implication from ambiguity to failure of cross-readings which Atlas needs. The semantic version is also interesting in itself. It would allow the possibility that two expressions of the same syntactic type might belong to distinct semantic categor ies, and then require that reductions are meaningful only if the two disparate parts were of the same seman­ tic category. Technically we would have to augment the specification of a language with what we may call (following Thomason and others) a sortal specification, some partitioning of possible semantic values for each part of speech, stipulated prior to the· definition of an interpre­ tation. Then in lieu of condition (ii) we would require: (iii)

the semantic correlate of T is def ined for the grammatical history of a conjunction f(g(A), g(B)) if and only if (a) the parallel occurrences of every expression in its common content have the same semantic value in that h istory, and (b) the interpretations of A and B in that history are of the same semantic category as defined in the sortal specification of the language.

268

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The idea behind the condition is itself somewhat ambiguous; the . notion of type involved may be construed as either syntactic or semantic. I think many of the examples used to il lustrate this condition in the literature do make sense when the type restriction is interpreted strict­ ly syntactically. Zwicky and Sadock's example in which reduction is allowed if both duck and swallow are verbs or both are nouns is a good case. L ikewise Chomsky seems to think that the differences in type of cooking will be exhibited in its various syntactic histories and that John's cooking really does fall into two parts of speech. But if the condition is merely syntactic, it is essentially captured already in requirement (i). Formation functions are defined relative to parts of speech and in particular the operation g used in f(g(A), g (B)) is defined over a single part of speech. Therefore if g(A) and g(B) are defined, A and ·B are of the same syntactic type.

NEGATION, AM BIGUITY , AN D T H E IDENTITY TEST This categorical restriction has been discussed in the philosophical literature under the topic of zeugma. 1 O Indeed , a zeugma may be defined as a conjunctive reduction that violates (b). It is granted by most that some such violation occurs, often with literary effect, but it is a matter of contention whether they are seriously deviant and if so how this deviance should be marked within semantic theory. First some examples. The f irst two are standardly given and the latter two are cited by Fowler: (32) (33) (34)

Most concede, I think, that such examples are grammatical. They differ on whether they are semantically deviant, and if so, how this deviance IS to be marked. Theories include classical accounts which admit that zeugmas contam category mistakes but argue that these should be represented as false in a classical two-valued semantics. On this v1ew, zeugmas are literally fa,lse. There are also various many­ valued approaches that conform better to the intention of (iii) and render zeugmas neither true nor false. .

The first point to make about the treatment of zeugma required in (iii) is that it is at odds with most for mal approaches. It is not really true that zeugmas are meaningless in formal accounts even though their deviance is marked. Whether th1s marking consists in assigning the classical truth-value false o r in assigning some more elaborate truth-value gap or non-classical value, the expression literally has an intenswn in the model theoretic sense. It is always interpreted by some function from poss1ble world to trut h-values, and the semantic operations generating interpretations of zeugmas would be defined for any argument. Another way to put this point is that the term ' meaningless' is itself ambiguous. It may mean that the semantic opera­ tions of the theory are really partial functions and that, though the interpretations of the parts are assigned, the whole has no semantic value because the appropriate semantic rule is undefined for those inputs. It is in this sense that (iii) ensures that zeugmas are meaningless. But the intensions assigning sentences non-classical values constitutes meaninglessness in a different sense. Such assignments usually assign set-theoretic meanings. Indeed there is an important techniCal reason why undefmed values for semantic operations are avoided. The usual theory of logical consequence is defined in terms of truth-values and presupposes that expressions always have intensions in the model theo­ retic sense. We may speculate that a theory of logical consequence for the surface forms represented in conjunctive reductions would be inappropriate or that it might somehow be managed even in the JS, vol. J , no.3

269

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(35)

She came in a flood of tears and a sedan chair. The room was not light but his fingers were. Half-clad stokers toasting in an atmosphere consisting of one part air to ten parts m ixed perspiration, coal dust, and profamty. Such frying, such barbecuing, and everyone dripping in a flood of sin and gravy.

JOHN N. MARTIN presence of partially defined semantic operations. But in doing so we should realize that w e 're taking a big step into unchart�d regions. As theories in formal semantics now stand, (iii) is actually implausible.

If A and B are ambiguous then their various meanings may be paired up m such a way that they fall together m different semantic categories. But thts idea is altogether too strong. Why couldn 't all the meanings of two terms fall into the same sort? Let us turn to the converse. Does failure of cross-readings contmue to entail ambiguity as before? Assume that some cross-readings for the conjunction are closed to the reductton and that therefore the semantic correlate of T is undefined for some argument. Then by (iii) there is either an a mbiguity in the common content or a sortal violation. To derive the conclusion that there must then be an ambiguity in either of these cases, we need again a supplementary pr inciple, this time saying something like the following: Whenever the interpretations of A and B fall into different semantic categor ies, at least one expression in A or B is ambiguous. But this notion is even less plausible than the last. Why couldn 't A and B just be univocal lexical expressions of different sorts? This excursion into semantic category restricttons on conjunctive True, reduction has amounted in effect to a kind of elaborate reductio. there are some sorts of category constraints that make sense for conJunctive reduction, but these are syntactic and have already been captured in (i). 270

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A second point to make about (iii) is that even if It is accepted as it is, it needs buttressing to provide any links to ambiguity. Let us call a strengthened conjunctive reduction any transformation meeting conditions (i) and (iii). We may now ask whether the additional structure imposed in (iii) ensures Atlas' converse. Let us suppose that f(g(A), g(B)) contains an ambiguous expression. Given :hat the expression is in deep structure and therefore is no� syntactically ambiguous, this ambiguity must be traceable to ei ther some of the lexical expressions in the com mon content of g(A) and g(B) or to A and B themselves. To show that ambigutty entails failure of some cross-reading, what we must then be able to prove ts that ambiguity in any of these, together wtth (iii), entails that the semantic correlate of T is undefined. But (tii) as it stands only entails ambiguity if the source of the ambiguity is in the common content. Suppose the ambiguity is in A and B. To use (iii-b) we must be able to show that ambiguity in A and B entails a violation of sortal specification. We need some principle like the following:

NEGATION, AM B IGU ITY, AND THE IDENTITY TEST

It might seem that we could just do away with the referent1al sense and revert to readings as mtensions. But to do so would be to ignore an 1mportant heuristiC motivation. There is more to the referential idea than its use in the rather trivial defmition of a general expressiOn. Its attraction is methodological and epistemic. Intuitions about which situations make a sentence true are clearer, more positiv­ istically palpable than intuitions about proper analysis or definition. This is especially true of disputes about amb1gu1ty. It is not at all clear, to use a Quine 's example, whether hard has two definitions or just covers a range of quite disparate things. It IS precisely because intui­ tions about mtensions are unclear that the identity test is supposed to be useful. It removes inquiry from the realm of meanings to judge­ ments about truth and falsity. On th1s methodological preference, lingu1stic practice seems nght. Intuitions about truth are clearer than intu itions about how many senses an expression has, and Atlas' examples about negat1on are quite convincing. But it IS essential to this method that 'reading' be understood in the referent1al sense, and in this sense ambigu1ty is perfectly compatible with acceptable cross-readings. We may speculate that on some other analysis of reading, perhaps one somewhat in between the intensional and referential, the notion would support both d 1rect10ns of the identity test and serve to define generality. Atlas has actually raised this interesting possibility with me m conversation. One approach might be to take the Identif ication of reading with state of affairs more seriously, and to unpack reading m terms of various set-theoretic constructs of facts or events developed for other purposes in the logical literature. The reading of a sentence would then be any fact that if it obtained would make the sentence true. Whether facts could be defined sufficient to Atlas' purposes is, however, an open question. 1 1 JS, vol. l , no.3

27 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The overall dialectic of the paper so far has not been simple. What we have been doing, in effect, is considering various elaborations of the theory behind the identity test. We would like to make 1t work, to make Atlas' converse work, and to maintain the referential notion of reading. Given the background assumptions I have adopted, we can conclude that the identity test itself with its referential usage of reading is justified, and that even with the intensional use of readmg, fa!lure of cross-readings marks ambiguity. We may also conclude that ambiguity entails failure of some cross-readings m the intensional, but not the referential sense. I suspect that 1t 1s the former entailment that Atlas may have in mind. If so, he is right insofar as I have justified it here. But it is wrong to assume the entailment for the referential notion of reading, as he does in the critique of presupposition. Both the identity test and Atlas' particular examples about negation are formulated in terms of the referential sense. These conclusions depend in part on the technical details of the background theory, and these no doubt are arbitrary in places and open to negotiation. But as far as I can see, no small change will alter the major crit1cal point.

JOHN N. M ARTIN Philosophy Department U n iversity of Cincinnati

Notes l would l ike to thank Jay A tlas and Merrie Bergmann for helpful discus­ sion of an earlier draft.

272

JS, vol. I , no.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

See especially Atlas ( 1 97 7) and Kempson ( 1 975: 99, 1 00). For an account of the history of the argument, see Atlas ( 1 978: 402, note 3). 2 Wilson ( 1 97 5) and Boer & Lycan ( 1 976). See also Martin ( 1 979). ' 3 There is · another sense in which an expression is ambiguous if there is some tree containing more than one occurrence of that expression and an interpretation for that tree that assigns different values to those occurrences. But given ambiguity in the first sense there are two trees each w ith the same maximal element assigned different semantic values. From these trees we can make up a larger one by feeding the two maximal elements as distinct occurrences into some formation function that takes this pair as arguments. Then the expres­ sion would be ambiguous in the second sense. Conversely, given ambigui­ ty in the second sense it is straightforward to break up the tree into two subtrees with the same sentence as maximal element but different interpretations. We may then define different interpretations over expressions, one that conforms to the interpretation relative to the first tree, and one to that relative to the second, and the expression is then ambiguous in the f irst sense. Note that if the expression is not in deep structure or cannot be repeated as part of the argument series of some formation operation, the implication from sense one to sense two fails. But the two notions are for practical purposes much the same, and l opt here for the former because it more closely conforms to usage in formal semantics and philosophy of language generally. 4 Chomsky ( 1 972: 32-35). 5 lakoff ( 1 970); Zwicky & Sadock ( 1 975); Atlas ( 1 977). 6 The observant reader may have noticed that the syntactic condition (i) doesn' t actually say anything about the shape or form of the simplification itself. In particular it doesn 't require that the common content be displayed in its previous form and that the disparate elements be yoked in some fashion with an 'and '. A complete account of conjunc­ tive reduction would indeed need these additional conditions howeve r they should really be spelled out. But for our purposes the actual shape of the resulting abbreviation is irrelevant, so l haven 't ventured to mention it. Syntactically, conjunctive reduction is one of a large family of transformations that eliminate redundancies. Other sorts, for example, reduce 'Tom wants Tom to com e ' to 'Toms wants to come ', 'Tom is as large as Bill is' to 'Tom is as large as Bil l ' , and ' Tom looks like Bill looks' to 'Tom looks like Bill ' . Varieties of specific conjunction

N EGATION, AMBIGUITY, AND THE IDENTITY TEST

Refererx:es Atlas, J.D., 1 977: Negation, ambiguity, presupposition. Linguistics and Philosophy 1 ; 3 2 1 -336. Atlas, J.D., 1 9 78: On presupposing. Mind 8 7. Bergmann, M., 1 977: Logic and sortal incorrectness. Review of Metaphysics 31; 6 1 -79. Bierwisch, M. and H eidolph, K .E. (eds.), 1 970: Progress in Linguistics. Mouton, The Hague. Blackburn, W.K., unpublished: Ambiguity and non-specificity: a reply to Jay David Atlas. Department of Philosophy, University of Toronto. Boer, St. and Lycan, W.G., 1 976: The myth of semantic presupposition. Indiana University Linguistics Club. Bloomington Indiana. Proceedings of the Berkeley Channon, R., 1 975: Acceptable ambiguity. Linguistic Society I; 37- 6. Chomsky, N., 1 957: Syntactic Structures. ( Janua Linguarum 4). Mouton, The Hague. Chomsky, N., 1 972: Language and M ind. Harcourt, New York. Grinder, J. & Postal, P.M., 1 9 7 1 : Missing antecedents. Linguistic Inquiry 2; 269-3 1 2. Hankamer, J., 1 973: Unacceptable ambiguity. Linguistic Inquiry 4; 1 7-68. Linguistic Inquiry 2; Jackendoff, R., 1 97 1 : Gapping and related rules. 2 1 -35. =

JS, vo1. 1 , no.3

273

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

reductions including and-so-does constructions have been studied since the early days of transformational grammar. See Chomsky ( 1 957: 26- 27, 65-67). One imp ortant variety of conjunctive abbreviation is known as gapping, the deletion of centrally embedded (verbal) material, as in ' Mary catches fish and Tom butterflies'. See Ross ( 1 967). The syntac­ tic literature on the varieties of conjunctive reductions is qUJte large. Cf. Ross ( 1 970), Lakoff & Ross ( 1 970), Jackendoff ( 1 9 7 1 ), Grinder & Postal ( 1 97 1 ) , H ankamer ( 1 9 7 3) , Channon ( 1 97 5), Stillings ( 1 9 7 5), Neijt ( 1 978). It is interesting that all such reductions, conjunctive and otherwise, eliminate cross-readings and are probably governed by the sorts of semantic constraints discussed in this paper. Another interesting paper ciritical of Atlas, one which I came upon too late to discuss here, is Blackburn (unpublished). 7 Zwicky & Sadock ( 1 97 5), esp. p.3, note 9. 8 Note that even if g(A) and g(B) share some ambiguous expressions; any other whole they are parts of, like f(g(A), g(B)) and T(f(g(A), g(B))) , might well be unambiguous because semantic operations as defined are allowed to give different combmations of arguments the same value. 9 Zwicky & Sadock ( 1 975: 1 8); Chomsky ( 1 972: 33). 10 See Thomason ( 1 972); Martin ( 1 97 5a); Bergmann ( 1 977). 1 1 See for example Van F raassen ( 1 969); Martin ( 1 97 5b); Martin (in press).

JOHN N. M ARTIN

·

ical Linguistics.

·

Neijt, A., 1 978: Constraints on gapping. Chicago Linguistic Society 1 4; 307-3 1 5. Ross, J.R., 1 96 7: Constraints on Variables in Syntax. Dissertation M IT. Ross, J.R., 1 970: Gapping and the order of constituents. In: Bierwisch & Heidolph (eds.); 249-259. Stillings, J.T., 1 975: The formulation of gapping in English as evidence Linguistic for variable types in syntactic transformations. Analysis 1 ; 247-274. Thomason, R., 1 972: A semantic theory of sortal incorrectness. Journal of Philosophical Logic 1; 209-258. Van Fraassen, B.C., 1 969: Facts and tautological entailment. Journal of Philosophy 66; 4 77-487. W i lson, D., 1 97 5: Presupposition and Non-Troth-Conditional Semantics. Academic Press, London - New York. Zwicky, A.M. & Sadock, J.M., 1 975: Ambiguity tests and how to fail them. In: Ki mball (ed.); 1 -36.

274

JS, vol. 1 , no.)

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Kempson, R . M ., 197 5: Presupposition and the Delimitation of Semantics. Cambridge University Press, Cambridge. Kimball, J.P. (ed .), 1 975: Syntax and Semantics IV. Academic Press, New York. Lakoff, G., 1 9 70: A note on vagueness and ambiguity. Linguistic Inquiry 1 ; 357-359. Lakoff , G., & Ross, J.R., 1 970: Two k inds of and. Linguistic Inquiry 1 ; 2 7 1 -272. Martin, J.N., 1 975a: A many-valued semantics for category mistakes. Synthese 31; 6 3-83. Martin, J .N ., 197 5b: Facts and the semantics of gerunds. Journal of Phil­ osophical Logic 4; 4 39-454. Martin, J.N., 1 979: Some misconceptions in the critique of semantic presupposition. Theoretical Linguistics 6; 235-282. Martin, J.N., in press: Facts and events as semantic constructs. Theoret-

DISJUNCTION

Stephen Read

Abstract

I . How many disj unctions are there? Even classically it has been common to distinguish exclusive from inclusive disjunction. More recently, w1th the development of the relevant logics, a distinction has been observed between (in both cases inclusive) extensional and intensional disJunctions. Formal calculi with the power of treatjng these connectives have been set up. showm g their logical behaviour and their interaction with other connectives. 1 (Of these calculi, I shall work for the purposes of this papaer in R, the calculus of relevant implication.) I wish here to consider whether such a distinction between two uses of or is present in everyday reasoning. For simply to be able formally to characterise a connective may have no relevance to our reasoning practice: consider, e.g., the many calculi for necessity (50.5, S l , 52, 54.2, t.ukasiewicz, K, etc.) which are most implausible as formal representations of any everyday notions. M y question is whether one finds in natural reasoning these two different notions of disjunction. 2. Extensional inclusive disjunction, • v , may be formally characterised by: A f- A V B (Addition) '

JOURNAL OF SEMAN TICS, Vol. ! , no.3, pp. 275-285

275

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

�

In relevant logics one can ormally distinguish two logical operators, Addition holds for ' v ' and Dis­ symbolised as r A v B -, and A + B ! junctive Syllogism for ' + ', but not vice versa. The question arises, whether this distinction between two different formal notions of disjunc­ tion can be found in natural reasoning. First it is necessary to rebut Quinean objections to any rival to classical logic, Grice 's claim that an intensional disjunction is not needed to e:rplain the everyday uses of or, and Kempson and Cormack's argument that there can be no a mbiguity between putative readings one of which entails the other (for A + B f- A v B). Finally Jackson's use of assertibility-conditions to defend the thesis that r if A then B I is equivalent to r A :J B I is rejected, but the notion of robustness which he introduces is usefully adapted to show why Disjunctive Syllogism must fail for ' v '• Having cleared the ground in this way, two uses of or are considered which have in everyday reasoning the inferential properties of ' v ' and ' + ', thus de­ fending the relevantist claim that or is a mbiguous.

5TEPHEN READ (A and

-+

C) &. (B

-+

CHI- (A v B)

-+

C ( Reasoning by Cases, and its converse)

� (A v B)-II- (� A &. � B)

(De Morgan)

(Here ' & ' is extensional conjunction, ' -+ ' is intensional (relevant) implication, and '--ll-' abbreviates two-way logical consequence - in R .) If we have available an intensional conjunction, 'x' (often written 'o' and called 'fusion'), we can characterise intensional disjunction '+ ' , in a similar way:

� (A + BH I- ( � A

X

� B)

C

(Disjunctive Syllogism) (Intensional Reaso11ing by Cases) (IntensiOnal De Morgan)

We have also A + B --ll-- � A

-+

B.

3.

It is immediately apparent that one difference between 'V ' and ' + ' is that where 'v ' pairs w ith ' & ' , ' + ' pairs with ' x ' . But there is ;mother important feature: in R Addition actually fails for ' + ' , and Dis­ junct ive Syllogism ( DS) fails for 'v'; that IS, and

A v B, � A II B.

l:ldeed, in order to distinguish '+' from ' v ' , the consequence relation in relevant logic must deviate from that in classical logic. One can of course attempt a Quinean point here: that to " save the obvious" validity of DS, one must infer that in relevant logic the sense of · � · and 'v ' must be different: "Here is the deviant logician 's predica­ ment: when he tries to deny the doctrine he changes the subject." (Quine 1 970: 8 1 ) But the classical paradigm, however obvious it may be to its proponents, is not thereby automatically correct: of course one can be w rong. about logic. Sheer logic is conclusive only m that, if correct, it IS necessarily so. It does not follow that all (classical) logical claims are true. Given standard meanings to ·� ', ' v ' and ' 1- ' , it is denied m R that, e.g., DS is valid: the consequence relation has a different (smaller) extension. How then do we give standard meanings to ' � · , ' v ' and ' I- ' ? What is it that shows what is meant by, e.g., ' v '? If one is to challenge one conception of logical consequence by another, then one answer - the 216

J�,

vol. l ,

r.o.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A + B, � A 1- B (A .... C) X (B .... CHI- (A + B) ....

DISJUNCTION wrong one - is not that all, but that enough of the logical relations in the one must be conserved in the other. And we do f ind that a good number of the i mportant classical relations of ' v ' are preserved in R (and E). We have A 1- A v B, B 1- A v B, and � A & - B 1-� (A v B). Thus the simple truth of A suffices for that of A v B (as does that of B), and the (extensional) conjunction of the falsity of both suffices for that o f r A v B -,.

' v ' is a rigid designator whose reference is fixed by a rigid descrip­ tion: it is that truth-function which has truth-line TTTF. Then in R and E one is claiming that OS fails for that connective: that a particular possibility for that connective is actual. • v ' is an extensional disjunction for which Addition holds and OS fails. '+' is an intensional disjunction for which Addition fails and OS holds. 4. Disjunctive Syllogism is certainly a very common form of argument, and therefore an understanding of ' + ' has a large part to play ir. exploring the logic of natural language. From ( I ) Every student studies either Latin or Logic, one can infer that if a student chooses not to read Latin he or she enrols in Logic. Addition clearly fails for such a use. The Lewis argu­ ment, from r A & � AI via r (A or B) &� A I , by Addition, to B, by OS, reveals the unacceptable consequence of combining Addition and OS in one connective. Accepting OS for r A or B-, shows that its sense is r if not-A then B I ; the above regulation is, we saw, tantamount to (2)

lf

any student does not study Latin, he or she studies Logic.

But the truth of (2) is not a necessary condition of every student's studying Latin (or for that matter, Logic), i.e., A does not entail r if not-A then B , . 3 What is more puzzling is whether any uses of or in English exhibit the JS, voi. J , no. 3

277

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The trouble with this, shall we say Fregean, view is that it seems not to per mit radical disagreements over the conception of logical consequence, just as its analogue for names does not permit considera­ tion of radically different histories. This thought may lead us to claim that the connectives are rigid designators, or at least can be used rigidly. z That is, that whatever one supposes true o f the connectives, it is the connective3 themselves which one supposes it of. Just as one may suppose that N ixon was innocent of Watergate, and supposes this of Nixon himself, not any counterpart of him, so too one can suppose that ' v ' and ·� ' do not satisfy Disjunctive Syllogism, and suppose this of ' v ' and ·� ' themselves, not of any counterparts of them, simply by stipulation: one stipulates that it is Nixon, and 'v ' and '� ', which are being considered .

STEPHEN READ logical behaviour the role ' v ' has on the conflation less, natural uses

of ' v ' . This may seem a surpnsmg doubt considering played in classical logic. But classical logic has rested of these two different d isjunctions. There are, nonethe­ of ' v '. Here is one:

(3) You qual ify for a grant if either you are over 65 or you earn less than $2000 a year. We have here a disjunctive condition - a condit ional with a disjunctive antecedent; and one can satisfy the condition by satisfying either disjunt - so Addit ion holds for it. Secondly, it is equivalent to

where the conjunctions admits Simplif ication, and so is extensional. Hence this use of or IS extensional. 5. Paul Grice ( 1 978) has aq�ued against there being two such senses, intensional and extensional, of or. However, his arguments depend on two prejudicial assumpt ions about the relation between the senses which do not apply to ' + ' and ' v '. He takes the two putative senses to be: (I) the 'weak ' sense given in ' v ; and (Ii) the 'strong' sense such that I A or B I contains two components, first the assertion of IA v B I , and secondly the assertion of a non-truthfunctional reason for acceptmg IA v B I. '

entails Now it is certainly true (at least in R and E) that I A + B I v B I ; but it does not follow that there is some identif iable claim or condition P such that I A + B I is equivalent to I (A v B) & P I , but not to P alone. (This is worthy of remark since many detractors, and even some proponents �of relevant logic have supposed that IA -+ BI means 1 (A => B) & Q 1 for some condition Q such as 'A and B are mutually relevant'. But relevance is not identif iable (m general) separate­ ly from the existence of an impl ication relation from A to B . If A really does imply B then what greater reason can we have for clai ming that A is relevant to B? IA

Further, the account of the sense of r A + Bl need not make reference to I A v B I . A + B is true j ust when if A is not true then B IS true. (Ct. the classical truth-condition, that IA => B I is true j ust when ei ther A is not true or B is true - using the extensional sense of or!) This truth­ condition is contained implicitly in the claim that DS holds for ' + ' given the analytic condition that Modus Ponens holds for if. 11f A then B-; is the material mode expression of the formal mode claim that B follows from A, 1 if A then, necessarily, B 1 is the material mode expres­ sion of the formal mode claim that A entails B.) Grice 's arguments are two: first , that the putative strong sense has 278

JS, vol. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(4) If you are over 65 you qualify for a grant and if you earn less than $2000 a year you qualify for a grant.

DISJUNCTION restrictet application, whereas the weak sense is universal. This sugges­ tion results from classical blinkering. Certainly the classical Weltan­ schauung has led us to believe that many l!Ses of or a r e e x t ens ional for it has claimed that all are. But I am arguing that the intensional sense is a common one. The weak sense occurs in such restricted area!> as the antecedents of conditionals. Thus Grice's first argument cannot even get started. The same happens with the second argument. It depends on the use of the extra condition P, which i tself involves the weak sense, to characterise the strong sense, urging us that this should lead us to look for a way - a conversational way - of explicating the stcong sense as derivative from the weak. But we have seen that the strong sense does not depend on the weak iri this way.

If this argument were sound, then my claim that I A or B I (also) ex­ hibits an ambiguity b�tween I A + B I and IA v B 1 would be wrong, for, as I said, lA + B ! entails I A v B I . However, it is not sound. J agree with Kempson and Cormack that being true in a var !ety of different circumstances is not a mark of ambiguity. All sentences are true in a vari�ty of c i rcumstances: e.g., Someone is com ing down the road is true whether it is a man or a woman, whether it is Fred or Charlie or Alice; and Fred is com ing down the road is true regardless of whether it is raining or not, and so on. The ability to construct, for any sentence S, further sentences P, Q, R etc. which further charac­ terise different sorts of circumstance in all of which S is true, so that S is equivalent to some disjur.cticn P or Q or R or ... , d o e s not �how that S is ambiguou5. :ndeed, the point this drives home is that ambigu i ty in a sentence is not a mat ter of its having a disjunctive truth-co:-�di tion. (Kempson and Cormack appear to recognise this on "an amb!guous expre:;sion cannot be characterised by a sing!? p. 263 "in the d:sjur.ctive set oi p.:>ss1biEties" - only to forget it on p. 264 face of this :!VIdence [against the assumpt1on c.f no ambigui ty], one �isht <�ttempt an alternat1ve cha-acter is.'ltion of ambiguity by setting -

-

::'5, vc.J. J ,

no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

6. More generally, Ruth Kempson and Annabel Cormack have argued that ambiguity should not be attributed to a sentence unless the two resulting readmgs are logically independent. 4 For if one purported reading entai!s the other, there will be no difference in their predictions about acceptability of the sentence in different conditions between that claim of ambiguity and the simpler claim that the sentence is unambiguous but true in a variety of d ifferent circumstances. Cons1der, e.g., exclusive versus inclusive extens10r.al disjunction: I A '!/. B I and I A v B I . I A 'f. B I entails I A v B I , the former being true in thf: two Circumstances TF and FT, the lat ter in both these and also for truth assignment TT. But I A '!/. B I is not ambiguous in being true if either A 1s true or B is true. So sim llarly, I A or B I should not be a.c­ counted ambiguous in being true either if I A '/. B I is true or if I A v B I is true. For the circumstance TT will not be accounted a case where I A or B I is false, since it is true then m virtue of the truth of I A v B I in that circumstance.

STEPH EN READ up a disjunction of two logical forms".)

•.

Indeed, it is equally plausible to say that S means both P and Q, when S is ambiguous. What we intend can be put in this way: some tokens of S mean P, some mean Q, and, a case which is arguably deriva­ tive, in puns some tokens of S mean both. That is why, in the last paragraph, I could say that S seems in some circumstances to be both true ans false: we can imagine cases where the utterance of S is true, and others where it 1s false, where the circumstances of utterance of S do not dilfer. In such circumstances S is both (in the one case) true, and (in the other case) false. Hence Kempson and Cormack are mistaken to claim that, e.g., r A or B I cannot be false when uttered in the case where A and B are both true. You may take cereal or fro.it juice is a canteen instruc­ tion which does not permit one to take both; as emphasised the following day when the notice reads; You· may take either cereal or fro.it juice. To counter the censure brought about by taking both, by claiming that r A or B I permits both, is as much evidence of misunderstanding of the injunction, as would be trying, when told she has no duck, to defend one 's inference that she is some sort of ornithologist, by pointing out that John sa w her duck contains a reference to a duck of hers. In both cases the sentence so understood is false (in the appropriate circumstances). It is to avoid admitting that a single sentence is both true and false that we resort to invocation of ambiguity. The keys are So too w ith r A + Bl and r A v B I : suppose I say, either in my pocket or in the car, when they are indeed in my pocket

but could easily have been in the kitchen. Then what I say is true 280

JS, vol . l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Let us consider then what it is that leads us to clai m that a word or sentence is ambiguous. One of Kempson and Cormack 's examples · is John sa w her duck. There are clear circumstances in which it is false: when John has his eyes shut, say, and others where 1t is indispu­ tably true: when John saw her flinch and duck when her duck flew over. But there are others where it seems to be both true and false: when she has no duck but ducked in John's view, or when she did not duck, but showed her duck to John. It IS facts like these about the circumstances in which such a sentence is true or false which reveal that it cannot have a single meaning. In particular, they show once again that its meaning cannot be disjunctive: John saw her duck does not mean "Either John saw her ducking or John saw a duck of hers". 5 It is tempting to suppose so: if S is ambiguous between P and Q, we say it means either P or Q. But even an unambiguous sentence S either means S or means T, say: so it signifies nothing to say that S either means P or means Q. While to say it means 1 P or Ql is again true of an unambiguous S: e.g., as we saw, Someone is coming down the road means "Either a man is com ing down the road or a woman is com ing down the road"

DISJUNCTION since it follows by Addition from the true sentence, The keys are in my pocket; but it is also false since it entails (as in S4 above) the false sentence, If the keys are not in my pocket they are in the car. To avoid

recognising a single sentence to be both true and false, we can follow standard practice in such a case, and claim that it is ambiguous, and fairly naturally, locate the origin of the ambiguity, not syntactically, but seman t ically. in the word or. There is an extensional sense of or, for which the truth of either disjunct suffices, and there is an intensional sense of or, which i mports an inferential connection between the disjuncts.

Jackson's plan is to defend what he calls the Equivalence Thesis, that I if A then B I is equivalent to I A :=� B I - w h i c h he expresses as the equivalence of r if A then B l and r not-A or B I , since he takes it as uncontroversial that or is extensional. (One aim of this paper is to make it clear that this is controversial.) But he wishes also to explain how this truth-condition leads to a different assertion-condition: that r if A then B I is assertible when Pr(B/ A) is high. The reason, he claims, is the existence of a conventional implicature in the assertion of I if A then B I that it is robust wo its antecedent (though not wo , its consequent - he attributes that robustness to the assertion of I A only if B I ). This robustness ensures the utility of the conditional in Modus Ponens inference. Hence, given that I if A then B l i s e q u i v a l e n t t o - has the same truth-condition a s - I A :=� B I , and Pr(A:=� B ) is high, and l if A then B l is robust wo A - so that Pr(A :=� B/A), b e i n g c l ose to Pr(A :=� B), is high too, then Pr(B/ A) Pr(A :=� B/ A) is high, and so I if A then Bl is (highly) assertible. =

8. I do not wish to characterise the worth of inference in terms of preservation of assertibility. This is what Jackson finds he has to do when he claims that if A then B, A 1- B is valid, but possibly useless, if, e.g. , the separate high probabilities ()f the premises do not combine into a jointly high probability. On learning that A , one 's belief in I if A then B I , given the Equivalence Thesis, may suffer, if, e.g., it was based on a belief that I not-Al was true. The separation of valid­ ity and utility is required by that of truth and assertibility. For if truth is distinguished from assertibility then, in the presence of the Deduction Theorem , we have to distinguish some new notion crresponding to validity as assertibility does to truth. Suppose P and Q are assertible, and P, Q 1- R is valid; then so is P 1- if Q then R (given a proper un­ derstanding of. P , Q 1- R). Now if 1- transmitted assertibility, r if Q then R I would be assertible, bu t , by Jackson's lights, it may not be, for it JS, vol. l , no.3

28i

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

7. Frank Jackson ( 1 980: 569-570) calls a sentence S robust with respect to new information I if the invocation of I does not sig nificantly reduce the probability of S. So, e.g., a negated conjunction r� (A &: B) I asserted on the belief that A is false is not robust wo (with respect to) A, for its high subjective probability collapses when it is learned that A is true.

STEPHEN READ can fail to be robust wo Q. Thus we can have P assertible and I If Q then R I not: hence 1- cannot transmit assertibility. What Jackson f inds he needs is a new notion which does transmit assertibil 1ty, viz. util1ty. What is needed to ensure the utility of P 1- if Q then R is the robustness of I if Q then R I wo Q.

