This excerpt from Presumptive Meanings. Stephen C. Levinson. © 2000 The MIT Press. is provided in screen-viewable form for personal use only by members of MIT CogNet. Unauthorized use or dissemination of this information is expressly forbidden. If you have any questions about this material, please contact
[email protected].
Introduction
This book is about meaning, but about a rather special subpart of general utterance meaning,l the pragmatic penumbra closely surrounding sentence -meaning: it is about utterance-type meaning, not the utterancetoken meaning that has been a major focus of recent work in pragmatics. Utterance-type meanings are matters of preferred interpretations- the presumptive meaningsof the title of this book- which are carried by the structure of utterances, given the structure of the language, and not by virtue of the particular contexts of utterance. In certain ways, the ideasin this book are conservative, and the majority of them are not new (the debt to early work by Atlas, Harnish, Horn , and Gazdar will be particularly evident) . But these ideas are in danger of being eclipsed and forgotten before their promise has beenproperly appreciated, so in the repetition of old information I remain unapologetic. Part of the motive of the book is therefore to collect theseprevious ideas and data together in one place, so that the systematicity of the observationscan be directly judged and their implications properly appreciated. In other respects, however, the ideas presentedhere extend the old ideasjust far enough for us to seethat they are potentially quite explosive: they threaten many of the orthodoxies about meaning and form in the languagesciences . I have tried to avoid technicalities. In part, this is becausehalf a dozen years interposed between the first and seconddrafts of this manuscript, and this time revealedto me the passingnature of many of the formalisms and formal theories to which the material in this book could be related.2 The observationsin this book seemto have a genuineindependencefrom such passingclouds, and so it seemswoI,thwhile to detach them from the technicalities. Nevertheless, I supposeit is inevitable that there will be passagesthat are rather hard going. I have therefore thought it worthwhile to sketch here in the introduction in a quite informal way the kind
2
Introduction
of picture of linguistic communication that will emerge from this study . I will do so partly through analogy . Consider the Rembrandt sketch in figure 0.1 (from Toney 1963, plate 94) . We interpret this sketch instantly and effortlessly as a gathering of people before a structure , probably a gateway ; the people are listening to a single declaiming figure in the center . You and I may see slightly dif ferent details adumbrated : for me, the declaiming character has a beard , and the man in the foreground to the left a hat , and the sitting figures seem female , but you may have your own interpretations . Knowing something about Rembrandt 's times and the subjects he was fond of , we can be fairly sure this is Christ telling parables to a crowd before a city gate . But all this is a miracle , for there is little
detailed
information
in
the lines or the shading (such as there is). Every line is a mere suggestion ; some lines seem to make no sense except as flourishes
of the pen ,
but yet we struggle to find some sense within them (see for example , the three little flourishes by Christ 's right " hand ," which suggest perhaps the heads of further listeners). So here's the miracle : from a merest, sketchiest squiggle of lines , you and I converge to find the adumbration of a coherent scene (that we may diverge , according to fantasy and our knowl edge of art history , is of course less miraculous ) . How is it possible? Although the theory of vision is considerably more advanced than the theory of the language capacity , current knowl -
edge cannot really explain our understanding of visual information as " degraded " or physically impoverished as this (but see Changeux 1994). What we do know is that there is tremendous visual sensitivity to outlines and shading and that we can use these (although normally reinforced by multiple further sources of information like parallax , stereopsis, color , texture , and reflection ) to extract three-dimensional models out of two dimensional retinal arrays . Somehow , then , Rembrandt is exploiting the fact that we are " built " to see three-dimensional scenes in squiggles on a plane surface. He is also relying of course on cultural conventions about how to represent scenes in drawings , using the post -Renaissance prefer ence for temporal instantaneousness and correct perspective , as opposed to , for example , the Central Australian cultural conventions in drawing a bird 's-eye view from above, which represents entire episodes over time . Another important ingredient is that the sketch is presented as a representation
, not , say , as scratch
some kind
of compact
marks
between
on linoleum
sketcher
, and
and viewer
as a result , there
that
warrants
is
the
4
Introduction pursuit of represented detail - there is an assumption that with proper attention
we should be able to recover the artist ' s intentions .
