No title

JOURNAL OF SEMANTICS AN l}.'TERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY S11JDY OF THE SEMANTICS OFNATURALLANGUAGE VOL...

Author: Oxford University Press

5 downloads 378 Views 18MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

JOURNAL OF SEMANTICS AN l}.'TERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY S11JDY OF THE SEMANTICS OFNATURALLANGUAGE

VOLUME 5, No. 2, 1986/87

R�pr1111td wills P''MWIC4 of Foru Pwbl.cottOISS Dordru.IU

by

SWETS & ZElTLINGER B.V. LISSE- THE NETHERLANDS· 1991

Journal of Semantics 5:

89-121

SEMANTICS AND TEXT MEANING: RETROSPECTS AND PROSPECTS ROBERT DE BEAUGRANDE

ABSTRACT

is a specialized mode of being meaningful. The relation

between that mode and the general mode of ordinary discourse should be clarified withm a framework: that integrates p!Ut trends in semantic modelling with each other and With possible future ones. This paper proposes to VIew language as a complex system of control levels with characteristic distnbutions of determinacy. Here, meaning 1s descnbed as a processing event whose structure

is formed by the maintenance of control and the limiting of indeterminacy.

Known properties of complex physical and biological systems offer some clues about how such operauons might occur.

I. SEMANTICS

1. 1.

AS A CHA LLENGE FOR A SCIENCE OF LANGUAGE

A classic issue whose investigation has remained tenaciously proble

matic is the relationship between language and meaning. When Saussure ( l966[1916]:xix) formulated many fundamental concepts and positions in linguistic theory, .. semantics" (as his editors remark) was "hardly touched upon". His passing reference to ..the accidental and peculiar character of semantic facts" (1966:93)1 may be a clue to his motive for the omission: semantics would have offered a considerably less auspicious domain for his central lines of argument. The abstractions and dichotomies he proposed between "social" and "individual", between "essential" and "accessory", and in particular between "language" and "speaking" ["langue" and "pa role"] (1966:14)- are much more workable for the study of sounds than for the study of meaning. Had Saussure inaugurated modern linguistics on the basis of semantics rather than of phonetics and phonology, the whole theore tical superstructure might have looked quite different.

1. 2.

In a very influential work for North American research (the area I will

focus on in this section), Bloomfield (1933) was more explicit on this point. He situated "meaning" close to "the speech-utterance" (which would corres pond to "parole" in the Saussurian framework), because "the meaning of a speech-form for any speaker is a product of the situations and contexts in which he has heard this form" (1933:27,407). Bloomfield complained that "our knowledge of the world" "is so imperfect that we can rarely make accurate statements about the meaning of a speech-form" (1933:74). Such

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

The activity of "doing semantics"

90

"meanings" "could be scientifically defined only if all branches of science, i ncluding, especially, psychology and p hysiology, were close to perfection" (1933:78). Meanwhile, "the study of language can be conducted without special assumptions only so long as we pay no attention to the meaning of what is spoken"- whereupon Bloomfield, like Saussure, turned to "phone tics" and "phonology" (1933:75). My case would therefore be t hat certain early conceptions in linguistic theory made the study of semantics hard to pursue, because the latter spills out beyond many of the major definitions and categorizations scholars took to be fundamental. The most common result was that research confronted semantics with strategies for elimination and reduction. In America, some linguists proposed to describe language structure without any reference at all to meaning (e.g. Harris 1951; Trager & Smith 1951; Chomsky 1957). But the elimination tactic was more a profession offaith than a practicable methodo logy. Even if the investigator could genuinely disregard meaning during the process of analysis, and I know of no conclusive evidence ever being adduced that this is feasible, the investigation could never honestly claim to reflect what ordinary speakers do in communication; we would obtain only the results of a remarkably specialized and atypical approach to language. 1.3.

These procedures seem reasonable enough for the intended purposes. The ultimate empirical basis for grammatical word-classes in a language like English, where many class distinctions are not marked in the word-form, is precisely the tendency of ordinary language users to place words of a given 1.5.


Hence, reduction became a far more prevalent strategy than elimina tion for dealing with meaning. Scholars would allow meaning some restricted place in their theorizing. Saussure (1966:105,186,106) used "meaning" as a criterion for recognizing "linguistic units", such as "roots", "preflXes", and "suffixes", though he offered only examples rather than a theoretical ratio nale. Bloomfield ( 1933:77,93,128) used "meaning" somewhat similarly as the means to tell "whether two uttered forms are 'the same' or 'different"' and which "features" are "distinctive". Fries (1952:56,294,74) singled out "structural meaning", warning that it "does not account for all the meanings in our utterances"; "the grammar of a language consists of the devices that signal structural meanings", these signals "consisting primarily of patterns of arrangement of classes of words" (1952:56,74). The members of a given class can be substituted for each other in an "utterance frame". If we "make certain whether with each substitution the structural meaning is the same ", we need not "define the structural meaning" itself (1952:74). Generative grammar adopted the comparable conception that syntactic transformations do not change the meaning of a sentence (Katz & Postall964). If formal rules can be designed on this assumption , the description of meaning itself might be postponed until the grammar is finished (should that ever happen).

1. 4.

91 class in some utterance positions or sequences and not in others. "Structural meaning" could be said to subsume grammar's contribution to meaning. In the same vein, Stockwell ( 1977:32) proposed for formal grammar a "semantic representation" that would "single out those aspects of the total meaning which are carried by the syntactic structure of sentences". Though this kind of meaning was given a privileged status for being the easiest to relate to issues of form , we have no compelling reason to expect that other aspects of meaning can be covered or included within the same approach. Lakofrs ( 197 l a: 269) principle for "generative semantics", namely that "semantic representations and syntactic phrase-markers are formal objects of the same kind", has not been terribly successful.

In retrospect, if we regard "semantics" as a whole and as an enterprise carried in philosophy as well as linguistics, we might arrive at a provisional classification of approaches that attempt to make the domain tractable in particular ways:

1.7.

In some abstract "lexicon" or ideal dictionary, the meanings of all words are conclusively defined ( cf. Gleason 1962; Chomsky 1965, 1972; Pete>fi 1980). Being meaningful would be like finding a word for a definition , or looking up a definition for a word. But since the whole lexicon isn't available to any single normal language user, we need to inquire how he or she creates and uses some version of it with the efficiency everyday communication reveals.2

1. 7.1. Lexical definition.

Meaning is a composite of basic minimal entities ("features") like "animate" versus "inanimate", etc. (cf. Greimas 1966; Nida 1975). Using a word-meaning would entail activating or recovering its sets of features. This conception is easy to combine· with the foregoing one by envisioning a "lexicon" built of semantic features (cf. Katz & Fodor 1963). That scheme would compact the lexicon, provided there are fewer features than lexical entries: but it might turn out that the number of features needed to fully characterize all definitions is as large as or far larger than the number of entries. Moreover, many features only become relevant to a word's mean ing within a particular context and would hardly be expressly stated in the lexicon (Bolinger 1965). Here too, the division between "language" and "speaking" becomes difficult to uphold ( l . l ). 1.7. 2. Semantic featurn


1.6. Moreover, the idea that "structural meaning" stays the same during "substitution" or "transformation" is problematic and hard to support with independent, non-circular, and realistic tests. People are habituated by nor mal communication to judge the sameness of meaning by looking at what is relevant for the context - including many aspects of meaning beyond the contribution of grammar or syntax. So the notion of "structural meaning" may be the exclusive property of specialized analysts.

92 I.7.3. Propositions.

Meaning involves assigning to a "predicate" one or

more "arguments", i.e., giving some property to one or more names of something (cf. Miller 1970; van Dijk 1977; Jarvella 1977). This approach treats language as a device for asserting (presenting as true) properties or relations, or making judgements, about objects. The difficulty is again that the context, not just the logical structure, of an utterance decides what is being "predicated" versus what is already presupposed; properties may be mentioned for many reasons other than to assert them, e.g. to single out one item from similar ones within the current visual field (Krauss & Weinheimer

1967; Osgood & Bock 1977). The meaning of each sentence is an "inter

1972; Stockwell 1977). Generativists disagreed about where the semantic interpretation should fit into the design of the grammar, e.g., within the ratio between "surface" and "deep" structure. Formalisms were proposed mainly in terms of the conceptions listed in 1. 7.1-3. Here too, the person doing the interpretation and the operations involved were not made the focus of investigation.

I. 7.5. Behavioral response.

The response of the text receiver equals the

meaning of the text; the receiver's behavior reveals and enacts that response

(cf. Bloomfield 1933; Skinner 1957; Quine 1960). Though this approach purports to deal solely with observed events, the observable substrate of textual, communication is typically impoverished in relation to the total information conveyed. Moreover, very different meanings can elicit the same outward response and vice-versa (cf. Chomsky 1959).

I.7. 6. Referenu.

Things or states in the "world" are "referred to" by

language; if the language denotes an actually existing state of affairs, it is "true" (cf. Frege 1892; Russell 1905; Strawson 1950; Linsky 1957). This approach also externalizes meaning, not as overt behaviour, but as the manifest world of objects and properties. Acute problems arise regarding the range of objects referred to- their existence, their number, their necessary or defining properties, and so on.

I.7. 7. Intention.

Meaning is what a text producer intends the receiver to

understand (cf. Gardiner 1932; Grice 1957; Searle 1969). The intention might be to make a predication, elicit a response to something, or whatever; but the intention is the ultimate recourse for knowing what is meant. Here, the problem is how an intention is formed and implemented, and how it may succeed or fail. Only a few clear-cut cases, such as making promises, have been analyzed in detail so far.

I.8.

Within these seven approaches, three main conceptions of meaning


I. 7. 4. Semantic interpr�tation.

pretation" of its logical forms and relations (cf. Chomsky 1972; Jackendoff

93 may be recognized:

representation. externalization. and intentionalization.

As

far as I know, the reconciliation of these conceptions has not been accomplis hed so far and certainly would not

be easy. The representational conception

relies on after-the-fact transcriptions of interpretive acts by trained analysts, and these acts differ both in behavior and in intention from the acts of everyday language users. The externalizing conception relies on debatable parallelisms between world and language, and ignores the way speakers and hearers decide what's going on and what to say about it. The intentionalizing conception tends to obscure the way logical conditions and the state of the world (among other factors) constrain the human will.

1.9.

abstractions widely practised and accepted by linguistics dilute or obscure the very factors and conditions a comprehensive study of meaning crucially entails. If, as I have suggested, meaning is strongly determined by contexts of communication, then the removal of contexts proposed since Saussure tends to make meaning into an anomalous phenomenon, of which only carefully selected aspects can be considered. Perhaps empirical research might place the rationale of conventional linguistics on a realistic footing (Section 2). Or, we may need to adopt quite different models wherein the status of semantics is made the center of concern from the start (Section 3).