However, by bringing the notion of robustness back to the language of truth-preservation, we can use it to understand the way or does in fact work, and so understand the difference between ' +' and 'v ' . 9. Whenever P 1- Q, belief m P can· be a ground for belief in Q. So let us say that Q is robust wo P if Q cannot be grounded on I -Pl 1.e., if P If Q. Then, since - A 1-- (A &. B) , -(A&B) is not robust wo A - let us say 1t is feeble wo A. This accords with Jackson 's example g1ven above - though 1t will not accord fully of course, for robustness as now characterised is, e.g., no longer a context-dependent not1on. -

Since - A 1- A ::> B, belief in I A ::> B I can be grounded on I-A l , and since B 1- A ::> B, similarly on B. So I A ::> B I is not robust wo A nor wo I- B I . But in general - A !I A -+ B. So only for those few mstances of A, B such that -A 1- A -+ B, e.g., B = Ax - A , does r A -+ B I fail to be robust wo A. In general, I A -+ Bl is robust wo A, and similarly wo I - B I . We can then say that a conditional is robust (tout court) if it is robust both wo its antecedent and wo the contradictory of its consequent. A feeble conditional, e.g. l A ::> Bl , 1s one belief in which can be 282

JS, vot. l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It is prima facie attractive to suppose that assertibility shoud be distinguished from truth - when, e.g., one reflects that one may wish to assert something which one thinks at least 9096 probable. But to assert P is to assert that P is true, not that P is probable - that is done by asserting that P is probable, i.e., by asserting that it is true that P is probable. Sim ilarly, to infer Q from P is to infer the truth of Q from that of P. Now of course some have argued that one must replace the notion of truth by one of justified or warranted assertibility. But that is a different matter from what Jackson (and others) are suggesting; the two notions are not then competing side by side. What happens when we admit both truth and asser tibll1ty - such that both what is assertible may not in fact be true, and what is true may not be asser­ tible - is that truth drops into the background, and no longer plays any useful role. What is the pomt in saying that an argument is valid i f one cannot reason in accord with it, or in saying that a sentence is true if one cannot correctly assert it? To continue to claim that I if A then B I is true even though the conventional impl1cature of its assertion - that it is robust wo A - forbids one from asserting it, 1s simply misguidedly to cling to truth-functional logic in the face of evidence from the way we use disjunctions and conditionals that they are not truth-functional.

DISJUNCTION grounded on the supposed falsity of its antecedent, or truth of its consequent, alone. Similarly, a robust disjunction, e.g., '+' , is one which is robust wo the contradictory of either disjunct. A feeble disjunction, e.g., • v ' , is one grounded on the truth of one or other disjunct. This was precisely the case with the disjunctive condition for the award of a grant. If someone has qualified for a grant, to learn that he is, say, under 65, does not warrant the inference that his income must be under $2000. His failure to satisfy one half of the disjunctive condi­ tion can just as well undercut his claim to a grant. For the disjunction here is too feeble to support the inference.

But that is not the point. The point is that in r A v B I we have lost information. Suppose we are given that r P & Q I is true. We simplify this, for some reason, to P. Then no one would suggest that from P we could infer Q. The ground for P indeed also supports Q: r P & Q len­ tails Q. But what was inferred from r P & Ql , namely P, does not. Similarly here: if rA v B I was inferred from B, then indeed the ground for r A v B I supports B. But the claim based on that ground, viz. r A v B I , does not. For if r A v B I w a s grounded instead on A then nothing in the ground for r A v B I supports B, and r A v B I can be grounded on eithe r. What we do when we adduce r �A I is to try to rule out the possibili­ ty that the ground for r A v B I was A: to force its ground to have been B, which of course easily supports B. But we cannot rule out the possibility of an inconsistent situation in this way. For the premises may simply amount to r A & � A I , from which B cannot be inferred. ( Remember that we are here in a paraconsistent situation, one in which A & � A II B .) we do not necessarily have r � A I , r A v B I as logical tn.J.ths, but possibly as assumptions in the course of a proof (a reductio for example) or as the consequences of some possibly inconsistent theory (naive set theory, or ZF, for example). I I . It is interesting to end by treating formally the way the inconsistent situation cannot be ruled out. For, one m ight start to counter, JS, vol. l , no.3

283

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

1 0 . So what exactly goes wrong when we feel tempted to infer that the grant-holder, being under 65, must have an income under $2000? Simply to say that the new information undercuts the claim to a grant suggests, wrongly, that it is not the validity of the inference which is being challenged, but rather its soundness. That is, to say that we are not entitled to infer B from r � A I and r A v B I because r A v B I is feeble, suggests that the addition of r� A I shows, not that B does not follow from r� AI and r A v B l , but that rA v Bl may be false, and so the inference, though valid, may not have true premises, whence there naturally is no guarantee that its conclusion is true.

STEPHEN READ do we not have the law of Non-Contradiction - or at least a special case of it, viz. 1- � (A & � A)? However, the argument then launches on a regress, never terminating in the simple conclusion B: A

v

B, � A 1- B

v

(A & � A).

(by Distribution)

I- � (A & ..vA), v

(A & � A), � ( A & � A)

A

v

B, � A 1- B

I-

v

1-

B v ((A & � A) &� ( A & � A)) (Distribution again)

((A & � A) & � (A & � A)).

� ((A & � A) & � (A & � A)),

ano so on. Let us add to the language two symbols t and f , to represent th� truth-values t and f , or equivalently, the (extensional) conJunction of all truths and the d tsjunction of all falsehoods. 6 We can specify as <m axiom A

-++

(t

-+

A)

and as a definition, f =df"-' t . Then we can show that � A, A v B 1- B v f. By the axiom for t, we have (t -+ t) -+ t, and so 1- t. So can we not now argue from t ( �t ) and I B v f I to B? No, for again we have only � · t , B v f 1- B v f. Indeed, if we do admit the inference from� f a nd ; B v f I to B, then we can show that, for any A, A & �A 1- B. =

Hence, within the context of relevant logic, and its rejection oi Consequentia Horribilis, A, �A 1- B, we must distinguish two disjt•nc­ tions, intensional and extensional, both of which are used in everyday reasoning. u niversity of St. Andrews

284

JS, vn! . l , n0.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

So

B

DISJUNCTION Notes

1 Anderson & Belnap ( 1 97 5) § 27. 1 .4. 2 See McGinn ( 1 982: 1 03- 1 04). McGinn restricts his comments to truth-functions, but this is clearly unnecessary. 3 This argument is developed in Read ( 1 98 1 ). 4 See (iv) in Kempson & Cormack ( 1 9 8 1 : 265). ' 5 See Priest & Read ( 1 9 8 1 ) , § V . 6 See Anderson & Belnap ( 1 97 5) S 27 . 1 .2, and Meyer ( 1 978), esp. S 3.

f

Anderon, A.R. and Belnap, N.D., 1 975: Entailment, Vol. ! . Princeton Uni­ versity Press, Princeton. Cole, P. (ed.), 1 978: Syntax and Semantics, 9: Pragmatics. Academ ic Press, New York. Grice, H.P., 1 978: Further notes on logic and conversation. In: Cole, P. (ed.); 1 1 3- 1 27. Jackson, F .c., 1 980: On assertion and indicative conditionals. Philosophi­ cal Review 88; 565-589. Kempson, R. : and Cormack, A., 1 98 1 : A mbiguity and quantificatio.-1. Linguistics and Philosophy 4; 259-309. McGinn, C., 1 982: Rigid designation and semantic value. The Philosophi­ cal Quarterly 32; 97- 1 1 5. Meyer, R.K., 1 978: Sentential Constants in R. A u s t r a l i a n National Uni­ versity Logic Group Research Paper, 2. Canberra, A.C.T. Priest, G . and Read, S . , 1 98 1 : Ockham's rejection of ampliation. Mind 90; 274-279. E n g l e w oo d Quine, W . V .O., 1 970: Philosophy of Logic. P r e n t ice H a l l , Cliffs, N . J . Read, S . , 1 98 1 : What is wrong with disjunctive syllogism? Analysis 4 1 : 6670.

JS, vol . l , no.3

285

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

References

8ook review

T. Ballmer &. W. Brennenstuhl, Speech

Act Classification. A the lexical analysis of English speech activity verbs.

Series in Language and Communication, Vol.&. Verlag, Berlin-Heidelberg-New York, 1 98 1 . X + Price: US $ 33.60. ISBN 3-.540- 1 0294-9.

study in

Springer Springer 274 pp.

Reviewed by: Jens Allwood

The posthumous publication of J.L. Austm's How to do things with words in 1 96 2 resulted in the creation of a new field of enquiry - which Austin sometimes referred to as linguistic phenomenology, i.e. the investigation of everyday phenomena through their codification in language. In How to do things with words Austin combined the idea of a linguistic phenomenology with the idea of speech as action. This combi­ nation has subsequently become extremely influential, especially through the publication of John Searle's contribution to the Austinian ideas in Speech Acts ( 1 969). Speech Acts have become popular not only in philosophy but also in linguistics (Sadock 1 974), literary criticism (Ohmann 1 970), anthropology (H ymes 1 974), sociology (Silverman 1 976) and psychology (Clark and Clark 1 977). In linguistics, one of the first applications of Austin's thinking came through the so-called performative hypothesis, put forth within the bounds of what was then known as generative semantics by linguists such as Ross ( 1 970) and Sadock ( 1 974). The main idea of the perform­ ative hypothesis was that .an abstract conceptual entity corresponding to a performative verb (a verb that could be used in the hereby formula e.g. I hereby baptize you) was an underlying component of the semantic structure of every sentence. During the mid-seventies, however, a different approach to the phenomena Austin had pointed out, was developed. In All wood ( 1 976) and Verschueren ( 1 979) it was claimed that the so-called 'speech act theory ' , the term used by Searle for the study of speech as action, primarily was a study of the semantic fields of terms designating different aspects of speech and that the most important rules and conventions involved m speech act theory were in fact lexical conven­ tions for the use of different lexical items denoting speech. In this approach, the analysis of speech activity terms is primarily a contribu­ tion to lexicography. The analysis can, however, secondarily give us important hints about what has proved important enough in speaking for the members of a certain speech community to codify in language. JOURNAL OF SEMANTICS, Vol. ! , no. 3, pp. 287-290

287

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

University of Gi:iteborg Department of Linguistics

JENS ALL WOOD This in turn can give us hints about what the important aspects of communication are.

In this way, the last two thirds of the book serve as a kind of thesaur­ us for the verbs of speaking which constitutes a part of a larger thesaur­ us of verbs and vocabulary in general. Part l of the book contains a theoretical discussion of lexicography (the practical and empirical study of lexical items and phrases) and lex icology (the theory of lexical phenomena). According to Ballmer and Brennenstuhl there have been too many studies that are either lexicographic but not lexicological or lexicological but not lex icographic . . They therefore try to steer a middle course, avoiding neither theory nor a sufficiently large amount of data. It would seem, on the whole, that they have been successful in doing this. The theoretical point of departure for the authors was the hope of testing Austin's and Searle ' s proposals for a taxonomy of speech acts on German verbs. In carrying out this test they found that they had to reject the previous proposals in favour of a proposal of their own. This change was partly caused by a consideration of a much larger body of verbs than had earlier been the case. The authors speak of their approach as being more holistic than the previous approaches, i.e. departing from a large number of verbs at the same time rather than departing from an analysis of one verb at a time. The authors' work on speech acitivity verbs is, in fact, conceived of as a part of a larger study of German verbs and ultimately of the whole German vocabulary. This means that not only speech acitivity verbs in general have been taken into consideration when proposing the taxonomy and classification of speech act verbs. The method of analysis described by the authors is the following: F irst a list of verbs denoting speech activities was collected from various dictionaries, then this list was divided in two stages into sub­ groups on the basis of similarity in meaning. There are 600 subgroups of this type. Each subgroup is called a category and is labelled by a verb 288

JS, vol . l , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The work by Ballmer and Brennenstuhl very definitely also takes a lex icological/lexicographic approach, but differs from that of the two preceding authors by taking a wider range of verbs (4800) into account. The book is d ivided into three roughly equal parts. I. A theoretical discussion of the lexicological and lexicographic approaches taken in the book. 2. A classification of verbs designating speech activity based on relations of synony my, entailment and presupposition yielding a taxonomy of the larger semantic field of speech activity verbs into smaller more specified subfields. 3. A list of the verbs in alphabetical order with indications about where to find them in the taxonomy given in part 2.

BOOK REVIEW

intensity

)

categones of a model

)

The various stages of this intensity curve is reflected in a great many of the models. Thus, in the struggle model we find the following stages marked by the occurrence of verbs ( l ) starting situation, (2) marking claims, dissent, (3) attack, (4) tactical phases, ( 5) making coalitions and finally, (6) victory, defeat and cooperation. The authors claim that this analysis of ' model hat structure' should enable us to throw light on problems of aspect and 'Aktionsarten ' . Thus, instead of counting only the inchoative, durative and terminative as Aktionsarten, we should also count the 'preparative ', the 'postparative ' , the 'intensifying', the 'cooling off', the 'preservative ', the 'laissez faire' and the 'destruc­ tive ' . Finally, I would l ike t o mention that Ballmer and Brennenstuhl in their book champion their own theory of meaning, i.e. what they call cogmtive behaviourism, an approach to meaning which is closely linked to Thomas Ballmer's theory of context change, Ballmer ( 1 979). The approach is an attempt to create a theory of meaning which will synthesize several different perspectives on the theory of meaning, doing justice to the natural, behavioural and humanitarian sciences. Although one can criticize Ballmer 's and Brennenstuhl's book for perhaps overestimating the role of linguistic categories for thinking and conceptualization, the f inal judgement is that they have produced an extremely interesting book which will surely have a place among the pioneering works of lexicography and lexicology. The book should be inspiring reading for all those who are concerned with both a descrip­ tive and an explanatory semantic approach to the lexicon.

JS, voi. J , no.3

289

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

which is typical of the group, e.g. verbs like accept, admit, and assent belong to the category of accepting while verbs like reject, decline, dis­ agree and dismiss belong to the category of rejecting. Using the notion of presupposition the categories are then grouped into 24 models and 8 modelgroups. The modelgroups are: emotion, enaction, struggle, institutional, valuation, discourse, text and theme. In this way, model­ groups in some cases cont9-in several models. For example, the valuation modelgroup contains the following models: valuation of an action, valuation of a person, valuation of oneself, valuation of objects. Another feature of the lexicological approach of Ballmer and Brennenstuhl is that many of the models can be given what the authors call a 'hat structure'. Intensity is low at the beginning of an action or process: then it grows in order to f inally decline at the end of any process.

JENS ALL WOOD References

·

290

JS, vol . 1 , no.3

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Allwood, J . , 1 976: L inguistic Communication as A ction and Cooperation, Gothenburg Monographs in Linguistics 2, Dept. of Linguistics, University of Goteborg. Austin, J.L., 1 962: How to do things with words, Oxford University Press, Oxford. Ballmer, T., 1 979: Context change, truth and competence. In: Bauerle, Egli &. von Stechow (eds.), Semantics from different points of view, Springer, Berlin-Heidelberg-New York, Pp. 2 1 -3 1 . Clark, H . &. Clark, E., 1 97 7 : Psychology and Language, H a rcourt B race Jovanovich (international edition), New York. H ymes, D., 1 97/f: Foundation in Sociolinguistics - an e thnographic ap­ proach, Tavistock publications, London. Ohmann, R., 1 970: Speech acts and the definition of literature. Philosophy and Rhetoric, University Park, Pennsylvania, 4. 1 ; 1 - 1 9 . Ross, J.R., 1 970: O n declarative sentences. In: R.A. Jacobs &. P.S. Rosen­ baum (eds.), Readings in English Transforma tional Gra m mar, Waltham, Mass. Ginn. Pp. 222-272. Sadock, J., 1 97/f: Toward a Linguistic Theory of Speech A cts, Academic Press, New York. Searle, J.R., 1 %9: Speech Acts, Cambridge University Press, Cambridge. Silverman, D., 1 976: The Text of the Master - From Locke to Austin. Unpubl. ms. Goldsmith's College, London. Verschueren, J., 1 9 79: What People say they do with Words. University of California Press, Ph.D. dissertation.

Announce ments

In order to enhance the value of the Journal, the Editors have decided to have occasiOnal issues centred around a particular topic. Thus, issue No. 4 of the first volume, scheduled for December 1 982, will contain var1ous contributions on problems of anaphora. Another thematic issue is planned for 1 983, focussing on problems of meaning and intona­ tion. Suggestions for more thematic issues are most welcome. The editorial tasks may be delegated to one person or a group of persons not necessarily already associated with the Journal. Plans are maturing also for a second Cleves Colloquium, organised by the N . I.S. Foundation. The colloquium will be on the lexicon, its relations with grammar, logic, memory, and its impl ications for truth-theories. The planned time is: August 30 - September 2, 1 983. And given our very positive experiences with Hotel Braam in Cleves (just acros:. the German border from Nijmegen) during the Discourse Colloquit.m in 1 98 1 , we are thinking of holding our Lexicon Colloquium at the same location. (By that time, the hotel will have a new and larger conference room.) Contributions, or prom 1ses of contributions, are invited. As from

Voi.2,-No. l , both the format and the type-face of the

Journal will be larger, thus contributing to improved appearance and

legibility.

The Editcr:;

JS, vo! . l , no.3

29 i

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The Journal of Semantics has attracted, in its first half year of exis:­ ence, widespread mterest in many parts of the world. If the trend:; wiil no that have manifested themselves so far persevere, the Journal doubt prosper.

EDITOR'S PREFACE

After transformational linguistics had its peak of pronoun-research !n the late Sixties and fundamental problems had cropped up for the notion that pronouns should be seen as derived from copies of their antecedents in deep structure, it took a good while before, almost in the late Seventies, pronouns again became a central topic of research.

But what makes anaphora such an interesting phenomenon in this new situation? If we take the notion of anaphora in a wide sense, perhaps in the sense that an anaphorically functioning expression is one that is understood in extension of or on the basis of what has already been understood, then anaphora very clearl y comes to occupy a position in which the interaction between an utterance and preceding discourse or background knowledge must become particularly obvious. And here we seem to be in the centre of one of the currently most exciting areas of language research. The turn from sentences to texts or discourse, the loosening up of the traditional structuralist dogma that (knowledge of) language should be kept separate from (knowledge of) the world, and the increased interest in linguistic processes rather than structures would seem important factors in the rise of this new area of interest. The six articles in this issue, none of them previously published, reflect a number of different tendencies currently present in anaphora research and emphasize different aspects of what many now think may well be one more or less unified phenomenon. Liliane Tasmowski-De Ryck and Paul Yerluyten represent what looks on the face of it like a conventional linguistic a"pproach: they argue against too wide a scope for 'pragmatic control' of pronouns and empha­ size the role of linguistic control. But what they call ' linguistic control'

29�

JOU RNAL OF SEMANTICS, vol. l , no.�

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

But meanwhile the scene had changed considerably. It had become much harder to insist that linguistics was concerned, in the first instance, with the unit of the sentence, rather than with texts or discourse. It had become much harder to maintain a strict division between know­ ledge of the language and knowledge of fact and corresponding compart­ ments of the human m ind. And the success of procedural approaches in psycholinguistics and artificial intelligence had made it much more difficult to propose models of description and explanation that ignore the process character of natura! language. - Of course, this is not to say that one could not continue to treat pronouns in some more conventional linguistic fashion or that we had reached the definite end of the fruitful period of a paradigm.

turns out to be a more general relation than conventional notions of control. And since they allow also for the controller itself to be 'prag­ matically controlled ' by a non-linguistic object, the proposed control mechanism stands a good chance of bringing about a great deal of unification in the notion of pronoun control, at least for the large class of pronouns Tasmowski-De Ryck and Verluyten call 'true pronouns'.

Barry R ichards discusses a problem typ1cal of the transition from a sentence-oriented approach to a discourse-oriented theory of pronouns: the interpretation of definite singular pronouns with indefmite singular antecedents. He demonstrates that the two classic ways of modellmg this relation, viz. 'binding' of the pronoun by its antecedent and corefer­ entiality of pronoun and antecedent, suffer from serious shortcommgs. The solution Richards presents is based on the interesting proposal that certain pronoun occurrences may profitably be regarded as function­ ing simultaneously as de1ctic expressions and as anaphors. Han Reichgelt links up to the problem discussed in the article by Richards and puts forward an alternative solutiOn that rests on what may be called an interactiOnal view of the function of linguistic utter­ ances. The question here is not any more in abstract 'how does this linguistic form function?' but rather 'what IS the function of this linguistic form with respect to what the speaker may assume the listener to know?'

Whereas in the papers mentioned so far pronouns have figured as Jack the central cases of anaphors, the article by Jaakko Hintikka and Kulas focusses on the anaphoric use of the definite article and provides a treatment in terms of Hintikka 's game-theoretical semantics, which may, in certain respects, be regarded as an essential vindication of Russell 's Theory of Oeser iptions, a! though, of course, the Russellian theory had nothing to compare with the discourse orientation and the dynamics of the game-theoretical approach. The only experimental study in th1s issue, the paper by Lorraine Komi­ sarjevsky Tyler and William Marslen-Wilson, focusses on the role of pro­ cesses of inference as they are needed in the interpretation of anaphors, and comes up with the striking result that on-line processing of utter­ ances is not automatically slowed down by those inferences. The signifiJS, vol. l , no.4

295

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The article by George Yule takes a different point of view. Yule questions the notion that the interpretation of pronouns is determined by (the referents of) their antecedents and discusses cases where pro­ nouns, though (in some sense) referentially used, need not be assigned a referent in order to be interpreted. A w1der perspective of the func­ tioning of pronouns in discourse would be needed in order to accommo­ date cases like these, which are new in the anaphora discussion and must be seen as a challenge to just about all theories about pronouns currently on the market.

cance of this for a theory of anaphora would seem to be that, as far as actual processes of understa[lding linguistic utterances are con­ cerned, notions like the plausibility or the pragmatic likelihood of an interpretation (which are inference-based notions) may actually play a far more central role than has so far been assumed in philosophi­ cal and linguistic models of interpretation, or could be assumed as long as we had to think of inferences as being very costly in terms of processing time.

Peter Bosch

296

JS, vol. I , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This short survey of the six articles probably emphasizes the diversity of approaches and methods we mentioned at the beginning. And then only very few of the many facets of anaphora are treated in this small collection of papers. The encouraging side of this is, and this would seem typical for the recent scene of anaphora research, that there is communication, interaction, and mutual stimulation between these various approaches. Probably none of the authors in this issue would regard the work of the others irrelevant to his or her own enter­ prise.

PROCESS ING UTTERANCES IN DISCOURSE ON-LINE RESOLUTION OF ANAPHORS

CONTEXTS:

Lorraine Kom isarjevsky Tyler and Will iam Marslen-Wilson

Abstract

Some recent research on spoken language understanding suggests that the processing of the speech input is, at all ttmes, conducted with reference to the listener ' s mental representatton of the discourse context in which it occurs. From the first word of an utterance, the lt stener is assumed to integrate together constraints derived from the specific dtscourse context, and from his general knowledge of the world, with his analysis of the linguistic properties of the utterance itsel f. The representation which results from the integration of these different sources of information provides a framework with respect to which subsequent words are analysed. A major source of evidence for this claim comes from the results of word monitoring experiments ( Marslen-Wi lson and Tyler, 197 5, 1 980) in whtch subjects listened to sentences containing target words specified in advance, and pressed a response button when they identified the critical word. When an utterance was preceded by a dtscourse context, monitoring reaction times to target words were facilitated right from the beginning of the utterances, compared to reaction times for words occurring in isolated utterances. Given that this facilitation occurred so early in an utterance, we argued that it could JOURNAL


297

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The on-line interpretation of utterances in discourse contexts was investigated by varying the type of dependency between an utterance and its context. Listeners heard short sequences of utterances ending in incomplete fragments. The fragments varied in length and in whether their anaphoric linkage to the context (by repeated names, pronouns or zero anaphors) required inferences to be resolved. The subject 's task was to name a visual continuation probe tha t appeared at the offset of the fragment. The differences between naming latencies to appropriate versus inappropriate probes was constant across condi­ tions, and irrespective of whether or not inference-based processes were required to determine this preference. This was interpreted as showing that on-line speech processing is not necessarily slowed down by the use of inference to link utterances to their contexts.

TYLER & M ARSLEN-WILSON only have derived from the prior discourse which provided a framework withm which the incoming speech could be rapidly interpreted.

We will use different kinds of anaphoric connections between utter­ ances and the1r discourse context for two main purposes. First, to obtain direct evidence that knowledge which could only have been denved from the discourse context is actively available to listeners during the on-line processing of an utterance. Second, to determme whether there are general restrictions on the types of relatiOnship between an utterance and its context that can be readily exploited durmg on-line processmg; in particular, whether relatiOnships that can only be based on inference are necessarily more ti me-consuming for the listener. Consider, for example, the followmg sequence: ( I ) As Phil1p was walking back from the shop, he saw an old woman trip and fall flat on her face. (2) She seemed unable to get up again. (3) a. Ph1lip ran towards .... . b. He ran towards .... . c. Running towards Th is sequence cons1sts of two context sentences, ( 1 ) and (2), followed by three different continuation fragments, (3a-c). Each of these frag­ ments contains an anaphoric device linking the fragment to the preceding discourse. In (3a) the device is simply the repit ition of the name of an antecedent individual 1; in ()b) it is an unambiguous personal pronoun; while in (3c), an example of ' zero' anaphora, there are no explicit lexical cues at all. It is clear that rather different procedures can be invoked to cope with these three cases. In each case, to interpret the fragment, it is first of all necessary to determine who is the agent of the action denoted by the verb, and to evaluate this with respect to the prior discourse. In (3a) and (Jb) the agent is lexically specified w ithin the fragment (Philip, He), and can be unambiguously related to possible antecedents in the discourse context just on the basis of this lexical 298

JS, voi. J , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The purpose of the present research is to clarify some of the proper­ ties of this connectiOn between the input and the discourse model, by examming the processing . of certain forms of anaphora. We choose this approach to the general problem because of the special properties of anaphors relative to other less explicitly anaphoric lexical items. Anaphor1c devices (such as pronouns and definite noun phrases) can be clearly located m the local and global structure of a text, so that one can· determine qUJte specifically what the listener has to be able to do in order to make a successful link with the appropriate antecedent in the discourse.

PROCESSING UTTERANCES information. That is, in (3a) there is a repetition of the unique name

Philip in the fragment, and in (3b) the pronoun He unambiguously specifies

Philip as the only male, singular, potential referent. However, (3c) presents a quite different case. The only way agency can be assigned is on the basis of some inference process which relates the properties of the verb phrase, RW111 ing towards , to the properties of the potential antecedents in the discourse model. It is necessary to infer, given the scenario set up by the preceding context sentence, who is most likely to be running towards whom. •••

The critical experimental question is, first, whether on-line prefer­ ences can be obtained at all, and second, whether the size of such a preference effect d iffers across the three types of fragment. All three fragments are functionally equivalent in that they contain enough information so that, given the context, the listener can eventually assign the correct agent for the verb. The question is whether and, in particaular, when the listener can exploit this functional equivalence during real-time speech processing. Both the AI and the cognitive psychology literature on anaphor resolution generally regard processes based on inference as being more complex and more time-consuming. In the AI literature, for example, the assumption is made that inference should be kept out of the resolu­ tion process until as late as possible because it is expensive in time and resources, and is liable to combinatorial explosions unless the inferencing search space is highly constrained (e.g., Sidner, 1 979; Charniak, 1 978). It was considerations of this sort that motivated, for example, Sidner 's ( 1 979) development of a computational model of anaphor resolution in which the inference mechanism is normally only used to confirm the suitability of a proposed antecedent after it has been put forward as a candidate by other components of the system. Claims that inferencing requires extra processing time have also been widely made in the psychological literature (e.g., Clark and Havi­ land, 1 977; Just and Carpenter, 1 978; Ehrlich, 1 980). Clark and Haviland JS, vol. l , no.4

299

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Adopting a paradigm we used in an earlier experiment (Tyler and Marslen-Wilson, 1 977), we can probe at the offset of each type of fragment (3a, b, c) with a probe word that is either an appropriate or inappropriate continuation given the context. Thus, for each of the three fragments above, her would be a more appropriate continuation than him. The crucial point, for each case, is that for the listener to prefer one continuation to the other, he must be interpreting the frag­ ment, and the anaphoric device it contains, with respect to the discourse context. If the fragment were initially processed without immediate reference to its context, then the listener would have no basis for such a preference, and both him and her would be equally preferred con­ tinuations.

TYLER & MA RSLEN-WILSON ( 1 977), for example, have claimed that when a reader encounters an anaphor, he looks for an explicit a11tecedent in his memory representa­ tion of the prior discourse. If he cannot locate one, he makes "bridging inferences" in order to establish an appropriate discourse linkage. These bridging inferences seem to make the comprehension process more complex, which is reflected in longer comprehension times in a reading task.

It is, then, not clear what the implications are of inference-induced increases in overall response-time to written materials, for the on­ line resolution of inference-based anaphoric linkages in spoken discourse. Nor, unfortunately, is this ambiguity resolved by the few genuinely on-line studies (as defined here) which have been done in this area. These are the experiments of Carpenter and Just ( 1 977; Just and Carpen­ ter, 1 978) in which readers' eye-movements were tracked as they read texts containing anaphors that required more or less direct infer­ ential linkage processes. The results show clear on-line effects of inference, with gaze duration and number of regressive eye-fixations (looking back to potential antecedents earlier in the text) tending to increase for the less direct anaphoric linkages. However, these effects were not distributed over the texts in such a way as to make it possible to specifically infer that the linkage process itself was slowed down relative to the control cases - which is the critical question for the present study . The increases in gaze duration and regressive fixations were observed not only as the critical anaphoric item was being read, but also over the rest of the sentence containing the item. This is consistent with a general increase in process­ ing difficulty for sentences containing less direct anaphoric linkages, but it is not clear just which aspects of the discourse integration process are being affected. ·

300

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Such results, however, can only be indirectly related to the present study. There are two main reasons for tl,is. First, all of the relevant studies in the area have used written rather than spoken texts. Second, these studies have not, with very few exceptions, used 'on-line' tasks of the type that are proposed here; that 1s, tasks which probe the subject's processing of a given input during the period that the relevant parts of the input are actually being heard or read. Most commonly, reading- or comprehension-time tasks are used, which measure the total amount of time it takes to read and/or comprehend a complete sentence. Such tasks are well suited to certain kinds of experimental questions, but they cannot directly mform us about the precise time­ course of different processing decisions as the material is being input. Sim ilar limitations apply to experiments where subjects are asked to make reference assignment judgements to anaphors (e.g., Caramazza, Grober, Garvey, and Yates, 1 977; Ehrlich, 1 980; H irst and Brill, 1 980). Here again the subjects see the entire sentence before responding, so that their responses cannot tell us directly what is happening when the anaphoric device itself is being read.

PROCESSING UTTERANCES The available research on spoken language processing also fails to provide a clear prediction for the contrast between anaphoric linkages which either do or do not require inference. There are, however, some general observations which imply that the use of inference need not, and often does not, slow down on-line comprehension processes.

Second, there is some indirect evidence from earlier experiments (e.g., Marslen-Wilson and Tyler, 1 980; Tyler and Marslen-Wilson, 1 977) that inferencing is part of the normal process of on-line speech under­ standing. For example, in the monitoring experiment mentioned earlier (Marslen-Wilson and Tyler, 1 980), words occurring right at the beginning of an utterance were identified faster when that utterance was heard in a discourse context compared to when it was heard in isolation. Take, for example, the facilitation of the sentence-initial sequence Some thieves , heard following the context sentence The church was broken into last night. It is difficult to see how recognition of the word thieves could be facilitated here without some form of inference process that determined the relevance of thieves to the situation set up in the previous sentence. Sim iliar arguments could be made for other research which has shown discourse effects on word recognition processes (e.g., Cole and Jakimik, 1 9 80). ••.

Putting both of these claims together, the on-line interactive approach both allows for a partially specified input, and for the possibility that the ' missing' information can be reconstructed on-line on the basis of inference. This provides a processing framework in which the resolu­ tion of zero anaphors on the basis of inference processes need not be more costly or unnatural for the listener than the resolution of more fully specified anaphoric devices. JS, vol. l , no.4

30 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

F irst, the on-line interactive approach to speech processing (Marslen­ Wilson and Tyler, 1 980, 1 9 8 1 ) assumes a cooperative integration of different sources of processing information with respect to the primary perceptual goal of interpreting an utterance in its discourse context. This means that the appropriate analysis of the input need not be fully specified with respect to each and every processing knowledge source; to the extent that the context for interpretation is itself more richly specified, then' a less fully specified speech input can be tolerated. For example, in the word-recognition domain, less acoustic-phonetic information needs to be provided by the speaker for the correct identifi­ cation of the word when an appropriately constraining context is avail­ able. Similarly, in the case of anaphor resolution, the lexical specificity of the anaphoric device can (and does) decrease as the constraining context increases (Marslen-Wilson, Levy, and Tyler, 1 98 1 ). In the present experiment, the discourse contexts are constructed in such a way that anaphor resolution is possible even when the gradient of lexical specificity of' the anaphoric device falls to zero. In other words, the processing approach we are proposing is designed to allow for the on-line exploitation of 'redundancy ' in the relationship between an utterance and its context.

TYLER .tx. M ARSLEN-WILSON This brief discussion of the relevant research leaves us, then, with a range of possible outcomes to the experiment. At one pole, applying the readmg data directly to the speech situation, one could predict condition ( 3c) to behave quite differently from the pronoun or repeated name cases ( 3a, b), with there being little or no difference between the appropriate and inappropnate probes in the zero anaphor condition (3c). At the opposite pole, adopting a strong interpretation of the on-line interactive approach, one could predict all three conditions to behave in the same way.