The problem of utterance understanding is not dissimilar to this little visual
miracle
. An
utterance
is not , as it
were , a veridical
model
or
" snapshot " of the scene it describes, although much talk of truth conditions might lead one to suppose the contrary (the work of the early Wittgenstein explored this initially attractive idea , which has never been quite washed out of our thinking since). Rather , an utterance is just as sketchy- as the Rembrandt drawing . Given the way we are built and the cultural conventions that specify a particular language and its appropriate
deployment, we are inexorably led, at least in most cases, to a common understanding (and , as in the Rembrandt sketch, we even have a clear sense of where we may diverge) . It is the apparently deterministic nature of
this
process
that
underlines
the
miracle
. A
sentence
like
" It
will
be
ready soon" is indeed sketchy like the Rembrandt drawing : it may refer to anything under the sun as long as it is not a human adult , will be specifies an infinite series of future temporal spans, and ready might be interpreted as 'cooked ' (as when the utterance is a response to the question " When is supper?" ) or 'surfaced ' (as in response to " When will the motorway be open?" ) or 'refuelled ' (in response to " When does the flight leave?" ) . And clearly , soon can take on quite different values in the context of serving a meal or finishing a doctoral dissertation . But in each of these differ ent contexts , " It will be ready soon" has its appropriate and apparently determinate interpretation effortlessly delivered to us. How ? That is the question that a theory of utterance comprehension has to answer. As with the visual processes behind the interpretation of the Rembrandt sketch, let us confess that we don 't really have the faintest idea how it works . First principles even seem to suggest that the recovery of a speaker's intentions on the basis of what he has said is in principle impossible ( Levinson 1995b) .3 Books like those by Sperber and Wilson ( 1986) or Atlas (1989) or Horn ( 1989), or the present effort , which attempt to spell out some of the pragmatic processes involved , are pretty much stabs
in the dark . What
I have
tried
to do in this
book
is detach
a little
part of this problem and argue that there must be powerful heuristics that give us preferred interpretations without too much calculation of such matters as speakers' intentions , encyclopedic knowledge of the domain being talked about , or calculations of others ' mental processes. Such preferred interpretations may be overridden by , and are certainly supplemented with , calculations of just this complex kind . But it seems implau -
5
Introduction
sible that the phenomenologyof instantaneous, determinateinterpretation could be delivered solely by reasoning about such matters as the potentially infinite regressof what the speakeris thinking the hearer will think that the speakeris thinking , and so on, ad infinitum .4 Issuesof cognitive processinghave played an explicit role in only one pragmatic theory so far: in Sperberand Wilson's (1986) Relevancetheory, the addresseeinfers as much as he or she can for the processingbuck, following a mini-max strategy. Somehow this must have some general correlation with the speaker's communicative intention. I do not think that the Sperberand Wilson framework can offer an account of the phenomena discussedin this book, precisely becausethe inferencesin question are relatively invariant over changes in context and background assumptions, despitebeing defeasible(i .e., they do not go through in the presenceof contrary assumptions). It is this relative invariance that gives these inferencestheir linguistic importance, as systematic feeders to semantic processesand language change, and I think the rich observations that have beencollected by Horn (1972, 1989) and others need to be preserved. Explicit processing considerations do not enter the framework offered here, but they do form part of the background, for the character of the inferencesin question as default inferencescan, I think , be understood best against the background of cognitive processing. The evidence, so far as it goes, from the psycholinguistic literature is that hypotheses about meaning are entertainedby the hearer incrementally- as the words come in, as it were. It seemsunlikely that this could be achievedwithout rich heuristics, becausethere will be early moments in utterance processing where propositional or clausal information for inferencing is simply not yet available. For this reason, the Sperber and Wilson framework is not psycholinguistically plausible, or at least it is not plausible for early