2. SEMANTICS AND EMPIRICISM: A CASE STUDY

2.1.

The appearance of text linguistics and discourse analysis on the scene

renewed the challenge of meaning, since here at last was a domain in which context could no longer be discounted. The text is a realistic entity, an actually occurring event of language. The functions of its elements- words, meanings, intentions, and so on -are determined not merely by their place in some general abstract system, but by the relationships whereby the text itself is made communicative. Just as Lyons (1977:643) tries to "maintain the distinction between sentence meaning and utterance meaning" despite the frequent confusion in linguistics, the meaning of the text is not to be account ed for by some addition of sentence meanings (Beaugrande 1980). The concern for the specialized analysis into abstract basic units, practised by linguists in the tradition of Saussure and Bloomfield, must be complemented with a concern for the everyday synthesis of concrete texts, practised by ordinary language users.

2.2.

We can appreciate the current dilemma by making a detailed case study

of one project for developing an approach to text meaning that would reconcile standard views in linguistics and philosophy with the exigencies and tools of empirical, i.e., psychological and social, research. In reviewing David


In sum, semantics remains an ongoing challenge which the science of

language is still trying to meet. I cannot resist the impression that many of the

94 Olson's work, I do not mean to single him out as a promulgator of erroneous ideas he has devised. Instead, I believe his work, and the problems it raises, are fully symptomatic of any attempt to retain established frameworks when moving to empirical inquiry. Given the conflicting pressures involved, such an outcome is only to be expected.

2. 4. This argument is intended to justify the slogan, "the meaning is in the text" (Olson 1977a:262,264; 1977b:l6; etc.) The interested speaker is displa ced by the disinterested writer who disappears behind the text, at least as an agent controlled by implicit societal influences such as values. Only "expli cit" materials are to be considered, rather than the mass of world-knowledge humans might use during text processing. Hence, text meaning is restricted to those indicators that can reliably serve to define the subset of "assumptions and premises" a strictly literal reading would favor. To draw a dichotomy between "speaker's meaning" and "sentence meaning" (prefigured in Grice 1957) is to uncouple representation from intention and to discount the notion of "writer's meaning" as anything besides what a text appears to say on the surface. 2.5. A further reduction limits us to one special type of written text, namely "the explicit written prose that serves as the primary tool of science and philosophy" (Olson 1977a:271). There, "the intended meaning is exactly represented by the sentence meaning". This domain is thus the "particular specialization of language" to which "Chomsky's theory applies". Such an argument may be something of a curiosity in science: an existing theory is judged correct, but its usual domain of application wrong. If instead of "speaker and hearer", Chomsky and his followers had said "writer and reader of logical prose", would the problem of meaning have become resolva ble in their terms? 2.6. Olson offers several modes of support for his argument: (a) evolutiona ry, (b) epistemological, and (c) experimental. In the �volutionary perspective,


2.3. One of Olson's major reductions is to limit "text" to written texts only, a domain he considers particularly well-structured. "To serve the require ments of written language", "all of the information relevant to the communi cation of intention must be present in the text"; "written text" is "largely responsible for permitting people to entertain sentence meaning per se, rather than merely using the sentence as a cue to the meaning entertained by speaker" (1977a:277). Moreover, "truth drops its ties to wisdom and to values, becoming the product of the disinterested search of the scientist"; indeed, "the bias of written language toward providing definitions, making all assumptions and premises explicit, and observing the formal rules of logic produces an instrument of considerable power for building an abstract and coherent theory of reality" (1977a:276f).

95 he avers that modern prose is the cu lmination of a trend initiated by the Greeks, whose invention of the alphabet "was responsible for the develop ment of the intellectual qualities found in classical Greece" ( 1 977a:266; cf. Havelock 1973). Or, to follow Goody's ( 1 977) diction expropriated from Levi-Strauss, writing systems helped to "domesticate the savage mind". Olson suggests that "the decrease in ambiguity of symbols" "permits a reader to assign the appropriate interpretation to a written statement even without highly tuned expectations"; hence, "written language became an instrument for the preservation of original statements that could violate readers' expectancies and commonsense knowledge" ( 1 977a:266). Later, "the rise of print literacy" brought "a further evolution in the explicitness of writing on the semantic level"; henceforth, the "writer's task:" was "to create an autonomous text - to write in such a manner that the sentence was an adequate, explicit representation of the meaning, relying on no implicit premises or personal interpretations" ( 1 977a:268). The main illustrations are the "texts" of "empiricist p hilosophy" and "empirical science" ( 1 977a:269). 2.7.

Olson's argument implies that cultures having not an alphabet, but a "logography" (e.g. Chinese) wherein each symbol represents a whole word or idea (cf. Taylor 198 1 ), have less chance to "preserve original statements"; and that pre-literate cultures should be beset by interminable ambiguities, being unable to construct "sentences with only one interpretation". By linking "the development of intellectual qualities" to an alphabet he happens to share, Olson projects a cultural chauvinism against communicative systems unfami liar to him, as if they were uniformly tools of a "savage mind". 2.9.

Yet the languages of pre-literate cultures often have vastly more complex and differentiated formal grammatical categories than does current English, witness the Amerindian languages studied by Whorf ( 1 956). Perhaps this complexity and differentiation help to maintain the precision and clarity

2.10.


2.8. Thus, Olson's evolutionary account foresees two "steps in the creation of explicit, autonomous meanings": the "invention of an alphabetic writing system", and the invention of p rinting that "allowed a given sentence to have only one interpretation". As a product of this evolution, "the attempt was to construct sentences for which the meaning was dictated by the lexical and syntactic features of the sentence itsetr' ( 1 977a:270) - precisely the sort of sentences that conventional linguistics is well equipped to handle. Olson depicts the overall historical and social trends of the centuries as leading directly to that mode of discourse he feels would best suit modern linguistics as a "normal science", which is exactly the typical retrospection Kuhn ( 1 970) diagnoses for "normal sciences" of all kinds. The specialized activities of contemporary investigators encourages a special interpretation of general movements in whole societies of the past.

96 that other cultures pursue with writing systems. Olson uses historical inven tions in postulating an unduly facile causality between the

dissemination of

meanings and their conception- a tendency we also find in the theorizings of Marshall McLuhan and Walter Ong, whom Olson ( 1977a:268) cites as autho rities.

2.11.

Olson's

epistemological evidence coincides with the evolutionary.

Olson proclaims writing to be a precondition for "literal truth", "critical analysis", "the abstraction of logical procedures that could serve as the rules for thinking", "the development of abstract categories", "formulating a definition", and "theoretical synthesis" at large (1977a:266-270). Disregard ing whole communities of scholars who steered the history of ideas - the East, the mediaeval scholastics, and many more - Olson singles out "the British essayists", notably John Locke, as "among the first" to "exploit writing for the purpose of formulating original theoretical knowledge" ( l977a:268), ostensibly because they wrote shortly after the rise of printing presses, but more likely because their prose best fits Olson's notion of "text".

2.12.

Thanks to "its demand for explicitness of meaning, its permanence as

a visible artifact compatible with repeated scrutiny and reflection, and its realignment of social and logical functions", writing "serves the intellect" as "an

�ssential means for the formulation of the abstract true statements that

constitute objective knowledge", and as a "criticaf' factor in "the particular mental achievements we designate as conceptual intelligence" (1977b:l l , emphasis added). Again, the implication i s that non-literate cultures or individuals should be markedly inferior in their cognitive and communicative capacities.

2.13.

For a counter-example to his notion of "text" as logical prose, Olson

depicts

po�try as a memory device for offsetting the evanescence of spoken

texts: "significant oral-language statements, to be memorable, must be cast into some oral poetic form"; "the written statement, constituting a more or less permanent artifact, no longer depended on its 'poeticized' form for its preservation" (1977a:264). When the course of history moved toward "the further development of explicitness at the semantic level", "poetic state ments" were "not permissible because they admitted more than one interpre tation, the appropriate one determined by the context of utterance" (1977a:270).

2.14.

Here, Olson uncritically incorporates another piece of folk-wisdom,

this one going back at least to Plato, whose famous condemnation of poetry is echoed in spirit (cf. Hildyard & Olson 1982:20). Yet Olson should predict that the advent of printing would greatly decrease the need for poetry, since the preserved text would obviate memory supports for oral statements; and that


mathematicians and geometricians of Greece, Egypt, and the Islamic Middle

97 written poetry would be less ambiguous than oral. Just the opposite is the case: the production and reception of poetry increased enormously after the rise of printing, and modern poetry, composed in a time of unprecedented printing activities, has moved steadily further away from explicit, logical statements.

2.15.

Besides, literary authorities would agree that poetry is precisely the

domain where we counter a particularly high ratio of "original statements that violate readers' expectancies and commonsense knowledge" (cf. 2. 5). Many poetic statements have had a profound impact on the intellectual development of our culture because they seem "true" in a special sense- not as reports of facts, but as abrupt revealings of what is generally true of the

significances beyond our established conventions of discourse (cf. Beau grande l978a, l978b; Coseriu 1971; Jakobson 1960, 1968; Lotman 1976; Mukarovsky 1964; Riffaterre 1978; Schmidt 1971). This potential is not "parasitic" upon literal discourse (as speech-act theorists aver), but essential for the process of adapting discourse to changing circumstances.

2.16.

Another problematic analogy in Olson's epistemological evidence is

his assumption that "commonsense knowledge is related to theoretical knowledge in the same way that" spoken "utterances are related to" written "text" ( l977b:l3ff). Commonsense is "coded for action", specialized in the particular and concrete", "inconsistent", based on "illustration", and "va lue-laden". Science "is coded for reflection, not for action", "seeks universal laws", "eliminates contradiction", and "is value-free in the sense that it is the consequence of the 'disinterested search for truth'". In writing, a statement is proven "by virtue of compelling arguments, not by its utility or the autority of the one who speaks it".

2.17.

This argument reflects popular views about science. But an alternative

view is gaining ground among the scientists themselves, above all in the human sciences (cf. Cicourel 1964). "Scientific rationality is

not in conflict

with commonsense reasoning"; instead, "scientific rationality is embedded in commonsense reasoning" (Jennings & Jennings 1974 :249). " Focussing on the researcher's actual behavior shows that the world of commonsense experience and understanding the researcher shares with his subject is a greater resource than his ideal and explicit methods" (ibid.). Besides, "the authority of the one who speaks" is at least as crucial in scientific discourse as in most other domains, witness the practice of "blind refereeing" to eliminate the bias in judging materials for publication.