A final variable involved the discourse structure of the sequence of stimuli. For each set, we manipulated whether or not the antecedent was foregrounded (cf., Chafe, 1 976) in a sentence which intervened between the first context sentence ( I ) and the fragment (3a, b or c). An example of an intervenin g sentence which foregrounds the ante­ cedent of the anaphor (i.e., Philip) would be: (2) a. He only hesitated for a moment. In contrast, sentence (2) above (She seemed Ullable to get up again) fore­ grounds the old woman rather than Philip. The reason for including this variable was, first , to keep control of the discourse structure of the materials, and, second, to find out whether there was an interaction between whether a referent is foregrounded and the ease with which it can be indexed by different types of anaphor. In the example above, the assignment of Philip as the agent in (3c) might be faster when it is preceded by (2a) than by (2). In summary, th1s experiment exploits the properties of three different types of anaphoric linkage to investigate two main questions. Is discourse­ based knowledge actively available to listeners during the on-line process­ ing of an u tterance, and is the use of inference to establish such dis­ course-links necessarily more time-consuming for the listener? 302

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To be able to fully evaluate not only these two extremes but also the various intermediate possibilities, a further manipulation was added to. the experi ment. This was a variation in the time delay between the anaphor and the point of testing. We did this by varying the length of the verb phrase, using verbs with and without particles and with and w1thout adverbs. If the effect in ( 3c) is smaller - that is, if the difference between appropriate and inappropriate probes is smaller than in either ( 3a) or (3b) - then th1s might interact with the amount of time which the listener has available from the onset of the fragment. For the longer fragments, where the listener has more time in which to work out the necessary relationships, the effect in ( 3c) might be larger, and closer to that observed in (3a) and (3b). In other words, if the time-course of anaphor resolution is indeed slowed down by the use of inference, then this further manipulation will enable us to estimate how far this effect reaches.

PROCESSING UTTERANCES Method

Subjects heard sequences consisting of two sentences followed by an incomplete fragment. Timing pulses, which the subject could not hear, were placed precisely at the offset of the fragment. These pulses served both to initiate the presentation of a visual probe word and to trigger a timer. The subject's task was to name the probe word as rapidly as possible, and this naming response activated a voice operated relay which stopped the timer. After naming the probe, the subject had 10 sec in which to write down whether it was a good or bad continuation of the fragment, and then the next trial began. Each session lasted approximately 30 minutes.

We constructed 48 stimulus sets, each consisting of 6 sentences. One of these sentences was a context sentence which provided a general scenario with respect to which subsequent sentences could be interpreted. Two further sentences foregrounded either one or the other of the two protagonists mentioned in the context sentence. We also had three sentence fragments, each of which contained one of three different anaphoric devices, always occurring in subject position. One type of anaphor was simply a repetition of a proper name initially mentioned in the context sentence; the second type was an unambiguous personal pronoun referring to one of the antecedents mentioned in the context sentence; the third type was a zero anaphor, with no lexical marking of the antecedent. These anaphoric devices were located at the beginning of fragments containing verb phrases that systematically varied in length. For 1 2 o f the 48 stimulus sets, the verb phrase consisted o f a single word, such as ' Punishing' or 'Opening '. The average length of this set of sentences, in their zero anaphor form, was 2.4 syllables. For another 1 2 the fragment was a verb plus a preposition or particle, such as 3.8); for another ' Running towards' (average number of syllables 1 2 it was a verb plus an adverb, such as 'Politely removing' , or 'Tear­ fully welcoming' (average number of syllables 5.25); or, finally, it could be a verb plus particle and/ or preposition and an adverb, as in 'Skillfully probing inside' or 'Carefully getting out of' (average number of syllables 6.4). 2 =

=

=

These materials were extensively pre-tested to ensure that all three fragments in each set successfully indexed the appropriate antecedent when there were no time-constraints on the subject. Six lists were compiled for this pre-testing, each consisting of 48 test sequences and 42 filler sequences. In each list a context sentence was paired with one or other of the two foregrounding sentences (such as 2 or 2a in the example above), and this in turn was followed by one of JS, vol. l , no.4

303

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Materials

TYLER & M ARSLEN-WILSON the three fragments (such as 3a, b, or c in the example set). Each Jist was typed on a sheet in random
Each of the six lists of test sequences, mixed pseudo-randoml y with the filler sequences, was recorded at a rate of 1 60 words per minute by an experienced female reader. After recording the stimuli, timing pulses were placed on the non-speech channel of the tape at the exact offset of each fragment. These pulses started a digital timer and, at the same time, opened a tachistoscopic shutter which projected the probe word onto a screen in front of the subject. The presentation of the probe word and the offset of the fragment occurred simultaneously. For each test sequence v.e had two probe words which were always pronouns. One probe word was an appropriate continuation of the frag­ ment, and the other was an inappropriate continuation. We used a selection of different pronouns, such as him, her, it, them, and equalised the number of times, across the 48 test sequences, that each pronoun functioned as an appropriate or inappropriate probe. This was to control for two potentially confounding effects. Firstly, to control for the voice operated relay 's differential sensitivity to different acoustic inputs, and secondly, for possible effects of word length on the time it takes to begin to utter a word (Sternberg, Monsell, Knoll & Wright, 1 979). Probe words for the f iller sentences varied considerably in their grammatical category in order to reduce the subject 's expectations that the probe word would be a pronoun on every trial. Design and Procedure

A mixed design was used here. The variable of Verb Type was nested within sentences, with 12 instances of each of the four verb types. The other variables were fully crossed with each other. Each of the two foregrounding sentences was paired with each of the three types of anaphor fragments, which were in turn paired with each of the two visual probes. This produced a 2(Foreground) by 3(Anaphor) by 304

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To obscure the regularities in the test sentences we constructed 42 filler sequences. The structure of these sequences varied considerably, as did the point at which the fragment ended. A further set of 20 sequences was constructed for use as practice.

PROCESSING UTTERANCE� 2(Probe) design, yielding 12 experimental conditions. The six experimen­ tal tapes were crossed with two probe sequences such that, across subjects, each sequence appeared with both probes.

Subjects

The 36 subjects were all students at University College, London, and native English speakers. They were paid for their participation in the experiment. A further 48 subjects were used to pre-test the stimuli. These subjects were adult native speakers of English who lived in the London area. Results

Mean naming latencies (in msecs) for the main results are shown in· Table l . The table also includes the differences between appropriate and inappropriate probes. On an statistical analysis 3 , the difference between overall mean reaction-times for appropriate and inapj ropriate probes (383 and 428 msec, respectively) was highly significant . Moreover, this preference for the appropriate probes was observed in all three anaphor conditions 5, although the size of the difference between appropriate and inappropri­ ate probes averaged 1 5 msec Jess in the zero anaphor case than the other two types of anaphor. This difference, which is not significantly different from the size of the difference for the other two anaphor cases, is nevertheless worth commenting upon. It probably reflects the fact that in the pre-test phase of the experiment, when subjects' continuations were collected to ensure that anaphors correctly indexed their antecedents, there was a small difference between the percentage of correct continuations for zero anaphors and for the other two types of anaphor. That is, although our criterion demanded that at least 80% of subjects produ�ed an appropriate continuation, the percentage JS, vol. 1 , no.4

305

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

36 subjects were tested, with three subjects in each of the 1 2 condi­ tions. Subjects were tested individually in a quiet room. The sentences were presented over headphones, and the visual probe was projected onto a screen in front of the subject. At the beginning of the experiment subjects were told that they would hear sequences of sentences, each of which ended in an incomplete sentence, and that this fragment would be followed by a single word which would appear on the screen in front of them. They were encouraged to listen to the sentences normally and to name the probe word as rapidly as possible, after which they were to write down whether it was a good or bad continua­ tion of the sequence. This second task was introduced to prevent the subjects from naming the word without l istening to the sentences.

TYLER & M ARSLEN-WILSON Mean Naming Latencies (msec): By foregrounding sentence, type of anaphor _and probe type.

l

App.

Repeated name Pronoun Zero

379 382 38 1

Overall standard error

=

Probe Type

Inapp.

/),

App.

427 432 417

48 50 36

378 388 388

2

Probe Type

lnapp.

/),

431 436 423

53 48 35

5.45 msec

* In foreground sentence l , the antecedent of the anaphor occurring in subject position in the fragment is foregrounded. In sentence 2, it 1s the other possible antecedent which is foregrounded.

Table l Mean Nam ing Latencies (msec): By verb type, type of anaphor and probe type. Type of anaphor

Verb Type

Repeated name

l 2 3 4

Pronoun

Zero

Overall standard error

=

Probe Type

lnapp.

/),

379 383 370 382

428 440 4 24 425

49

l 2 3 4

389 395 382 373

428 442 432 434

39 47 50 61

1 2 3 4

386 381 387 384

422 421 421 417

36 40 34 33

App.

57

54 43

7 . 8 2 5 msec Table 2

306

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Type of anaphor

ForeCiroUnd sentence *

PROCESSING UTTERANCES of correct responses was, in fact, slightly higher for anaphoric repeated names and pronouns than for cases of zero anaphor. This inevitably meant that there was a sharper distinction between inappropriate and appropriate continuations for anaphoric repeated names and pronouns than for zero anaphors.

.

The length of the verb phrase also had no significant main effect 7 , nor did it interact with any other variable. Table 2 presents a breakdown by Verb Type of naming latencies to appropriate and inappropriate probes for the three types of anaphor. There is clearly no trend, even in the zero anaphor cases, for the difference between a ppropriate and inappropriate probes to increase as the length of the verb phrase increases (from verb type 1 to verb type 4). 8 Discussion

The first goal of the experiment was to obtain direct evidence that knowledge which could only have been derived from the discourse context is actively available to listeners during the on-line processing of an untterance. The results here are completely clearcut. For all types of anaphor, the appropriate probes were named faster than the inappropriate probes. To have developed such a preference, the listeners had to have worked out who was the agent of the action denoted by the verb in the fragment, and to have related this action to the frame­ work established in the prior context. For example, in the sequence given earlier, her could only have been considered a more appropriate continuation than him , given the information provided in the prior dis­ course about the properties of Philip and the old woman. The results of the Verb Type manipulation provide strong evidence, furthermore, for the continuous nature of the processing linkage between an incoming utterance and its context. The listener's preferences for appropriate probes were equally strong irrespective of the length of the verb phrase in the fragment. Even when we tested only one word after the anaph or, listeners had already developed their preferences for the appropriate probe. This result confirms the conclusions we drew elsewhere (Marslen-Wilson and Tyler, 1 980) from the effects of discourse context on word-monitoring responses very early in an utterance, as reflecting the same continuous assessment of the input against a representation of the discourse context. JS, vol. l , no.4

307

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The difference between appropriate and inappropriate probes was also unaffected by the other variables we manipulated. The foregrounding variable - that is, whether or not the antecedent of the anaphor was foregrounded in the intervening sentence - produced no effect 6 • As Table I shows, the difference between appropriate and inappropriate probes is very similar in both foregrounding conditions for each of the three types of anaphor.

TYLER &. M ARSLEN-WILSON The second goal of the experiment was to determ ine whether there are restrictions on the type of rel�tionship between an utterance and its context that can be readily exploited during on-line processing; in particular, whether relationships that can only be based on inference are necessarily more time-consuming for the listener. We investigated this by using three different types of anaphoric linkage - repeated names, unambiguous pronouns, and zero anaphors - which differed in the types of mapping processes that their interpretation required.

This result is consistent with the claim of the on-line interactive model that the recognition of spoken words involves the assessment of the pragmatic plausibility of potential word-candidates against the available context. Part of the on-line process of recognising nmning towards m (3c), for example, would involve the parallel assessment of the plausibility of th1s particular verb relative to the various potential antecedents. Given the observed speed of word-recognition processes m context (cf., Marslen-Wilson, 1 980; Marslen-Wilson and Tyler, 1 980), this provides a mechanism for the rapid resolution of anaphoric reference even in the zero anaphor cases. The results also support the clai m that different sources of processing mformation are integrated together on-line during the interpretation of the speech input. When one source of processing information in the present case, the lexical specif icity of the anaphor - is less fully specified in the input, this can be tolerated when other sources of information are sufficiently well represented. In the present experi­ ment, the discourse provided a sufficiently rich interpretative context so that the listener could correctly interpret the anaphor regardless of whether or not it was explicitly lexically realised. However, despite this i mportance of the discourse context, the explicit manipulation in the experiment of discourse structure had no effect on performance. The difference between appropriate and inappropriate probes remained the same irrespective of which of the two protagonists set up in the scene-setting sentence had been 'fore­ grounded' in the intervening sentence. This, it turns out, is because the particular manipulation used here was too weak to produce the 308

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The results show, in particular, that the inference processes required to interpret the zero anaphors did not cost the listener more in terms of time and success than the pronoun or nominal anaph or, which do not necessitate the use of inference. The difference between appropriate and inappropnate probes was very similar for the three types of anaphor­ ic device. Moreover, the size of the effect did not vary, for any of the three types, as function of the length of the fragment (Verb Type). Thus, even when inferencing is mvolved, the listener is able to assign the correct antecedent to an anaphor as the verb followmg the anaphor is being heard, and this process is essentially complete by the point at which we probed (at the end of the verb).

PROCESSING UTTERANCES intended effect. Recent work by Sanford and Garrod ( 1 9 & l) shows that such effects require at least two sentences to intervene between the scene-setting sentence containing the protagonists and a further sentence containing an anaphoric pronoun indexing one of these individ­ uals. In the present experiment, w ith only one intervening sentence, the unforegrounded protagonist still remained readily accessible for anaphoric reference.

Secondly, given that we could establish from the post-test which protagonist was in reality the more salient, we could then re-examine the effects of discourse salience on the responses to the two types of probe. We contrasted here the cases where there was a strong bias towards the protagonist who was the subject of the continuation frag­ ment (in the experiment), with the cases where there was the opposite bias. We found no effect of this contrast on the size of the difference between appropriate and inappropriate probes. This lack of effect is consistent with one 's intuition, for all our materials, that it would have been perfectly natural to pronominalise either potential antecedent in subject position in the continuation sentence. Thirdly, the post-test enables us to exclude one possible alternative account of the general results; namely, that the critical protagonist, who occurred in subject position in the continuation fragment, was always the most salient in the discourse. If this had been the case, then it would have been possible to argue that in all conditions anaphoric reference was primarily determined on the basis of this discourse bias. This is clearly excluded, since, as we reported above, the critical protagonist was not always the most discourse salient. The relationship between discourse bias and pragmatic inference based on the properties of the verb is further clarified in a subsequent experiment (Tyler, Marslen-Wilson and Koster, in preparation), whose results both replicate and extend the findings reported here. This follow-up experiment, using the same naming technique, factorially JS, vol . l , no.4

309

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This explanation is supported by the results of a post-test we carried out to assess the extent to which the foregrounded manipulation did have any effect. Subjects were presented with written lists of sentence­ pairs, where each pair consisted of a scene-setting sentence (such as ( 1 ) in the example set above) paired with one or the other of the two foregrounding sentences (such as (2) and (2a) in the example set). The subject 's task was to write down an appropriate continuation sen­ tence for each stimuli pair. We could then determine which protagonist was in fact salient in the discourse by measuring the probability that one protagonist or the other was placed in subject position, as agent of the continuation sentence. The results showed, first, that the 'fore­ grounding' manipulation had lit tle effect on the subjects' choices; that is, the shift from (2) to (2a) only rarely succeeded in strongly switching preferences from one protagonist to the other.9

TYLER

&.

M A RSLEN-WILSON

contrasted the effects of (i) discourse focus, (ii) the lexical con­ straints carried by a pronoun, and _ (iii) pragmatic inference based on the properties of the verb. The results showed, first, that each of these three sources contributed to anaphor resolutions, since any single source operating in isolation (with the other two held natural) was sufficient to produce rapid on-line resolutions, as measured by the differences in naming-t1me for appropriate and inappropriate probes. _

Concluding Remarks

Two concluding points need to be discussed. The first concerns the implications of the results for a theory of anaphoric reference. The fact that inference-based resolution processes can be so effective does not mean, of course, that only inference is used to resolve anaphors. It must certamly make a d ifference if one encounters a repeated name or a pronoun, and not just a zero anaphor. In fact, it is conceivable that subjects' responses in the repeated name and pronoun cases did not involve the use of inference at all. It seems to us, however, to be more plausible, given the speed with which the listeners can use inference to establish anaphoric linkages, that the use of pragmatic inference is a normal part of the resolution process even when other sources of information are also present. More specif ically, we propose that part of the normal anaphor resolution process is what can be called 'pragmatic checking' (cf. , Sidner, 1 979), in which the properties predicated of an anaphor in an utterance (for example, in the subsequent verb phrase) are checked against the proper­ ties attached to potential antecedents in the discourse model . Thus even though - as in the repeated name and pronoun cases the lexical information carried by the anaphor is, strictly speaking, sufficient to uniquely deter mine the correct antecedent, the assignment process remains incomplete without the further information, carried here by the verb phrase, which allows an inferential check on the interpretative plausibility of the proposed antecedent. In the example used here, the acceptability of Philip as the antecedent is confirmed by the likeli­ hood that it is he, rather than the old woman, who is performing ti-e 310

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Secondly, the results confirmed the on-line i mportance of pragmatic inference; both in selecting antecedents and in the process of pragmatic checking. In particular, the antecedent consistent w ith the verb was chosen in the zero anaphor cases even when this conflicted with the choice of antecedent dictated by the discourse focus set up in the preceding context. When an unambiguous personal pronoun occurred in subject position m the continuation fragment, then subsequent resolu­ tion processes were severely disrupted in cases where the pragmatic implications of the verb conflicted with the antecedent indexed by the pronoun.

PROCESSING UTTERANCES action of running. This would mean that all three types of continuation tested here have in common the use of inference as an essential and normal part of the resolution process (cf., Marslen-Wilson and Tyler, 1 980; Marslen-Wilson, Levy and Tyler, 1 98 1 ; Tyler, Note 1 ) . What the results for the zero anaphor case show is that inferencing without the constraints provided by a lexically more specific anaphor need not create special problems for the listener. The type of processing relationship between an utterance and its context that can be exploited on-line by the listener is clearly less restricted than earlier approaches mtght have led one to expect.

The situation i s quite different i n listening, since the listener does not have the same sort of control as the reader over the speed w ith whtch he can move through a given input. That is, he has no control over the rate at which the input is presented. No matter how slowly he might interpret a given word or phrase, the listener does not get to hear the next portion of the input any later. Whereas the reader can jump to the next fixation point as late or as early as it is convenient for him to do so. The point, then, for the listener, is that increased processing complex­ ity need not show up as increased processing time, so long as the end-point of the process in question is nonetheless reached soon enough not to interfere with the analysis of the subsequent input. And note that the question being asked in the present experiment was not whether inference-based resolution was more complex than other possible types of resolution, but rather whether it slowed down the resolution process, as tapped into by our on-line probes at the end of the verb following the anaphor. Thus in interpreting the present results, we can safely say that, testing as early as was practicable, inference-based anaphor resolution processes do not significantly disadvantage the listener in terms of JS, vol. l , no.4

31 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This leads us to the final point; namely, the divergence between the present result and earlier research on the effects of inference in reading tasks. This divergence should, first, be a warning that experi­ ments in the spoken and written modalities should not be assumed to be directly comparable. The second, more complex point, involves the relationship between processing complexity and processing time in listening as opposed to reading tasks. In the reading situation the relationship seems relatively straightforward. A given sentence will take a certain necessary amount of time to read, determined by the minimal time it takes to make the necessary number of fixations across the string. Any increase in processing complexity which produces an increase in fixation time, above the irreducible physiological minimum, will thus show up in an experiment as an increase in total reading time. Essentially the same analysis applies to tasks measuring overall 'comprehension time ' .

TYLER & M A RSL EN-WILSON the time-course of resolution. What we cannot say directly is whether inference-based processes are or are not more complex for the listener. But an answer to that question would require a quite different sort of experiment. In the meantime, the divergence between the reading and the listening research is hardly very surprising, given that the two types of research study such different processes in such very different ways. Max Planck Institut fur Psycholin£uistik, NIJ MEGEN

I Of course, a proper name is not anaphoric in the same sense as a pronoun or defimte noun phrase. However, when repeated in this way, the proper name mentioned in the fragment clearly must be linked to the prwr mention m the discourse of the individual to which the name is attached. 2 These examples are all given m the zero anaphor form. In the re­ peated name or pronoun form the fragments would be "He opened ... " , o r "Mary carefully disposed o f . . . ", etc. 3 Followmg Clark's ( 1 973) recommendation, two analyses of variance, one with subjects and one with sentences as random variables, were computed on the raw data, with missing and extreme values (less than 2%) replaced (Winer, 197 1 ). In the sentence analysis, sentences were nested within Verb Type and crossed with Foreground, Type of Anaphor and Probe Type. In the subject analysis, subjects were fully crossed with all variables. Subject and sentence analysis were subsequent­ ly combined to form Mm F ' values. 4 Min F '( l , 73) = 53.3, p< .001 5 There was no significant main effect of Type of Anaphor on a Min F ' analysis (Min F ' < 1 ), nor on either the subject ( F 1 (2, 70) = 2.24, p=. l ) or sentence (F 2 (2, 88)= 1 . 52, p=.22) analysis. Nor, as can be seen from Tables 1 and 2, were there any interactions of Type of Anaphor with the other variables. 6 Min F'< 1; F , ( l , 35)=2.88, p=.09; F 2 ( 1 , 44)= 1 . 6 1 , p= .20 7 Min F ' < 1; F , (3, 1 05)= 1 .74; p= . 1 6; F 2 (3, 44)=0. 56; p=.64 8 It is worth adding that we in fact ran the experiment twice; once on the population reported here, and a second time on a much more heterogeneous group - the native speakers of English who happened to be in the Nijmegen area. The results were exactly the same as for London college students. There were no significant differences between types of continuation, and there were significant preferences for the appropriate probe in each case (the difference for the repeated name case was 45 msec, for the pronoun case 65 msec, and for the 312

JS, vol. 1 , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Notes

PROCESSING U TTERANCES zero case it was 54 msec). 9 Note that there were two separate sets of sentence-pairs, so that no subject saw both the foregrounding sentences for a given scene­ setting sentence.

Acknowledgements

References

Caramazza, A., Grober, E., Garvey, C. & Yates, J., 1 977: Comprehen­ sion of anaphoric pronouns. Joumal of Verbal Leaming andVer­ bal Behavior, 1 6; 6 0 1 -609. Carpenter, P.A. & Just, M.A., 1 977: Reading comprehension as the eyes see it. In: M.A. Just & P.A. Carpenter (Eds.), Cognitive processes in comprehell3ion. Law rence Erlbaum Associates, Hillsdale, N.J. Chafe, W , 1 976: Givenness, contrastiveness, definiteness, subject, topics, and point of view. In: C.N. Li (Ed.), Subject and topic Academ­ ic Press, New York. Charniak, E . , 1 978: On the use of framed knowledge in language compre­ hension. Artificial Intelligence, 1 1 ; 225-265. Clark, H .H . & Haviland, S.E., 1 977: Comprehension and the givennew contract. In: R.O. Freedle (Ed.), D iscourse comprehell3ion and production. Ablex, Norwood, N.J. Cole, R.A. & Jakimik, J., 1 980: A model of speech perception. In: R.A. Cole (Ed.), Perception and production of fluent speech. Law­ rence Erlbaum Associates, Norwood, N.J. Ehrlich, K . , 1 980: Comprehension of pronouns. Quarterly Joumal o f Ex­ perimental Psychology, 32; 247-255. H irst, W . & Brill, G., 1 980: Contextual aspects of pronoun assignment. Joumal of Verbal Leaming and Verbal Behavior, 1 9; 1 68- 1 75. Just, M.A. & Carpenter, P.A., 1 978: Inference processes during reading: Reflections from eye fixations. In: J . W . Senders, D.F. Fischer & R.A. Monty (Eds.), Eye movements and the higher psychological functioll3. E r l b au m , Hillsdale, New Jersey. Marslen-Wilson, W.D., 1 980: Speech understanding as a psychological process. In: J.C. Simon (Ed.), Spoken language generation and understanding. D. Reidel, Dordrecht, Holland. Marslen-Wilson, W.D. & Tyler, L . K . , 1 975: Processing structure of _

.

JS, vol. l , no.4

313

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The order of authorship is arbit;ary. Both authors contributed equally to the research reported here. We wish to thank Simon Garrod and Merrill Garrett for their insightful comments on an earlier version of this paper, and David Green for kindly making available testing facilities at University College, London.

TYLER &. MARSLEN-WILSON

314

JS, vol. 1 , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

sentence perception. Na ture, 2 5 7; 784-786. Marslen-Wilson, W.O. &. Tyler, L . K_. , 1 980: The temporal structure of spoken language understanding. Cognition, 8; 1 -7 1 . Marslen-Wilson, W.O. &. Tyler, L . K . , 1 98 1 : Central processes in speech understanding. Philosophica l Transactions of the Royal Socie­ ty, London B, 295; 3 1 7-332. Marslen-Wilson, W.O., Levy, E . &. Tyler, L.K., 1 982: Producing interpreSpeech, table discourse. In: W. Klein &. R. T. Jarvella (Eds.), place and action. Wiley, Chichester. Sanford, J.A. &. G<;�rrod, S., 1 982: Understanding written language: exploWiley, Chirations of comprehension beyond the sentence. chester. Sidner , C.L., 1 979: Towards a computational theory of definite anaphor comprehension in English discourse. M.I.T. AI Lab. Memo, TR-537. Sternberg, S., Monsell, S., Knoll, R.L. &. Wright, C.E., 1 979: The latency and duration of rapid movement sequences: comparisons of speech and typewriting. In: G. Stelmach (Ed.). Information processing in motor control and learning. Academic Press, New York. Tyler, L.K.: The development of discourse mapping processes: The on-line interpretation of anaphoric expressions. Cognition (in press). Tyler, L.K. &. Marslen-Wilson, W.O., 1 977: The on-line effects of seman­ tic context on syntactic processing. Journal of Verbal Learn­ ing and Verbal Behavior, 1 6; 683-692. Tyler, L.K., Marslen-Wilson, W . O. &. Koster, C.: Integrative processes in discourse comprehension. (in preparation). Winer, B.J., 1 9 7 1 : Statistical principles in experimental design. McGraw­ Hill, New York.

INTERPRETING

ANAPHORA

WITHOUT

IDENTIFYING

REFERENCE

George Yule

Abstract

The most widely held view of how anaphoric pronominals receive their interpretations is based on a 'substitution' or 'referring back ' relation­ ship between the anaphor and its antecedent nominal. This essentially linguistic view, presented most strongly in Halliday &. Hasan ( 1 976), has had a powerful influence on the way psycholinguists have approached the interpretation of pronouns as a mental process. Carpenter &. Just ( 1 977) describe pronouns as 'referring back ' to their antecedent nominals, Clark &. Clark ( 1 977) present their 'given-new strategy' as a process of first determining the unique antecedent for the pronoun encountered, and Garrod &. Sanford ( 1 982) seem to suggest that, without an identi­ fiable antecedent, pronouns present extremely difficult processing problems. In fact, Garrod &. Sanford propose that "pronouns can only be used anaphorically to refer to explicit representations" ( 1 982:29). This is a very strong claim and one, I w i l l suggest, which is difficult to maintain. Yet, this type of claim is actually a natural outcome of a generally misleading view of the way anaphoric pronominals receive their interpretation, in all circumstances.

JOURNAL OF SEMANTICS, vol. l , no.4, pp. 3 1 5-322

315

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It is proposed that, by adopting an antecedent-determined acco!Dlt of the interpretation of anaphoric pronominals, we may m isrepresent what is actually required in the interpretive process. If we adopt an antecedent plus predicate(s)-determined account, we may arrive a t massively over-specified representations which would seem to create a substantial processing load. Since it can be observed that one of the characteristics of conversational speech is the occurrence of ante­ cedentless pronominals, it is suggested that the analytically required referential identity for anaphora resolution may not actually be an on-line processing requirement. What hearers may do is focus their attention on what is predicated of (a t least some) pronominals, following the focus-marking of the speaker, and simply accept that there is, for the speaker, some referent or referential set for the pronominals encountered. Consequently, for the hearer, identifying the reference of those pronominals need not be a requirement in the interpretation of the speaker's utterance. That is, the interpretation of anaphora need not be, in some c ircumstances, a referential issue at all.

GEORGE YULE The 'antecedent-determined' interpretation of anaphoric pronominals would have to take place in such a way that ( I a) would represent the interpretation of the second conjunct in ( ! ). (!)

Buzzard 's Bay was f itted with blinkers for the first time and he won the race convincingly. ( I a) Buzzard's Bay won the race convincingly.

The correctness of this approach to the interpretation of anaphoric pronominals seems, on first consideration, to be very plausible. Let us consider a second example, once again with the anaphoric pronominal in (2) replaced by its antecedent nominal in (2a).

In example (2), it may be more obvious that (2a) represents a misleading interpretation of what we understand from the instruction in the second conjunct in (2). Surely (2b) would be a more accurate representation. (2b) The onion, cut into small pieces, ts fried slowly in the fat. The nature of the object which is fried is crucially different in (2b) from that in ( 2a). Returning to example ( ! ), we can suggest that ( i b) is a more accurate representation than ( I a) of the second conjunct in ( ! ). ( ! b) Buzzard 's Bay , fitted with blinkers for the first time, won the race convincingly. If we take a simple subject-predicate view of the structure of the first conjuncts in ( I ) and (2), we can identify the antecedent as the subject in each case and note that the interpretation of the anaphoric pronominals in the second conjuncts, as shown in ( I b) and (2b), is not 'antecedent-determ ined ' . It is, in some sense, also 'predicate-deter mined '. A very similar view is expressed, though not in these terms, by Webber ( 1 978) and some experimental work on the influence of verbs in the interpretation of pronouns by Garvey et a!. ( 1 97 5) and Caramazza et a!. ( 1 9 77 ) would support this approach. A fuller discussion and exemplifi­ cation of the influence of 'antecedent predicates' on the interpretation of anaphoric pronominals is presented in Brown & Yule (in press). The difficulty of maintaining the 'antecedent-determ ined' view of pronoun interpretation is even more apparent when we consider some examples of the use of pronouns in conversational speech. If the inter­ pretation of the pronominal is made on the basis of the 'antecedent ' , then whatever i s predicated o f the pronominal, a s subject o f i t s own clause, should not play a part in that interpretation. This is precisely the claim made by Haviland & Clark ( 1 974) and Clark & Clark ( 1 977) 316

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 2} The onion is cut into small pieces and it is fried slowly in the fat. (2a) The onion is fried slowly in the fat.

ANAPHORA WITHOUT REFERENCE in their accounts of the priority of referential assignment to a 'given ' element (e.g. a pronoun) before the 'new' element (e.g. the predicate) can be processed. Such a process should allow us to assign a 'correct' or 'uniquely identifying' interpretation to the pronominal in each of the following examples, where the predicates have been omitted. (3) (�) (5)

the car 's we could the last he heard

coming up to the junction and he . . . say of Caithness that they ... time he was here they got antlers - and he was writing that this was the time they •..

(3a) and he starts to turn right. No direct antecedent for the pronominal is available in example (�). To interpret this fragment, we may indeed provide an inferred anteced­ ent predicate such as 'has people in it' attached to a region name like Caithness, before going on to consider the new predicate shown in (�a). (�a)

that they are lazy speakers.

W ith the inferred antecedent predicate and confirmation in the 'new' predicate, we can interpret the they as something like 'the people of Caithness'. W ith example ( 5), we may suggest that mention of antlers brings 'deer' into the implicit representation for the interpretation of they (perhaps not automatically), and the 'new' predicate cast, shown in (5a), confirms the 'dee r ' interpretation. (5a) this was the time they cast. Several points should be made about examples (3)-(5). Notice that Garrod & Sanford's view of pronoun-use, quoted earlier, specifically predicts the non-occurrence of these examples. We should also note that the antecedent-nominal I anaphoric pronominal basis for the inter­ pretation of pronouns would be inoperable. Finally, the information contained in the 'new ' predicate is involved in the interpretation of the pronominal. JS, vol. I , no.�

317

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In example (3), we have no grammatically compatible antecedent nominal which could be used as a basis for the interpretation of the pronominal he . If we guess that mention of the car provides a co-referent which is implicit or 'inferrable' (cf. Prince, 1 98 1 ) - that is, 'the driver of the car ' - then we have a potential means of interpreting the pronominal. We may be more confident of this interpretation if the 'new' predicate describes an action which is pragmatically compatible with the behaviour of our 'inferred' referent. The completed clause is presented as (3a).

GEORGE YULE What should be noted about the interpretation of the anaphors in examples ( 1 )-(5), and this is also true of Webber 's ( 1 978) treatment of discourse anaphors, is that, once one goes beyond the simple anteced­ ent-anaphor 'substitution' basis of interpretation, there is an explosion of elements (e.g. predicates, both explicit and implicit) required in determining the reference of the anaphor. The 'antecedent-determ ined ' principle may be too simplistic, yet the 'predicate-determined' principle must lead to unwieldy representations and involve rather extended processing time.

Yet, it is typically the case that pronominals, in spoken discourse, are used by speakers for the expression of information which is treated as 'given' (cf. Hall iday, 1 967; Chafe, 1 976; Yule, 1 9 80). Pronominals, uttered with low pitch and amplitude, are the typical means by which a speaker indicates that part of his message is not to be treated by the hearer as requiring the hearer's focus of attention. The focus of attention should be directed to the 'new' information, that part of the message which is not lexically and phonologically attenuated, but is marked with raised pitch, pitch movement and amplitude. In the subject-predicate structure of the examples considered so far, the focus of attention indicated to the hearer should be on the predicate. That is, what the speakers of examples (3)-(5) are indicating to their hearers is that it is 'starting to turn right' in ( 3), 'being lazy speakers' in (If), and 'casting' in (5) which are the elements to be attended to in their interpretations. Why, then, as analysts, do we direct our atten­ tion to the requirements of interpreting what the speaker treats as not requiring major interpretive attention? It may be, of course, that the speaker misjudges what information he may justifiably treat as 'given' and so 'known to ' or 'shared with ' the hearer. In such a case, the hearer will indeed have to work (possibly through some backwards inferencin g) at interpreting what it is he is assumed to know. Yet, it is difficult to believe that such cases are the norm , that hearers are continually busy with working out speak­ ers' 'givens'. Much more plausible is an interpretive strategy which operates on the assumptions the speaker has made about the hearer ' s knowledge and pays attention to what the speaker i s marking out for attention. If we admit such a strategy, we m ay have to consider the possibility 3 1 8.