moments in utteranceprocessing. In contrast, someof the heuristicsto be offered here can proceedon a word-by-word basis- for example, a scalar quantifier like some will , as I will show, already invoke default enrichments before the predicate is available. Close observersof this line of thinking have noticed that it constitutesa renaissanceof information -theoretic ideas: Thesenotionswentout of fashionin theoreticallinguistics - andI think thisis the right way to put it- whenChomsky(1956, 1959, etc.) criticized(correctly) their association with radical, tabularasa, behaviourism andfinite-statemodelsof languageacquisitionand of grammaticality . Their rehabilitationwithin the framework of more satisfactorymodelsof the structureand useof languageis very muchto bewelcomed . (Lyons 1995:239- 240)
6
Introduction Let me return now to the kernel idea behind the present work , which is disarmingly simple. The central background fact, an information theoretic observation, is that human speechencoding is relatively very slow: the actual processof phonetic articulation is a bottleneck in a system that can otherwise run about four times faster (a point taken up in chap. 1). The pressuresthat this exerts on language are easy to see- for example, the pressurefor frequent words to be reduced, as has been long documented (e.g., by Zipf 1949). Zipf (and, more recently, Horn 1984, 1989) saw the resulting pattern as a balance between two forces: a speaker's desire for economy and an auditor's need for sufficient infor mation. This tension, real enough, is not the focus of the current work . Instead, I assumea kind of coincidence of interests, treating linguistic communication as a " game of pure coordination~' in the game-theoretic senseof Schelling (1960), a picture I think presupposedby Grice (1957): the speaker is trying to find an economical means of invoking specific ideas in the hearer, knowing that the hearer has exactly this expectation. Now , the solution to the encodingbottleneck, I suggest, is just this: let not only the content but also the metalinguistic properties of the utterance (e.g., its form) carry the message . Or, find a way to piggyback meaning on top of the meaning. How can this be piggybacking be achieved? Only by utilizing the form, the structure, and the pattern of choiceswithin the utterance to signal the extra information beyond the meanings of its constituents. So, here I proposejust three simple heuristics, which will serveto amplify utterance
content: 1. If the utteranceis constructedusing simple, brief, unmarkedforms, this signalsbusinessas usual, that the describedsituationhas all the expected , stereotypical properties ; 2. If, in contrast, the utteranceis constructedusingmarked, prolix, or unusualfonns, this signalsthat the describedsituationis itselfunusual or unexpectedor has specialproperties; 3. Where an utterance contains an expressiondrawn from a set of contrasting expressions , assume that the chosen expressionsdescribe a world that itself contrastswith those rival worlds that would have been describedby the contrasting expressions. This description is impreciseof course. However, one can imagine formal treatments of this IIJeaningamplification along the following lines. Let the metalinguistic information control the model in which the utter-
7
Introduction
ance is interpreted: let it pick out of the set of possible worlds a proper subset that has the kinds of properties signaled by the metalinguistic properties of the utterance ; let it determine the domain of discourse; let it restrict the model to one in which certain extra properties obtain . In the body of the book , especially in chapter 2, many examples will be given that indicate the more precise directions in which these constraints on interpretation must be formulated . The essential point here is that no special, exotic , hermeneutic principles are being proposed : metalinguistic scrutiny by the addressee of utterance form , including the consideration of salient alternates , will be sufficient to amplify the content of utterances and ease the pressure on the encoding bottleneck ; knowing which , the speaker is bound to conform to the same heuristics . Approaches to meaning are so diverse and fractionated that it may help the reader
to know
where
the writer
stands
on a wider
range
of issues than
are directly addressed in this book . So, let me declare my cards: 1. The distinction between semantics and pragmatics is one of a number of essential distinctions in the study of meaning ; it may be that in the long run the distinction will dissolve into a larger set of distinctions , but nothing is gained by lumping in our attempt to understand human communication
.