2.18.

Olson's

experimental evidence further illustrates the problems of re

lying on traditional theories of semantics. Assisted by Angela Hildyard, he


human situation. By opening up and constantly re-negotiating the relations between structure and meaning, poetry forces us to confront and create

98 has probed one implication of his proposal, namely: "in written language, the words and syntax, the 'sentence meaning,' is preserved by the artifact of writing, and mental recall becomes the precise reproduction of that artifact" (Hildyard & Olson 1 982:20). Here also, the writer's act of production is excluded from direct investigation, perhaps in the belief that the meaning is "in the text", not in the writer. The studies focus instead on text reception, listening and reading, plus such subsequent memory tasks as inferencing and verifying (Hildyard & Olson 1978, 1982; Olson & Hildyard 1 980).

2.19.

If writing elicits "an increase in attention to the 'sentence meaning,'"

then readers should "recall more of the surface structure of verbatim fea tures", whereas listeners should "recall more of the gist" (Hildyard & Olson

focus more on "pragmatic inferences" that "follow from non-explicit aspects of speaker's meanings" (Hildyard & Olson 1 978:96). Thus, people can be tested to see how well they recall which aspects of a story, and how well they can distinguish between what was actually said and what they supplied by inferencing.

2. 20.

Hildyard and Olson did find on verification tasks that the "listeners

performed better than the readers in judging the truth-falsity of those state ments which bore directly on the structure or meaning of the story as a whole", while "the readers were superior to the listeners on those statements which referred to incidental verbatim details" ( 1978: 1 1 6; cf. 1 982:30). Those findings are taken as evidence for a developmental projection: the "compe tence" of "preserving the semantic structure" and of "constructing a possible world stipulated by" the sentence "in the derivation of literal meaning" "develops only when the child is well into the school years and perhaps, only under the impact of schooling in an literate society" (Olson & Hildyard

1 980: 39). The implication is that pre-literate and non-schooled people lack cognitive "competence", which seems doubtful, unless "literal" meaning is by definition a product of "literate" culture, which makes the dependency circular rather than causal.

2. 21.

On the whole, does the experimental evidence actually support Ol

son's original theory of text meaning? The contention that Chomsky's gram mar is "completely appropriate for the analysis of logical prose" ( 1 977a: 1 9) was not in fact tested, though it is quoted whitout reservation in the 1 982 paper. In the 1 977 papers, all text examples were either invented "John-and Mary" sentences or brief children's utterances. Later, the empirical studies were done not on logical essays, but on stories. No doubt the prose of Locke is not so convenient to analyze; nor could many test subjects be found who habitually read it, let alone listen to oral presentations of it. But this type of text cannot be made the key domain for a semantic theory until it is specifical ly addressed.


1 982: 20). Similarly, readers should focus more on "propositional inferences" that "follow necessarily from sentence meaning", whereas listeners should

99 Stories differ from logical essays in ways directly relevant to the class of eviden.:e the Olson team brings forward about the recall of texts. Narrative organization provides "schematic" parameters of time, causality, and pur pose (goal-planning) people have been consistently found to use when com prehending and remembering stories (survey of research in Beaugrande 1 982). Such knowledge is applied whether the stories are presented orally (e. g. Stein & Glenn 1979) or in writing (e.g. Rumelhart 1977). Logical essays, in contrast, are organized according to principles for relating argument to evidence, general to specific, superclass to subclass, and so forth. In some of my own studies, students proved much less distracted by intentionally bad or obscure writing styles when the text they were asked to recall was narrative rather than expository; and this effect was found for both oral and written presentation (reported in Beaugrande 1979, 1 980). 2.22.

"Billy's Lunch. Th1ck. handleless cups on the wet oilcloth-covered counter. An odor of onions and the smoke of hot lard. In the doorway a young man audibly sucking a toothpick.."

Linguistic semantics would presume the reader to be converting these surface components into well-formed sentence frames of the "deep structure". Yet Hildyard and Olson's ( 1 982: 20) approach associates "sentence meaning" with "surface structure" and "verbatim features", which is not the level Chomsky's standard model ( 1 965) proposed as the input for semantic inter pretation . To the degree that the experiments of the Olson team presuppose the collapsing of deep with surface structure, they cannot bear directly on Chomsky's theory. Nor could they handle texts like Lewis's except with doubtful special provisions, e.g. , that the text is spoken language accidentally written down, or is somehow very hard to read because it fails to "provide all definitions, make all assumptions and premises explicit, and observe· the formal rules of logic" (cf. 2.3). 2.24. The more we try to determine just what "sentence meaning" includes or excludes, the less it resembles the straightforward semantic basis it is advertized to be. Compare these varying quotes (from Hildyard & Olson 1982:20; and Olson & Hildyard 1 980: 1 3ft): 1 . In written language, the words and syntax, the "sentence meaning", is preserved by the artifact. 2. The semantic structure is what is preserved in "the very words" or at least in the linguistic structure of the sentence.


So far, the stories used in Olson's experiments were composed exclusi vely of complete, fully grammatical sentences (examples in Hildyard & Olson 1 978 : 1 07, and 1 982:24). But suppose people were reading about Carol's tour of Gopher Prairie in Main Street by Sinclair Lewis ( 1 920:34-37). The passage contains some 73 sentence fragments, e.g.: 2.23.

100 3 . A sentence meaning is a function o f a possible world into a speaker's meaning or intention.

4. A semantic structure is a function from a context into a set of truth conditions.

5 . Sentence meaning as we use the term is roughly equivalent to the literal meaning of that sentence and equivalent to the logical or propositional structure underlying and expressed by that sentence.

6. Literal meaning [ . . . ] does not correspond to linguistic sentence meaning, but, rather, to an utterance spoken by a particular individual in a particu lar context on a particular occasion in such a way as to determine a set of truth conditions. Quotes

I

and 2 would equate "sentence meaning" with "semantic structure",

"sentence meaning" is equivalent to "literal meaning", and quote 6 says it isn't. Such vacillations cloud the opportunity for bringing evidence in sup port of the original theses about semantic theory.

2.25.

Similar inconsistencies appear when the narrow line of argument in

Olson's earlier papers is expanded in the later ones. 3 For example, "the determination of the set of truth conditions" is allowed to draw upon "background assumptions or knowledge of the world which is not marked in the semantic structure of the sentence" (Olson & Hildyard 1 980: 1 3). Testimo ny is gathered from Searle "that even in literal utterances, where speaker's meaning coincides with sentence meaning, the speaker must contribute more to the literal utterance than just the semantic content of the sentence"; and from Johnson-Laird that "there is no such thing as the literal meaning of a sentence, only the literal meaning that a given listener places on a given utterance of it" (Olson & Hildyard 1980: 1 3). At one point, Olson and Hildyard ( 1 980: 17) adopt the stance that "all meanings, literal or otherwise, are speaker's meanings"; and go beyond Searle, who "takes literal meaning as the unmarked case", to "suggest, rather, that casual speech be taken as the unmarked case and that literal, indirect, and metaphorical speech all be treated as special cases". 2. 26.

All these concessions undercut the strict opposition between "sen

tence meanings" versus "speaker's meanings" whereby the continuity of semantic theory was supposed to be maintained. In effect, the limitations I cited as the heritage of conventional linguistics and semantics proved unduly restraining and were relaxed, but apparently without realizing the full impli cations of the shift. The psychological experiment is a special situation in the sense that the focus of interest is placed on carefully selected factors of the context; the "variables" are deemed specific to this context, and the findings are deemed generally true for human beings in all comparable contexts. Thus. the division linguistics made between "language" and "speaking" or "com petence" and "performance" adopts a different format: experimenters scruti-


whereas quotes 3 and 4 would make them different notions. Quote 5 says

101 nize the performance in search of factors that would constitute the "unity" of "speech" that Saussure ( 1966:9) claimed "we cannot discover". He in fact had to assume that the "studies" of"the psychologist" do "not reach beyond individual execution" ( 1966: 1 7).

2.28. In sum, the attempts of Olson and co-workers to describe text mea ning upon a broad empirical and realistic basis without sacrificing central conceptions of conventional linguistic semantics is beset with problems and vacillations. The desire for confining research on text meaning to some wel l-formed sub-domain of overall communicative meaning is unders"Landa ble, but apparently not workable without introducing so many relaxations and qualifications that the formal grounds of the theory are seriously put in question.

It would of course be hasty to generalize from one case study. None theless, we have here a sincere and concerted atte mpt to do something often neglected by bridging the broad gap between theory and evidence in the difficult domain of meaning. Irrespective of particular experimental designs, sample texts. and test populations, I suspect that such projects would typical ly be forced to weaken or restate their original restrictions, idealizations, and abstractions, in ways roughly comparable to those we have seen here. We might profit therefore by considering whether traditional semantic theories such as those described in Section I may not be directly transferable into realistic models of meaning for everyday communication. Instead, they may represent a subset of factors such models should attempt to subsume and reformulate in a different framework. Section 3 outlines one possible candi date for making a start in that direction. 2. 29.

3. TEXT MEANING AS A DYNAMICS OF CONTROL

3. I. One plausible source for a model of text meaning is systems theory, which has been applied in a wide range of disciplines (cf. Bertalanffy 1962).


2.27. Recent research has witnessed an upsurge of interest in the compari son of spoken and written language (survey in Beaugrande l984a). Earlier work receives renewed attention alongside recent studies; but no clear consensus has emerged about the similarities or differences. Spoken texts can have a surface organization as complete and well-formed as written texts, e.g., in the language of radio broadcasters (Goldman-Eisler 1972) and public orators (Blankenship 1962). Even using a fine-grained analysis of linguistic structures, Blankenship found few differences between the speeches versus the prose essays of Adlai Stevenson, and concluded that "syntactic structure is determined by an individual's style" rather than by the "written/oral" dimension ( 1962:422).

102 Due to this generality, we are forced to specify the approach to fit the conditions of understanding and using texts; yet we can profit by bearing in mind some conceptions that are familiar in theoretical physics and biology (Yates & Beau grande 1987)_4 Language is certainly a "complex system", i.e., a system possessing more than one "level" distinguishable in terms of its operational functions. The major division I would postulate for complex systems is between levels where the system seeks to maintain control over itself or its "world" versus levels where it does not. " Control" can be defined here not in the sense of engineering, but as a hypothesis of limited ind�terminacy (Beaugrande 1987a). That is, to control something is to proceed as if the number of its possible states were limited to a predictable range. Since however the outcome might have occurred without any agency from the controlling system, control remains a hypothesis that can be strengthened for particular cases but not conclusively verified for all cases.

3. 2.