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

So far I have attempted to discuss the examples considered as if identifying the reference of the pronominal was the key element in the interpretation of the sentences involved. This, of course, is very much in line with the typical analysis of sentences containing anaphoric pronommals. This tendency in the analytic approach, always to treat the referential issue as the key to interpretation, may have led us to require, and even force, pronouns to have identifying referential assignments, in all circumstances.

ANAPHORA WITHOUT REFERENCE that, on occasions, hearers do not actually expend any effort working out the referential assignments of anaphoric pronominals. They may not even consider identifying reference at all. The principle may be simply one whereby, if the speaker assumes that the ' referent' in question is not to be paid as much attention as the predicated infor­ mation, then the hearer will do the same in his interpretation. The hearer will interpret the speaker's message in terms of some information marked for attention predicated of some individual or group, the referen­ tial identity of which is not an issue.

(6) oh everything they do in Edinburgh - they do it far too slowly (7) well I saw a demolition order they said they were going to is · a pity - I don 't know what - as long as they don ' t do what

there actually - a few months ago demolish some of the flats - which they 're doing with Edingburgh though they did with Glasgow

The use of they, as in (6) and (7), with indeterminate reference, seems to serve as an indication of some anonymous group-agent. The identity of the group or its members is not a relevant issue. It may be that, in written language, the expression of the type of information in (6) and ( 7) is more typically conveyed by agentless passive forms and so the pronoun they, with no obvious referential assignment, is rarely encountered. Note that the interpretation of examples (6) and (7) can be made via the predicated information without a determinable referential assignment for the pronominals. One aspect of the use of pronominals with indeterminate reference is the occurrence in conversational speech of an apparent switching from one referential set to another through the use of different pronom­ inal forms, yet without the implication that there are, in fact, separate referential sets. In extract (8), the speaker is generally talking about 'the situation of the average working class man in South Edinburgh in the 1 920s'. Each of the 'new' predicates serves to elaborate on this topic. If a referential analysis is required, then we could say that the reference of the pronominals, despite their different forms we, they, you, is a constant, if vague, set. Each member of this set (of a necessarily pragmatically determined cardinality) must be within the referential range designated by the topic member and implicit predicates attached to that member. (This latter point is required to bring the family of the 'average working class man' within the referential range of we, they and you.) In extract (8), it is rather difficult to discuss the plural pronouns as substitutes for the singular antecedent nominal.

JS, vol. l , no.4

319

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Once one sees the interpretive process as operating in this way, at least on occasion, then one has a fairly natural account of a certain type of pronominal which rarely has an antecedent in spoken discourse. Examples (6) and (7) are taken from casual conversational discourse.

GEORGE YULE

(8) the average working class man - the wages were very small - the rents would run from anything � rom about five shillings to - seven shillings which was about all they could've possibly afforded in these days - we j ust had to live - so it didn't matter how many of a family you had - if it was two rooms well - devil take the hindmost - and you couldn 't get out of your environment you see you just had to suffer it and make the most of it - and they all survived that was the great thing

In the following extract (9), we could also provide a strategy for arriving at a referential assignment for the pronominal she. That is, the speaker is talking about a family and in a family, there is only one female to whom the predicate lost eight sons could be applied, hence the referent of she is 'the mother ' . It may be that the hearer d id go through such an inferential process when he heard this piece of conversation. ( The speaker is talking about the large number of people from Edin­ burgh who died in the First World War. ) (9)

I used to go about with a chap - I don 't know - whether he's still alive now or not - but - there was nine ten eleven in the family altogether - two girls - and nine boys - and she lost e ight sons one after the other

Consider an alternative view of what the speaker is asking his hearer to do in the interpretation of this conversational contribution. If it was of primary importance in the message being conveyed by the speaker that the hearer should ' identify' the referent of she, then he makes no effort to abide by the Gricean ( 1 975) max i m of manner and 'Be perspicious'. If, however, the speaker is mainly concerned with communicating information about the number of people who died in the War, then it is the information lost eight sons, predicated of some individual or other, that is of primary importance. Consequently, the ' identity' of this individual is not a real matter for concern. ( Would the message be essentially different if the speaker had used he, intending the father, rather than she?) The larger structure of the discourse contribution within which this pronominal she is used seems to be organised around the speaker's establishing some personal 'credentials' for being able to assert that, in fact, a lot of men died in the F irst World War. Those personal credentials involve knowing someone from whose family eight sons were lost, but the identity of that family or its members does not have to be established beyond the fact that they existed and the speaker 320

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In this extract, the speaker's purpose seems less a matter of ensuring that his hearer will succeed in making a referential assignment, in any identifying sense, and more a matter of conveying to the hearer a set of details regarding a situation. The hearer's focus of attention I S directed to considering not 'who' was involved, but what life was like.

ANAPHORA WITHOUT REFERENCE knew {or knew of) them. If the hearer is interpreting the message as it is being presented, then not only may he not make a referential assignment for she, he may not even notice or hear it. Such a claim obviously requires testing. Such testing is surely as justified as that centred on pronominals appearing in the second sentence of a construc­ ted, contextless, sentence pair.

University of Minnesota Minneapolis, U .S.A.

Acknowledgement The opportunity to do the research reported in this paper, and the data-base from which the examples were taken, were provided by SSRC Project HR65 1 9 , under the direction of Dr. Gillian Brown, Edin­ burgh University, without whose help the paper would not have been written.

Refer-ences

Brown, G. & Yule, G., in press: Discourse Analysis. Cambridge University Press. Caramazza, A., Grober, E., Garvey, C. & Yates, J., 1 977: Comprehension of anaphoric pronouns. Journal of Verbal Learning and Verbal Behavior, 1 6; 60 1 -609. Carpenter, P.A. & Just, M.A., 1 977: Integrative processes in compre­ hension. In: D. Laberge & S.J. Samuels {Eds.), Basic Processes in Reading: Perception and Comprehension. Lawrence Erlbaum, Hillsdale. JS, vol. l , no.4

321

·

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

I have argued that pronominals without antecedents can receive a non-identifying referential assignment via the interpretation of the information predicated of them. I have also proposed that some pronom­ mals may not receive referential assignments at all, if the interpretation of the discourse does not depend on such ·assignments. This is not intended as an account of all anaphoric pronominals. It is intended as a caveat on any theory of anaphora which attempts to describe the use of pronominals in English without taking speaker 's and hearers' purposes in speaking into account. The phenomenon of anaphora in discourse is a partial reflection of how speakers organise their messages so that hearers can arrive at the intended interpretation. It may be a disturbing fact, but we may have to conclude that, in terms of how the language is used on occasion, anaphora need not be a referential issue at all.

GEORGE YULE

322

JS, vol. 1 , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Chafe, W .L., 1 976: Given ness, contrastiveness, definiteness, subjects, topics, and point of view. In : C.N. Li (Ed.), Subject and Topic. Academic Press. Clark, H . H . & Clark, E.V., 1 977: Psychology and Language. Harcourt Brace, Jovanovich, New York. Garrod, S.C. & Sanford, A.J., 1 982: The mental representation of dis­ course in a focussed memory system: implications for the interpretation of anaphoric noun phrases. Journal of Semantics 1 ; 2 1 -4 1 . Garvey, C., Caramazza, A . & Yates, J., 1 975: Factors influencing assignment of pronoun antecedents. Cognition, 3; 2 27-243. Halliday, M.A.K., 1 967: Notes on transitivity and theme in English: Part 2 . Journal of Linguistics, 3; 1 99-244, . Halliday, M.A.K. & Hasan, R., 1 976: Cohesion in English. Longman,London. Haviland, S. & Clark, H.H., 1974: What 's new? Acquiring new information as a process in comprehension. Journal of Verbal Leaming and Verbal Behavior, 1 3; 5 1 2-52 1 . Prince, E.F., 1 98 1 : Toward a taxonomy o f given-new information. In: P. Cole (Ed.), Radical Pragmatics. Academic Press, New York. Webber, B . L . , 1978: A formal approach to discourse anaphora. Report No. 376 1 . Bolt, Beranek &: Newman Inc. Yule, G., 1 980: Intonation and givenness in spoken discourse. Studies in Lcmguage, 4; 27 1 -286.

LINGUISTIC CONTROL OF PRONOUNS

Liliane Tasmowski-De Ryck and S. Paul Verluyten

Abstract

1. The main issue w e want to raise i n this paper goes back t o the early days of generat1ve grammar, and it can be stated as the following question: what is the control 1 mechanism for a pronoun that occurs in a sentence which does not contain an antecedent for that pronoun? Two opposing viewpoints have been expressed on this problem. McCawley states that in such cases the pronoun is still syntactically controlled, unless it is clearly deictic: "If a personal pronoun occurs in a sentence which does not contain an antecedent for that pronoun, then either the pronoun has an antecedent in some preceding sentence in the discourse (possibly a sentence uttered by someone other than the speaker) or that pronoun is used deictically (i.e., is a direct reference to someone or something physically present as the sentence is uttered) and is stressed and accompanied by a gesture." (McCaw ley I 970: 1 78). Postal on the contrary holds the view that in these cases the pronoun must be pragmatically controlled: "The idea that a form like she in she dances well is a 'replacement' or 'substitute' for some other noun, say in 'discourse contexts' or the like, seems to me completely without basis. Such an assumption JOU RNAL OF SEMANTICS, vol. l , no.4, pp. 323-346

323

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We argue again.st the claim put forward by Lasnik (1 9 76) that pronoun.s are, in all cases, pragmatically controlled, i.e. that they directly refer to objects or situation.s in the world. In fact, the generalization we defend is exactly the opposite: all pronoun.s are linguistically controlled, i.e. they have a linguistic antecedent in all cases. Ewn in those in­ stances where no antecedent is present in uttered discourse, the neces­ sity of postulating such an antecedent, and the poss i bility of identifying i t, can be demon.strated. An antecedent which is not present "in uttered discourse is subject to particular recoverability condition.s, both prag­ matic (it must be controlled by a salient object) and linguistic (the pronoun which is controlled by such an absentee antecedent can only occur in a restrained class of discourse contexts).

T ASMOWSKI-DE R YCK & VERLUYTEN explains nothing for the quite simple reason that there is nothing really to explain. It is quite sufficient to indicate precisely that such forms refer to object-types whose particular referents are assumed by the speaker to be known to the person spoken to." (Postal 1 966 fn.3). The problem has been taken up again more recently by Lasnik ( 1 976), who argues that McCawley 's position is untenable, given that there clearly are counterexamples to it:

Lasnik then goes on to propose a unified treatment of pronominal control, which can be summarized as follows: given that there are clear-cut cases such as the sentence above, where only pragmatic control of the pronoun can be at work, no additional mechanism should be hypothesized for other o:1currences of pronouns: in all cases, pronouns are pragmatically controlled (cf. also Lyons 1 977:657-77). Our contention in this paper will be that an extension of McCawley ' s views i s preferable t o Postal 's. W e will show that, in examples similar to the sentence given above by Lasnik, direct reference to object­ types (i.e. pragmatic control of the pronoun) as suggested by the Postal­ Lasnik hypothesis is not a possible mechanism, that a noun phrase for the operation of a formal grammatical process must be provided, and that this noun phrase can be identified without dragging the theory into mysticism. As such, the unification of the theory of pronominal control proposed in this paper goes in a direction exactly opposite to Lasnik's: we claim indeed that pronouns generally considered as being pragmatically controlled do, in fact, have a linguistic antecedent in all cases.

32 4

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

"Surely Postal 's claim, with its consequence that at least some pronouns are present in the base, is preferable to McCawley 's. For example, there are trivial cases which are inconsistent with the latter. Consider a situation in which an unpopular man is p1esent at a party. He is there for an hour during which period he is avoided by all, no one even mentioning his name. Finally, he stor ms out in a huff. It would be neither unacceptable, nor incomprehensible nor bizarre for someone at this pomt to remark, 'Well, he's left ' . But McCawley's theory explicitly excludes such a use of a personal pronoun. Postal, on the other hand, could provide a natural way of explaining this fact, since the speaker could reasonably expect his audience to know who he was talking about. McCawley's theory could be weakened so as to encompass the relevant example. However, such a weakening would either remove f rom the theory all content in excess op Postal 's theory; or it would drag the theory in a sort of mysticism in which the content of an idea or belief or memory or supposition, etc., can provide a noun phrase for the operation of a formal gram matical process." (Lasnik 1 97 6:2)

L INGU ISTIC CONTROL OF PRONOUNS 2. Before exposing our own argumention, w e wish t o mention Evans's ( I 980) criticism of Lasnik's paper, which constitutes a first step in our direction. Evans proposes the following classification of the uses of pronouns: Pronouns used to make a reference to an object (or objects) present in the shared perceptual environment or rendered salient in some other way, as in ( 1) He's up early. (said of a man passing in the street), or (2) I'm glad he's left. (said of a man who has just walked out of the room).

(ii)

Pronouns intended to be understood as being coreferential with a referring expression occurring elsewhere in the sentence, as the personal pronoun contained in his in one of the readings of (3) John loves his mother.

(iii)

Pronouns which have quantifier expressions as antecedents, and are used in such a way as to be strictly analogous to the bound variables of the logician, as in (4) Every man loves his mother.

(iv)

"E-type pronouns", i.e. pronouns which have quantifier expressions as antecedents but are not bound by those quantifiers, as in (5) Few M P 's came to the party but they had a good time.

One of the main topics of Evans's paper is the introduction of this new, fourth, category of pronouns. The other main topic, which concerns us more directly, is the reduction of the mechanism underlying (iii) to the mechanism underlying (ii). Evans observes that " . . . there is this very striking connection between pronouns in cate­ gories (ii) and (iii): whenever we substitute a singular term for a quantifier binding a pronoun, we arrive at a sentence in which the pronoun can be interpreted as coreferential with that singular term. This surely suggests that some common principle underlies the use of pronouns in categories (ii) and (iii) - that the capacity we have to understand sentences like (3) is, at the very least, connected with the capacity we have to understand sentences like (4). [ ] •.•

The semantic problem posed by those dual occurrences [of singular and quantified sentences] can be solved quite generally if we provide a semantic account of quantified sentences which proceeds by way of a semantic account of their singular instances." ( 1 980:338) In order to reduce (4) to ( 3), Evans appeals to the Fregean explanation JS, vol. l , no.4

325

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(i)

TASMOWSKI-DE RYCK & VERLUYTEN of satisfaction: "[Sentence (4) Every man loves his · mother] is true iff every man satisfies the complex predicate ( B ) loves his mother. By the Fregean explanation, we know that an arbitrary object, x, satisfies that pred­ icate iff, taking a " as referring to x, the sentence (B ) loves his moth­ er is true. Now we can apply the coreference rule quite properly, and Jearn that such a sentence is true iff the referent of " " loves the mother of the referent of " B ", and hence iff x loves the mother of x. Hence, x satisfies the complex predicate iff x loves x's mother, and so the whole sentence is true iff for every man x, x loves x's moth­ er . " ( 1 980: 346-7) "

Therefore, if one admits the obvious need for a unif1ed treatment of classes (ii) and (iii) above, and given that Lasnik's theory cannot account for all cases of pronominal reference in quantified sentences ( class (iii)), it follows that Lasnik's pragmatic theory of pronominal control becomes generally untenable. =

Thus, Evans aims at unifying the treatment of pronouns of classes (ii) and (iii); but in his view this can only be done by isolating class (i), where no antecedent is available in uttered discourse: pronouns of class (i) can, at least as it would seem, only be pragmatically con­ trolled. G1ving up Lasnik's complete generalization is, for Evans, " . . . the price of being able t o recognize the obvious semantic connection between pronouns with singular and quantified antecedents " ( 1 980:352), i.e. between classes (ii) and ( iii). We want to argue that it is unneces­ sary to pay this price, by demonstrating that pronouns of class (i) fall under the same basic mechanism as outlined by Evans: they too are controlled by a linguistic antecedent. •••

3. By extending Evans's line o f reasoning, it i s possible t o show that cases under (i) can be reduced to (ii) just as cases under (iii) can. Take a simple case of so-called pragmatic control (corresponding to 326

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Now note that Lasnik's pragmatic theory of coreference is unable to account for those instances of quantified sentences where the pronoun is coreferential to the quantified NP. 3 In Lasn i k ' s view indeed, as Evans convincingly argues, "it does not make sense ( . . . ) to enquire about the truth value of a sentence of the form ( s ) loves his mother in­ dependently of information about a particular context of utterance ( ... )". It is of course true that the original quantified sentence will be uttered in a context; but facts about that context will only "enable us to settle upon a preferred interpretation of a given expression which is constant in all the substitution instances; there is no way these facts can determine a different referent for the pronoun in each substi­ tution instance" (Evans 1 980: 3 5 1 ).

LINGUISTIC CONTROL OF PRONOUNS Evans's case (i) above): the pronoun refers to an object which is present in the perceptual environment, as in the situation where John and Mary are walking in the country and they suddenly see a rattlesnake three feet away. The following sentence can then be uttered (without stress on the pronoun or a gesture): (6) Watch out, it bites without warning! ' Sentence (7) below, however, seems to be equally appropriate in the same situation:

(7) Watch out, they bite without warning!

At this point, a paradox see111 s to arise. Above we established that the pronoun in sentences like (6) and, undoubtedly, ( 7), cannot be prag­ matically controlled. The pronoun should then be linguistically con­ trolled; but how can that be, given that there is no antecedent available, either in the sentence or in preceding discourse (in the situation as described, there may not be any preceding discourse at all)? This is a question to which we shall return. 4. First, we wish to provide a different, and stronger, argument in favour of an intervening antecedent in sentences like (6) and (7). At this point, proponents of the theory of pragmatic control could still refuse the analogies we rely on and maintain that at least in those cases JS, vol. l , no.4

327

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In (6), one may hesitate whether it refers to the token or to the type; if the latter is the case, then the pronoun is not pragmatically controlled by the object present in the perceptual environment. In (7), it is un­ questionable that the plural pronoun they does not directly refer to the object in the perceptual environment, given that there is only one rattle­ snake present. The mechanism of pronominal control which is at work here appears to be simi lar to the one proposed by Evans for quantified sentences, as outlmed above: they may refer to rattlesnakes in general, but for each substitution-object (in the Fregean sense), the pronoun refers to that particular object. It is obvious, then, that the pronoun they in (7) is not pragmatically controlled by the object, at least not in a direct manner. Now recall the analogy Evans drew between examples like (3) and (4), which was the basis for arguing that an identical mech­ anism of pronominal control must underly both these sentences. The analogy observed here, between the reading of (6) where it refers to the token on the one hand, and the reading of (6) where it refers to the type, as well as the reading of (7) on the other hand, is at least equally strong. But we pointed out before that on the second reading of (6) and in (7) the pronoun cannot be pragmatically controlled; therefore (if one accepts the analogy just mentioned) it is not pragmatically con­ trolled in the first reading of (6) either, where it refers to the token.

TASMOWSKI-DE RYCK & VERLU YTEN where the pronoun refers to the token, it is controlled by the object which is perceptually present. But it can be demonstrated that also this position is untenable. ·

We provided evidence for our views in a recent squib (Tasmowski and Verluyten 1 98 1 ); for the sake of convenience, we briefly repeat it here. The crucial examples against the view that a pronoun is directly controlled by the obJect it supposedly refers to are sentences where the pronoun agrees (in gender and /or number) with what can only be a linguistic antecedent (the question of how to identify this antecedent will be treated below):

}

}

(9) (same situation, but with a shirt:) to me, please? Could you hand it *them

{

Clearly, the real objects ( pants and shirt) do not have anything inherently singular or plural (in French, they are both rendered by words that are singular). The choice of the singular or plural form of the pronoun can only be dictated by the linguistic features of the English words pants (plural) and shirt (singular) respectively. In his paper, Lasnik suggests that the pronoun he m ight mean ' male human being' ( 1 976:7). The examples presented above show that, as far as the singular/plural distinction in pronouns goes, a similar meaning is impossible: them in (8)-(9) does not mean 'more than one object ', but 'antecedent is a plural word ' . Gender agreement, on the other hand, i s readily observable i n French:

( I O) (John is trying to stuff a large table (!a table, feminine) in the trunk of his car; Mary says:) Ia faire entrer dans Ia voiture. Tu n 'arriveras jamais a �le into the car.' ) ['You will never manage to get it(fem.) *it(masc.)

{

}

{ }

{

}

( I I ) (same situation, but with a desk (le bureau, masculme)) le faire entrer dans Ia voiture. Tu n'arriveras jamais a *Ia into the cac.' ) ['You will never manage to get it(masc.) *it(fem.)

{

}

Here again, the real objects (table and desk) are neither masculine nor feminine; only the fact that the French words la table and le bureau, 3 28

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

{

(8) (John wants his pants that are on a chatr and he says to Mary: Could you hand � hem to me, please? *tt

LINGUISTIC CONTROL OF PRONOUNS in some way or other, function as linguistic antecedents can explain the difference. We therefore conclude that the theory of direct pragmatic control of pronouns is incorrect. In fact, all true pronouns are linguistically con­ trolled. 4

5. Let us then assume that in cases like ( 1 0) or ( 1 1 ) there is a linguistic antecedent, although it is not expressed in discourse; its recoverability conditions will be examined below.

Pragmatic control is the relation between a (perceived, or otherwise salient) object (the controller) and a linguistic item; the latter can be an N P (e.g. a proper name or a definite description) or a deictic "pronoun" (which, in fact, is not a pro-noun at all: a deictic does not have any kind of relation to a noun; it does not "replace" a noun, nor is it controlled by one). Let us think of a situation where John and Mary have been talking to one of Mary's professors at a party; after he has left, she can say to John:

}

( 1 2) Sandy is a very good instructor. ( 1 3) a. That man works too hard . b. The ( 1 4) The professor we just talked to flunked me last year. ( 1 5) That b1tch flunked me last year. The ( 1 6) That one is a real bitch. ( 1 7) Lui, c'est une vraie vache. 5

}.

Sentences ( 1 2) to ( 1 7) exemplify cases of pragmatic control. N ot ice that several of these sentences contain an overt deictic marker (that). It must be stressed that sentences like ( 1 7), although they appear to contain what is commonly called a "pronoun" (such as lui) are not on a par with sentences like ( l O) and ( 1 1 ). It can be shown that in ( 1 7) lui is pragmatically controlled by a perceived or salient object (the professor in the situation described). Whereas in ( 1 0) and ( 1 1 ) , the pronoun is lingUJst1cally controlled by an antecedent noun (Ia table and le bureau respectively). The paradigm of lui-elle-eu:c-elles, anteposed, stress­ ed ,and often accompanied by a gesture, is distributionally and functional­ ly distinct from paradigms of true pronouns such as the clitics le -la-les ; our claim about the necessity of linguistic control only concerns the latter. 6 We do not wish to claim that all the paradigms considered JS, vo! . l , no.4

329

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

If the existence of a linguistic antecedent is assumed, cases like ( l O) or ( 1 1 ) can be fairly transparently explained by the 'telescoping' of two straightforward (but quite distinct) control mechanisms, which we call pragmatic control and linguistic control respectively.

TASMOWSKI-DE RYCK

&:

VERLUYTEN

here, which are not all paradigms of true pronouns, are necessarily deictic in the narrow sense (accofl_'lpanied by a gesture, etc.); we do maintain, however, that these paradigms can easily be distinguished from true pronouns and that they behave differently. 7

( 1 8) (John is looking at a shirt (la chemise, feminine) in a department store:) a. *EIIe, c'est une tre s belle chemise. b. c;:a, c 'est une tres belle chemise . ['That is a very nice shirt.'] As one can see, forms of the lui-elle-eu.r-elles paradigms cannot agree with a presumed linguistic antecedent. For sexless objects, another form must be used: neuter t;a . Conversely, it can be shown that pronouns of the le - la - les paradigm cannot be pragmatically controlled. Let us rule out the possibility of an intervening linguistic antecedent by imagining a situation where two discourse participants see an indistinct, unnameable object (e.g. something between a UFO and a ghost, for which there is no word in English or French). In such a situation, we obtain the following: ( 1 9) (John to Mary) a. *Tu l : vois?

:

I

b. Tu vois c;a?

['Do you see it?'] [ ' Do you see that?']

or, if there were two such objects: (20) a. *Tu les vois? b. Tu vois c;a?

['Do yoy see them?'] ['Do you see those?']

Examples ( 1 9a) and (20a) show that, in the absence of a possible linguistic antecedent, true pronouns, such as the forms from the le-Ia-les paradigm, cannot be used.9 Riddles provide additional evidence for this claim. In riddles, the antecedent to the "pronouns" used must remain unknown for obvious reasons; hence, true pronouns cannot occur:

330

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

For instance, we can predict from the distinction we have introduced that forms of the lui-elle-eur-;-elles paradigm , which are not true pro­ nouns, cannot exhibit agreement in gender: agreement in gender indeed is conditioned by a linguistic antecedent, and we claim that these forms are not linguistically controlled. In other words, we claim that in sen­ tences like ( 1 7), lui "agrees" (if this is the appropriate term) with the sex of the controlling salient object (the professor), not with the gender of some linguistic antecedent. 6 To illustrate this, consider the following example:

L INGU ISTIC CONTROL OF PRONOUNS ( 2 1 ) a. C'est petit, c 'est rond, c 'est vert, c;a monte et c;a descend; qu 'est-ce que c'est? b. * � II I Elle est petit( e), rond(e), ... [•it is sma I, round, green, and goes up and down; what is it?']

l

--

-

-

....___-__

_

r �;n�:;,

-

l

l ���:� �� j

perceived or salient pragmatically o bJ·ec t__,�. controls

J

linguistic e

_

[fig. A]

pronoun I l inguistically �==:..r-:= controls : L-.....

_

6. Of course, the value of the proposals developed here depends on the validity of our hypothesis that there may be a l inguistic antecedent which is not present in uttered d iscourse (we shall henceforth call this an "absentee antecedent" 1 0). We demonstrated above that postulating this is necessary in any case in order to explain gender and number agreement as in (8), (9), ( I 0), and ( 1 1 ) . In the remainder of our paper, we shall address the question of the recoverability of the absentee ante­ cedent. Recovering a noun that is not uttered by the speaker is likely to be problematic for the hearer. It will appear that, indeed, the class of possible absentee antecedents is severely constrained by two different sets of conditions: a set of conditions constraining the possibility of linguistic control between an absentee antecedent and a pronoun (stat­ able, as we shall see below, as conditions on the possible discourse contexts in which the pronoun may appear), and a set of conditions constraining the possibility of pragmatic control of the absentee ante­ cedent by a real-world object. We shall first examine the latter, prag­ matic, recoverability conditions. To begin with, we observe that abstract nouns are not recoverable. Thus, in a situation where the speaker sees someone giving a large bill to a beggar, (22) is felicitous because the abstract noun is present in discourse (is not an absentee antecedent); (23), where the pronoun elle would supposedly be linguistically controlled by the absentee ante­ cedent Ia charite, which would itself be pragmatically controlled by the real-world context described, is not felicitous: JS, vol. l , no.4

331

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In sum, we claim that in the crucial cases (as exemplified by Lasnik's sentence "Well, he's left", and by our examples (8), (9), ( 1 0), and ( 1 1 )) both pragmatic control (the relation between a real-world object and a noun) and linguistic control (the relation between an antecedent NP containing this noun and a true pronoun) are at work, and that the "telescoping" of the two control relations explains why the use of a true pronoun is possible. This telescoping is schematically repre­ sented in fig. A:

TASMOWSKI-DE RIJCK

&:

VERLUYTEN

(22) Que Ia charite est belle! (23) Qu'elle est belle!

['How beautiful charity is!'] ['How beautiful it is!']

(24) Tu ne Ia verras pas, elle est en vacances [ ' You won 't see her, she is on a holiday ' ] The linguistically controlled pronouns (!a, e!!e) i n (24) call for a n antece­ dent ma fille 'my daughter ' which should be pragmatically controlled by a real-world object. But in the situation described, the object which is present and salient is not your daughter herself; rather, it is the teddy bear which metonymically represents her and brings about the noun fille referring to your absent daughter. Notice that, in the case of metonymy, the object which is physically present (let us call this object (A)) brings about the noun which refers to the other object, (B), and that only this latter noun can function as an absentee antecedent. Thus, in the situation just described, you cannot felicitously . utter: (25) *Tu ne le verras pas, il est en vacances where !e and il would be linguistically controlled by ! 'ours en peluche (masculine), referring to the teddy bear which itself would metonymically represent your daughter. (25) cannot convey this meaning; it could only mean that the teddy bear itself is on a holiday. Similarly, in some interesting examples given by Nunberg ( 1 978), where ! 'entrecote (feminine) 'the steak ' refers to the object (A) and le client (masculine) 'the customer' to object (B), the following judg­ ments hold: (26) (in a restaurant, one waiter is informing his colleague :) Cette entrecote, elle est assise a Ia table 5. [ ' This steak is sitting at table 5!] (27) *Cette entrecote, it est assise a Ia table 5. (28) II est assis a Ia table 5.