2. Semantics is not to be confused with " conceptual structure " or the " language of thought " ; the semantics of a language is a language-specific phenomenon , and the mapping into a " language of thought ," a nontrivial relation between nonisomorphic structures ( Levinson 1997) . 3. Aspects of semantic content (usually only when enriched by prag matic specifications ) can be specified by the apparatus of a recursive truth definition , but this is unlikely to have a direct cognitive counterpart . Consider the analogy to vision : we can objectively specify the relation between the world that is viewed and the signals reaching the visual cortex ,
and compare the subjective visual experience. But this psychophysical mapping is not always truth -preserving : there are numerous conditions under which we see things that do not exist, fail to see things that do , and
so on- the stock of illusions that psychophysiciststake as a central part of their job to explain . Similarly , truth -conditional semantics viewed in the realist way - as a direct veridical mapping of semantic structures onto states of affairs (bypassing the head as it were)- is useful as a yardstick of human performance . Something like this is what the cognitive processes must do , but as in the case of visual illusions , they may fail to do so,
and how they generally do so will be unrelated to the machinery of truth -
8
Introduction conditional
semantics
. So we can have
our cake
and eat it , too : we can use
the insights of truth -conditional semantics without buying Realism , and without caring that it obviously fails to meet any criteria for adequacy as a cognitive model . It is a bit like buying a pair of shoes: we are glad that the salesperson can measure our feet and get the size roughly right , but in the end, it 's a subjective question whether the shoes fit . 4. There is no algorithm that , given a syntactic string in a language , cranks out its unique logical form or semantic structure . The view that there is such an algorithm forms the basis of much linguistic theorizing , from Montague to Chomsky 's most recent views . But it is patently absurd to hold such a view . First , there is the enormous range of ambiguities in natural language (requiring at least a one-to -many correspondence) . Second , syntactic structures may actually be indeterminate in certain respects ( Matthews 1981: 17- 21) . Third , pragmatic resolution is crucial before semantic interpretation or the assignment of semantic structure : nobody disputes the role of deixis here, but there are many other aspects of prag matic resolution from the determination of anaphoric reference to the assignment of scope. The overall picture sketched by Sadock ( 1991) or Jackendoff ( 1997) is here at least correct : phonology , syntax , and semantics are areas each with their own generative capacities , and there are significant mismatches between the structural strings in each representa-
tion that, in the end, come to ~.e associatedwith one another. What puts them into association with one another are correspondence rules or pro cesses, which
are not detenninistic
in character . J ackendoff
suggests that
the picture with lexical items is not in principle different : they are corre spondences across phonological , syntactic , and semantic representations , but the correspondences may be partial and not one to one (compare idi oms, or the mismatch between phonological and syntactic words ) . 5. Insofar as we can get from syntactic structures and lexical material directly to a semantic representation (which in the most part , I argue, we cannot ) , then such semantic representations are only partially specified; templates of partial information far too unspecified to determine truth conditions
.
6. The overall role that pragmatics plays in such a picture is funda mentally different from that sketched in the textbooks (including mine of
1983). There is no schemeof the kind that syntactic structuresare mapped onto semantic structures which themselves represent full -fledged proposi tions , these semantic structures being the input into pragmatics , which yields additional inferences or restrictions on meaning . Rather , pragmatic
9
Introduction
processesplaya crucial role in the correspondencerules mapping syntactic structuresonto semanticrepresentations; and again mapping semantic representationsonto communicated thoughts or utterance meanings. The picture of the overall theory of meaning that will emerge from this book is thus radically different from the starting point from which Gricean pragmatics began. The distinction between semanticsand pragmatics was construed in the standard theory as the distinction between sentence -meaning and utterance-meaning, with the output of the semantics being the input to the pragmatics. For reasonsdetailed at length in chapter 3, this cannot be right . Instead, I argue, we should stop thinking of the distinction in terms of levels of representation. Instead, we should think about both semanticsand pragmatics as being component processes that offer their own distinctive contributions to a single level of representation . The processesremain distinct in kind , and thus the distinction betweensemanticsand pragmatics must be retained. This new view of the architecture of a theory of meaning is one of the explosive consequences , mentioned at the beginning, of careful observations about the apparently minor phenomenaof preferred interpretations. Another far-reaching set of consequenceshas to do with constraints on syntactic form, explored in chapter 4 through consideration of anaphors, where it is argued that patterns of preferred interpretation explain the distribution and typology of anaphoric expressions . As background to all this, chapter 1 lays out the reasonsto suppose that preferred interpretations do indeed exist and have a life of their own. Chapter 2 offers a catalog of these phenomenain terms of three simple neo-Gricean principles, and sketchestheir interaction. The reader will find many familiar facts here assembledin perhapsunfamiliar ways. A catalog may make tedious reading, but I think there is no other place where the student can find all thesephenomenalaid out, and it is important for the thesis of this book that the theory of preferred interpretations is seento have perfectly generalapplication. This then is the structure of the book: a chapter that introduces the idea of presumptive meanings, a chapter that lays out many diverseexamples and a schemethat organizes them, a chapter that explores the consequencesfor theories of semantics, and a fourth that explores the impli cations for syntax.
This excerpt from Presumptive Meanings. Stephen C. Levinson. © 2000 The MIT Press. is provided in screen-viewable form for personal use only by members of MIT CogNet. Unauthorized use or dissemination of this information is expressly forbidden. If you have any questions about this material, please contact
[email protected].