3. 4. In each of the various sub-systems of language- for the time being, we can retain the conventional terms "phonemic, morphemic, syntactic, and semantic" -a control level coordinates fairly determinau targets with fairly indeterminate execution. That is, a speaker or writer has a determinate version of the intended targets, e.g., sounds to be uttered, letters to be inscribed, meanings to be expressed, etc. , whereas the actual execution of the target is assigned to probabilistic but skilled action programs. The hearer or reader picks up the resulting input probabilistically and relates it to a configuration of determinate targets. This ratio allows the systems of both producer and receiver to conserve resources. Only the more determinate levels demand significant energy; the system can economize on the less determinate levels. In support of this notion, I would adduce the fact that communication is fairly immune to surface variations or disturbances- differences in voice or print, omissions, minor errors, etc.- because they are generally allowable within a probabilistic range. But for the same reason, a total elimination of such occurrences is not feasible; even a near approximation of invariant or error free performance could impose a considerable strain on the human proces-


Even extremely complex systems can operate within limiting environ ments. Text processing, for instance, must respect the constraints of perform ance, attention, and memory by allotting its resources in strategic ways (Beaugrande 1984a). An anecdotal illustration is the seemingly effortless transition people make between the text-artifact and their own understan ding of its meaning. Apparently, processing is so designed that despite the complexity of the activities involved, overload is an exceptional occurrence. Presumably, text processing has a flexible design capable of adapting to ongoing demands, so that complexity elicits integration rather than disinte gration (Beaugrande 1987a). 3.3.

103 The revision I propose diversifies each "level" of traditional linguistics 3.5. into a complex system with at least three entities, as shown in Fig. l . We have a "control level", its "dependent" or "controlled level", and a ratio that becomes more determinate at the former's end and less determinate at the latter's. The control level has "active targets"5 carrying specifications that regulate the probabilities on the dependent level.

.±

indeterminacy low CONT ROL LEVEL

high

....

+

T

�-------,

DEPENDENT LEVEL

high

L-------�

+ r + I

low

determinacy Fig. 1 A language level as a complex system

.c. at .c.

""

Q ... .. a ,.. .. ., •

Cl ""

;-

� i

,.. u • c

,_

ii �I .. �-

�

i

,..

�

�

...

� �

""

SPEAKER'S PHONEMIC SYSTEM 2.

�

�

co --

Fig.

.. n ...,

i

HEARER'S PHONEMIC SYSTEM

The phonemic level of speaker and hearer


3.6. For example, to articulate a word, the control level activates a determi nate target describing the attributes that should be respected during execu tion. A phoneme-matrix of features might be one representation for such specifications. The latter are transmitted to the dependent level as vectors governing the articulation program, but not strictly determining its execu tion. On the hearer's side, the articulated output is retransmitted from a probabilistic identification level to a determinate target on the control level, and the word is "recognized". The transaction is diagrammed in Fig. 2. I f the linkage between speaker's target and the hearer's is determinate and

1 04 causal only toward its two end points, but indeterminate and probabilistic in between, the transfer could tolerate considerable variations in tone, quality, and volume of voices, or in acoustic conditions such as bodily proximity, environmental noise, visibility of the articulatory organs, and so on. In the written modality of "graphemes", the same principle would accommodate variations in size, shape, style, and legibility of print and handwriting. A strictly determinate transfer would have to process all such variations as additive complexities, which is clearly not the case in everyday communica tion. Within this approach, "phonemics" and "graphemics" are "low-level'' systems not because their units themselves are small in size ("size" is a rather metaphorical notion apart from written transcripts), but because the com pactness and well-structuredness of the repertoire of units (sounds and letters) strongly enhances the probabilities of coordinating targets within either system as well as between the two systems of the text producer and the text recipient. Still, sounds and letters can be mispronounced or misheard, just as letters can be miswritten or misread. These errors margins, recently an object of much inquiry in linguistic and reading research,6 are by-products of the indeterminate coupling between target and execution. 3.7.

3.9. As the considerations of "automatic" versus "attentional" processing indicate, my basic model could be justified on the grounds that the system's complexity can rise without requiring major additions of resources from the outside. Such a system could be designated "self-complexing", a familiar concept in engineering. 7 The general principle should be that upward shifts of control to a higher level consume fewer resources, whereas downward shifts to a lower level consume more. The evidence for this thesis on the upward gradient is not very massive. Some experiments, such as those of Richard Shiffrin and Walter Schneider ( 1 977), have been able to track the automatiza tion of learning, but a staggering number of trials is needed even for fairly simple recognition or recall tasks with sparsely contextualized materials (such as numeric or non-meaningful visual arrays) . The automatization of language would be vastly harder to observe in controlled settings. 3.10.

However, evidence for the downward gradient is easier to obtain. A


3.8. An operation could become "automatic" by establishing a higher control level to oversee actions and by organizing a configuration of relays to a now dependent level of reliably distributed probabilities. "Attention" is therefore not needed in the sense that such an operation does not compete for higher-level resources (cf. Keele 1 973). Predictably, an automatic operation is hard to consciously monitor and change, presumably because control must be downshifted to raise determinacy, and the added drain on resources may encourage overload.

1 05 person who tries to devote detailed conscious attention to the lower-level movements of articulation or inscription would incur a disintegration of performance on other levels of speaking or writing, such as phrase formation and semantic coherence. In general, any attempt to closely observe oneself using language and to interpret every event as causally determinate would produce a severe strain. Modernist art, which tends to interfere with automa tic perception so as to encourage this multi-controlled processing, is often rejected as "too difficult" by naive audiences.

3.11.

If determinacy

is highest at the two end points of speaker's and

hearer's (or writer's and reader's) respective targets, the midway point of their actual contact should be conversely the most indeterminate. This corol

is surprising for the sounds and letters of a language, because they are the

easiest entities to objectify and thus seem to carry their own maximal determi nacy right inside them. Recordings and inscriptions can preserve the artifacts for unlimited repetitions of perception. Moreover, transfers between sound and letter are highly skilled, automatized operations literate culture takes for granted. s All these factors encourage us to imagine that the contact point is the very sound or letter transmitted between the two participants in commu nication, rather than a probabilistic configuration of signals people have to synthesize in order to recognize.

3. 12.

Yet my rise-fall indeterminacy gradient appears far more reasonable

when applied to the "level" of meaning or "semantics", as sketched in Fig.

3.

This time, the control level activates meanings as relatively determinate targets at the two ends of speaker and hearer. The indeterminacy of their

s:. ao ... s:.

,.. u .. c:

indeterminacy

::r

i

I�

� i

,_

�

.. ... .. '0

�

I

SPEAKER'S SEMANTIC SYSTEM Fig.

3.

�

i

s

...

� �

HEARER'S SEMANTIC SYSTEM

The semantic level of speaker and hearer


lary

106 point of contact now seems quite plausible, because meaning is notoriously hard to objectify or artifactualize. Even in the deceptively simple-seeming case of referring to a perceived physical object , the object itself rather than its meaning is the shared perceptual point of contact. In semantics, indetermina cy can readily be recognized to be not some accident or breakdown, but the the very "ground" against which determinacy is composed as the "figure". Again in the semantic system, the indeterminacy interval enables a tolerance of wide variations, this time the ones among different people's stores of knowledge and experience that count as "meaningful" . The reperto ry of meaning is neither compact nor tightly organized in comparison to those of sounds and letters. Hence, the rise and fall of indeterminacy presu mably marks a rather steep gradient whose end points are still less determi nate than those of, say, phonemic targets. Even so, a reasonable symmetry between the speaker's end point and the hearer's is apparently attainable in contexts where, for current purposes, the hearer "knows" what the speaker "means". Such instances entail by no means trivial processes, and semantics cannot afford to disregard the dynamics of the event. 3. 13.

Within the total scheme, the systems of the language are no longer related merely in terms of unit size, which was the standard perspective of descriptive linguistics. The systems are more vitally interrelated in terms of the control operations needed to manage and coordinate the relative com plexity, diversity, and productivity of the systemic repertoires. Meaning is on a "higher level" than the others because it manages greater complexity, supports higher ratios and steeper gradients of indeterminacy, and thereby attains higher returns and more elaborated results from the resources it

3.15.


For an overall model, language can be depicted an "intersystem" or "polycentric system" (cf. Beaugrande and Dressler 198 1 ). Still keeping the traditional terms, we would assign a system to each of the levels of phonemic (or graphemic), morphemic (word-formation), syntactic (word-sequencing), and semantic (meaning). Fig. 4 presents an overview. These systems are not mere black boxes in a fixed relay chain , but mutually interactive and consul tative control centers comparing and consolidating their results all throu ghout their operations (whence the multiple links drawn in Fig. 4). The more simultaneous, matrix-like aspect of the phonemic and graphemic systems must be coordinated with the more sequential, string-like aspect of the morphemic and syntactic systems. And both these aspects must somehow correlate with the still undecided aspect(s) of the semantic dimension; the question of whether meanings should be visualized as matrices or strings is hard to answer and may well be falsely posed in the first place. Meanings might be better compared to sub-atomic particles, whose identities and trajectories remain undecided except during interactions with each other and during observations of those interactions. 3.14.

107

> u � c

.c ca

E c.. Cl)

-

....

Cl) '0

>u � c ...

e L Ql

....

�·s

.....

.....