332

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Generally speaking, one could say that the absentee antecedent can only be pragmatically controlled by an object which is physically present or salient, either because the object that the antecedent refers to is itself present, or because it can be made salient through a chain of physical associations. This constraint explains why abstract nouns cannot function as absentee antecedents. It also explains why an absen­ tee can be pragmatically controlled through metonymy (involving a chain of physical associations), but not through metaphor. Let us illustrate this with a few examples. Suppose that, in your living room, a friend picks up the teddy bear your little daughter usually plays with; it is then felicitous for you to utter:

LINGUISTIC CONTROL OF PRONOUN S In (26), the pronoun elle i s linguistically controlled by a feminine antece­ dent which is not absent: l 'entrecote; therefore, a masculine form of the pronoun makes the sentence ungrammatical (cf . (27)). In (28), however, only the noun referring to object (B), le client, can function as an absen­ tee antecedent. The noun referring to object (A), l 'entrecote, cannot be an absentee antecedent; the following sentence is ungram matical in the situation described: (29) *EIIe est assise a Ia table 5

•

(30) La vache m'a recale l 'annee passee. ['The bitch flunked me last yeat'] ( 3 1 ) *EIIe m'a recale l'annee passee . As one can see, elle in (3 1 ) cannot be controlled by the metaphorical expression Ia vache, for the latter cannot be an absentee antecedent. The regularities observed with respect to the use of metony my and metaphor lead to an interesting conclusion: each object appears to have its own proper name. This proper name is the only denomination of the object which can function as an absentee antecedent referring to it. Recoverability restrictions thus limit the possible absentee antece­ dent of a pronoun to a unique world; all other descriptions of the same object (metaphors, definite descriptions, etc.) are excluded from the function of "naming" the object in the context described. 1 1 The preceding is strong evidence against the view of Lyons ( 1 977:664), who talks about "the unverifiable, and unnecessary, assumption that individuals, even when we first encounter them , are invariably cate­ gorized in terms of a proper name or descriptive expression ('that man ', 'that thing ', etc.)". The data presented here in connection with linguistic control by an absentee antecedent show that, on the contrary, it is necessary, and verifiable, to assume that generally individuals and objects are categorized in terms of such a proper name. 7. We will now address the other face of the recoverability question. As a preliminary, let us emphasize that we distinguish between discourse as context and situation as context. Obviously, even in the absence of the real-world object, discourse itself can provide whatever antece­ dent is needed, and any appropriate definite description can do the JS, vol. 1 , no.4

333

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We already stated that metaphorical expressions, on the other hand, cannot function as absentee antecedents to a linguistically controlled pronoun. Compare (JO) below (the French equivalent of ( 1 5)), where the pragmatically controlled item Ia wche 'the bitch ' (literally: ' the cow ? is present in utter ed discourse, with (3 1), where it supposedly would be an absentee antecedent:

T ASMOWSKI-DE R IJ CK & V ERLU YTEN job of defining the kind of object referred to - desk, table, or unicorn thus introducing it as an object of discourse by its very enunciation. Note that in that case French affords to indicate if the object referred to is considered as belon g ing extensionally to the world the speaker is situating himself into (as in (32) below), or, rather, if the object is intensionally defined (cf. (33)): ( 32) Saint Nicolas, ['Santa Claus, ( 33) Saint Nicolas, ['Santa Claus,

il est carrement genial . he is really great. ' ] c 'est carrement genial. that is really great. ']

(34) Saint Nicolas, c;a n'existe pas vraiment, tu sais. ( 35) Saint Nicolas, il existe vraiment, tu sais. In common discourse, (36) and (37) are less felicitous: ( 36) Saint N icolas, c;a existe vraiment, tu sais. (37) Saint Nicolas, il n'existe pas vraiment, tu sais. Both these sentences involve some kind of near-contradiction: in (36), � alludes to a concept (that is only intensionally, not extensionally predicates extensional defined) of which the predicate e:xiste vraiment existence in the real world; the opposite is true for (37). If (37) is uttered felicitously, it can only be interpreted as meaning that the speaker (the non-believer), by his use of il, momentarily adopts the point of view of the believer, for whom Saint Nicolas has an extensional exist­ ence in the real world. In fact, � cannot be used for an NP referring to an actual object; in that case, a pronoun of the paradigm il-elle-ils-elles must be used: (38) • Regarde! La chemise, c;a n'a pas de manches! (39) Regarde! La chemise, elle n'a pas de manches! [' Look! The shirt, it has no sleeves!'] If discursive structure makes this distinction, one may ask whether an analogous distinction can be observed in those cases where linguistical­ ly controlled pronouns on the one hand, and deictics on the other hand, are ultimately to be linked to an object given in the non-linguistic situation. We saw above that there are severe constraints on the recoverability of the absentee antecedent from the "pragmatic'' side. One m ight be tempted to think that these constraints are not only necessary, but also sufficient ones. A priori, it would seem rather easy for the hearer to infer the noun N x of object x if this object x is sufficiently salient 334

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The following roughly complementary paradigm holds:

L INGUISTIC CONTROL OF PRONOUNS in the pragmatic context. However, more careful investigation of utterances where a true pronoun is controlled by an absentee antecedent shows that the pragmatic constraints which we outlined in the preceding section are not sufficient ones. Let us imagine a situation where the pragmatic constraints are clearly respected. John gets ready to read a book; he opens the case where he always puts his glasses ( les lunettes, feminine plural), but does not find them there. So he starts looking for them, and he finally f inds them in his little daughter 's toy box. He picks them up and he can then triumphantly exclaim: [ ' Here they are! ' ] [ ' There they are!'] je dois les retrouver dans une boite a jouets! got to find them in a toy box! ']

Now the real world object which serves as a controller for the absentee antecedent les lunettes is certainly most salient in the situation de­ scribed: John may well be holding his glasses above his head and show them to everyone. Nevertheless, some utterances with an occurrence of the linguistically controlled pronoun les (which exhibits agreement in number with the absentee antecedent lunettes) are ungrammatical: ( 4 1 ) a. *Les ici! b. *Les dans une bofte a jouets! Sentences like those under (4 1 ) are ungrammatical only with a linguis­ tically controlled pronoup. Deictics, with their inherent direct pragmatic control by a real-world object, render such sentences grammatical. Thus, in an appropriate context, one may utter either one of the sen­ tences under (42): (42) a. <;a ici! b. Elles ici! 1 2 The crucial difference between sentences like those under (40), which allow for a linguistically controlled pronoun, and those under (4 1 ), which do not, appears to lie in the fact that the former contain a verbal predicate, whereas the latter are predicateless sentences. This explains why there is a difference in grammaticality between the two sentences of the near-minimal pair Les voici! vs. *Les ici!: voici counts as a verbal predicate, ici does not. 1 3 So there appears to be at least one constraint of a l inguistic (as opposed to pragmatic) nature upon the occurrence of pronouns that are controlled by an absentee antecedent: they can only occur in a clause that contains a verbal predicate. Further investigation shows that not just any predicate will do, nor will every complement of the verb. Let us f irst demonstrate the latter. Recall the situation of example ( 1 0), where John is trying to stuff a large table (la table) fernJS, vol. l , no.4

33}

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(40) a. Les voici! b. L e s voila! c. Ah, dire que [ ' To say I ' ve

TASMOWSKI-DE RIJCK & VERLUYTEN inine) in the trunk of his car. He can then not felicitously utter (43) or (44): (43) *Je ferais peut-etre mieux de commander une camionnette pour elle. ['Maybe I'd better order a van for it.'] (44) *J'ai passe tant d ' heures a travailler sur elle. ['I've spent so many hours working at it. '] But (45), with the linguistically controlled pronoun in argument-position, is felicitous, as well as (46), with a deictic:

Often, but not always, prepositional complements are to be considered optional. Studying a corpus, Danell ( 1 973) discovered that the so-called ''tonic pronouns" (!ui-elle-eu.r-elles) are not normally used to refer to ''things", i.e. non-sexed objects. This is in full agreement with the claims we made above: the so-called " pronouns" of this paradigm are, in fact, deictics, which agree with the sex of the object they refer to; for non-sexed objects, only � may be used. DaneII observes, however, that the so-called tonic pronouns may be used after a preposition in some particular instances: when the preposition is "colourless", i.e. when its appearance is completely predictable and its semantic content close to zero. Thus the following utterance may be entirely felicitous even if elle refers, through the absentee antecedent la voiture 'the car ' , to a car: (47) Je sais que je peux compter sur elle. ['I know I can rely on her.'] But clearly, in examples such as (47), the prepositional complement is an essential verb-complement, and the "tonic" pronoun may be a phonologically conditioned variant of the preverbal clitic; it is a true pronoun anyway, as it agrees with the feminine gender of an absentee noun (voiture). 14 Let us now turn to the examination of the other point we stated above: not every predicate allows for the occurrence of a pronoun which is linguistically controlled by an absentee antecedent. First we wish to consider the class of verbs of the "effizierend" type, such as create, compose, Ialit, etc. In their present tense use, these verbs i mply that the object that is named by their complement (their direct object for instance) does not exist (yet). Indeed in these cases a logical contradicition would arise between what is asserted by the proposition in which the effizierend-verb is used, and an existential 336

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(45) Elle est trop grande pour le coffre de rna voiture . ['It is too large for the trunk of my car.'] (46) Je ferais peut-etre m ieux de commander une camionnete pour c;a.. [ 'Maybe I'd better order a van for that']

LINGUISTIC CONTROL OF PRONOUNS presupposition bearing on the object that is named by their second argument, at least in the present (cf. (48)); no such contradiction arises in the past (cf. (49)): (48) there exists now an object x such that I am now knitting x (49) there exists now an object x such that I knitted x in the past

(.50) *C 'est toi qui Ia tricotes? ['Are you knitting it?'] ( .5 1 ) C'est toi qui ! 'as tricotee? ['Did you knit it?'] Naturally, a deictic, which does not need linguistic control by an ab­ sentee antecedent, is possible even with a present tense use of tricoter: (52) C'est toi qui tricotes �a? [ 'Are you knitting that?'] The infelicity of (50) may be explained by a violation of the pr�gmatic recoverability condition:' there is no salient object that could prag­ matically control the absentee antecedent !a robe. But we observe that intensional predicates, which do not impose preconditions of existence upon the object named by one of their arguments, behave in a similar way. They cannot be used with a pronoun which is linguistically con­ trolled by an absentee antecedent, even if the pragmatic recoverability conditions are clearly fulfilled (i.e., even if there is a salient object in the perceived environment which could very well pragmatically control the desired absentee antecedent). Consider, for instance, French il y a [ 'there is'], which is typically used to introduce a topic of future discourse (Tasmowski 1 97 3). Thus, while (.53) is grammatical, the examples under (54), where the pronoun would be controlled by the absentee antecedent !a broche · (feminin), are not: (53) Madame, il y a votre broche qui se d«kroche . [' Madam, you brooch is opening. ' (lit. ' there is your brooch which is opening.')] (54) *Madame, il l 'y a qui se decroche. * Madame, il y a elle qui se dec roche. [ 'Madam, it is opening. ' (lit. ' there is it which is opening.')] JS, voi . J , no.4

337

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In full agreement with the semantics of such verbs, and with the re­ coverability conditions on the pragmatic side that we outlined above, infelicity arises when these verbs are used with a pronoun which should be controlled by an absentee antecedent in the present tense but not in the past. Imagine a situation where John is just starting to knit a dress (la robe, feminine) (examples (.50) and (52)), or has finished knitting it ( .5 1 ) , <;Uld .M ary asks:

TASMOWSKI-DE RYCK & V ERLUYTEN Or take e.rister, another predicate that does not impose a precondition of existence upon its (subject) argum�nt (in fact, the predicate asserts its existence. Imagine a situation where suddenly and unexpectedly a unicorn (la l!come, feminine) appears in Mary's garden, and she says to John:

(55) ??Impossible! Je sais bien que je dois ['Impossible! I know I must imagine (56) Impossible! Je sais bien que je dois ['Impossible! I know I must imagine

l 'imaginer! it!'] imaginer c;a! that!']

(57) Elle existe! (58) <:;a existe!

1['1t exists!'] ['That exists!']

In fact, a natural continuation of (57) would be: "Je le savais bien!" [ ' I knew it!'], and for (58): "C'est inoui!" ['It's unheard of! ']. This is certainly reminiscent of the use of the clitic pronoun with an explicit antecedent referring to an intensionally defined object (cf. examples (32) to (3 7) above). To sum up our findings, there appear to be constraints on the possible discourse contexts in which a pronoun which is controlled by an ab­ sentee antecedent may occur. We observed that the occurrence of such a pronoun is permitted only i f the pronoun is an essential com­ plement of a verb belonging to a certain class of predicates which, in some way or other, presuppose the existence of the object that the absentee antecedent refers to. Let us call this class of predicates PREDpe ("predicates with presupposition of existence").1 6 It is tempting to establish a link between our findings and some of the views first expressed by Seuren ( 1 979), who dissociates the problem of existential presuppositions from the occurrence of a definite description. Thus, in sentences such as the well-known Le rot de France est chauve ['The K ing of France is bald'], the existential presupposition that there is a King of France would be brought about by the meaning of the pred­ icate. Some predicates, among which etre chauve ['be bald' ], impose as a precondition on their felicitous use that the object their (first) argu­ ment refers to exists. 1 7 In the case at hand, Seuren 's analysis thus ex­ plains the various constraints on the occurrence of a pronoun which 338

JS, voi. J, no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Judgments about the infelicity of (55) may be less than categorical; but there undoubtedly is a difference between (55) and (56) as far as their respective felicity in the situation described is concerned. 1 5 When comparing sentences (57) and (58) below (uttered in the same situation), it appears that (57) is more likely to be said by someone who believes in unicorns (who "has unicorns on his mind"), and who finds confirmation of his beliefs in the appearance of an actual unicorn, whereas (58) is more likely to be uttered by someone who does not believe in unicorns, or who has not given the question any thought, and who can hardly believe his eyes seeing one:

LINGUISTIC CONTROL OF PRONOUNS is controlled by an absentee antecedent

as we observed them :

- the necessary presence of a predicate (compare (40) and (4 1)); - the observation that the pronoun must be an essential complement of that predicate (cf. (43)-(44) vs. (45)); - the fact that the predicate must belong to PREDpe (cf. for instance (54) vs. ( 4 5)).

(B) An NP containing the name of an object x, N x , can be the absentee antecedent of a pronoun referring to x only if that pronoun occurs in a discourse context that allows for an existential presupposition of x. Our views do, however, differ from Seuren's on what seems to be a rather metaphysical issue at first glance. Seuren is concerned with presuppositions of existence in the real world. Thus The King of France i3 bald is really true only if there is a King of France in the real world. We however are led not to give such a privileged status to the real world, as opposed to other possible worlds (see e.g. Fauconnier 1 979). In this matter we share the viewpoint expressed, for instance, by Rorty ( 1 980:292), for whom "[we can] not only [ ... ] talk about non­ existent entities, but we can be discovered to have really been talkin11. a­ bout nonexistent entities". Rorty continues: "In the view I am putting forward, 'Sherlock Holmes roomed with Dr. Watson' is as true, and as little in need of 'philosophical analysis' as 'Snow is white.' [ ... ] The general line I want to take is that there are true statements about fictions, values, and numbers, as well as about cats on mats, and that the attempt to find something like 'correspondence' in terms of which to 'analyze' the former truths on the model of the latter is pointless." ( 1 980:300 - 30 I , note 34) For us, then, there is no principled difference between the (a) and the (b) sentences below, although both k inds of sentences obviously pertain to different worlds, and only the (a) sentences pertain to the real world: (59) (60)

a . Fidel Castro has a beard. b. Santa Claus has a beard. a. A bull has two horns.

JS, voi . I , no.4

339

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

8. If we are right in establishing a link between Seuren's re-analysis of existential presuppositions and our observation about the linguistic constraints to be imposed upon the occurrence of pronuns which are controlled by an absentee antecedent, the linguistic recoverability condition of the absentee antecedent can be stated in its most general form as follows:

T ASMOWSK I-DE R YCK &. VERLUYTE N

(60)

b . A unicorn has but one horn.

If the predicates that belong to PREDpe impose preconditions upon the existence of (some of) their arguments in some possible world, but not necessarily in the real world, this amounts to nothing more, as we see it, than stating that those predicates require, as a precondition to their felicitous use, that an intensional definition o f the o b j e c t t o which their argument(s) refer(s) be available: at least some o f its salient properties must be known. Let us call the class of predicates which we redefined in this manner "PR EDpid" ("presupposing the inten­ sional definition of ). •.. "

(6 1 ) *Quarks are either red, green, or blue. In a similar situation, the use of a predicate that does not belong to PREDpid renders the utterance felicitous:

(62) Quarks are constituents of elementary particles (and they are either red, green, or blue) Sim ilarly, for instance after enumerating all the elementary particles, the teacher could introduce the totally new and unknown object by il y a:

(63) Ensuite, il y a les quarks. [' And then there are quark s ' ] Replacing the notion of PREDpe by PREDpid i n t h e constraints on the possible occurrence of a pronoun which is controlled by an absentee antecedent leads to different predictions in some crucial instances. If we appeal to the notion of PREDpe, we run into problems with sentences such as the following:

(64) (situat ion: a mother wants to show to her little daughter that Santa Claus does not exist by pulling the white beard of the man playing Santa Claus at the school fair; at the same time, she utters:) Tu vois bien qu'il n'existe pas! ['See, he doesn 't exist!']

340

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To illustrate our point, let us imagine a situation where a predicate that belongs to PRE Dpid is used without an intensional definition of the object(s) to which its subject-argument refers being av a ilable: a teacher starts a course for students with little current knowledge of subnuclear physics and who, as he is well aware, have certainly never heard about quarks. He can then not start his lecture by using a pred­ icate from PREDpid in conjunction with quarks as its subject-a �u ment: the following utterance is not felicitous in the context described:

L INGUISTIC CONTROL OF PRONOUNS Now e:rister in the sentence obviously means 'to exist in the real world '; a common paraphrase for this sentence is (65): (65) Tu vois bien qu'il n'existe pas n!ellement! [ 'See, he doesn't really exist!']

These problems disappear once we replace the notion of PREDpe by PREDpid in the constraints on the possible occurrences of the pro­ nouns we are here concerned with. All we need to claim in that case is that n 'e:riste pas belongs to PREDpid for a sentence such as (46) to be felicitous. This claim seems unproblematic to sustain: as far as we can tell, non-existence in the real world is normally only pred­ icated of objects of which an intensional definition is available (of which some salient characteristics are known), such as Santa Claus or unicorns. With the notion of PREDpid then, we correctly predict that (64) is felicitous in the context described. We therefore propose to replace (B) by a new linguistic recoverability condition of the absentee antecedent: (C) An NP containing the name of an object x, N x , can be the absentee antecedent of a pronoun referring to x only if that pronoun occurs in a discourse context where the availability of an intensional definition of x can be presupposed. In the normal case, the presupposition that an intensional definition of the object x is indeed available will be brought about by the use of a predicate that belongs to PREDpid, and by placing the pronoun referring to x in the desired argument position of this predicate. We do not wish to exclude the possibility that in some rather marginal cases (as exemplified in ( 57)), the intensional definition of the object x can come about in a different manner. In fact, a condition such as (C) may ultimately turn out to be but a special case of a more general condition on the use of true ( non-
JS, vol. l , no.4

341

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Therefore e:rister in this sentence cannot belong to PREDpe in its use with a negation, for that would yield a logical contradiction: on the one hand non-existence (in the real world) would be predicated of an object, and on the other hand, the existence (in the real world) of the same object would be presupposed. The alternative is to accept that n 'e.riste pas does not belong to PREDpe. But then, there would be no existential presupposition of the object referred to by the subject­ argument of this predicate, and the constraints we discovered above lead us to predict that no pronoun which is linguistically controlled by an absentee antecedent can occur as a subject of the predicate n 'e:riste pas. Yet (64) shows that this prediction is incorrect.

TASMOWSKI-DE RYCK & VERLUYTEN (ct. Bosch ( 1 982: par.2.3 and passim). 9. To sum up, we hope to have established a complete unification of the control mechanism for true pronouns: all true pronouns are controlled by a linguistic antecedent, whether the antecedent is superficially present in discourse or not.

The normal direction of the control relations, pragmatic and ltnguistic, is different in the two cases. One must suppose that the normal strategy for establishing a link between a noun N x and the correspondtng object x is for the noun to select, by means of its phonetic shape, the correct object x in the real world. In the case where N x is part of an ab­ sentee antecedent, however, it is the real-world object x which, under certain pragmatic conditions as outlined above, allows for the recovery of its name N . On the linguistic side, the normal situation presumably is that the g lobal context (including the uttered antecedent containing N x> yields sufficient familiarity with the object x (or provides a suffi­ cient intensional definition of it) for the pronoun referring to x to be used felicitously. In the case of an absentee antecedent on the other hand, the hearer must be able to infer from uttered discourse (e.g. through the occurrence of a true pronoun and a predicate from PREDpid) that the speaker is talking about an object x that is familiar (intensionally defined), and can then start to look for its name. University of Antwerp (UIA) Romance Department 26 1 0 W ilrijk Belgium

Notes * Many friends and colleagues would have to be thanked for their constructive comments on earlier versions of this paper; not all of them can be mentioned here. We are particularly indebted to Pieter Seuren, and gratefully acknowledge remarks and comments by Seraphin Devriendt, Gi lles Fauconnier, Susumo Kuno, Jacques Noel and an anony­ mous referee of the Journal of Semantics. Of course, none of the above necessarily share all of the views expressed in this paper.

342

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The recoverability condttions of the absentee antecedent are of two kinds: some are pragmatic conditions on the salience of the real­ world object that the antecedent refers to, and some are linguistic conditions on the discourse contexts in which the pronoun that is con­ trolled by an absentee antecedent may occur.

LINGUISTIC CONTROL OF PRONOUNS For the terminology of pragmatic control vs. linguistic control, d. Hankamer and Sag ( 1 976). In brief, an item is linguistically controlled when it is anaphoric to a linguistic antecedent, and pragmatically controlled when it directly refers to an object or a situation in the world. In the present paper, we will be interested not so much in the nature of the two control relations, but rather in Identifying the two items that are linked by the pragmatic control relation and the linguistic control relation respectively. 2 L asnik's paper is the basis for the distinction made by Chomsky ( 1 980) between pronouns and anaphors. If our claim that pronouns are anaphors is correct, Chomsky's syntactic constraints would need to be reformulated. 3 In French, there are instances where coreferentiality of a pronoun to an NP is obligatory and syntactically conditioned; cf.: (i) Je pense de Jules qu' il est vraiment trop maigre . [ ' I think of Julius that he really is too thin ' ] ( i i ) Le meuble dont je crois q u ' i l conviendrait l e mieux, c'est celui-ci. [' The piece of furniture of which I think that it would fit best is this one '] These cases falsify the Lasnik-Chomsky hypothesis that only non corefer­ ence is syntactically conditioned. 4 The term "true pronouns" will be clarified below . Our class of linguistically controlled pronouns is not to be confused with what Bosch ( 1 982) calls syntactic pronouns (or pronoun uses); Bosch's anaphoMc pro­ nouns are also linguistically controlled pronouns in our sense. 5 Henceforth, we will use French examples mainly. However, most of our findings, though not necessar ily all, should be valid for languages like English too. 6 At f irst glance, one might object that such obviously deictJC items as ce­ lui-ci, celui-la fall in the wrong category. They indeed appear to undergo gender agreement, hence must be controlled by a linguistic antecedent: (i) (John order a k ilogram of appels (les p<m1mes, feminine plural) in a grocery store): Je voudrais un kilo de celles-ci, s ' i l vous plait. ['I would like one kilogram of these, please . ' ] I n the absence o f a linguistic antecedent, celui-ci, celui-la cannot be used. Suppose John is ordering one of those new kinds of exotic fruit one does not even know the name of: (ii) Je voudrais un kilo de *ceux-c i , *celles-ci, s'il vous plait c;:a, A straightforward analysis is that in celui-ci and celui-la. -ci and -Ia are deictics (just as the simple French adverbs lei 'here' and Ia 'there'), and celui-celle-ceux-celles are linguistically controlled pronouns. Without -ci or Ia these pronouns may indeed appear in unstressed position and need not be accompanied by a gesture (and, of course, they exhibit gender agreement): (iii) Celles qui coGtent le plus cher ne sont pas toujours les meilleures. [' Those that are more expensive are not always the best . ' ] 7 The French linguist Gustave Guillaume, correctly in our opinion, l

JS, vol. l , no.4

}

343

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

{

T ASMOWSKI-DE R YCK &. VERLUYTEN

"persons" "things"

direct complements

essential preposi tiona! complements

optional prepositional complements

clitics clitics

tonic forms tonic forms

tonic forms *

1 5 For the situational description it is important that the appearance of the unicorn be sudden and unexpected, so that the speaker does not presuppose its existence, not even in his dreams, at the time 344

JS, vol. 1 , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

called the forms of the lui-elle-ew::- elles paradigm nOWtS, and used the term existential pronouns for clitics; _cf. Moignet ( 1 965) Our notion of deixis is, as we see it, identical to that of Bosch ( 1 982). 8 There are reasons to think that the pragmatic control mechanism is acquired by children prior to linguistic control. Karmiloff-Smith ( 1 980) notes that younger children use pronouns deictically, and have them "agree" with the sex of the object they refer to rather than with the gender of the expression describing it. On the other hand, at this point, we cannot exclude the possibility that future research might show that even in these cases there is a linguistic antecedent (like homme in ( 1 7), for mstance). If this were so, forms of the Iui-elle­ ew:: - elles paradigm would be true pronouns, and the mechanism of pronominal control we propose would be completely general. At the moment however , the hypothesis involved is neither corroborated nor falsifiable as far as we can see. 9 In ( 1 9b) and (20b), the English translations show that deictics may exhibit what looks like number agreement in some cases. But this type of "agreement" with the number of objects observed in the real world should, of course, be distinguished from grammatical number agreement with a singular or plural noun. I 0 In a different context, the existence of cognitive antecedents is argued for in Kuroda ( 1 979). 1 1 This is not to say that the proper name of each object can always be readily identified. We used clear-cut examples where this is the case, but one may ask what will happen when, for instance, two words are about equally well established to name the same object (maybe this is the case for das Auto (neuter) and der Wagen (masculine) 'the car ' in German). In other instances, the proper name of the object may well depend on the speaker. A fisherman is probably more likely to call a trout truite (feminine) in French, whereas a layman in the field may call it poisson ' fish' (masculine). The question of the identifi­ cation of the proper name of each object requires detailed investigation which is beyond the scope of the present paper. 1 2 It should be clear that in (42), elles can only refer to females, not to obJects with a feminine name. 1 3 lei [ 'here ' ] is an adverb; voici and voilO are composed of the verbal imperative vois ['see ' ] and the adverbs ci ['here '], !a [ ' there']. 1 4 In summary:

L INGUISTIC CONTROL OF PRONOUNS of the utterance. Otherwise, (55) may become felicitous for the same reasons (57) is. 16 The notion of PREDpe asks to be refined. In some cases, it is a class of uses of predicates rather than a class of pregicates: effizierend verbs belong to PREDpe in their past tense use, but not in the present. Also, some predicates may belong to PREDpe for one of their arguments (say, their subject) but not for others. It is not our purpose to establish a list of predicates that belong to PREDpe; we only rely on clear-cut examples in this paper. 17 Investigations by Th. Ballmer on the semantic structure of verbs tend to corroborate this (communication at the Colloquium on Discourse Representation, Cleves, September 1 98 1 ). The lexicon would be composed of disjoint verb classes Ao, A 1 , , A n , where A1 implies the existen�e of class A 1 _ 1 • ln that manner, existential verbs are "ante­ rior• to other classes. 1 8 Of course, it is possible to exploit the resulting infelicity stylistically; for instance, in order to make the hearer curious and hence more eager to listen. Still, the obvious difference in felicity between ( 6 1 ) and (62) remains. 19 It should be clear that in our view true pronouns can only "refer" to an object in an indirect fashion: their referring is always mediated through a linguistic antecedent. •••

Bosch, P., 1 98 2: A�Jreement and Anaphora. Revised version of unpublished Dr. phil. thesis ( to appear at Academic Press , 1 983 ) Chomsky, N., 1 980: On binding. Linguistic Inquiry 1 1 , 1 ; 1 -lf6. Danell, K., 1 973: L 'emploi des formes fortes des pronoms personnels pour designer des chases en fram;ais modeme, U ppsala, Acta Universitatis Upsaliensia, Studia romanica upsaliensia 1 3. Evans, G . , 1 980: Pronouns. Linguistic Inquiry 1 1 , 2; 337-362. Fauconnier, G., 1 979: Mental spaces. A discourse-processing approach to natural language lo g ic. Unpublished paper, Universite de Paos Vlll, Vincennes (to appear at Routledge and Keagan Paul, London). Hankamer, J. & Sag, I., 1 976: Deep and surface anaphora. L inguistic In­ quiry 7, 3;39 1 -lf26. Karmiloff-Smith, A . , 1 980: Psychological processes underly ing pronom i­ nalization and non-pronominalization in children 's connected discourse. Papers from the parasession on pronouns and anaphora, Chicago Linguistic Society, Chicago; 2 3 1 -250. Kuroda, S.-Y . , 1 979: Concealed anaphora and pragmatic presupposition. Papers in Japanese linguistics 6; 1 77- 1 93. Lasnik, H., 1 976: Remarks on coreference. L inquistic Analvsis 2. 1 · l -22. Lyons, J., 1 977: Semantics, vol .II. Cambridge University Press, Cambridge. .

JS, vol. l , no.4

•

345

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Refer-ences

TASMOWSKI-DE RYCK & V ERLUYTEN

346

JS, vo1. 1 , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

McCawley, J., 1 970: Where do noun phrases come from? In: R. Jacobs & P. Rosenbaum (eds.), � eadings in English transformational gra m mar, Ginn, Waltham, Mass.; 1 66-1 83. , Moignet, G., 1 965: Le pronom personnel fran� is. Essai de psycho - syste­ matique historique. Klincksieck, Par is. Nunberg, G., 1 978: The pragmatics of reference. Indiana University Lin­ guistics Club, Bloomington, Indiana. Postal, P . , 1 966: On so-called pronouns in English. In: D. Reibel & Prentices. Schane (eds.), 1 969: Modem Studies in English. Hall, Englewood Cliffs ; 201 -224. Rorty, R., 1 980: Philosophy and the mirror of nature. Princeton Univer­ sity Pres'>, Princeton, New Jersey. Seuren, P., 1 979: The logic of presuppositional semantics. Mimeo, Nij megen University. Tasmowski-De Ryck , L., 1 973: Sujet-predicat vs. the me-commentaire. Zeitschrift fiir Romanische Philologie 89, 1/3; 88-96. Tasmowski-De Ryck, L. & Verluyten, S.P., 1 98 1 : Pragmatically controlled anaphora and linguistic for m . Linguistic Inquiry 1 2, 1 ; 1 53-4.

DISCOURSE AND DEIXIS

Barry Richards

Abstract

Introduction

Three distinctions might be seen to play a role in interpreting discourse. Where definite descriptions are involved, one might see ambiguity arising from the possibility of two different uses, the one referential and the other attributive. The latter would seem to be independent of contextual factors while the former is plainly not so: it is essentially deictic. A somewhat similar , though not identical, distinction is often said to apply to indefinite noun phrases. Such expressions can be used either specifically or non-specifically, and significantly the former is deictic while the latter is not. Pronouns too would seem to admit of a related ambiguity. They can be used either deictically or anaphorical­ ly, and again the latter, but obviously not the for mer, is context inde­ pendent. It is tempting to conjecture that there may be an interesting connection among the three 'non-deictic' uses, i.e. among the anaphoric use of pronouns, the non-specific use of indefinite noun phrases and the attributive use of descriptions. It is also tempting to think that these uses might have some special relation to their deictic counter­ parts, especially for indefinite noun phrases where the 'deictic' use is perhaps least tractable. We have argued elsewhere that there may be an important connection between the interpretation of certain anaphor ic pronouns and the JOU RNAL OF SEMANTICS, vol. l , no.4, pp. 347- 369

347

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

There is an inclination to think that pronouns, when they have singular indefinite antecedents, adm it of only two kinds of interpretation. They can be seen either as 'bound ' by their antecedents or as corefer­ ential with them. Which is appropriate on a given occasion will depend upon how the antecedent is used. We shall argue that neither account can be paradigmatic of the relation bet ween pronouns and singular indefinite antecedents, at least not as this relation is realized in dis­ course. 'Binding ' would seem to m isrepresent the typical structure of discourse; and indefinite noun phrases do not refer, despite U1e fact that people sometimes use them with some particular in m ind. We shall suggest an alternative account which satisfies some of the intuitive desiderata.

BARRY RICHARDS semantic characterization of attributive descriptions. 1 We suggested that when anaphoric pronouns have indefinite antecedents, i.e. antecedents that are indefinite noun phrases, they can sometimes only be interpreted as involving a special sort of restricted generalization. We also noted that this same phenomenon can be seen to be involved in attributive descriptions. For such cases we observed that there is no requirement that the description be uniquely satisfied, only that its range of quantifi­ cation be non-empty. Below we shall indicate how this reading is to be characterized and how it applies to pronouns. Moreover, we shall explore its role in the interpretation of discourse.

To focus the range of speculation it may be helpful to consider the following discourse. ( I ) Halfway along Princes Street at the foot of the Mound you will see an interesting neo-classical building. (2) It was commissioned to house a collection of paintings and its design was meant to reflect the importance of the holding. (3) This can be plainly seen, even though the building is rather small. (4) Its special interest, however, is to be found elsewhere. (5) It was intended, among other things, to stand as a continuing indictment of the design of its i mmediate neighbour. (6) And this too can be plainly seen. There is strong tendency to interpret this discourse as being about a certain building, e.g. one standing at the foot of the Mound in Edin­ burgh. It would seem natural to say that this building is, in effect, introduced into the discourse by the phrase an interesting neo-classical building, albeit indefinitely, and that it forms the locus of all the assertions. Significantly, however, there are two interesting neo-classical build­ ings standing at the base of the Mound, one just in front of the other . This may lead one to ask which of the two is indefinitely referred to. The matter is not inconsequential. If reference is to the National Gallery, all of the assertions will be true; but if it is to the Royal Scottish Academy, at least the penultimate one will be false. 348

JS, vol. l, no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

No one would seem to doubt that pronouns can be either deictic or anaphoric. 2 Few have conjectured, however, that they m ight occasion­ ally be both. We shall argue that there is in fact a 'hybrid' use of pro­ nouns, one that is simultaneously deictic and anaphoric. Once this phenomenon becomes clear the specific use of indefinite noun phrases would seem to become less problematic. Strictly speaking, there is really no such use; indefinite phrases are used univocally as existential quantifiers. Nevertheless, the aspect of specificity is not chimerical. It arises as a consequence of anaphora: when a pronoun is tied anaphor­ ically to an indefinite phrase, the phrase becomes 'specific ' if and only if the pronoun is also used deictically. This at least is what we shall propose.

DISCOU RSE AND DEIXIS

We shall argue that this interpretation, although perhaps possible, cannot be paradigmatic of discourse. We shall also argue that the given discourse does in fact admit of a 'specific' interpretation, one that allows for a certain referential ambiguity. We shall attempt to explain this as arising from deixis. As a consequence, however, we seem bound to accept that there must be a third reading. This would seem to result from the need to interpret the discourse as about some­ thing. What it appears to assert might, in general, be characterized in this way: whatever building happens to satisfy the first assertion is one that satisfies all the others. The fact that no building is deictical­ ly picked out does r.ot necessarily alter this basic structure. But it does require the discourse to be reinterpreted, or so we shall argue. The 'boUlld ' interpreta tion

Let us begin by looking at the 'bound ' interpretation. Henceforward it will be convenient te refer to the discourse as D. As we have noted, the 'bound' reading of D arises as a consequence of taking the phrase an interesting neo-classical building to be an existential quantifier phrase that binds all the occurrences of the neuter pronoun. Since the binding is inter-sentential, this analysis cannot of course be rendered within first-order syntax. Nevertheless, an ad hoc adjustment can be invoked, one that may in fact prove to be generalizable to cover more complex cases. Basically the idea is to introduce the notion 'of a discourse representation and then to redefine the concept of truth over it. For our purposes it will be sufficient to illustrate the strategy and leave a more formal characterization for another time. In constructing a representation for D we shall use all the resources of first-order syntax, except the quantifiers all and some. These will be handled in a rather special way; that is, they will not be represented syntactically but will be introduced only at the level of the truth definition. Accordingly, the first sentence of the discourse will be JS, vol. l , no.4

349

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Some philosophers would claim that there is really no ambiguity to be resolved here because there is no such thing as indefinite reference. The quantifier phrase an interesting neo-classical building is to be ana­ lysed in the usual first-order way and thus as not referring at all. Its role in the rest of the discourse is to bind all the occurrences of the neuter pronoun. On this 'bound ' interpretation the discourse is no more about one of the buildings than it is about the other. There is, as it were, only an illusion of aboutness. Of course the discourse can still be seen to be true of an object, for there is something which · has all the properties distinguished. But it would be a mistake to think that this is sufficient to yield the intuitive idea of aboutness. On the 'bound ' reading there is no object which is introduced in the first assertion, and hence none that the other assertions can be said to be about.

BARRY R ICHA RDS rendered by a formula which is its first-order 'translation ' minus the quantifier. We shall adopt the de vic� of using quasi-first-order notation. ( I 1 ) Halfway along Princes Street at the foot of the Mound you will see ;r and ;r is an interesting neo-classical building. Here the two occurrences of ';r ' are to be regarded as free variables and hence as available for binding.

(2 1 )

;r

was commissioned to house a collection of paintings and the design of ;r was meant to reflect the importance of the holding.