Ql '0

u "' 'Q) �

...,

3

::::J c. Ill

;::, Ql n ' R

�}

-

P A Q (and�introduction)

ergo: R SW reply that the conclusion R can be reached without and-introduction as follows: ( I I)

(P A Q) :::> R p ergo: Q :::> R Q ergo : R


up this elimination. Nor would one wish to exclude, e.g., the inference of 'Someone really existing escaped from prison' from 'The murderer escaped from prison'. Elimination of the murderer (existential generalization) would give 'Someone escaped from prison', but this does not suffice to get actual existence of that someone. That entailment hinges on the fact that the predicate escape is extensional with respect to its subject (every constant subject tenn of escape must refer to an actually existing entity for the proposition with escape as its highest predicate to be true). The predicate imaginary, for example, is not extensional with respect to its subject: if I say truthfully 'The murderer is imaginary', then elimination of the murderer may give 'Someone is imaginary', but actual existence for this someone is not derivable. This means that entailments of actual existence, though no doubt natural and cognitively real, cannot be secured by means of an elimination rule. Another stumbling block is the rule of and�introduction. SW maintain (pp. 98�9) that this rule plays no role in natural inferencing. This is clearly counterintuitive. Suppose, in some context, Bennie must not be left alone in the house because he will immediately start making costly calls to his girl friend in Hawaii. You are responsible for the telephone bill and I am telling you that Bennie is at home. But you think your son Jimmy is at home as well, though nobody else is. Then I let on that Jimmy has left for a good long game of snooker with the boys. Would you not now realize that Bennie is at home and nobody else is there? SW themselves consider the following argument against their position that and� introduction does not occur in cognition. They let their imaginary opponent say (p.98):(a) modus ponens is a natural form of entailment, and (b) and�introduction occurs in modus ponens arguments of the following format:

1 36


This is correct, but it must be noted that, if the deductive procedure is syntactic, the equivalence is required of(P 1\ Q) :::> R and P :::> (Q :::> R), whence modus ponens gives Q :::> R.In order to get this equivalence other than just elimination rules are required. Note that SW, apparently, do have in mind syntactic logical derivation: 'The device we envisage is an automaton with a memory and the ability to read, write and erase logical forms . . . . '(pp. 94-5). But if the derivation given in ( I I) is done semantically, it still remains true that ( I I ) is a great deal more complex than ( 1 0). This is admitted by SW (p.99), which means that in their view cognition goes for the more complex procedures, surely a most unattractive position. For similar comments see Gazdar & Good ( 1 982: 89-20). Meanwhile the question of the criteria by which natural cognitive entail ments can be singled out from the total set of logically valid entailments remains unsolved. Moreover, the logical apparatus that is postulated fails to satisfy even the mildest criteria that apply to a logical system. SW claim (p. l 03) that what they 'have done is merely place an upper bound on the set of implications that could in principle be derived from a given set of assump tions', thereby referring, apparently, to their claim that t he set of'final theses' derivable from a finite set of premises is finite. But, as we have seen, the concept of 'final thesis' is as yet without any filling in its logical entry. We are gradually getting closer, or so SW attempt to make us think, to the main target of the book, the characterization of the notion of relevance. The next station on this tortuous and longwinded path is the notion of contextual implication , introduced on p. l 07. I shall paraphrase the rather turgid text as faithfully as possible, stressing the relevant points. A set of premises PR can, for pragmatic reasons, be split up into two mutually exclusive subsets P and C. C represents 'old' knowledge, i.e. (to the extent that the text is clear on this) either encyclopaedic knowledge or knowledge already processed (p. l 07). P represents 'new' knowledge or infor mation, i.e., 'assumptions derived from perception or linguistic decoding' (ib.). When such a split is made, P is said to be contextualiud in C. A contextual implication is an entailment q from PR, such that neither P alone nor C alone entails q, and only natural cognitive deductive processes are involved. The entailment q as well as the information that was new until a moment ago have now been processed and pass into what will be C for any possible new P. The addition of q to the store of old information is called a contextual effect (pp. l 08-9). No definition is given of the notion of 'contextual effect'. The reader must make do with 'the intuitive idea' that 'to modify and improve a context is to have some effect on that context', whereby he is given to understand that the modification must be both related to old information and new (p. l 09). 'Intuitively', again, 'there should be two more types of contextual effect. On the one hand, new information may provide further evidence for, and there fore strengthen, old assumptions; or it may provide evidence against, and perhaps lead to the abandonment of, old assumptions' (ib.). The latter occurs

1 37 (p. 1 1 4) 'when there is a contradiction between new and old information'. In such cases the weaker of the two contradictory assumptions will be abando ned.2 As regards the 'strenghtening' effects, it is not made clear whether rela tedness to old information is sufficient for the title of contextual effect, or whether there must also be actual strengthening, and certainly no weakening. In the latter interpretation, a conjuction of the form : ( 1 2)

Stocks in New York are plum'meting, but oil prices are rising.

'Reltvaflce' An assumption i5 relevant in a context if and only if it has some contextual effect in that context'.

SW then take up three pages of text to explain that this is a gradable notion, since the expression 'some contextual effect' allows for degrees. Unaccounta bly, the definition is then extended with an extra condition to do with the principle of minimal effort. The result is found on p. 1 25: 'Re/tvQflce Exteflt co!lditiofl 1 :

an assumption is relevant in a context to the extent that its contextual effects in th1s context are large.

Exteflt coflditiofl 2:

an aS!Iumption i5 relevant in a context to the extent that the elTon required to process it in this context i5 small. •

Given this undeserving formal workmanship it will come as no surprise that it is easy to present examples with awkward consequences. We have already considered example ( 1 2): if strenghtening is a condition, the second conjunct


should have a second conjunct without contextual effect even though it is related to the fLrSt conj unct. The criterion of relatedness is not elaborated, so that one must rely on one's intuition in applying it. Or it should be that a new utterance is considered relevant when it, together with some body of back ground knowledge, gives rise to a contextual implication. But even this charitable interpretation does not help much. For there will be few utterances that fail to give rise to some inference together with some background knowledge. But most of these SW will want to rule out as irrelevant, as indeed we would. So what is needed is some further criterion of relevant background know ledge. The same applies to cases of contradiction: often enough a new utterance will be logically incompatible with some bit of existing knowledge, but, of course, that will not suffice to make it relevant in any desired sense of the term. SW discuss the question of 'context location' at length (pp. 1 3 2142), but, as we shall see, in vain. The concept of relevance is then, finally, defined in chapter 3. The defini tion proceeds in stages. On p. l 22 we read:

1 38 is not relevant, and if mere relatedness is a condition then we are left without any definition. Or take the well-known joke of the man eating in a restaurant and exclaiming: ( 1 3)

Waiter, there is a dead fly in my soup.

Wh�rf>•_•l'on the waiter quips: ( 14)

I know, sir, it's the heat that kills them .

It should be noted that SW insist (p. l l 9) that they 'are not trying to define the ordinary English word 'relevance'. 'Relevance' is a fuzzy term, used differently by different people. or by the same people at different times. It does not have a translation in every human language. There is no reason to thin I. that a proper semantic analysis of the English word 'relevance' would also characterize a concept of scientific psychology. We do believe, though, that .... there is an important psychological propeny - a propeny of mental processes - which the ordinary notion of relevance roughly approximates, and which it is therefore appropriate to call relevance too, using the term now in a technical 5CI\5C. What we are trying to do is . . . to define relevance as a useful theoretical concept.'

Useful theoretical concepts, however, are useful, among other reasons, be cause they fit into a deductively ordered set of theoretical statements or hypotheses yielding 'theorems' which, upon interpretation, reflect possible facts. If intuitions of relevance are not the factual basis required, what are? Normally speaking, one would expect that the theoretical analysis of the concept of relevance is meant as an explicitation of the intuitive notion. But if the intuitive notion is said not to be the object of investigation (though


Clearly, the waiter's reply is irrelevant here, but it is highly relevant in a very different situation where the 'concern' is about what killed the fly. SW's analysis provides no way of making this distinction, since the union of whatever concern parameters make ( 1 4) relevant and whatever such parame ters would make a really relevant reply relevant makes them both relevant. Clark ( 1 982: 1 27), in his critique of Sperber & Wilson ( 1 982), makes the same point: 'Surprisingly, the principle of relevance is mute on a crucial point about relevance itself: relevance with respect to what?'. As regards the 'extent condition 2', which claims that smaller processing effort corresponds with greater relevance, the reader is referred to the tho rough discussion of this point by Gazdar & Good ( 1 982: 92-98), who show that (a) so little is known about cognitive 'processing cost' t hat any claim in this direction is, in fact, empirically vacuous, and (b) any such relation of inverse proportionality, no matter the empirical content of the notion of 'processing cost', will clash with natural intuitions and thus become implausi ble.

1 39

'In verbal commumcation, the hearer ts generally led to accept an ass u mption as true or probably true on the basis of a guarantee given by the speaker. Part of the hearer's task is to find out which assumptions the speaker 15 guaranteemg as true. Our hypothesis IS that the hearer is guided by the principle of relevance in carrying out thts task. He expecu the information the speaker intended to convey, when processed in the context the speaker expected it to be contextualized in, to be relevant: that is, to have substantial contextual effect, at a low process ing cost. Thus, if the hearer assumes (91 ), (9 1 ) The speaker intends to assert P and P turns out to be relevant in the expected way, assumption (9 1 ) is strengthened; moreover, if the hearer trusts the speaker to be truthful, assumption P is strengthened too. If P turns (out] to be relevant in the expected way only when assumption Q is added to the context, then assumption (92) is strengthened: (92) The speaker intends the hearer to assume Q and again, if the hearer trusts the speaker, then assumption Q is strengthened'.

The first thing that strikes the reader is, of course, the disarmingly candid statement that strengthening is a result of relevance, and not the other way around, as the definitions want it to be. But the point here is that this quotation is confused in that it makes the degree to which the hearer trusts the speaker play a systematic role. It is trivially true that a speaker's convincing power covaries with his authority and with the hearer's pre-existing know ledge. It is true that relevance guides interpretation. It is likewise true that the reconstruction of a relevant context or inference itself helps determining whether an interpretation is correct: a fitting hypothesis may be circular, it confirms itself by its fit. But the trustworthiness of the speaker does not, in general, play a role in spotting the correct interpretation of an utterance . The truth or falsity of P in SW's (9 1 ) is generally irrelevant. and it matters not a


perhaps vaguely related}, then one is entitled to know what is the intended empirical basis for the intended 'theoretical concept' of relevance. When discussing SW's notions of 'ostensive-inferential communication' and of 'assumption', we concluded that the most probable interpretation of the hopelessly confused text was to attribute to SW the view that in communi cation a speaker tries to convince his hearer of the truth of what he says. More specifically, communication is successful to the extent that the speaker transfers his own degree of certainty to his audience. We have criticised this 'rhetorical' view of communication on the grounds that what counts in successful communication is the recognition on the part of the hearer of the kind and degree of commitment the speaker intends to take on, and not at all whether, in the case of assertions, the hearer allows himself to be actually convinced by the speaker. Such a view centers too much on assertions, and it is empirically inadequate. On p. l 03 SW go even further. There they claim that communication is helped, not just by a transfer of degree of conviction, but also by the actual truth of what is asserted : 'the relevance of new information to an individual is to be assessed in terms of the improvements it brings to his representation of the world' . This claim is repeated on p. l 08 and on p. l l 4. On p . l l 6 SW mix the rhetorical with the commitment view of communication:

140 whit whether P is strengthened or not. What may matter, for a correct interpretation, is whether SW's (92) is true, but then, whether or not Q is 'strenghtened' is nothing to do with the process of comprehen�ion. Further more, (92) fails as a general condition. The general condition is more like 'the truth of (9 1) depends on the assumption that the speaker assumes Q'. But SW throw everything into one bag.