Since we are only concerned with the import of anaphora, we need not analyse (2) any further. Among the puzzling aspects of D is the role of the phrase the building in (3). Does it function as a description (as might appear from the definite article) or is it some kind of anaphor for an interesting neo-clas­ sical building (as might appear from its semantic role)? If the latter, the question of reference need not arise since the building would be ap­ propriately treated as having the same status as the occurrences of the pronoun, in effect as being a 'bound ' variable. If the former, how­ ever, the building must be regarded as a referring expression and hence as having a semantic role which is independent of that of the pronouns. This would allow a curious anomaly to arise in D: it would admit the posstbility that the referent of the building is not one among the objects that satisfy ( I 1 ) and (2 1 ). Although some building may satisfy these two formulas, it need not be the one denoted by the building. Since such a possibility can only be seen as counterintuitive, we shall assume that the building must be regarded as having the same semantic role as the pronouns, in the circumstance as being similarly 'bound'. Accordingly (3), (4) and (5) will be represented respectively in the following way. (3 1 ) This can be plainly seen, even though ;r is rather small. (4 1 ) The special interest of ;r , however, is to be found elsewhere. (51) ;r was intended, among other things, to stand as a continuing indictment of the design of the immediate neighbour of ;r. Since (6) contains no relevant occurrence of anaphora, it can be taken to represent itself. The set of formulas consisting of ( 1 1)-(51) and (6) will be said to be a 350

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The 'binding' relation between an interesting neo-classical building in ( I ) and the two pronouns in (2) can be captured simply by using the same variable in the representation. Apart Irom a small stylistic change it is sufficient just to replace the pronouns with the variable.

DISCOURSE AND DEIXIS representation of D; we shall call this set DR 1 • It remains now to cap­ ture the fact that all the occurrences of '.r ' are to be read as bound by the same existential quantifier. To render this we need only redefine the concept of truth, specifying it for discourse representations rather than single sentences. The appropriate clause can be exemplified as follows: the discourse D will be true in a model if and only if there is some object in the domain of the model which satisfies all the formulas of DR 1 • Since the occurrences of '.:c ' in DR 1 are aile free, any assignment from the domain will be uniform throughout. G iven that there is at least one such assignment satisfying all the members of DRI , D is then true.

One of the apparently fundamental features of a discourse is that each of its constituent assertions is available for denial. Although a discourse may be held together by certain principles of coherence, it would seem that an interlocutor can justly envisage querying any single assertion. In D, for example, one who would wish to dispute the truth of ( 5) would not seem to be challenging any canon of interpre ­ tation. He might appropriately interject, although perhaps pedantically, ( 7) It wasn 't intended to stand as a continuing indictment of the design of its immediate neighbour. Here the denier 's intention would seem plain enough: he wishes to deny of some entity what (5) asserts of it. Moreover, he wishes to do so without calling ( 1)-(4) into question. In the circumstance it would seem appropriate to represent (7) as

JS, vol. l , no.4

35 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In the case of the actual world we can assume that the National Gallery will be a truth-generating assignment. Hence, as would befit the existential reading, it is irrelevant to the truth of D that the Royal Scottish Academy is not similarly so, given that it fails to satisfy (5 1 ). The narrator of D is only comm i tted to there being at least one such assignment and thus, relative to the actual world he can rightly claim to have told the truth. Many would find this consequence to be intuitive, if not indubitable. Nevertheless, there is reason to think that it may not be entirely compatible with one's natural understanding of D. Intuitively one is disposed to interpret D as being about some particular entity, one that is the locus of all its assertions. On the 'bound' reading this can only be seen to be an illusion. What DR I is true or false of is not appropriate to determine what D is about, since DR 1 is technically about everything that falls withing the range of its free variables. Thus, D is no more about the National Gallery than it is about the Royal Scottish Academy. Were the intuition of aboutness merely an isolated phenomenon, it might perhaps be dispensed with. However, it seems to be connected with another 'deeply embedded ' intuition.

BARRY RICHARDS (7,)

::r.

was not intended to stand as a continuing indictment of the design of the immediate neighbovr of ::r..

To realize this on the ' bound' reading it would seem that the denier viz. must be interpreted as involving a clai m about the whole of DR ' that it is unsatisfiable. When he asserts (7) as a denial of (5), h e must be taken as saying not merely that DR � is satisfiable but also that DR1 is not. To do so would rule out the possibility that both speakers 'could be telling the truth. However, it would not disallow that they might both be speaking falsely: it is possible, relative to a given model, that neither DR 1 nor DR � is satisfiable. Hence again, (7) remains to be interpreted as the contradictory of ( 5), in effect as denying of an object what (5) asserts of it. Havi ng failed to capture this intUitiOn, one may be inclined to enter­ tain either of two doubts. One may wonder, on the one hand, whether (7) has been misconstrued. Perhaps its interjection was not intended to contradict the assertion of (5). Indeed, on the 'bound ' reading of D such an intention would seem to be anomalous; this interpretation may be seen to i mply that D can only be regarded as forming a narrative from which no assertion can be singly extracted, at least not for dispute. Hence, the utterance of (7) appears to be inappropriate, and to that extent infelicitous. One who would wish to argue to the contrary is bound to show how the assertion of (7), in the context of D, can be plausibly interpreted. On the other hand, one might wonder whether there is truly a 'bound' reading of D. Two related points would seem to stand against it. First, it implies that there is nothing in particular that D can be said to be about, despite an intuition to the contrary. Second, there is no apparently natural way to construe (7) as the contradictory of (5), despite a strong intuition to do so. Still the 'bound' reading is not

352

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Clearly, if (7 1 ) were added to DR 1 , the result would be contradictory: there can be no object satisfying both ( 5 1 ) and (7 1). The question, however, is whether this is sufficient to capture the intuitive content of the denial. Given that there is a real contradiction between (5) and (7), one might expect that it must be possible to represent the narrator as saying something false if and only if the denier is represented as saying some­ thing true. To this end one might consider resorting to two discourse representations, one for the narrator and the other for the denier. Since DR 1 represents the former, it would seem natural to suppose that the result of replacing (5 1 ) with (71) will represent the latter. But it can be readily seen that this set of for mulas, call it DR � , is not incom­ patible with DRI: they can be jointly satisfied in the same model (although of course not by the same objects). In fact, the Royal Scottish Academy, while not satisfying DR I , does satisfy DR l . Strictly speaking then, D R I and DR � do not represent the narrator and denier as making contradictory assertions.

DISCOU RSE AND DEIXIS without intuitive support. Many would regard the narrator of D to have told the truth in the world as it is, and the 'bound' interpretation would seem a natural way to justify their intuition. As for the balance of intuitive points, one seems bound to admit that the situation is indecisive. The 'specific' interpretation

This interpretation conforms paradigmatically to the intuition that D is to be read as about something. Moreover , it respects the intui ti \£e feeling that each of the assertions in D is available for dispute. One who would intend to contradict (5) by interjecting (7) would succeed, provided that all the pronouns coincide in reference. Clearl y , (5) and (7) can only be understood as contradictory when the pronouns are coreferential, notwithstanding the parenthetical phrase a mong other things. While few, if any, would doubt that D can have a 'specific' reading, some would query that it can be traced to the referential status of an interesting neo-classical building. H ow , . they might ask, can this phrase be seen to pick out a particular individual? In the context we have envisaged, there are two buildings that can be taken to satisfy the property of being an interesting neo-classical building; hence, more is required to discriminate the referent. Here it is natural to say that the narrator ' s intention is the additional, and decisive, factor. If he intends to refer to the National Gallery, then that is what the indefinite phrase refers to; and alternatively, if he intends to refer to the Royal Scottish Academy, that is its referent. This use of the indefinite phrase may be seen to be similar to the referential use of a definite description. It will be recalled that when a description is used referentially, the referent is fixed as a function of the speaker ' s intention. His intention, Donnellan argues, is pivotal in that it may not only supplement but override semantic factors: even though an intended referent may fail to satisfy the description, JS, vol. l , no.4

353

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

To· gain some additional perspective let us consider another reading of D. Many would be inclined to say that the occurrence of an interesting neo-classical building in ( 1 ) is perhaps most naturally (although not necessarily) interpreted as some sort of referring expression. What it can be seen to do here is introduce a certain interesting neo-classical building into the discourse, and to do so by referring to it. Thus, ( I ) i s not t o be understood a s having the form of a n existential assertion but rather as being a singular statement. As a result, the anaphoric connection between the indefinite noun phrase and the subsequent occurrences of the pronoun is to be interpreted in terms of coreference, and not ' binding' . The semantic role of the pronouns is to pick up the referent of an interesting neo-classical building.

BARRY RICHARDS the speaker 's intention may be sufficient to secure it as the referent. If we were to adopt an analogous view of the referential use of indefi ­ nite phrases, the logical f o r m of ( I ) would be represented in the follow­ ing way, with 'b' as a constant term. ( 1 1) Halfway along Princes see b.

Street at the foot of the Mound you will

It will be noted that ( I I) does not imply that b is an interesting neo-clas­ sical building. There is no semantic, nor indeed pragmatic, requirement that it be so.

Donnellan suggests that this is just the situation when definite de­ scriptions are used referentially: they are 'semantically free' and thus have the status of deictic pronouns. It is significant that the conditions of appropriate use can be seen to be similar in the two cases, i .e. in at least some circumstances. Not untypically, the role of a descrip­ tion; when used referentially, can be served as well by a deictic pronoun. The factors which allow the referent to be identified will be uniformly pragmatic, and will depend essentially upon what the speaker believes about the circumstances of his inter locutor. How far the use of deictics and referential descriptions actually coincide is of course a moot point. Nevertheless, they do seem to have a common semantic role, viz. to refer purely, and in both cases reference is determined wholly by pragmatic factors. The suggestion that indefinite phrases may also have a purely refer­ ential use would seem less plausible and thus more difficult to defend . The issue is whether it is ever right or necessary to regard non-specif­ icity as a defeasable aspect of indefinites. Intuitively, i t would seem that indefinites are by their very nature non-specific. Consider the following example. (8) A member of the Botany Department smuggles drugs. If the Botany Department were to have many members, one who asserted (8) would not normally be taken to have referred to any particular 354

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Such an interpretation of indefinite reference may, however, seem rather extreme. While the property of being an interesting neo-classical building is not suff1cient to discriminate the referent, one might argue that it must at least be a necessary condition of any admissible referent. If the speaker 's referential intention were wholly unconstrained by the semantic content of an interesting neo-classical building, the phrase would have the status of a purely deictic expression, e.g. like the demonstrative pronoun that. What that refers to on a given occasion de­ pends solely upon what the speaker intends to refer to. The expression places no semantic restrictions on what the speaker can intend to refer to and hence on the range of admissible referents.

DISCOU RSE AND DEIXIS member, even though one may have had a certain individual in mind. Should one have had somebody in mind, it would be natural to conclude that one had chosen to say Jess than one believed. Indeed, this would doubtless have been the point of having used the indefinite phrase a member of the Botany Department, rather than some name or definite description.

The question remains, however, whether this would amount to a special use of the indefinite phrase. Would it have been used in a way which would warrant the conclusion that it is ambiguous? Surely not. We need only note that no adjustment in the analysis is required to take account of the peculiar state of the world and one 's knowledge of it. The standard f irst-order account, where a member of the Botany Department is treated as an existential quantifier phrase, appears to characterize the truth conditions appropriately. It says that (8) will be true if and only if there is some member of the Botany Depart­ ment and he smuggles drugs. Clearly, the proposition specified in the definiens respects what could only have been the intention in asserting (8), viz. the intention to avoid implying specificity, despite one ' s knowledge. Analy sing a member of the Botany Department as a re­ ferring expression would lose this essential ingredient of its use. It would require (8) to have the logical structure (9) b smuggles drugs

and hence to be true if and only if b smuggles drugs. In the circumstance it would be a mystery why one had chosen to use an indefinite phrase, unless of course one had simply m isused the phrase. The mystery would become more acute if we were to assume, as we have not yet done, that one had actually intended to refer to b in as­ serting (8). Such an intention might, in the context, be thought to override the conventional meaning of the indefinite phrase. That is, it might be thought to be sufficient to render a member of the Botany Department as a referring expression, perhaps with a status not unlike that of a deictic expression. But why, we must ask, would one have JS, VQI . l , no.4

355

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

But now Jet us suppose that the Department were to have only a single member. One might then be said to have referred to him in asserting (8). Since the indefinite phrase would be uniquely satisfied, there can be no ambiguity and hence reference would be as definite as if made by a description. Of course one might legitimately complain that it was no part of one 's intention to refer definitely. That the world would contrive to distinguish a particular person cannot ' alter the fact that one had used the indefimte phrase with the intention of being indefmite. On the other hand, if one were to know that the Botany Department had just one member, there may be a strong tempta­ tion to say that one could only have intended to make specific reference, i .e. to use a member of the Botany Department referentially.

BA RRY RICHARDS

The analysis we are envisaging is one that would embrace two, apparently incompatible, claims. First, there is the claim that the indef inite phrase an interest ing neo-classical building cannot be a refer­ rmg expression, and hence must be an existential quantifier-phrase. Second, there is the suggestion that when D is read 'specifically ' , the pronouns are used deictically. Prima facie these two suggestions seem to generate a problem. To capture the coherence of D there must be some way of interpreting the anaphoric connection between an interesting neo-classica l building and the succeeding pronouns. Given that the latter occur referentially and the former does not, there is an obvious incom mensurability. The connection cannot be interpreted in terms of bindmg, because the pronouns do not have the status of variables; nor can it be interpreted in terms of coreference, since the indefinite phrase is non-referential. The solution is to be found in how deixis is restricted by anaphora. Typical ly, the range of objects to which a deictic expression might be correctly applied is not semantically restricted. That is, a deictic can be used to refer to whatever one would have it refer to, independent of any semantic constraints arising from the language. What a deictic refers to is usually a pragmatic matter, not a semantic one. But this is not always so. Someti mes deictic pronouns are restricted in the range of objects they can be correctly applied to, in particular when they are tied anaphoncal ly to an indefinite phrase. Assuming that the occurrences of it in D are used deictically (as they will be on the 'speci fic ' reading), we would argue that they will be used correctly only if the intended referent is an interesting neo-classical building. Because each of the occurrences is related anaphorically to an interest­ ing neo-classical building, its proper use is to pick out one among the entities that satisfy this property. 356

JS, vo! . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

chosen to be so careless of its conventional meaning? Given that one had mtended to refer specifically, it can be assumed that one could have chosen to use an appropriate expression, i.e. one whose conventional meaning determ ines It as specifically referential. Indeed, one can only have intended to use the indefinite phrase referentially if one could al­ ternatively have used a definite phrase to do so. This would suggest a most curious paradox, namely that one had decided to flout convention to do just what one might have done conventionally. Fortunately, unl ike some paradoxes, there is an easy way with this one. We need only concede that indefinite phrases do not refer, notwith­ standmg any intention to the contrary. Such a concession need not, moreover, force us to give up the 'specific' reading of our discourse D. To be sure, it would require that we abandon the idea that sentence ( l ) can be interpreted 'specifical ly'. But this may be of no consequence smce the source of specificity can be seen to lie elsewhere. It may be attr ibuted to the pronouns which occur in the succeeding sentences. That is, D can be understood to acquire a 'specific ' interpretation when, and only when, the pronouns are used deictically.

DISCOURSE AND DEIXIS The implication is that the pronoun occurs uniformly with a dual role. It is at once ' bound ' by its antecedent and 'free ' to refer. The question is how to represent this situation formally. Let us consider adapting D R I . Suppose that we introduce a distinguished parameter 'u' in­ to our quasi-first-order language and treat it as reserved for deixis. That is, suppose we regard the denotation of 'u' to be determined accord­ ing to deictic intent. To capture the dual role of it, say in (2), we m ight then add the formula ' .r is identical to u ' as a conjunct to (21). ( 2 1)

.r

The amended discourse representation DR I will then substitute (21 ) for Since ( 1) contains no pronouns, ( 1 1) will go over unchanged, as will (6). Given that all the pronouns in D have a common antecedent and are to be read coreferentially, DR I can be filled out simply by adding the result of adjoining '.r is identical to u' as a conjunct to each of (� )-(51 ). Accordingly D will be true in a model only if the denotation of ' u ' satis­ fies all the open formulas in DRI. Since ' .r is an interesting neo-classical building ' is among them , the desired restriction on 'u' seems to be real­ ized. It is not enough that 'u' denote; it must denote an interesting neo­ classical building.

(21).

It is arguable, moreover, that DR I respects the basic intuition that each of the assertions of D is available for denial. Consider the representation of (7) when it is asserted as a denial of ( 5). (71)

.r

was not intended to stand as a continuing indictment of the design of the immediate neighbour of .r and .r is identical to u.

To reflect the fact that none of the othef assertions of D is in question we let the situation be represented by D R * which is like DR I except in hav­ ing (71) in place of (51). It will be immediately apparent that DR and DR I cannot be both true in a model, even though the free varrables in each case are regarded as bound by different quantifiers. Since ' u ' has a fixed denotation in a model, there will be an assignment satisfying ( 7 1 ) if and only if there is no assignment satisfying. (51)

.r

was intended, among other things, to stand as a continuing indictment of the design of the i m mediate neighbour of .r and .r is i­ dentical to u.

And of course vice versa. Note that '.r is identical to u ' will always be sat­ isfied by some assignment in a model, because the denotation of 'u' must be drawn from the domain of the model. As a result, (5� and (71 ) are represented as strictly contradictory. Although 'u' will have a denotation in every model, there is no JS, vol. l , no.4

357

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

was commissioned to house a collection of paintings and the design of .r was meant to reflect the importance of the holding and .r is identical to u.

BARRY RICHARDS

Clearly the role which ( I) has as mtroductory utterance is significant. Indeed, th1s 1s so patently obvious that 1 t should go without saying. Hence, it is important to note that that DRI does not represent it as so. Since DRI is JUSt a set of formulas, the order in which they are listed is irrelevant. We can shuffle the order without effecting the identity of DR I . Thus, (I 1 ) cannot be seen as having any greater claim to repre­ sent the introductory assertiOn than any of the other formulas, except of course (6). To put it another way, ( 1 1) is not represented as having any preferred status over the other formulas in deter mining the range of subsequent deixis in D, i.e. in restricting what can be a proper referent of 'u'. For example, consider the subformula consisting of the first two conjuncts of (21 ) , in effect (2 1 ). From the point of view of DR L , (21) is as much a restnct10n on what is a suitable referent of 'u' as is ( 1 ;}: they are equally relevant to the truth conditions of DRI, which is the only status they are portrayed as having. For this reason failure of the denotation of 'u' to satisfy ( 2 1) IS represented as having the same im­ port as failure to satisfy ( 1 1 ). If the referent of 'u' were to satisfy (I 1 ) but not (2 1), the significance for D would be given to be the same as if the opposite were to be so, i.e. if the referent were to satisfy (2 1 ) but not ( 1 1). To this extent DR I misrepresents the structure and content of D. If we are to give a fair rendering of the special status of ( 1 ), we shall have to entertain a rather different view of discourse represen­ tations. We m ight discern what may be required by trying to paraphrase the i mport of the anaphor ic pronouns. For convenience take the pronoun 358

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

requirement that it be an interesting neo-classical building. It is not impossible that the denotation in .a given model will fail to be so and hence to satisfy ( I 1). In the c ircumstance DR I will be false and thus it might be said that D is false of the referent of 'u'. Would it be right to say , however, that D is about the referent of 'u'? There is an inclina­ tion to think not. We have noted that the role of sentence ( 1) , at least intuitively, is to introduce into the discourse a certain range of objects. We have supposed this to be determined by the indefinite phrase and thus to involve the property of being an interesting neo­ classical building. But this is not the only property mvolved. The set in­ troduced is actually fixed by both conjuncts of ( I 1); that is, it is the set consisting of those interesting neo-classical buildings which you will see halfway along Princes Street at the foot of the Mound. Why this should be so is something we shall try to explain later. For the moment it 1s sufficient to note the fact. Now if the denotation of 'u' is not a member of this set, say because it is not an interesting neo-classical building, there IS a sense in which D cannot be about it. The reason is plain enough to see: there would arise an incoherence in the inter­ pretation in which ( I ) will be isolated from the rest of the discourse. What ( 1 ) will be true of will be distinct from what the rest of the discourse may be true of. Here we are inclined to say that D cannot rightly be seen to be about any individual.

DISCOURSE AND DEIXIS in the first conjunct of

(2 °)

(2),

i.e. in

It was commissioned to house a collection of paintings.

We want to claim that it is used deictically here and that its use will be correct only if the intended referent satisfies ( l 1). We m ight real­ ize these desiderata in the following paraphrase: the interesting neo clas­

sical building which you will see halfway along Princes Street at the foot of the Mound and which is identical to u. Obviously this is not a

phrase we are likely to use in the context, hence the occurrence of the pronoun. Nevertheless, it does arguably capture just what we would mean to say. What we would intend to assert is that The interesting neo-classical building which you will see halfway along Princes Street at the foot of the Mound and which is identical to u was commissioned to house a collection of paintings.

Given that the is read as implying uniqueness, the description in will denote u if and only if u satisfies (i)

x is an interesting neo-classical building and you will see x hal fway

along Princes Street at the foot of the : Mound and x is identical to u.

Since u is the only thing that can satisfy (i), if u satisfies both (i) and (ii)

(2r)

(2�)

will be true if and only

x was commissioned to house a collection of paintings.

From this it can be seen that (2�) gives a 'strong' characterization of the anaphora linking (2°) and ( 1 ): (2�) can only be true if ( l ) is. Perhaps that is as it should be.

There is, nevertheless, a weaker characterization of anaphora which arises if we read the deter miner the, not as implying uniqueness, but rather in the sense of any. Lest �his be thought to be generally i mpossi­ ble, we would note that the sentence ( 1 0)

The man who dares to open Pandora's Box is foolish.

is perhaps most naturally read as meaning ( 1 1)

Any man who dares to open Pandora's Box is foolish.

To be sure, substituting any for the in (2�) may not yield a similarly typical reading; but then (2�) is not a typical utterance. It is sufficient for our purpose that the reading be entertained for argument, since ultimately nothing will depend upon the admissibility of so construing the. Accordingly (2�) will be true in a model on this interpretation if JS, vol. l , no.4

359

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(2�)

BARRY RICHARDS

and only if every assi§nment from the domain which satisfies (i) also satisfies (ii). Hence, (2 k ) can be true in a model even though ( I ) is not. If nothin& satisfies ( II ), then ( 1 ) is false, which in turn is sufficient to make (2 k ) true. Since (i) implies ( 1 1 ), it follows that there can be no assignment which satisfies (i) and not (ii), and thus (2r) must be true.

(5k)

The interesting neo-classical building which along Princes Street at the foot of the identical to u was intended, among other a continuing condictment of the design of its

you will see halfway Mound and which is things, to stand as I mmediate neighbour.

The anaphoric connection between (5) and ( I ) is, as before, evident in the descr iption, i.e. just as it is in (2�). Moreover, the link between (5) and (2 °) is recorded in the parameter 'u' , thus indicating that the two occurrences of it are used coreferentially. However, the matrix (ii) is not portrayed as having any status in determ ining what is a proper referent for the parameter. Although the referent of 'u' will have to sat­ isfy (ii) 1f (2°) is to be true, its status as a proper referent of 'u' is not given to depend upon this being so. Provided that ( 1 ) is true and that u satisfies (i) and thus ( 1 1), we would claim that D can rightly be said to be about u. To test this claim let us examine the import of denying (5) by asserting (7). We would expect the latter to be characterized as denying of u what (5) asserts of it. If we take the possessive its aside, (7) will be paraphrased as

(7 k)

The interesting neo-classical building which you will see halfway along Princes Street at the foot of the Mound and which is identical to u was not intended to stand as a continuing indictment of the design of its im mediate neighbour.

In both (5 k ) and (7 k ) we regard the description as havinj;( wider scope relative to negation. On the assumption that u satisfies (i) we can see that (5k) will be true if and only if (7 k) is false, and vice versa. In effect, we can see just what we would expect, that (7) is represented as denyinJ?: of u what (5) asserts of 1t. Of course 360

(5k) and (7k) are not

strictly contradictory, since they might JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This possibility notwithstanding, it can be seen that the 'weak' characterization of (2 � ) still captures the anaphoric connection between (2°) and ( 1 ). Since the relevant matrix underlyin � ( 1 ), viz. ( 1 1� also under­ lies the description in (2r), in that (i) implies ( 1 1), it is plainly represen­ ted as a condition restricting the range of proper referents for 'u'. That the matrix of the main clause of (2f), i.e. (ii), does not play a similar role can be seen if we consider (5J again. It will be helpful if we leave the possessive its aside for the moment. Given that the interpretation of it in (5) is to coincide with the interpretation in (2°), we must para­ phrase (5) in the following way.

DISCOURSE AN D DEIXIS be

true. Should u not satisfy (i), this would be sufficient to verify (7 k). It would also be sufficient to verify all the other para­ phrases involved in representing D, i.e. of the second conjunct of (2) and of (3) and (4). Note, however, that ( 1 ) may still be either true or false. This would indicate that ( 1 ) will be, in effect, semantically irrelevant to the rest of the discourse. What could be a more apt characterization of the breakdown of coherence in D than this: i f u does not satisfy (i), then (2)-(5) become true independent o f the truth of ( 1 ). This would imply that D cannot be seen to be about any individual in particular. Whatever ( 1) may be about, i t certainly cannot be what (2)-(5) are about. Surely this is a clear manifestation of inco­ herence. both

(5k) and

It is illuminating to examine the basic categorical structure of ( 1 ). According to Montague's treatment in PTQ ( 1 974) the final step in the construction of ( 1 ) would be that of inserting the phrase an inte resting neo-classical building in place of '.r ' in the matrix (iii) Halfway along Princes see .r .

Street at the foot of the Mound you will

The structure of the indefinite phrase would itself be analysed as the result of adjoining the quantifier a to the complex common noun interesting neo-classical building. From a semantic point of view this has the effect of representing a as having scope over the matrix (iv)

.r

is an interesting neo-classical building.

It is also represented as having scope over (iii) in virtue of the operation of term insertion. The analysis of the description used in the paraphrase would proceed in a similar fashion. The determiner the would be seen as adjoined to the common noun interesting neo-classical building which you will

see halfway along Princess Street a t the foot of the Mound and which is identical to u. The effect of this would be to represent the as having scope over (i). Despite the fact that the adverbial halfway a long Princes Street at the foot of the Mound is preposed in (iii) and postposed in

the second conjunct of (i), these two 'formulas are equivalent. As a JS, vo!. l , no.4

36 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The 'weak ' interpretation of the paraphrases would seem to y1eld all that might be asked here of an adequate characterization of ana­ phora. This is not to suggest that the 'strong' interpretation is inade­ quate. At the moment both might be seen to be equally acceptable. Nevertheless, there are reasons for preferring the weaker version, one or two of which we shall consider in the next section. More imme­ diately, however, we must try to explain why the particular description used in the paraphrases is the correct one. Intuitively it would seem to be so, but it remains to be seen how it can be structurally justified.

BARRY RICHA RDS consequence, we can perceive that the two matrices comprehended by the quantifier a are also comprehended by the determiner the. This, we would argue, is the structural manifestation of anaphora: since an interesting neo-classical building is the antecedent of it, they have con­ tent in common but they do not share a common determiner. The anaphora here might be said to be ' matrix' anaphora.

Having explained the origin of the matrices which characterize the content of it, we would seem bound to say something about the choice of determiner, i.e. about the. Among the issues to be resolved is whether the should be read in the sense of the uniqueness quantifier or the universal quantifier. Fortunately, we can sidestep this problem, and indeed all the problems associated with the, by opting for a neutral alternative. To this end we shall eschew natural language in favour of the resources of first-order logic. Let us introduce a new bit of logical notation. Where A is any formula of first-order logic, we shall stipulate that the result of bracketing A and prefixing a variable will be a term. Thus, given that x is variable, x[A] is a term. We shall regard any occurrence of x which is free in A as bound in x[A]. Semantically, however, we shall not treat x(A] as a denoting expression. It will instead be handled contextually by supplementing the truth definition (for sentences) with the following clause: where B(x[AD is any formula containing an occurrence of x[A], an assignment g will satisfy B(x[AD if and only if for any assignment g' which differs from g at most in what it assigns to x, if g' satisfies A then g' satisfies B(x). Here we adopt the syntactic convention of replacing the occurrence of x[A] in B(x[A ] by the prefixed variable x. If the schematic term had been y[A], then the appropriate open formula in the definiens would have been B(y), instead of B(x). It must be stressed that in all other respects the standard truth definition will remain the same; in particular, truth will be defined only for sentences and in the usual way. We can now proceed to build an appropriate discourse representation for D using the new notation. We shall continue to think of discourse representations as consisting of formulas, i .e. as being sets of formulas. H owever, the concept of truth will not be adjusted to cover discourse. It will be assumed to be defined only for sentences. The role of a d iscourse representation will be solely to characterize the inter-senten­ tial connections which arise in respect of anaphora and deixis.

362

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

But of course there is more to the ' matrix' content of it since the pro­ noun is used not merely anaphorically but also deictically. To reflect this it is necessary to add the conjunct '.r is identical to u'. We would em­ phasize that there is no syntactic justification for the addition, as there is in the case of the other two conjuncts. This is as we should expect since the import of deixis can only be rendered semantically.

DISCOURSE AND DEIXIS

·

Let DR 1 be our representation. Since the initial assertion of D contains no instances of anaphora or deixis, it can be taken to represent itself. In the case of ( 2) there are two anaphoric pronouns it and its. To capture their import we shall rewrite its design as the design of it and otherwise follow the analysis indicated in ( 2 � ); that is, we shall use the underlying matrix (i) to form an appropriate first-order term . The representation will replace each occurrence of it with the term (v)

x [x is an interesting neo-classical building and you will see x half­ way along Princes Street at the foot of the Mound and x is identi­

cal to u].

x [x is an interesting neo-classical building and you will see x half­ way along Princes Street at the foo t of the Mound and x is identi­

cal to u] was com missioned AND the design of x [x is an you will see x halfway along Mound and x is identical to u] of the holding.

to house a collection of paintings interesting neo-classical building and Princes Street at the foot of the was meant to reflect the importance

Despite the length of (2 1 ) its truth conditions can be spelt out briefly: (21) will be satisfied by an assignment g if and only if for every assign­ ment g ' differing at most from g on x if g' satisfies (i) then it satisfies both (ii)

x was commissioned to house a collection of paintings

and (vi)

The design of x was holding.

meant

to

reflect

the

importance

of

the

The fact that (v) occurs twice in (2 1 ) does not require that consideration be given to scope ambiguities and hence both occurrences can be regarded as having widest scope. It will not have g one unnoticed that (v) is treated here just like the description in (2 � ) when the is read as any. Since this analysis was seen to lead to an Interpretation of D which respected our fundamental intuitions about its structure, so too will the analysis based upon (v). In particular, (2 1 ) will be rightly taken to be about u if and only if u is an interesting neo-classical building which you will see halfway along Princes Street at the foot of the Mound. However, it will be clear that (2 1 ) represents each of the conjuncts in (2) as available for denial in just the way that they ought to be, i.e. in the way exemplified by the discussion of (5k) and (7 k ). JS, vol. l , no. 4

363

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Since the resulting formula is excessively long, we shall indicate the main connective and by capitalizing it.

BARRY RICHARDS Having illustrated how the pronouns in (2) are to be handled, i .e. how they give rise to (2 1 ), we need only say that the other occurrences in D will be treated in a sim ilar way. Since they all have the same antecedent, they will be replaced uniformly by (v). In the case of (3) we noted above that the building, although not strictly an anaphor , is perhaps best treated as one and hence this too will give way to 1 (v). The operation of the substituting (v) in D will map D into DR whose members jointly characterize the 'specific' reading of the discourse. Since it would seem rather tedious and unnecessary to display the remaining members, viz. (3 1 )-(5 1 ), we sha:J leave it to the reader.

We are now in a position to consider how D is to be interpreted when the pronouns are used in a purely anaphoric way. We shall say that a pronoun is so used if it lacks any deictic role, in effect, if its seman­ tic content is determined solely by its antecedent. Clearly, there is no requirement that the pronouns of D have any deictic significance; and in the absence of deixis their semantic import can only be fixed by anaphora. Given that they all have an interesting neo-classical build­ ing as their antecedent, and that this is also the 'antecedent' of the building in (3), we might naturally conjecture that the requisite interpretation of D, i.e. the 'attributive' one, will differ in only one respect from the 'specific' reading: it will invoke no deictic condition of the form 'I is identical to u' . That is, the characterization of the 1 'attributive' reading, let us call it DR m , will be exactly like DR except in having a shorter substituent term, i.e. instead of (v) (vii) I [I is an interesting neo-classical building and you w ill see I half­ way along Princes Street at the foot of the Mound]. The justification for (vii) will derive, as in the case of (v), from the scope of the determiner of the antecedent. Here the determiner will comprehend all that is semantically relevant since the pronouns will have no deictic role. It is interesting to note what DR m will portray D as being about. In the circumstance we have envisaged, there are two buildings that satisfy the matrix underlying the initial assertion, i .e. the matrix Halfway along Princes Street at the foot of the Mound you will see I and I is an interesting neo-classical building. These two buildings, namely the National Gallery and the Royal Scottish Academy, can be regarded as having been introduced into D by ( 1 ). That is, insofar as ( 1 ) introduces anything, it can only be seen to intro­ duce them both. It is obvious, moreover, that (vii) will not discrim inate between them since the matrix in (vii) is equivalent to ( 1 1 ). Thus (vii) 364

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The 'attributive ' interpretation

DISCOURSE A N D DEIXIS will carry both objects forward as being jointly the focus of D. Each assertion, with the exception of (6), will be represented in DR m as mak­ ing a claim about all the objects that satisfy ( I 1), in effect about both the National Gallery and the Royal Scottish Academy.

In this connection it may be useful to examine the i mport of DR m in respect of denials. Let us again take (5) as an example, which is repre­ sented in DR m as

x [x is an interesting neo-classical building and you will see x half­

way along Princes Street at the foot of the Mound] was intended, among other things, to stand as a continuing indictment of the design of the immediate neighbour of x [x is an interesting neo­ classical building and you will see x halfway along Princes Street at the foot of the Mound]. Since the two occurrences of the substituent term in (5 m ) can only have widest scope, (5 m ) will be true in a model if and only if every assign­ ment from the domain which satisfies (viii) x is an interesting neo-classical building and you will see x halfway along Princes Street at the foot of the Mound also satisfies ( 51)

x was intended, among other things, to stand as a continuing indictment of the design of the immediate neighbour of x.

JS, vol. l , no.4

36 5

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This may be seen to be somewhat paradoxical in view of the fact that all the relevant ter ms in D are singular in number. Does their singularity not restrict D to adm itting only singular interpretations? . We would argue that the restriction, if there is one, can only be one of implicature. Although one m ight be naturally disposed to think of D as invoking a single object, there is no semantic requirement that there be just one relevant entity. Should one discover that i n fact two entities have been introduced b y ( 1 ) , one would have n o seman­ tic grounds for complaint as long as both objects have all the properties ascribed in the other assertions. The narrator has committed himself to there being at least one such object. In the event of there being two, one can hardly be justified in accusing him of having spoken falsely. After all, he is supposed to have used the pronouns 'attributive­ ly' and hen�e may not have been in a position to give an undertaking of uniqueness. What better way for him to signal this than by opting for an indefinite phrase, thereby confining his commitment to the minimal claim that there is at least one appropriate object. Singularity is to be taken to indicate a lower bound, not an upper one. That this may sometim es lead to a presupposition of uniqueness is not a semantic phenomenon. It can only arise as a product of pragmatic considerations.

BARRY RICH A RDS How then are we to interpret the denial of (5), i.e. the assertion of m ( 7)? It is represented in DR as (7 m> ;r [x is an interesting neo-classical building and you will see�;r half­ way along Princes Street at the foot of the Mound] was not intended to stand as a continuing indictment of the design of the i mmediate neighbour of ;r [x is an interesting neo-classical building and you will see x halfway along Princes Street at the foot of the Mound].

x was not intended to stand as a continuing indictment of the design of the immediate neighbour of x.

Clearly (7 m ) , when so read, will be false in the particular model we have in m ind, i.e. in the real world: although the National Gallery satisfies (vii i), it fails to satisfy ( 7 1). This does not imply, however, that ( 5 m ) is true, for the Royal Scottish Academy, which satisfies (viii), does not satisfy (5 1 ). Hence, (5) and (7) are not strictly represented as contradictories, but rather as contraries. When the negation operator is read as having widest scope in (7 m >, the situation is a little different. So interpreted, (7 m ) will be true in a mod­ el if and only if it is not the case that every assignment which satisfies (viii) also satisfies (51). Accordingly (7 m > will be true in the intended model: since the National Gallery falsifies ( 5 m ) , it follows that not every assignment satisfying (viii) satisfies (5 1 ) , and this is sufficient to verify ( 7 m>· It will be noted, moreover, that (5 m ) and (7 m ) are now represented as strictly contradictory, and not merely as contraries. Since their relation here can be seen to generalize to the representations of (2)­ m (4) and their denials, it is apparent that DR respects the intuition that each of the assertions of D is available singly for denial. In passing we would draw attention to what the effect of denying ( 1) would be. Suppose one were to assert, in replying to ( ! ),

( 1 2)

That 's not so.

Given that the intention is to deny the truth of ( 1) , we would represent the import of ( 1 2) in this way. ( 1 2 m ) It is not the case that there is an x such that halfway along Prin­ ces Street at the foot of the M ound you will see x and x is an in­ teresting neo-classical building.

366

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Note that from the point of view of first-order analysis ( 7m) is structural­ ly ambiguous: the negation operator can have either wider or narrower scope relative to the substituent terms. When it has narrower scope, (7 m> will be true in a model if and only if every assignment from the domain that satisfies (viii) also satisfies

DISCOURSE AND DEIXIS If ( 1 2 m > is true, one would expect that the rest of D should be repre­ sented as vacuous, since the discourse cannot be m seen to be about anything in particular. In fact, this is just what DR w o u l d i m ply. The truth of ( 1 2 "1 ) entails that each of (2 m )-( 5 m ) is true but in the circum­ stance it w1ll be vacuously true since nothing satisfies (viii). When this is so, the assertions (2)..(5) can only be seen as aimless, i.e. as themselves vacuous. Having placed such a lot of emphasis on negation, we seem bound to consider the following kind of discourse. Suppose that Tom says, Halfway along Princes Street at the foot of the Mound you will see an interesting neo-classical building.

Dick continues, ( 1 3) I'll bet that it was designed by Playfair. And then Harry interjects, ( 1 4) I'll bet that it was not. On

the assumption that it is used purely anaphorically and ( 1 4) we would represent them as follows.

in

both ( 1 3)

I'll bet that x [x is an interesting neo-classical building and you will see x halfway along Princes Street at the foot of the Mound] was designed by Playfair. I'll bet that x [x i� an interesting neo-classical building and you will see x halfway along Princes Street at the foot of the Mound] was not designed by Playfair. Let us suppose that the substituent term occurs within the scope of the opaque operator I'll bet that . It can be immediately seen that ( 1 3 n ) will render Dick's bet as false. Since Playfair did not design the Royal Scottish Academy, not everything which satisfies (viii) also satisfies (ix)

x was designed by Playfair.

This is not sufficient to imply, however, that Dick has lost the bet to Harry. If the negation operator in ( 1 4 n ) is read as having narrower scope relative to the substituent term, Harry 's bet is also false. Since Playfair did design the National Gallery, not everything which satisfies (viii) also satisfies (x)

x was not designed by Playfair.

In

the circumstance, neither Dick nor Harry would be entitled to claim the wager.

JS, vol. l , no.4

367

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1)

BARRY RICH ARDS If, on the other hand, the neF.ation operator is taken to have wider scope, Harry will then be portraye_d as having won the bet. He will have bet that it is not the case that everything which satisfies (viii) satisfies (x); and in this he will be right. There is a question, however, as to whether ( 1 4) can actually be read in this sense of ( l 4 n). Some would argue that if Harry were a gentleman, he would never claim to have won the bet. Indeed, he might well be disposed to concede that Dick has won the bet, semantics notwithstanding. After all one of the build­ ings was designed by Playfair and it may seem rather unsportsmanlike to be too literal minded. What a gentleman should do should not neces­ sarily be dictated by semantics.

Edinburgh University Edinburgh, Scotland

Notes

1 See Richards 1 982 and 1 983. 2 In an effort to forestall confusion we shall characterize the distinction between anaphoric and deictic pronouns in the following way. An anaphoric occurrence is one whose interpretation is determined solely by its antecedent, where its antecedent is some linguistic expression either in the sentence or in the d iscourse. To say this is not of course to indicate how the interpretation is to be rendered. Nevertheless, it does rule out contextual factors as relevant. Once the antecedent is fixed, so too is the interpretation, whatever it may be. In contrast, a deictic pronoun is one whose interpretation is essentially dependent on context. Deixis is a referential phenomenon where the object of reference is determined at least partly by contextual factors. Which are the relevant factors may not be entirely clear but that there are some is beyond dispute. 3 For discussion I should like to thank my recent classes in formal semantics, Keith Brown, Kit Fine, Gary Matthews, Han Reichgelt and Nigel Shadbolt.

368

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

But surely semantics cannot be irrelevant. Perhaps a fairer way to characterize Harry's situation would be as follows: it would be unsportsmanlike of him to claim the wager since the 'winning' reading of ( 1 4) is not unique, nor e ven preferred. The proper thing to do is to take the less favourable reading, which in this case renders his bet as false. As a consequence, neither he nor Dick will be justified in claiming the wager. To be fair Dick too must concede to the less generous interpretation of his bet, viz. the one which is semantically dictated. 3

DISCOURSE A N D DEIXIS Refer-ences

Philosophy).

JS, vol. l , no.4

369

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Donnellan, K . , 1 966: Reference and Definite Descr iptions. Philosophical R eview 75; 28 1 -304. Evans, G., 1 977: Pronouns, Quantifiers and Relative Clauses I. Canadian Journal of Philosophy 7; 467-536. Hintikka, J. & Carlson, L., 1 978: Conditionals, Generic Quantifiers, and Other Applications of Subgames. In: Saarinen, E. (ed.): Game-Theoretical Semantics. D. Reidel, Dordrecht. Kamp, H . , 1 980: A Theory of Truth and Semantic Representation. Manuscript, Bedford College, London. Montague, R., 1974: The Proper Treatment of Quantification in Ordinary English. In: R. Montague: Formal Philosophy (ed. R. Thomason). Yale UP, New Haven. Richards, B., 1 9 79: Pronouns, Reference and Semantic Leziness. In: Dancy, J. (ed.): Papers in Language and Logic. Keele; reprint­ ed in: Heny, F., (ed.): A mbiguities in Intensional Contexts. D. Reidel, Dordrecht. Richards, B., 1 982: Anaphora, Descriptions and Discourse Representa­ tions (to appear in S)'lthese). Richards, B., 1 983: On Interpreting Pronouns (to appear in L inguistics and

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

MENTAL MODELS AND DISCOURSE

Han Reichgelt

Abstract

Introduction: S-R Semantics

Shadbolt and Reichgelt (forthcoming) present a descriptive system which we feel enables us to describe a number of referential phenomena more accurately. In this paper I will use this system, let us call it S-R Semantics, to try to solve a number of the problems raised by Barry Richards (this Journal, pp. 347-369). S-R SemantiCS has been developed within a general approach in semantics that has become popular over the last ten years or so. The basic assumption is that using natural language involves the construction of 'mental models ' based on the state of affairs described in the dis­ course. There is an increasing number of advocates of this position (cf. Karttunen, I 976; Stenning, I 977; Grosz, I 979; Fauconnier, I 979; Johnson-Laird &: Garnham, I 980; Kamp, I 98 1 ; Seuren, 1 982; Bosch, 1 983). On this assumption, it is no invoke the world. If semantics and using language involves semantic theory has to do, is these 'mental models' and the them on the basis of discourse.

longer clear that semantic theory should is the study of how people use language, constructing ' mental models', then all to give an account of the structure of principles by which people construct

JOU RNAL OF SEMANTICS , vol. 1 , no.4, pp. 37 1 -386

37 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In this paper I take the vie w that using language amounts to constro.cting 'mental models '. Accordingly, semantics has to e:rplain the stro.�ture of these mental models and the principles by which people constro.ct them. The system proposed, which was developed jointly with Nigel Shadbolt, is called S-R Semantics. A mong the fundamental features of the system is a functional distinction drawn between two sorts of mental object: epistemic objects, which are supposed to model the long-term established knowledge a processor brings to a discourse, and discourse objects, which model the objects introduced into a dis­ course and the predications made about them. The system is used to solve a number of problems which come up in connection with some uses of singular indefinite noun phrases, and disputes.

H A N REICHGEL T As a consequence, linguistic referring expressions have to be seen as being used to refer to ' mental objects ' . Therefore, reference is a relation between uses of referring linguistic expressions and ' mental objects ' , or 'cogn1tive reference' as it is called in S-R Semantics. Now, it is beyond doubt that people sometimes use language with the intention of talking about real-world objects. The question then arises how we are to account for this fact.

Of course, a processor can assume of a mental object that it is a mental representation of a real-world object, even if he is not able to identify the real-world object in question. Consider for instance the following utterance: ( I ) Gerry says to N igel: "My father found a new job." Gerry has formed the mental object which is his representation of his father on the basis of direct perceptual interaction with the world. Therefore, he is in a position to identify the real-world object that is his father. N igel, on the other hand, if he takes Gerry seriously and authorita­ tively, will believe that there is a real-world object that is Gerry's father . Therefore, he will suppose that the 'mental object' he has created in response to Gerry's use of the referring expression is the representation of a real-world object. As a consequence, he will assume that Gerry uses the referring expression my fa ther to refer ecologically. H owever, on the assumption that Nigel has created his 'mental object' on the basis of mfor mation provided by Gerry, he need not have anything like a sufficient procedure to identify the ecological referent. The information he has is simply not enough. So, although Nigel assumes

372

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This puzzle can be solved once we realise that a processor supposes of some of his ' mental objects' that they are mental representations of real-world objects. We agree with Locke's thesis of the ' veil of perception ', which says that humans are trapped in their own minds and that they can have no direct knowledge of real-world objects. All they have is their mental representations of objects, or ' mental objects'. Sometimes, however, people assume that their mental objects stand for real-world objects. Now a processor can assume that referring linguistic expressions are used to refer to real-world objects, as they are used' to cognitively refer to mental objects that he supposes to be mental representations of these real-world objects. S-R Semantics calls the relation between uses of linguistic expressions and real-world objects 'ecological reference '. Note that ecological reference presupposes cognitive reference, or to put it in other words, we can only get to the real-world obJects through our mental representations of them.

M ENTAL MODELS A N D DISCOURSE that Gerry is talking about a real-world object, he does not know which actual person Gerry is (ecologically) referring to. I will use the term direct ecological reference for those cases where a processor assumes that he is able to identify the real-world object which the referring expression is used to to refer to ecologically. Indirect ecological reference will be used for the cases in which a processor assumes that there is an ecological referent, although he does not believe that he has sufficient knowledge to identify it.

Discourse and epistemic objects

Let us now return to the basic assumption that using language amounts to constructing ' mental models'. We argue that theories developed within this approach often suffer from a certain 'functional ' confusion as to the role of these 'mental models'. On the one hand, they are used to represent the objects introduced into the discourse and the predications made about them. On the other hand, they model the established knowledge a discourse participant brings to the discourse. So, on the one hand, a processor is supposed to set up a new mental object in response to a referring expression and create a new mental model in response to a discourse, whereas, on the other hand, referring expressions are thought to refer to mental objects which are somehow already there prior to the discourse. We want to argue that it is func­ tionally convenient to distinguish between these two roles of mental models in general, and of mental objects in particular. Accordingly, objects we distinguish between discourse objects (DOs) and epistemic (EOs). We must emphasize the fact that the distinction between EOs and DOs is a functional one. They have the same ontological status. We help ourselves to a tractarian notion of object, in the spirit of Wittgen­ stein ( 1 9 2 1 ), and regard an object as a 'hook ' to which properties can be attached. The same applies to 'mental objects'. A ' mental object' can then be seen as a hook with a set of 'mental properties ', or beliefs, attached to it. We use the ter m cognitive object to denote the mental objects in the 'hook ' sense. The term constituted cognitive object is used to denote a cognitive object plus the set of beliefs associated with it. We want to argue that this notion of cognitive object is useful for a number of reasons. It allows us to make sense, for one thing, of the notion of 'shared mental object' but I will not go into that argument here. JS, vol. l , no.4

373

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It is clear that whether a referring expression is used for direct or indirect ecological reference, is completely processor-dependent. So, a use of a referring expression in a discourse can be directly ecologi­ cally referring for one discourse participant, whereas the same use can be indirectly ecologically referring for another interlocutor.

H AN REICHGEL T It seems reasonable to distinguish functionally between two sorts of beliefs that can be attached to cognitive objects. On the one hand, there are the beliefs a processor h·as before he participates in a dis­ course. On the other hand, there are the predications made about the object in the discourse. If we look at a constituted cognitive object from the former perspective we get an EO; if we look at it from the latter perspective, we get a DO. So DOs and EOs both exist as con­ stituted cognitive objects in a processor's mind.

(2) Sam says to Robert: "Nigel crashed into another car yesterday." Robert says to Sam: "I know, the car he crashed into belongs to a policeman." Robert would not be co-operative if he were to use the definite descrip­ tion the policeman's car to refer to the object Nigel crashed into. If he wants to use a singular definite description, he must employ the property ascribed to the obJeCt in the discourse. Discourse and epistemic reference

At this stage, it is convenient to distinguish between two roles. The interlocutor who introduces a new object into the discourse will be called the instigator. The interlocutor who is faced with the task of interpreter. interpreting this introductory utterance, will be called the Note that this is not the same distinction as the one between speaker and hearer. Although at one stage in the discourse the instigator was the speaker, and the interpreter was the hearer, the interpreter can refer anaphorically to an object introduced into the discourse by the instigator. As long as everyone talks about the same object, the roles of instigator and interpreter remain the same, although the roles of speaker and hearer may change. Consider for instance discourse (3): ( 3) Sam

[ + l n s t , +s p J

says to Robert

[+ l n t e r , +h r ]

:

" I read a good book yesterday." Robert

374

[ + l n t e r ,+ S p ]

says to Sam

[+l ns t ,+hr]

:

" What i s i t s title?"

JS, vo!. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

This functional distinction is relevant for language. A processor has to keep a record of the DOs and the predications made about them. And this record partly determines what objects are available for pronominalization. 1 The predications have to be recorded because they partly deter mine what defimte descriptions he can use to refer to the objects introduced into the discourse anaphorically. Consider example (2):

MEN TAL MODELS AND DISCOURSE We said that DOs exist in a processor's mind as constituted cognitive objects which may have existed prior to the discourse. Because of this, DOs can act as pointers to EOs. Of course, this is not always the case. The instigator can introduce an object into the discourse about which the interpreter had no beliefs prior to his utterance. Therefore, there is no EO for the interpreter and the DO cannot act as a pointer to an EO for him. If the instigator introduces an object with the intention of providing a piece of information, he has to have beliefs about this object prior to the utterance. Therefore, for him the DO must act as a pointer to an EO.

Sometimes, however, as has been said before, DOs act as pointers to EOs. If that is the case, one can refer to an EO by referring to a DO that points to this EO. This relation between the use of a referring expression and an EO is called epistemic reference. Since both DOs and EOs exist as constituted cognitive objects, terms which are used to refer both in discourse and epistemically are used to refer to constituted cognitive objects. So, the distinction between discourse and epistemic reference is a functional distinction between two kinds of cognitive reference. ·

Since DOs and EOs exist in the minds of discourse participants, it is only relative to a processor that a particular referring expression is used as discourse-referring, or as both discourse-referring and epi­ stemically referring. If the instigator tries to provide the Interpreter with information about an object about which the interpreter had no beliefs prior to the discourse, then the use of the relevant linguistic expression will be both discourse-referring and ep1stemically referring for the instigator, but only discourse-referring for the interpreter . This means that all types of reference we have distinguished are com­ pletely processor-dependent. Given the basic assumption underlying our approach, this is as it should be. Intentions

S-R Semantics makes two further assumptions which more or less naturally follow from the theory outlined before. Firstly, discourse involves more than one participant, and the difJS, vol. l , no.4

375

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Given this distinction between DOs and EOs, there are two func­ tionally distinct ways in which one can be said to refer to a mental object. A referring expression will always be used to refer to a DO, independent of whether it was constructed prior to the utterance in which the referring expression occurred or whether it was set up in response to that utterance. S-R Semantics uses the label discourse refer­ ence for the relation between a use of a referring expression and a DO.

H A N REICHGEL T ferent interlocutors may have different beliefs about the topic of conversation. People are sensitive to the fact that their fellow-inter­ locutors can bring different beliefs · to a discourse and they also take the beliefs into account they suppose the other discourse participants to have. This fact, trivial as it may seem , is very important and it allows one to solve a number of the problems Barry Richards raises in "Discourse and Deixis". A second assumption is that language has to be set against the intention of language users to affect the models they and their address­ ees have in their heads. As a consequence, language is not properly studied if one looks at sentences independently of speaker and hearer.

The instigator I S represented as looking at his knowledge base. On the basis of that search he comes up w 1 th an object he wants to talk about. He then chooses a particular referring expression to introduce that object mto the d1scourse. He does so on the basis of the beliefs he supposes the interpreter to have. He then utters the linguistic ex­ pression. He sets up a DO which points to the EO he wants to talk about. The interpreter, on the other hand, upon hearing the referring expression, is represented as setting up a DO. He then searches his knowledge base, and on the bas1s of that search he decides whether the DO points to an EO or not. Singular indefinite N Ps and uniqueness

Let us now see how S- R Semantics accounts for some of the problems raised by Barry Richards. Consider the following utterances: (4) Gerry says to Nigel: " Yesterday John bought a new car. It is a black MG." a ne w Formal semanticists usually interpret singular indefmite NPs like car as existentially quanti fied expressions. Given an ad hoc adjustment to allow for inter-sentential binding, one may then regard this expression as binding the pronoun in the second sentence Gerry utters. This 'bound ' interpretation, as Barry Richards calls it, would be read from a quasi­ formula as in (5):

376

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

On the basis of the philosophical assumptions outlined above we have developed a descriptive system that models the referential aspects of the human language processor. I will not go into the fine details of this system here. As a consequence, I cannot give the exact analyses we provide for utterances containing referring expressions. For the present purpose, which is to show how one can account for a number of the problems Barry Richards raises, a very rough sketch of the analyses S-R Semantics provides for utterances is sufficient.

MENTAL MODEL S AND DISCOU RSE (5) (E.r)[new-car(x) & bought(j,x) & black-MG(x) ] Obviously, this analysis does not account for the intuition that after Gerry has uttered his first sentence there is exactly one object he and Nigel appear to be talking about. The reason is that ( 5) can be informally rendered as (6), a,nd in (6) there is no requirement that there be exactly one object which makes the sentence true. Given that (5) tells the whole story there is to be told about the meaning of (4), it is impossible to account for the above intuition. ( 6) There is at least one object such that it is a new car such that John bought it and such that it is a black MG.

Gerry wants to provide information about a certain object. In the first sentence he uses a smgular indefinite NP to introduce the object into the discourse. In S-R Semantics, he is represented as setting up a DO that for him acts as a pointer to an EO. After all, Gerry wants to provide new information about an object. Therefore, he has to have beliefs about this objects prior to the utterance. Nigel, on the other hand, provided he had no specific beliefs prior to the discourse about the object Gerry has introduced, is represented as setting up a DO which for him will not act as a pointer to an EO. At this stage, it might be useful to draw attention to the fact that singular indefinite N Ps can be used as full-blooded referring ex­ pressions, because the instigator can without doubt use a singular indefinite NP to introduce an object into the discourse. Since singular indefm1te N Ps can be used to set up a DO, they can be used to dis­ course-refer. However, in the above case, Gerry uses the singular indefinite NP to introduce an object into the discourse he wants to provide information about. Because he can only do so if he has some beliefs about this object prior to the discourse, this DO for him acts as a pointer to an EO. Therefore, for him the singular indefinite N P also has an epistemically referring use. If the E O is a constituted cognitive object, which Gerry assumes to be a - mental representation of a real-world object, then he may believe that he is able to identify that real-world object. In that case, the singular indefinite NP for Gerry refers directly ecologically to this real-world object by cognitive­ ly referring to the constituted cognitive object. Given that singular indefinite NPs can be used as referring expressions in any of the three senses S-R Semantics distinguishes, the question arises why one uses singular indefinite N Ps to introduce an object which one had beliefs about prior to the discourse. Barry Richards claims that this assumption leads to a paradox. If one wants to introduce JS, vol. l , no.4

377

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

S-R Semantics, like virtually any theory which takes seriously the idea that using language involves constructing mental models, avoids this problem.

HAN REICH GEL T a specific object into the discourse, why use a singular indefinite NP? Why not use an expression whose conventional meaning determines it as specifically referring, such as a proper name or a singular definite description? Why does one flout convention to do what could also be done conventionally?

If the instigator assumes that the interpreter has a constituted cognJtJve object prior to the discourse, and, moreover, that the inter­ preter knows a name for that obJect, then he will use a proper name. In the case of singular definite descriptions, the instigator may make either of two assumptiOns: He may suppose that the interpreter has a constituted cogn1tive object prior to the discourse, as in (7); or he may suppose that the interpreter has what we call a ' cognitive script ' z with exactly one slot with which he can associate the newly set up DO, as in (8): (7) Gerry says to Nigel: "The paper you wrote was really good." (8) Gerry says to Nigel: "The alternator in my car is not working." In the case of singular indefinite NP's, the instigator may again make either of two assumptions: Once more he may assume that the interpreter has a constituted cognitive object prior to the discourse, as in (9), or he may assume that the interpreter has a cognitive script w ith more than one slot with which he can associdte the newly set up oo J as in ( 1 0): (9) Gerry says to Nigel: "You know that Ken submitted a paper." ( 1 0) Gerry says to Nigel: "A member of the Botany Department smuggles drugs." The only difference between 'script uses' of singular indefinite N Ps and singular definite descriptions is the number of slots in the cognitive script the instigator supposes the interpreter to have. This 378

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

It is here that the full force of the assumption becomes apparent that discourse participants are sensitive to the fact that their fellow interlocutors bring d1fferent beliefs to a d iscourse. This is why they choose linguistic expressions on the basis of the beliefs they believe their interpreters to have. Shadbolt and Reichgelt (forthcoming) discuss a number of cases where the instigator mtroduces an object into the discourse with the intention of providing or eliciting information about it.

M ENTAL MODELS A N D DISCOURSE leads to some clear predictions. For instance, our cognitive scripts for university departments have one slot for the Head of Department and a number of slots for the members. As a consequence, utterances ( 1 1 ) and ( 1 2) sound odd in that they go against the normal expectation we have about university departments and may create some confusion in the interpreter. ( 1 0) and ( 1 3), on the other hand, are fine and readily understandable: ( 1 1 ) Gerry says to Nigel: "The member of the Botany Department smuggles drugs." ( 1 2) Gerry says to Nigel:

"The Head of the Botany Department smuggles drugs." It seems to me that this answers the question why one would use a singular indefinite N P to introduce an object into the discourse, although the DO he sets up points to an EO which he believes to be the mental representation of a real world object. It is simply a matter of the beliefs he supposes his interpreter to have. Let us now turn to how discourse (4) can be said to be about exactly one object. Since, in S-R Semantics, reference is primarily cognitive reference, there are as many constituted cognitive objects one could be said to be ' talking about' as there are discourse participants. After all, constituted cognitive objects exist in the minds of the interlocutors. However, people assume that the constituted cognitive objects they have in mind are in some sense the same as those others have in their mind. For instance, because they represent the same real-world object. But people make this assumption not only with respect to constituted cognitive objects which were formed on the basis of direct perceptual interaction with the world, but also with respect to constituted cognitive objects that have been constructed on the basis of the information received from other language users or on the basis of a novel they read. In Shadbolt and Reichgelt (forthcoming), we provide som e philosoph­ ical arguments showing why people are usually right in making this assumption. For present purposes however, it is sufficient to say that the interlocutors suppose that they share a cognitive object. As a consequence, we can say that they are talking about whatever that cognitive object stands for. Note that it follows from the view of reference outlined above that successful reference has to be defined entirely in terms of the assumptions d iscourse participants make. A referential act is successful if the various interlocutors think that they share a cognitive object, and that somehow the objects they have in mind are the same. We shall see, however, that there are cases in which the assumptions JS, vol. l , no.4

379

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

"A Head of the Botany Department smuggles drugs." ( 1 3) Gerry says to Nigel:

H AN REICHGEL T people make about the identity or non-identity of cognitive objects lead to problems. Specificity

The above analysis of singular indefinite NPs obliviates the need for a treatment of pronouns like the one Barry Richards proposes in order to get the 'specific' reading of discourse (4). He has to assume a dual role for pronouns, functioning either deict1cally or as bound by existen­ tial quantifiers.

( 1 4) Gerry says to Nigel: "I heard that Henry has a new girlfriend. What does she look like?" Neither of these senses of specificity forces us to postulate a dual role for pronouns. The second sense of specificity presupposes the first sense. Moreover, have made clear how one can account for that sense of specificity in S-R Semantics. The instigator searches his knowledge base and comes up with an object he wants to talk about. He can then introduce that object into the discourse by using a singular indefinite N P. Note that one cannot account for the first sense of specificity by supposing a dual role for pronouns. This has to do with the nature of deixis. A language user can only be said to use a pronoun deictically, if he is able to identify the real world object he intends to refer to ecologically. In the terminology developed above, the uses of the refer­ ring expression must for him be directly ecologically referring. It is clear that under this condition on deixis, the pronouns do not have to be used deictically to make the singular indefinite N Ps specific in the first sense of 'specific ' . In utterance (4) there is no requirement that Gerry has sufficient information to identify the object he wants to talk about. 380

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

There seem to be at least two senses in which a smgular indefinite NP can be said to be used 'specifically'. First, a singular indefinite NP can be said to be used specifically if the instigator has a specific object m mind he wants to introduce into the discourse by his use of a smgular mdefinite NP, although he does not assume that the interpreter has a shared cognitive object prior to his utterance. Second­ ly, a singular indefinite NP can be said to be used specifically if the instigator assumes that the mterpreter has a constituted cognitive object prior to the discourse and that the DO the interpreter has set up acts as a pointer to this EO in his mind, as in example (7), or in cases where the instigator introduces an object into the discourse and assumes that the interpreter already has some information about that object, as in ( 1 4):

M EN TAL MODELS AND DISCOURSE Moreover, a deictic pronoun is usually said to be used successfully if, in consequence, all discourse participants are able to identify the real-world object that was ecologically referred to by the pronoun. It is clear that this need not be the case if a singular indefinite N P i s used successfully.

( 1 5) Nigel says to Gerry: "Zeus was married to a very jealous goddess." One might of course say that this argument does not go against the way Barry Richards uses deixis. To account for the specific reading of discourse ( 1 5), he could translate it into formula ( 1 6), where 'u ' is an individual constant. ( 1 6) (Ex)[jealous(x)

&.

goddess(x)

&.

married(z, x)

&.

x

=

u]

Since ' u ' is an individual constant, it is interpreted as an individual in a model, and one can populate one's model w1th all sorts of fictional and mythological entities. Although this may be a consistent way out, it seems to me that we see a direct pay-off of the approach to language which takes ' mental models' as basic. One of the arguments for this approach is that the traditional formal semantics approach to language leads to a number of counter-intuitive and unacceptable consequences. One of the more undesirable consequences is that one has to postulate a very rich ontol­ ogy. One has to populate one's universe with entities which do not exist. The status the instigator ascribes to the EO which the DO points to makes no difference as far as the understanding of the discourse is concerned. It does not make any difference whether he assumes that he is able to identify the real-world object of which his constituted cognitive object is thought to be a mental representation. That this should be the case, is clear from the interpreter 's point of view. The interpreter is supposed to have no EO which the DO he sets up can point to. Therefore, the assumption the instigator makes can make no difference as far as he is concerned. He had no EO prior to the discourse. Therefore, he had no beliefs about the beliefs the instigator has about that object. The instigator, on the other hand, did not assume that the interpreter shared a cognitive object before his utterance. JS, vol. l , no.4

38 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

A third point is that deictic uses of expressions are usually seen as establishing a relation between the use of a linguistic expression and a real-world object. Again, it is clear that this is not necessarily the case if one uses a singular indefinite NP in the first sense. So, the singular indefinite NP in discourse ( 1 5) is used specifically in this sense, although neither discourse participant assumes that there is a real-world object they are talking about.

H AN REICH GEL T Therefore, he had to assum e that the interpreter had a cogmt1ve script. It was this assumption that determined which type of referring expres­ sion he would use. This assumption is independent of whether he thinks that he is able to identify the real-world object. Therefore, the status he supposes his EO to have, is not important for the choice of the type of referrmg expression he will use.

( 1 6) Allan says to Nigel: "Have a pint." The second sense of specificity can be accounted for if it is recalled that there were two assu mptions one can make in connection with the use of a singular indef inite NP. Firstly, one may assume the exist­ ence of a cognitive script in the interpreter. Secondly, one may assume that the interpreter already had a constituted cognitive object prior to the discourse. It is clear that if one makes the second assumption, one uses the singular indefinite NP spec ifically in the second sense. Note that 1t is not clear that one can account for specific uses of indefm1te NPs in this sense by postulating a dual role for pronouns. In the first place, one can get specific uses of indefinite NPs in this sense while talking about objects of which the discourse participants do not suppose that they are real-world objects. Consider discourse ( 1 7): ( 1 7) Gerry says to Nigel: "Sherlock Holmes had an assistent. What was his name again?" Secondly, since it is the pronoun that is responsible for the specificity of the singular indefinite NP, one cannot use singular indefinite N Ps specifically if they occur alone. Discourse (9) provides a clear example. D isputes

Another point Barry Richards raises against the 'bound' interpretation has to do with disputes. Consider d iscourse (4) again. Suppose that N igel, contrary to what Gerry believes, believed already before he heard Gerry's utterance that John had bought a car . Moreover, suppose that he thinks that it is a white Porsche. He m ight then reply to Gerry's utterance by uttering ( 1 8): 382

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

F inally, it must be noted that it follows from this account that, if an instigator uses a singular indefinite N P to introduce an object i nto the discourse, his use of it is 'specific' in the first sense. This is not to say that every use of a singular indefinite NP is specific. One can easily think of examples where one uses a singular indefinite NP without having the mtention to provide or elicit information about a certain object. Example ( 1 6) is a clear case.