provide. Let us look again at the little episode of the flag-seller and the passer-by, described in (6) and (7) above (pp . 1 2 1-2 of the book). A crucial premise for the conclusion is (7e ): 'someone who has no need of the services of a charity cannot be expected to subscribe to t hat charity'. Hopefully, this is not part of the background knowledge of anyone interpreting the passer-by's refusal to contribute. As SW themselves say (p . l 2 l ) , 'the hearer must be able to supply' the premises, and in this case this means that he must hypothesize, or if you like 'assu me' , that this is a premise in the passer-by's m ind. On this hypothesis the passer-by's utterance makes some sense. But now one wonders if (7e) should be considered 'old' or 'new' information. It can hardly be old, for the hearer that is, since the flag-selling hearer is not likely to sell flags if he thinks (7e) is true. If it is new, one wonders how it can be regarded as an 'improvement . . . to his representations of the world' (p. l 03). Clearly, (7e) itself plays no part in the hearer's world representation, but the fact that the speaker apparently assumes (7e) is i mportant for a correct interpretation of the speaker's refusal to contribute. S W fail to make this distinction and thus get entangled in absurdities. For example, if (7e) is relevant in SW's sense (or at least in terms of their definitions to the extent that these are interpretable), it must not be 'elimina ted' in favour of the hearer's own conviction to the contrary. Yet, his well-established conviction that charities are worth contributing to for other than purely egotistical reasons will no doubt prevail. Despite this, he inter prets the passer-by's reply correctly. In fact, his correct interpretation (always assum ing that the interpretation as sketched by SW is the one that took place) depends on the 'assumption' (7e) in so far as he- must 'assume' that the speaker 'assumes' (7e). SW cross metaborders unawares: they fail to distin guish, for a proposition p, between takingp as true, and taking p as believed to be true by an interlocutor. Yet p has a very different status in one case and in the other. A problem that has been touched upon but not dealt with is that of the selection of the proper context or set of background assumptions. Wilks ( 1986) points out that the answer provided by SW in their 1982 paper makes the definition of relevance circular. In 1982 SW stated that 'the search for the interpretation on which an utterance will be most relevant involves a search for the context which will make this interpretation possible. In other words, determination of the context is not a prerequisite to the comprehension process, but a part of it' (p. 76). Wilks replies ( 1 986:273) that this makes nonsense of the condition, quoted above ('extent condition 2'): 'an assump-

141

'We assume that a crucial step

m

the process ing of new information, and in particular of

verbally communtcated information, 15 to combine it with an adequately selected set of background assu mptions - which then constitutes the context - m the memory of the deductive device'.

One would now, naturally, wish to be instructed as to the meaning of the expression 'adequately selected'. And, after a great deal of dodging, SW finally inform us (p. l 4 l ): 'Our answer is that the selection of a particular context is determined by the search for relevance'. So nothing has changed. The problem has only been tucked away in a larger amount of text. The problem is, as it was before, mainly located in the search for a suitable context, or, as we called it at the outset, 'concern'. And there is nothing in the book under review that has brought this problem nearer to a solution. There is thus hardly any justification for the regular use of the term 'relevance theory', from p . l 30 onwards. What S W have offered comes no where near the status of a theory. The use of this term is an improper appropriation of prestige. The last chapter (Aspects of verbal communication) is mainly a loose collection of thoughts on the nature of communication and its relation with language, on implicatures, and on literary and stylistic aspects of communi cation. (Presuppositions, as we said earlier, are treated under 'style'). There is little of interest to be found in these pages, and I shall refrain from comment on all issues but one: at the beginning of the chapter S W present some, largely philosophical, ideas about the relation between language and communica tion. On p. l 72 we read: 'Languages are indispensable not for communica tion, but for information processing; this is their essential function'. One would have thought that information processing is anyhow i ndispensable for any kind of communication, so that, if language is necessary for information processing, it is likewise for communication, necessity being a transitive


tion is relevant in a context to the extent that the effort required t o process it in this context is small'. For if the determination of the context C for an assumption A is part of the processing of A, then the relevance of A may involve a search for a suitable C, so that A is relevant, but it makes no sense to ' speak of the relevance of A in C. The objection seems valid, and one would have expected SW to do something about this point (e.g. by dropping 'extent condition 2', which would have been a wise move anyway in view of Gazdar & Good's critique mentioned above): Wilks' point, though published as late as 1986, has been known to the authors since 1 982. Yet, as we have seen, the offending condition has not been removed or remedied in the 1 986 version under review, and, after some searching in the verbose text, one discovers that the circularity has not been removed either. The selection of the proper context, in which some 'assumption' to be processed is relevant, is still part of the processing of that assumption, as it was in 1982. On pp. l 37-8 we read:

1 42 relation. On p . l 74

SW make their meaning a little clearer: 'Language is not a

medium

for communication : non-coded communication exists'

necessary

{italics mine). therefore,

A communicating device must have 'an internal language', and

SW conclude, the essential function of language is not its use in

verbal communication but to make information p rocessing possible. Thus, on p. l 73 : 'Our po i n r

is precisely that the propcny of be i n g a grammar-governed representational

system and the propeny of being used for communication are not systematically linked. They are found together in the odd case of human natural languages, just as the propeny of bemg an olfactory organ and the propcny of being a prehensile organ, though not systematically linked in nature, happen to be found together in the odd case of the

One hears echoes here of Chomsky's ramblings, in the late

'60s and the early

'70s, on the same subject. There is a simple confusion, here, between overt languages, with their gram mars and their lexicons, as they exist in the world, on the one hand, and internal computional and represen tational systems on the other.

SW prefer to call such internal computational systems likewise

'languages', like computer 'languages ' . But nothing shows that this is not a metaphor. In any case, it is reasonable to surmise that actual, overt languages are highly functional for human communication: in what way other than by the use of grammar-governed and lexically elaborated representational sys tems would communication

be possible about the extraordinarily wide range

of topics that humans can and do communicate about? In the internal systems we find a coupling of representational and computational fu nctions. In the actual languages of the world we find a coupling of representational and communicative functions. In neither case is the coupling 'odd'. On the contrary, i t is not difficult to argue that in both cases the couplings are highly functional. I t would be a pity not to mention one last gem, found on pp. 205-6. Here SW discuss the sentence 'John invited Lucy', and assign it the 'logical form' :

something is the case

( 1 5)

/

"-.,_

someone

I John

did something

/

invited

""'

someone

I

Lucy Not surprisingly,

SW fail to enlighten their readers on the logic in terms of

which ( 1 5) can be used for deductions.


elephant's trunk'.

143 NOTES

I.

It is tempting to think that this is an Implicit admiSSion of the failure of Wilson's attempt

( 1 975) at givmg a pragmatic account of presupposition. 2 This is different from all or most exisung theories of acceptable discourse construction, where contradiction, if spotted, leads to unacceptability of the discourse. (See, e.g., Van der Sandt 1 982; in press). Ni;m�g�n UntvUJIIY Philo1ophy !nJtituu P.O. Box 9108 6500 HK Nijm�g�n - Holland

Brown, G., and G. Yul� 1983: Di1couru A. nalym. Cambridge University Press, Cambridge.

Clark, H.H. 1 982: The relevance of common ground: comments on Sperber and Wilson's paper. In: Smith (ed.), 1 24-- 1 27. Clark, H. H. and S. Haviland 1977: Comprehension and the given-new contract. In: R. Freedle (ed.), Ducour1� Producnon and Compr�h�nJion. Ablex, Norwood, N.J.: 1-40. Fauconnier, G. 1985: M�ntal Spac�J. A1�cu ofM�aning ConJtruction in Natural Lallguag�. MIT Press, Cambridge, Mass. Fodor, J.A. 1983: Th� Modu/anty of Mind. MIT Press , Cambndge, Mass. Gazdar, G. and D. Good 1 982: On a notion of relevance. Comments on Sperber and Wilson's paper. In: Smith (ed.): 88- 1 00. Haviland, S. and H.H. Clark 1 974: What's new? Acquiring new information

as

a process in

comprehension. Joldllal of Vtrbal Learnmg and V�rbal B�havior 1 3: 5 1 2-52 1 . bard, St. 1975: Changing the context. I n : E. Keenan (ed.), Formal s�mantics of Natural Lallguag�. Cambndge University Press, Cambridge: 287-296.

Kantunen, L. 1 974: Presupposition and linguistic context. Th�or�tical Linguutics 1: 1 8 1- 194. McCawley, J . D. 1979: Presupposition and discourse structure. In: Ch.-K. Oh &: D.A. Dinneen

(eds.), Pruuppo1ition. (= Syntax and Semantics Vol. I I ) AcademiC Press , New York-San Francisco-London: 37 1-388. Moore, T. 1982: Comments on Sperber and Wilson's paper. In: Smith (ed.), 1 1 1- 1 1 2 Reichgelt, H. 1 982: Mental models and discourse. Journal of &mantles 1: 371-386. Sanford, A.J. and S.C. Garrod 1 98 1 : Understanding Wrltt�n Lallguag�. ExplorationJ in Compr�h�ff.Jion beyond the S�nunc�. Wiley &: Sons, Chichester. Seuren, P.A.M. 1985: Discours� s�mantics. Blackwell, Oxford. Shadbolt, N. 1 983: Proccssmg reference. Journal of S�mantics 2: 63-98. Smith, N. V. (ed.) 1982: Mutual Knowl�dg�. Academic Press, London. Sperber, D. and D. Wilson, 1 982: Mutual knowledge and relevance in theories of comprehension. In: Smith (ed.), 6 1 -85. Stalnaker, R. 1 978: Assertion. In: P. Cole (ed.), Pragmatics. (=SyntaX and Semantics Vol.9) Academic Press , New York-San Francisco-London: 3 1 5-332. Van der Sandt, R.A. 1 982: Kont�kst �n Prt:rupposm�. E�n studi� van h�t proj�kll�problum �n de pre:rupposition�l� ��g�chappen van de logisch� koM�kti��n. Ph.D. thesis, N ijmegen University. Van der Sandt, R.A. in press : Conuxt and Pr�supposition. Croom Helm, London. Wilks, Y. 1982: Comments on Sperber and Wilson's paper. In: S mith (ed.), 1 1 3- 1 1 7. Wilks, Y. 1 986: Relevance and beliefs. In: T. Myers, K. Brown & B. McGonigle (eds.), R�aJoning and Discouru Proasus. Academic Press , London: 265-289. Wilson, D. 1975: Pruuppo111ionJ and Non-Truth-Conditional S�mantic1. Academic Press , London.


REFERENCES


Journal of s�mantic:r 5: 145- 1 62

D I S C U S S I O N THE SELF-APPOINTMENT OF SEUREN AS CENSOR A Reply to Pieter Seuren•

D E I RD R E WILSON and DAN SPERBER

"Pieter A.M. Sc:uren: ''The Sc:lf-Styling of Relevance:". thts issue.