M ENTAL MODELS AND DISCOURSE ( 1 8) Nigel says to Gerry: "No, it is not. It is a white Porsche." It is intuitively clear that Nigel 's utterance should be construed as a contrary of the second sentence Gerry uttered. On the 'bound' reading, where ( 1 8) is rendered as ( 1 9), whose i mport can be informally rendered as (20), this is impossible: ( 1 9) (E.r){new -car(.r) &. bought(j, .r) &. white-Porsche(.r)]

' Of course both (5) and ( 1 9) can be simultaneously true in a model. Therefore, Barry Richards concludes, ( 1 9) is not construed as a contrary of Gerry's second sentence. It should · be clear that the explanation of this intuition poses no problems for S-R Semantics. Gerry and Nigel assume that they share a cognitive object. They disagree, however, about the beliefs attached to that object, and it is because of this disagreement that they have ; a dispute. So, there does not seem to be any need for postulating a dual role for pronouns. Moreover, if one did so, one could not account for the following phenomenon. S!Jppose that Gerry's and Nigel's dispute gets out of hand and that they decide to consult John. John then tells them that he bought two cars, a black MG and a white Porsche. It seems to me that one would be inclined to say that both Gerry and Nigel had told the truth. Contrary to what they assumed, the cognitive objects they thought they shared were not the same, in the sense that they were not mental representations of the same real-world object. One can of course say that Gerry and N igel did not have a real dispute after all. Since they were talking about different objects, their utterances were not contrary. However, if semantics is about the principles by which people use and understand language, it has to account for the fact that in the above situation Gerry and Nigel at least think that they have a dispute. One can also think of examples where people have a dispute because they think that they are talking about two different objects, whereas it may be the case that, unbeknownst to them, the constituted cognitive objects are mental representations of the same real-world object. For instance, consider discourse ( 2 1 ) , which occurred in free speech: ( 2 1 ) A says to B: "I have been to Italy and I really like Rome. But the most beautiful town in Italy is Florence." JS, vol. l , no.4

383

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(20) There is at least one object such that it is a new car, such that John bought it and such that it is a white Porsche.

HAN REICHGEL T B says to A: "No, i t is not.

H

is Firenze."

Significantly, Firenze is the Italian name for Florence. So, although A and B are actually talking about the same real-world object, they think that they are talking about different towns. How can one account for this phenomenon if all there is to the meaning of a proper name is its reference, in the sense of ecological reference? Obviously, if semantic theory has to account for the above discourse, one has to take into account what A and B think they are talking about.

I have tried to make clear how one can use a par ticular descriptive system to account for some of the questions raised by Barry Richards. The main difference between this system and other systems developed within the same general approach is the distinction between EOs and DOs. The distinction helps one to account for a number of the problems. Therefore, the paper is in part also an argument for this distinction. However, some of the solutions I propose, are independent of that particular descriptive system. As a consequence, this paper also is an argument for cognition-oriented semantics. School of Epistemics University of Edinburgh Edinburgh, Scotland

Acknowledgements

I would like to thank the following people for their help and encourage­ ment: P. Bosch, G. Brown, B. Richards, P. Seuren, N. Shadbolt, and finally, N ikki Cowie. The research was made financially possible by the Niels Stensen Stichting in Amsterdam. Notes

1 One has to admit that the presence of an object in a model of the previous discourse is neither a sufficient nor a necessary condition for pronominalization. It is not sufficient because there are numerous other factors which are important in determining what can be pronomi­ nalized. Garrod &. Sanford ( 1 982), for instance, argue that 'focussing' is a relevant factor. On the other hand, the condition is not necessary either because people sometimes pronominalize even if the antecedent

384

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Conclusion

MENTAL MODELS AND DISCOU RSE has not been explicitly introduced before. Gill Brown observed the following utterance in free speech: (i) "The car stopped. He wanted to turn left."

Refer-ences

Bosch, P. ( 1 983 ): Agreement and Anaphora. (to appear; Academic Press, London) Garrod, S. &. Sanford, A., 1 982: The mental representation of discourse in a focussed memory system: Implications for the interpreta­ tion of anaphoric Noun Phrases. Journa l of S e m a n tics, 1 : 2 1 -42. Fauconnier, G . , 1 979: Mental Spaces: A discourse processing view to natural language logic. Unpublished paper, Universite de Paris. Grosz, B.J., 1 979: Utterance and objective: Issues in natural language communication. Technical Note 1 88 SRI. Johnson-Laird, Ph. &. Garnham, A., 1 980: Descriptions and discourse models. L inguistics and Philosophy, 3: 37 1 -93. Kamp, H., 1 9 8 1 : A theory of truth and semantic representation. In: J. Groenendijk, T. Janssen &. M . Stokhof (eds.) Formal Meth­ ods in the study of language. Mathematisch Centrum, Amster­ dam. Syntax Karttunen, L., 1 976: Discourse Referents. In: J. McCawley (ed.) and Semantics, volume 7. Academic Press, New York. Minsky, M., 1 97 5: A framework for representing knowledge. In: P. McGraw-H ill, Winston (ed.) Psychology of computer vision. New York. Richards, B., Discourse and Deixis. Journal of Semantics, this issue. Schank, R., &. Abelson, R., 1 9 7 5: Scripts, plans and knowledge. ProceedJS, vol . l , no.4

385

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

where he seems to stand for the driver of the car. 2 The notion of script (Cf. Schank &. Abelson, 1 9 7 5), or frame (Cf. Minsky, 1 975) is borrowed from A.!. Scripts or frames are stereotypic · representations of recurrent and regular events or situations. These representations contain slots, which are representations of the individuals (of whatever appropriate sort), that participate in the represented event or situation. Since these representations are stereotypic, we have certain expectations about the individuals which can fill the slots in the script. 3 Gill Brown drew our attention to the fact that this formulation can lead to a certain confusion. We do not assume that a cognitive script for a school with 700 pupils contains 700 slots for the pupils. When we talk about a script with more than one slot, we do not want to i mply that the slots are physically there. One could think of these mul tiple slots as slots which are marked for the fact that usually there is more than one filler for that slot.

HAN REICHGELT ings of the fourth international joint conference on A.I. Tblisi. Seuren, P ., 1 982: The construction of discourse domains through accumu­ lated increments. Paper delivered at the Edinburgh conference

on Language, Reasoning and Inference, April 1 982. Shadbolt, N . &. Reichgelt, H., (forthcoming): Discourse, semantics cognition: a theory of reference.

and

Stenning, K . , 1 977: Articles, quantifiers and their encoding in textual adcomprehension. In: R. Feedle (ed.) D iscourse processes: vances in research and theory. Norwood, New Jersey. Wittgenstein, L, 1 9 2 1 : Tractatus logicrrphilosophicus. Kegan Paul, London. Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

386

JS, vol . l , no.4

RUSSELL

VINDICATED :

TOWARDS A GENERAL THEORY OF DEFINITE DESCRIPTIONS

Jaakko Hintikka and Jack Kulas

It seems to us that time is ripe to turn the tables on Russell's critics. This paper is a first step towards a general theory of definite descriptions which borrows its inspiration from Russell's theory, even though the conceptual tools it uses go beyond Russell ' s methods. Because of this spiritual affinity, whatever success our theory may enjoy serves also as a partial vindication of Russell 's original theory. We shall assume that the reader is familiar with the main ideas of Russell 's theory. Nevertheless, we recall here what it is all about. Russell's theory is directly applicable to expressions which are couched in the familiar logical notation. In this notation, which is not unproblem­ atical, the definite description 'the individual (call it x) such that D(x)' is expressed by '(tx)Di x)'. What Russell 's ' theor y ' (better, treatment) of definite descriptions does is to analyse away each occurrence of a definite description '(ix)D(x) ' in a given context (say, in the context C[ D by interpreting the overall sentence to assert the existence of a unique individual x of which it is true that D(x) and of which it is true that C[x]. In brief, ( 1 ) C[(tx)D(x)] is taken to be equivalent to (2) (Ex)[D(x)

&:

(y)(D(y)

-+

y =x)

&:

C[x TI.

In this way, the rules governing definite descriptions can all be thought of as being consequences of the general logical rules governing quanti­ fiers and other such logical notions, and they turn out not to incur any ontologic al commitments or model-theoretical problems over and

JOURNAL OF SEMANTICS, vol. l , no.4, pp. 387- 397

387

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

called "the Russell's so-called theory of definite descriptions 1 has been paradigm of philosophy" (Ramsey 1 93 1 : 263). Its historical significance is due to its being a part of Russell 's largely successful campaign to reduce many different kinds of discourse to what in our days would be called the language of first-order logic. 2 In view of the overall success of the Russellian program, it is ironic that its paradigm case, Russell ' s theory of definite descriptions, has come under heavy c r iticism , and that i ts applicability looks quite limited.

H IN TIKKA

&:

KULAS

above those already involved i n the use of these basic logical no­ tions. This is an important part of. the Russellian strategy mentioned earlier.

following:

·

( 3) The boy who was fooling her kissed the girl who loved him. This sentence cannot be translated into Russell 's iota-notation, for reasons closely related to those that have made it hard for syntacticians to generate it by means of a natural pronom inalization rule. These problems will not be treated in this paper. 4 Even more important - and even more frustrating - is the fact that Russell 's theory does not begin to do justice to several other uses of the-phrases in natural languages. They include the uses which we will call the anaphoric use and the so-called generic use. These are illustrated by the following examples: A NA PHORIC

(4) A man was seen walking down a path, but soon the man turned back. (5) If Bill owns a donkey, he beats the donkey. G ENERIC

(6) The beaver builds dams. (7) The modern girl wore trousers. The modern woman wore trousers. Miss Cragg herself wore trousers. 5 These two types of uses of the-phrases are not amenable to a Russellian treatment. In cases of anaphoric uses of the-phrases, uniqueness of reference is not required in the sense of there being a unique individual of which the the-phrase is true. Instead, such a use of a the-phrase typically picks 388

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The attractiveness of Russell's theory makes the criticisms that have been levelled at it at the same time interesting and frustrating to a theoretician. A great deal of this criticism concerns those uses of Russell's treatment that he had in mind himself. 3They are applications to those occurrences of the-phrases in English which are, in some obvious sense, context-independent and in which we are in some (perhaps slightly less obvious) sense attributing or presupposing uniqueness. They are the We shall call such occurrences of the-phrases RU3Sellian. uses of the-phrases to which the philosopher's term 'definite descrip­ tions' applies most happily. Even when applied to such occurrences of the-phrases, Russell 's theory is controversial, especially when it comes to the treatment of those cases in which the definite description is empty. There are even natural-language uses of the-phrases where the very Russellian notation proves inadequate. Cases in point are so-called Bach-Peters sentences with crossing anaphoric relations, such . as the

RUSSELL VIN DICATED up the reference made by the use of an earl ier occurring noun phrase, rather as a pronoun would do. Indeed, in many sentences, an anaphgrical­ ly used the-phrase can be replaced by a pronoun, or vice versa, without any change of meaning . For instance, in (4) we could thus replace the man by he, and in (5) we could replace the donkey by it, without changing the meaning. This affinity between anaphorically used the-phrases and other anaphorically used expressions (e.g., pronouns) puts our enterprise in an interesting light. It suggests that if we can successfully deal with anaphorically used the-phrases, we can perhaps better under­ stand anaphora in general. This adds a great deal of potential interest to the enterprise of the paper.

It turns out to be surprisingly easy to treat anaphorically used the phrases in this spirit. The means of doing so are found in the game-theo­ retical semantics developed by Jaakko Hintikka and his associates in the seventies. Its basic ideas are assumed to be familiar to the reader. 6 Game-theoretical semantics allows for a strikingly simple treatment of anaphoric the-phrases. We shall i n effect stick to Russell's theory, but restrict the range of quantifiers used in it to a set, which we shall call I, of individuals which have figured in the game to that point. This restriction relies essentially on the dynam ic, stage-by-stage character of semantical games, for the set I changes in the course of the game. An approximation to embodying th1s idea is the following game rule: (G. anaphoric the) When the game has reached a sentence of the form (*) X - the Y (who Z) - W an individual, say b, may be chosen by MYSELF from the set I, which consists of those individuals so far chosen by the players, plus individ­ uals specified by proper names occurring in the imtial sentence of the game. Whereupon N ATURE chooses a different individual , say d, from the same set. The game is then continued with respect to (**) X - b - W, b is a Y, (b Z) and d is not a Y (who Z). (parentheses indicate here optional parts of the expressions considered.) As an example of an application of (G. anaphoric the) let us the following sentences:

JS, vol. l , no.4

consider

389

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

In view of the failure of Russell's original treatment to deal with anaphoric or generic uses of the-phrases, one possible desideratum for a theory of the-phrases might be to find a theory of such phrases which would cover these uses and yet somehow preserve the elegance of Russell ' s original theory. Hilbert once characterized his foundational efforts as an attempt to work his way back to Cantor 's set-theoretical paradise, from which paradoxes had banished mathematicians. Perhaps we can see our programme here as an attempt to remstate the paradise (or at least the paradigm) which Russell's 1 905 theory truly was for philosophical and linguistic semanticists.

HINTIKKA & KULAS (8)

A stockbroker and his mother Jive in our village. Every day the stockbroker goes to London.

For the purposes of the present illustration, we can think of this pair To handle it, the first of sentences as a conjunction of its members. 7 move is made by MYSELF in accordance with (G. a(n)), i.e., an individual A stock­ is chosen by M YSELF to be inserted for the quantifier phrase broker. Suppose, for the sake of argument, that the individual chosen is Kenneth Widmerpool. Then the application in question yields (9)

Suppose that at the next move N A TU R E chooses the last clause in (9). The resul t i s ( 1 0) Every day the stockbroker goes to London. For the sake of s1mplicity, we should assume without further discus­ sion that (G. anaphoric the) is here applied before (G. every). On fact i t i s , but a discussion o f the reasons a t this point would take u s too far afield.) Considering ( l 0) in isolation, we cannot expect uniqueness of reference of the the-phrase. For there are many stockbrokers in the world and perhaps even many in our village. However, the intended When it is ap­ contextual uniqueness is captured by (G. anaphoric the). plied to ( 1 0), an md1vidual is chosen by MYSELF from the set I, which in th1s case has as its only members Kenneth Widmerpool, Esq . , and his mother, since the players must have dealt with the genitive his moth­ er before arriving at ( 1 0). Here M YSEL F ' s best strategy is again to choose Kenneth Widmerpool, whereupon NATURE must choose Mrs. Widmerpool, who happens to be the only other member of I. Then the task of defending the following output sentence (cf. (n)) confronts M YSELF: ( I I ) Every day Kenneth W idmerpool goes to London, Kenneth W idmer­ pool is a stockbroker , and Mrs. Widmerpool is not a stockbroker. The rest of the game connected with ( 8) is trivial. It is obvious that in this way we in fact get the right truth conditions for (8). Our rule (G. anaphoric the), which will be somewhat refined later, deserves - and needs - a number of comments. (a) The rule (G. anaphor ic the) as formulated above is only a special case of a more general rule. As we have formulated it, i t is subject to several restrictions which can be eliminated, but whose elimination will not be discussed here. Among them are the following: \i) In the optional relative clause who Z, who is assumed to occupy the subject position. If this assumption is eliminated, we (the players) 390

JS, vol . l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Kenneth Widmerpool and his mother Jive in our village, and Kenneth Widmerpool is a stockbroker. Every day the stockbroker goes to London.

RUSSELL V IN DICATED

(b) Even though paper in any depth, it ly has priority over example, (G. and) and

we will not be discussing rule ordering in this is relevant to note that (G. anaphoric the) obvious­ such rules for propositional connectives as, for (G. or).

(c) The rule (G. anaphoric the) above is formulated for a natural language (English). It is very easy to form ulate a similar rule for the semantical games associated with formal (but interpreted) first-order sentences: (G. iota) When a semantical game has reached a sentence of the form (*) C[(tx)D(x)], an individual, say b, is chosen by M YSELF from a certain set I, where­ upon N ATU RE chooses a different individual, say d, from I. The game is then continued with respect to (**) C[b] & D[b] & D(d]. ,..,.

Here I is the set of all individuals chosen by the two players up to the time of the application of this rule. 9 This rule is identical with the rule which would codify Russell's treatment of definite descriptfons in game-theoretical semantics, except that for Russell, I would have to be the whole universe of dis­ course. This illustrates vividly · the spiritual closeness of our theory to Russell's. (d) The rule (G. anaphoric the) is applicable also on the discourse level, not only on the sentence level. This can be done by means of the general strategy (which we have to undertake in any case) of dealing with discourse in terms of a progression of subgames associated with its several sentences. (See note 7 above.) (e) The set I can change in the course of a game. It is largely through this dependence of I on the current stage of the game that the dynamic character of semantical games enters into the picture. There is obviously nothing comparable to this changing set in the conventional JS, vol. l , no.4

39 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

have to know where the wh-word, as it were, comes from in z. Then b, instead of being prefixed to Z, will in (**) go into the original address of that wh-word. (ii) More has to be said of the treatment of anaphora in the transi­ tion from (*) to (**). (iii) Obviously, the choices of individuals mentioned in (G. anaphoric the) have to be restricted further to a suitable subdomain of the total domain of individuals. How this subdomain is determmed is studied by Jaakko Hintikka elsewhere.B ( iv) If I is a unit set , then NATURE cannot choose an individual from it different from b. ln that case, the game will be continued with respect to (***) X - b - W, b is a Y (and b Z).

H IN TIKKA &. KULAS syntactical or in the Russellian treatments of the-phrases. This is as it should be. If examples are needed, �he rules of many games and rituals offer good examples. In them , phrases like the ba tter. the player carrying the ball, the oldest man, etc., can be used completely correctly even when they pick out different persons at different occurrences of the same the-phrase at di fferent stages of the same discourse (or even of the same sentence).

(f) This point is reinforced by the fact that the 'head' of an ana­ phoric the-phrase may actually refer to more than one individual, only one of which 1s the reference of the the-phrase. Such cases are typical ly handled automatically by our theory. The following examples serve to illustrate this: ( 1 2) A couple was sitting on a bench. The man stood up. ( 1 3) Two men were walking down the road. The taller man carried a rucksack. These observations suggest that an unanalys ed notion of an antedendence relation (a relation of a the-phrase to its 'head ') is not very useful for a satisfactory theory of the-phrases. (g) This is connected with a more general point. On our theory, the anaphoric use of the-phrases is an essentially semantical phenome­ non, not a syntactical one. This fact has two aspects. First, our theory is quite as semantical as Russell's theory. Indeed, our rule (G. anaphoric the) serves to deal with the Russellian uses of the-phrases, the only dif­ ference being that in his theory, I is the whole universe of discourse. Secondly, examples show that what counts is not which individuals are mentioned earlier in the sentence or discourse, but which individuals have been introduced into the game by players' choices. In order to see this, witness examples of the following sort (adapted from Kart­ tunen 's examples): ( 1 4) If John buys a car or a motorcycle, he w ill take good care of the vehicle. 392

JS, vol. l , no. 4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

Thus the semantical mechanism of anaphoric the-phrases is not really comparable to what anaphora is usually thought of as being like. Ana­ phoric uses of the-phrases, one might be tempted to say, are not 'really' anaphoric.1 0 They don ' t operate by means of an earlier syntactic ante­ cedent, but they rely on the temporary semantical situation in a game, wh1ch includes the set of all individuals chosen by the players up to that stage of the game. An 'anaphoric' the-phrase thus does not just pick up the individual referred to by its 'head ' . The very notion of a head becomes dubious. What the so-called head does is introduce an individual into the set I, viz., the ind ividual which eventually happens to be picked up by the the-phrase in question. However, that this 1s the indi vidual referred to by the the-phrase does not depend on the head alone, but also on other noun phrases which introduce individuals into I.

RUSSELL V IN DICATED ( 1 5) If John buys a car and a motorcycle, he will take good care of the vehicle. With respect to relations of antecedence, these are on a par. Yet only the former one is acceptable. What makes the difference is explain­ ed in H intikka & Carlson ( 1 979). The explanation is semantical rather than syntactical. (h) In the same direction, it may be pointed out that the the-p h r a s e does not have to match the description given of the individual in question by i ts head. An example of this is offered by ( 1 3), as well as by sentences like the following:

From this, there is only a short step to the observation that our theory enables us to deal with the so-called epithetic uses of the-phrases, illus­ trated by the following sentences: ( 1 8) John borrowed my book, but the bastard never returned it. ( 1 9) H1s wife walked slowly by his side. The old woman stooped slightly. These are the uses where, prima facie, a the-phrase serves not only to repeat a reference to an individual mentioned earlier, but also to attribute a previOusly unmentioned attribute to that individual. One of the virtues of our theory is that it can handle epithetic the -phrases without any adjustments or modifications. For instance, if (G. anaphoric the) is applied to ( 1 8), I consists of John alone. Then the output sentence is (20) John borrowed my book, but John never returned it, and John is a bastard. What this amounts to within our approach is that the force of the use of the the-phrase is to air the view that John is a bastard. And this, of course, is precisely what in general epithetic the-phrases express. Over and above using the-phrases epithetically· to attribute predicates to the bearers of the phrases, they can be used to introduce unspoken background assumptions. In a sense, epithetic the-phrases need not really be epithetic in that they can serve to introduce background information rather than attributes of individuals introduced earlier. As examples of this sort we m ight consider the following: ( 2 1 ) Surely there is night life in Tallahassee. weekend the lady is in Tampa.

Unfortunately,

this

All this is understandable on the basis of our theory. In it, ( 2 1 ) is true only if the night life in Tallahassee consists of one and only one lady. __

JS, voJ . l , no.4

393

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

( 1 6) A tall man was walking along the road. The man .. . ( 1 7) A man was walking along the road. The tall man . . .

HINTIKKA & KULAS (i) It is important to realize that the individuals in I may have been introduced by NATURE's choice rather than by choices by MYSELF. Cases in point are the following: ( 22) Every soldier in the company was given a rifle. Immediately, the captain f ired his. (23) John saw every orchestra member, but only recognized the first violin.

The need for this extension is seen from the following example: (24) If you give each boy a Christmas gift now , some boy will open the gift which he was given today. When it comes time to produce the the-phrase in (24), I will contain some function F which associates with each boy the particular Christmas gift given to him. If no such F exists, the antecedent clause of (24) is false, so there would be no need to process the the-phrase: Supposing i t true, we can then use F to find the relevant gift for any boy. Processing the consequent clause of ( 24) involves first processing some boy using (G . some). When an individual is selected, F then comes into play when processing the the-phrase. It, as it were, picks out for the boy selected, the gift he was given. If the selected boy will not open the gift picked out by F, the consequent clause is false, making the whole sentence false. Details omitted in the above description can be fairly easily supplied. Another change is needed in (G. anaphoric the). It is due to the fact 394

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

At this point a few general issues have to be raised. F irst, there is an extension which is needed in the formulation of our game rule (G. anaphoric the). The idea underlying the rule is that we stick to the Russellian treatment of the-phrases but let our quantifiers range only over such individuals as are available to the players at the stage of the semantical game at which the the-phrase is 'processed ' . This idea is Now it is time to codified in the set I mentioned in (G . anaphoric the). sharpen the definition of I. At the time (*) is dealt with in a game, the players certainly have 'available' to them (i) the individuals referred to by proper names in relevant preceding sentences and (ii) the individ­ uals picked out earlier in the game by the players and 'remembered' (in the strict sense that can be spelled out in terms of the concept of subgame) at the stage which the game has reached. However, at that stage, the players also have available to them certain functions and functionals that encode strategies used by them in completed, earlier subgames and 'remembered ' at the time of the application of (G. anaphoric the). The set I must obviously be closed with respect to all these functions and functionals. It can be defined as the smallest set so closed that contains (i) and (ii). This change in the definition of I is the extension needed here.

RUSSELL V IN DICATED - as i t seems to us - that natural language requires an 'exclusive inter­ pretation ' of quantifiers and quantifier-like operators occurring in the same clause. 1 1 Here an exclusive interpretation means one on which different quantified variables must have different values. It means in semantical games that certain choices of indi viduals are made 'without replacement ' , as probability theorists would say. This means that the set I mentioned in (G. anaphoric the) must not contain in­ dividuals introduced by names or substitution-values for quantifier phrases occurring in the same clause as a the-phrase. The need for this modification of the definition of I in (G. the) is seen from examples like the following:

anaphoric

In (25), the the-phrase The poor girl can refer, and given the context, virtually has to refer, to Jane. In contrast, in (26) the the-phrase cannot be coreferential with Jane. What makes the difference is that in (26) the the-phrase is in the same clause as Jane, whereas in (25) it is not. The third example shows the possibility of a the-phrase being coreferential with another noun phrase (vJz. , Jane) in the same sentence, but outside its own clause. Another small modification is needed which goes in the opposite d1rection. It is to be noted that our treatment of subgames will entail omitting certain individuals from I, viz., those chosen in earlier sub­ games and not determined by those strategies which are 'remembered ' at the time the rule (G. anaphoric the) is applied. This has in effect been presupposed in our treatment of (24) just sketched. One can also treat the the-phrase in (5) by means of (G. anaphoric the) and obtain the intended interpretation of this sentence. The rest of the treatment of ( 5) can follow the guidelines of H intikka & Carlson (see note 7 ) .

This result is more of a novelty than it might first seem to be. The paradigmatic example of H intikka & Carlson is not (5) , but (28) If Bill owns a donkey, he beats it. This d iffers from (5) in containing the pronoun i t in the place of the the-phrase the donkey. Its meaning is precisely the same as that of (5), however. Now it has been justly objected to Hintikka & Carlson that, even though their use of the subgame idea is suggestive, they do not give any explicit rules for the treatment of pronouns (pronominal anaphora) and that their treatment of the problem is therefore seriously incomplete. This is a telling objection. However, it is largely met JS, vol. l , no.4

395

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(25) Jane remembered her misfortunes. The poor girl was in tears. (26) Jane r�membered the poor girl. (27) Jane remembered someone whom the poor girl could not stand.

HINTIKKA &. K U L AS by our present observation. Whatever can be said of the treatment of (28), we have reached a completely explicit, successful treatment of (5). This strongly suggests that H intikka &. Carlson are on the right track. What remains to be done is to extend our treatment of anaphoric the-phrases to anaphoric pronouns. This extension, which can be expected to vindicate H intikka &. Carlson completely , will be attempted in a separate paper.

Florida State University Tallahassee, F lorida

Notes

1 See especiaily Russell ( 1 905). 2 The main development in Russell's views from the time of his Principles of Mathematics onwards is that sentences contammg what he called 'denoting phrases' can all be paraphrased in the quantifi­ cational idiom . Russell 's treatment of definite descriptions was, in his eyes, the most difficult part of the reduction, and has hence gained most of the attention. In the proper historical perspective, however, the other parts of the reduction are undoubtedly much more important. 3 Cf. , e.g., Strawson ( 1 950). 4 Cf. here Jaakko Hintikka &. Esa Saarinen ( 1 97 5). Lauri Karttunen ( 1 97 1 ) demonstrates the untranslatability when considering McCawley 's treatment of Bach-Peters sentences (pp. 17 3-76). The problem here may have more to do with pronominalization, however, than with the ­ phrases themselves. Karttunen says, "Pronouns [in natural language] are sometimes used in a way that is not possible with variables in the more restricted syntax of predicate calculus" (n. 1 2, p. 1 76). 5 Clinton-Baddely ( 1 98 1 : 1 23). 6 For expositions of its basic ideas, the reader is invited to see Saarinen (ed .) ( 1 979), or Hintikka's survey paper in the semantics number of Notre Dame Journal of Formal Logic, forthcoming in 1 982. 7 As shown in Jaakko Hintikka &. Lauri Carlson ( 1 979). One of the advantages of game-theoretical semantics is that it can be extended to discourse semantics. This extension uses the idea of a subgame which is explained in Zoe. cit. In the p_resent case, it amounts essentially to the conjunctive treatment presupposed here. 396

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

We can thus obtain a most satisfactory theory of anaphorically used the-phrases. In subsequent papers, we will try to show that the other main uses of the-phrases , including the Russell ian use and the generic use, can be understood as pragmatically motivated extensions of the anaphoric use. In this way, our analysis of the anaphoric use can serve as a step toward a truly general theory of the-phrases, and hen­ ce toward a vindication of Russell's basic ideas.

RUSSELL VIN DICATED 8 9

See Hintikka (forthcoming). A special case analogous to (a)(iv) above must be adm itted in this rule, too, when I is a unit set. 10 Of course, the really interesting question is whether other anaphoric expressions, especially anaphoric pronouns, behave 'anaphorically' in the currently received sense of relying essentially on a syntactic relation to their heads. 1 1 For this interpretation of quantifiers, see H intikka ( 1 956); ( 1 973: Ch. l )

References

JS, vol . l , no.4

397

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

C1inton-Baddely, V.C., 1 98 1 : My Foe Outstretched Beneath the Tree. Dell Pub!. Co., New York. H intikka, .J., 1 956: Identity, variables, and impredicative definitions. Journal of Symbolic Logic 21; 225-245. Hintikka, J., 1 973: Logic, Language-Gam es, and Information. Clarendon Press, Oxford. Hintikka, J., forthcomin�?,: Semantical games, the alle�ed am�ibuity of 'is', and Aristotelian categories. Synthese. Hintikka, J., & Carlson, L., 1 979: Conditionals, generic quantifiers, and other applications of subgames. In: Saarinen (ed .); 1 79-2 1 4 . Hintikka, J., & Saarinen, E., 1 97 5 : Semantical games and the Bach­ Peters paradox. Theoretical Linguistics 2; 1 -20. Karttunen, L . , 1 9 7 1 : Definite descriptions and crossing coreference. Foundations of Language 7; 1 57- 1 82. · Ramsey, F.P., 1 93 1 : The Foundations of Mathematics and Other Logical Essays. Kegan Paul, London. Russell, B., 1 905: On denoting. Mind 14; 479-493. (Also in: R.C. Marsh (ed.), Bertrand Russell: Logic and Knowledge. Essays 1 90 1 1 95 0. Macmillan, London, 1 956. Pp. 4 1 -56) Saarinen, E., (ed.), 1 979: Game-Theoretical Semantics. Reidel, Dordrecht. Strawson, P.F., 1 950: On referring. Mind 59; 320-344.

Discussion

AGAINST "AGAINST CONVERSATIONAL IMPLICATURE"

Johan van der Auwera

1.1 For some reason or other Sterelny seems to equate 'gram mar ' with 'generative gram mar ' . This equivocation takes care of the first objec­ tion. Sterelny claims it is hard to consider maxims as part of any grammar. A grammar should be something intrinsically linguistic and maxims do not fit in this picture, for they are special cases of general principles of conversation. This objection presupposes a conception of 'autonomous' linguistics, which is typical for generative grammar. But nobody interested in Gricean pragmatics is forced to be a genera­ tivist. Probably most Griceans aren't generativists. 1 .2 Syntactic behaviour is both arbitrary and rule-governed. Conversational behaviour is also rule-governed, but it is implausible to claim that it is arbitrary. My reply: so what ? 1 .3 Sterelny claims that Gricean rules of conversation do not define lan­ guage and that rules of syntax and semantics do. The alleged reason - or at least part of it - is that if one seriously violates the syntax and semantics of English, one does not produce any English, while i f one lies, for example, one 's utterance is still English. My answer goes like this: if I confuse the semantics of 'cat ' and 'dog' and think that 'dog' means 'cat ', any 'dog'-sentence could still be perfect English. In principle, I could associate my perfectly grammatical utterances of English with an entirely m istaken semantics. The minimal conclusion is that semantics does not define a language in the same way as syntax does. I cannot find fault with the claim that pragmatics does not define language in the same way as syntax and semantics do. 1.4 Sterelny correctly claims that, given the right context, sentence types can implicate anything whatsoever. He wrongly concludes that there are no interesting relations between the meaning and form of sentence­ types and what they i mplicate. I disagree and the reason for disagreeJOU RNAL OF SEMANTICS, vol. l , no.4, pp. 399-400

399

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

0 In a short notice in this Journal ( 1 : 1 87- 1 94) Kim Sterelny raises some objections to the notion of conversational i mplicature. In an even shorter notice I want to rebut these objections.

JOHAN V AN DER AUWERA ment is the fact that Sterelny fails to mention the essential distinction between particularized and generalized conversational implicatures. The generalized i mplicatures do bear an interesting relation to features of sentence-types. Not surprisingly, Gricean linguists have been focussing their research on the generalized implicatures.

Belgian National Science Foundation

400

JS, vol. l , no.4

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

2 Sterelny claims that the Gricean hypothesis does not have the explanato­ ry power it is supposed to have: when a hearer computes a conversational implicature, he should not rely on any general principle of relevance. He should rather use the occasion-specific estimate that the speaker is being relevant. Answer: in order to study the linguistically interesting conversational i mplicatures, i.e. the generalized ones, lin­ guists do not need to operate w ith uniquely specific expectations of hearers. ( Fortunately,) general conversational principles will do.

Announcements

The topic of the Colloquium is: Meaning and

the Lexicon .

Price: around DM 300 or 350, including registration, hotel accommoda­ tion, and full board. Further information is given on a separate sheet, enclosed with this issue of the Journal. The announcement-sheet can also be ordered from the Nij megen Institute of Semantics P.O. Box 1 454 650 1 BL NlJMEGEN - Holland

REASON, MUSIC, EMOTION. A com mon structure for arts, sciences and philosophies? International Symposium organized by C O M M UNICA TION AND COGNITION at the University of Ghent, Belgium. December 1 2 - 1 6, 1 983.

The organizers want to raise the question whether a conceptual and terminological framework could be designed which could be relevant to the description of music and other arts in the light of scientific and/or philosophical systef"::S· Information can be obtained from: Communication and Cognition Blandijnberg 2 B-9000 GHENT (Belgium)

JS, vol. l , no.4

40 1

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

The Nijmegen Institute of Semantics N IS organizes the 2nd International Colloquium on the Interdisciplinary Study of the Semantics of Natural Language, August 30th - September 2nd, 1 983, in Cleves, West-Germany.