Watching Seuren ridicule ideas that no sane person would hold is only mildly amusing; having to discuss this exercise in shadow boxing is no fun at all. Yet we must discuss it, since some readers may feel that they have learnt, from Seuren's purported review of our book Relevance, all they need to know about its contents. Seuren demonstrates a striking inability to grasp the structure of our book, its central arguments and themes. His 'summaries' are wild distortions. His textual analyses are laughable. His 'objections' are directed at non-existent targets. A student of the irrational would be intrigued by the motivations behind this vast expense of wasted effort. Here, we will refrain from diag nosis, and simply try to put right the most serious of Seuren's mistakes. In Relevance, we distinguish three levels of analysis, three domains in which significant generalisations can be sought: the level of cognition, the level of communication , and the level of verbal communication. The first chapter is about communication; the second and third chapters are about cognition; the fourth chapter is about verbal communication, and in particu lar about utterance interpretation, or 'pragmatics'. Seuren takes the central theme of our book to be 'the gap that exists between linguistically provided information and ful ly integrated comprehen sion'. That is, he has chosen to ignore our careful distinction among levels of analysis, and to treat Relevance as a book about pragmatics. More precisely, he has chosen to treat the first three chapters (which are not about pragma tics) as about pragmatics, and entirely ignore the last chapter, which is about pragmatics. Thus, the fundamental notions of our pragmatic theory are not even mentioned, and chapter 4 is dismissed with the comment 'There is little of interest to be found in these pages'. Well, not unless you want to know our views on pragmatics. Before turning to Seuren's detailed comments, we will try to do what he notably fails to do: give some idea of the overall structure of the book. In chapter 1 , we distinguish two models of communication, a code model and an inferential model; we argue that neither is reducible to the other, and that either can occur independently of the other, though human verbal communi cation liwolves them both. Inferential communication, we argue, exploits two fundamental facts about human cognition: that it involves the ability to perform inferences, and that it is geared to the search for relevance.

1 46 In chapter 2 , we describe the inferential abilities, both deductive and non-demonstrative, which lie at the heart of human central thought pro cesses: we show how these abilities, when applied to newly presented infor mation. may modify the individual's existing representation of the world by giving rise to contextual effects. In chapter 3, still at the level of cognition, we characterise various notions of relevance - relevance in a context, relevance to an individual, relevance of a

stimulus - in terms of the notions of contextual effect and processing effort. Up to this point, we have said nothing about the specific processes by which verbal and non-verbal communication are understood. We go on to characterise a notion of optimal relevance and a principle of relevance which, we argue, do yield a strict and rigorous criterion for the interpretation of all tested and found consistent with the principle of relevance is the only inter pretation consistent with the principle of relevance. We go on, in the fourth and final chapter, to show how this criterion applies to various aspects of utterance interpretation - disambiguation, reference assignment, enrichment, the recovery of implicatures, presupposi tional and stylistic effects, metaphor, irony, speech acts and non-declarative sentences. This 80-page chapter, as stated in the introduction to our book, is a condensed version of a much larger work on pragmatics and rhetoric which has been in circulation for some years, extracts from which have already been published; 1 and on which we are still working. Seuren, who takes the book as a whole to be about pragmatics, does not even mention the notion of optimal relevance, the principle of relevance, or the criterion of consistency with the principle of relevance, around which our pragmatic theory is built, and finds 'little of interest' in our sketch of a pragmatic theory in chapter 4. Many of his objections, then, are simply beside the point. For example, he finds our neglect of presuppositions particularly deplorable, yet he defines presuppositions as 'systematic properties of

sen

tences' (our italics) - in which case there is little point in expecting them to be treated in chapters dealing with fundamental properties of communication (both verbal and non-verbal) and cognition. 2 Such mistaken expectations pervade his review. On the other hand, we have searched the review in vain for valid - indeed, for remotely plausible - criticisms. We will now consider Seuren's 'objections', chapter by chapter, trying, at the same time, to give a fuller idea of what our book is about. In chapter l, we distinguish coded communication from inferential communication. While coded communication is quite easy to describe, inferential communication, which involves the use of central thought processes and the recognition of a communicator's intention, is notoriously difficult, and our aim is to sketch an explanatory account of how such communication takes place. Inferential communication, we argue, involves two types of intention: the

informativ� intenrion - to make a certain set of assumptions manifest, or more manifest, to an audience; and the communicativ� intention to make the informative -

intention mutually manifest.


inferential communication, verbal and non-verbal: the first interpretation

147 MANI FEST FACTS AND ASSUMPTIONS

In

Relevance, fact, assumption

and

manifest

are technical terms. A fact is a

true proposition. An assu mption is a thought which is taken to represent a fact . A fact is manifest to an individual on two conditions: fi rst, he must be capable of representing it; and second , he must have enough evidence to accept this representation as true or possibly true, i.e. as an assu mption. By a quite standard process of extension (using predicates which take propositions as arguments to apply, as well, to utterances or thoughts which represent those propositions), we use 'manifest' (and other predicates such as 'infer able') to apply not only to facts but to assumptions. Our reason for doing so (which is again standard) is that, fro m a psychological point of view, false tions play the same psychological role as true perceptions), and we want the refore to

be able talk in the same terms of all assumptions, whether they

actually represent fact s or not. Seuren strenuously objects to our using 'manifest' for both facts and assumptions. Yet he himself talks of 'true facts' and of 'true thoughts'; how can this standard kind of extension

be

all right for him and not for us? Why

does he have trouble in grasping our technical use of 'fact ' and 'assumption' (which is not particularly original)? His 'powers of comprehension are stretched even more', he ironizes, by the fact that we define 'manifest' in such a way that facts and assumptions can be manifest (i.e. evidenced and accessi ble via perception or inference) without actually being entertained. Yet the notion of a possible thought should place no great mental strain on someone familiar with the notion of a possible world. Indeed Seuren, stretching not just his powers of comprehension but also his altruism, ends up saying: 'my most charitable interpretation is that what SW wish to regard as manifest assumptions are possible justified assumptions as well as just ified assu mptions

actually made'.

What charity? What powers

of comprehension? Seuren is merely paraph rasing a point we repeatedly make (pp. 39-4 1 ). At his 'most charitable', he consents, after much sneering, to understand what we are saying more or less as we are saying it. Most of the time, feeling less generous, he stops short at the sneers.

COMMUNICATING AND CONVINCING

The definition of inferential communication is theoretically comple x , but cognitively simple: for successful communication to take place , all the addressee has to do is recognise the commu nicator's informative intention: he does not have to believe what he is being told, and he does not have to recognise the communicator's commu nicative intention at all. Seuren's failure to grasp the notion of manifestness leads him to misun derstand the informative intention too. I t fo llows from the informative


assumptions play essentially the same role as true ones (just as mispercep

1 48

COGNITIVE ISSUES

The question raised by this account of communication is: how are the communicator's informative intentions recognised? An informal answer is easy enough: addressees assume that communicators aim at certain stan dards, and use t his assumption in evaluating hypotheses about the communi cator's informative intentions. But what are the standards? Where do they come from? How are alternative hypotheses arrived at and evaluated? In an attempt to answer these questions, we move to t he level of cognition in chapters 2 and 3. In chapter 2, we develop an accoun t of t he central inferential abilities that underlie spontaneous human t hought. Deductive i nferences, we claim, are performed by a deductive device which takes as input a set of assumptions, and derives as output the full set of its non-trivial implications (a technical term discussed below). Non-demonstrative 'inferences' are a heterogeneous set of processes by which assumptions are assigned an initial degree of strength depending on their source (e.g. in perception. communication or


intention, he decides, that the speaker 'must try to convince the hearer, (or reinforce the hearer's conviction) of the truth of what he intends to say' - a conclusion which he rejects on the familiar ground that communication can be successful when the speaker is not trying to convince the hearer of the truth of what is said. There are three mistakes here. First, since an assumption can be made manifest without actually being entertained, the informative intention is not an intention to convince. Second. as we state explicitly (pp. 29-30), communi cation is successful when the informative intention is recognised, not when it is fulfilled: in other words, when the addressee knows what set of assumptions the communicator wants to make manifest, whether or not he is actually led to believe them. And third, the set of assumptions which the speaker wants to make manifest or more manifest may or may not include the proposition explicitly expressed by the utterance - in Gricean terms, what was said. This is true, rather obviously, in the case of metaphor and irony, but also in the less obvious case where, for example, Mary knows that Peter will not believe what she says, but wants to communicate that she believes it nonetheless. Seuren believes we have missed cases of this kind, and tries to use them against us. If he had found the time to read our fourth chapter, he might have discovered that we analyse such a case ourselves (p. l 8 l ), not as a threatening counter example, but, on the contrary, as a possibility following quite naturally from our theory. This is because the widespread idea that an utterance can be relevant only through the proposition explicitly expressed has no place in our framework. An utterance, like any other phenomenon , makes manifest a variety of assumptions and can achieve relevance through any combination of these.

1 49 deductive inference), and this initial degree of strength is modified by their su bsequent p rocessing history. When a newly presented assumption is added to a set of existing assu m p tions - a context - in the memory of the deductive device, it may significantly modify the context on one of three ways. It may strengthen an existing assumption; it may contradict and eliminate an existing assumption; or it may combine with existing assumptions to yield a contextual implication that is, a non-trivial implication based on at least one premise drawn from the context and at least one premise drawn from the newly presented informa tion. These three types of contextual modification we call contextual effects; they form the basis, in chapter 3, for our characterisation of relevance. Seuren takes chapter 2 to be 'devoted to the inferential element in compre hension: the fact that tacit premises often have to be invoked in order to show the relevance of an utterance (often a reply)'. He comments: 'S W hold that such inferential chains are formal deductive procedures, and not some "loose form of inferencing" . . . . This is perhaps so, though plausibility-based a nd default procedures cannot be ruled out.' In fact . the chapter is about neither utterances nor the processes by which 'tacit premises' (i.e. contexts) are chosen: those topics are left to chapters 3 and 4. The chapter is about the inferential p rocesses which apply to a given set of premises when some newly presented information is added to them. The idea that contexts can be deduced from newly presented information is absurd. A fundamental assumption of our book, as of much recent work in cognitive science, is that humans have a conceptual representation system, or 'language of thought', which differs from human natural language in its capacity for unambiguous, referentially unambivalent expression. Some conceptual representations have a unique sense and reference: in our terms, they a re fully propositional, and capable as they stand of being true or false. We suggest, however, that not all conceptual representations have this p ro perty: humans are capable of entertaining and processing incomplete assump tion schemas, which require the addition of further conceptual material to become fully p ropositional. Assumption schemas, we suggest, are an impor tant source of hypotheses in human thought. Seuren has difficulty with the distinction between propositional represen tations and assumption schemas. We claim, for example, that since the sense of a (natural language) referential expression does not uniquely determine its referent, the sense of a sentence containing referential expressions is an assumption schema rather than a fully p ropositional representation. We use a sentence containing pronouns to illustrate. Seuren seems in doubt about whether our claim extends to definite NPs. That the answer is 'yes' should go without saying: any sentence whose sense does not uniquely determine its reference corresponds to an assumption schema. which requires the addition of further conceptual material to become fully propositionaJ . 3 This decision i s n o t arbitrary. It follows from some basic assu mptions that we share with many others working in the field of cognitive science: that it is -


1 50 primarily thoughts, not sentences, that have semantic interpretations; that humans have no access to the semantic interpretations of their own thoughts; and hence that distinct thoughts, representing distinct states of affairs, must be fo rmally distinct. It follows trivially that utterances which express distinct thoughts, representing different states of affairs, must be assigned distinct conceptual representations. The fact that these assumptions are incompatible with much current practice in formal semantics merely suggests that much current practice in formal semantics will have to be modified if it is to interact fruitfully with cognitive science.

Fundamental to our notion of contextual mod ification, and in particular of contextual implication, is our characterisation of the deductive inference rules available to the deductive device. We claim that the only such rules are

elimination rules. M o re generally, we claim that the deductive device com putes only non-trivial implications, w here a set of assumptions {P} non-tri vially implies an assumption Q iff, when {P} is the set of initial theses in a derivation involving only elimination rules, Q belongs to the set of final theses. Seuren has d ifficulty understanding what we mean by 'final theses ' . We mean, as is made clear in our description of the deductive device the set of theses contained in the memory of the deductive device at the end of a deductive process: that is. the set of initial theses, plus the set of derived theses obtained by exhaustive application of the available elimination rules. Seuren pretends to think that we mean neither the set of initial theses, nor the set of derived theses, but the subset of derived theses to which no further elimi nation rules can be applied: since he does not know which these are, he dismisses the notion of non-trivial implication - which is fu ndamental to our framework - as vacuous. How can Seuren reconcile this interpretation , which has no basis whatsoever in our text, with the fact that we point out (p. l 04) that every assumption analytically, and therefore, by our definitions, non-tri vially, implies itself? The answer seems to be that Seuren is so sure that our ideas are inconsistent that any inconsistency in his i nterpretation of our ideas seems to him a confirmation of his view.

ELI MINATION RULES S euren also o bjects to o u r empirical claim that spontaneous human thought has access only to elimination rules. I n the book, we consider an objection once made to this proposal by Gazdar and Good ( 1 982), who claimed that the rules of or-introduction and and-introduction were needed to cope with certain intuitively valid derivations - for example,

( 1 ) and (2):


FIN A L THESES

151

{ I)

(2)

a. b.

I f (P o r Q ) , then R

c.

R

a. b. c.

p Q

d.

R

p

If (P and Q), then R

What we actually say (p. 99) is that derivations involving introduction rules are 'the simplest derivations in most standard logics using primitive rules alone ' [ou r italics]. We point out, however, that any standard logic would permit the use of conj unctive and disjunctive modus ponens as derived rules, and go on to make the empirical claim t hat these rules are directly accessible to the deductive device. As noted above, the resulting derivations - which we work through in detail (pp. 98- 1 00) are no more complex - and much more psychologically plausible (see pp. 99- 1 03) than the standard derivations using introduction rules. 4 -


In our original reply to Gazdar and Good (S perber and Wilson 1982), we showed how ( l ) and (2) could be dealt with in a framework without introduc tion rules. It is quite simple to conceive of an elimination rule (we call it disj u nctive modus ponens) taking ( I a) ( ! b) as premises, and yielding ( ! c) directly as conclusion; it is quite simple to conceive of an elimination rule (we call it conj unctive modus ponens) taking (2a) and (2b) as premises, and yielding 'If Q t hen R' as conclusion - from which (2d) would follow by a simple step of modus ponens based on (2c). We suggested that these deriva tions for ( I ) and (2) were psychologically more plausible than the 'standard' derivations using introduction rules. S ubsequently, we found independent empirical suppo rt for our proposed derivation of ( I ) in the work of the psychologist Lance Rips ( 1 983); this was cited and discussed in Relevance. Though Rips did not consider derivations like (2), we offered a number of reasons for thinking that ( I ) and (2) would be handled along similar lines, using elimination rules of the type we proposed. Seuren ignores our discussion of ( I ) entirely, and wildly misinterprets our discussion of (2). We claim that the human mind has direct access to a nile of conjunctive modus ponens, which takes as input premises of the form ' If P and Q, then R' and 'P', and yields as output conclusions of the form ' l f Q then R '. The resulting derivation of (2) is no longer than the standard derivation using and-introduction - and a derivation of ( I ) using disjunctive modus ponens is actually shorter than t he standard derivation. Seuren claims, however, that t he derivations we propose are more complex than the stan dard derivations. He takes us to have admitted that this is so, and concludes that in our framework 'cognition goes for the more complex procedures, surely a most unattractive position'.

1 52 CONTEXTUAL EFFECTS Seuren turns next to contextual effects. 'No definition is given', he claims, 'of the notion of contextual effect'. In fact, the notion is informally defined on page 1 1 7 and formally defined in footnote 26, p. 260. Moreover, we describe at length the three significant ways in which newly presented information may have a contextual effect: (a} by combining with the context to yield a contextual implication; (b) by strengthening an existing contextual assump tion; and (c) by contradicting and eliminating an existing contextual assump tion. Seuren feels that a criterion of 'relatedness' between newly presented information and existing assumptions must somehow be involved. He adds that since this criterion is not elaborated, 'one must rely on one's intuitions in

contrary: the characterisation of contextual effects is meant to explain pre theoretical intuitions of relatedness via its role in the characterisation of relevance, to which we turn in chapter 3.

EFFECT AND EFFORT A theory of human cognition must describe not only the inferential processes by which assumptions about the world are manipulated, but also the goals of cognition itself. Why do humans attend to one phenomenon rather than another, assign this phenomenon one conceptual representation rather than another, and process this representation in one context rather than another? We argue that human cognition is relevance-oriented: humans tend to pay attention to the most relevant phenomena, assign them the most relevant possible conceptual representations, and process them in contexts that maxi mise their relevance. Relevance is characterised in terms of contextual effects and processing effort: the greater the contextual effects, the greater the relevance, and the smaller the processing effort required to obtain these effects - that is, to assign the phenomenon a particular conceptual represen tation, to access a particular context, and to derive the contextual effects of the chosen representation in the chosen context - the greater the relevance.

lf communication takes place against this cognitive background, it follows that a communicator, by demanding an addressee's attention, suggests that the information she wants to communicate, when processed in a context she believes her addressee to have accessible, will be relevant enough to be worth his attention. This general principle, that inferential communication creates a presumption of relevance, we call the principle of relevance. At the end of chapter 3, we argue that it yields a strict and rigorous criterion for the interpretation of utterances: the first interpretation tested and found consist ent with the principle of relevance is the only interpretation consistent with the principle of relevance, and is the one the hearer should choose. The


applying it'. Seuren is free to do what he likes. The fact is, though, that a 'criterion of relatedness' plays nor role whatsoever in our theory. On the

153

(3) (4)

I have no siblings. I have no brothers or sisters.

It is clear that by any criteria, (3) is shorter than (4), which should thus, in a framework with a maxim of brevity, be predicted as less stylistically appro priate than (3). Yet most English informants, on most occasions, would judge (4), not (3), to be the more stylistically appropriate. A consideration of processing effort suggests an explanation. The English word 'sibling',though short, is also very rare, and may thus be quite costly to process - more costly, it seems, than the longer expression 'brothers or sisters'. In other words, the intuitions underlying Grice's maxim of brevity might be more accurately described as intuitions - rather stable and subtle intuitions, as these examples show - about processing effort. If we are right, then intuitions about stylistic appropriateness may eventually shed new light on the factors affecting processing effort itself.


principle o f relevance, and the criterion o f consistency with the principle of relevance, lie at the heart of our theory of inferential communication. Seuren does not even mention them. What he does is make some transparently silly remarks about our characte risation of relevance. For example, we say (p. 1 22) that 'it might be tempting' to define relevance in terms of contextual effects alone. We go on to explain (pp. 1 23-5) why this temptation should be resisted: why both contextual effects and processing effort need to be taken into account. Seuren gives the impression that we endorse the earlier definition, and then accuses us of 'unaccountably' replacing it with the second. It is his own behaviour which seems unaccountable to us. Seuren goes on to claim that appeals to processing effort are illegitimate in any framework. He echoes with approval a remark attributed to Gazdar and Good ( 1 982), that 'so little is known about "processing cost" that any claim in this direction is, in fact, empirically vacuous'. Many psycholinguists and cognitive psychologists would be surprised to hear this. It is a commonplace of the cognitive literature that, for example, frequency and recurrency of use - of words, constructions, processing strategies, concepts and assumptions affect processing effort. Processing time is an important indicator of proces sing effort, which is measured and manipulated with some sophistication in experimental work on language perception and comprehension. Any human knows that some phenomena - linguistic structures, logical structures, objects, events - are easier to understand than others ; that some contextual assumptions are easier to access in some circumstances than in others. Considerations of effort play a major role in all human activity, both physical and mental, and cannot be ignored.5 In the case of verbal communication at least, it is relatively easy to show that intuitions about processing effort must be taken into account. Here is an example. Grice has a maxim of brevity which, however brevity is measured, runs into difficulty with utterances like the following:

1 54 ORCULARITY Seuren turns next to our remarks about context selection, echoing with approval a claim made by Wilks ( 1 986), that the method of context selection we

envisage makes our definition of relevance circular. He says: 'For if the

determ ination of the context

C for an assumption A is part of the processing

of A, then the relevance of A may involve a search for a suitable C, sot that A is relevant, but it makes no sense to speak of the relevance of A in

C. The

objection seems valid . . .' It may seem valid to Seuren: to us it merely seems hard to parse. To see that the argument underlying the objection is invalid consider the formally identical argument where 'relevance' in replaced by

in a liquid' is given and used to

describe the behaviour of a chemist who is trying to dissolve various sub stances and, for each, looks for the most effective liquid; then, Wilks and

if the L for a substance S is part of the processing of S, then the solubility of S may involve a search for a suitable L, so that S is soluble, but it makes no sense to speak of the solubility of S i n L.' This is silly, Seuren should argue, the defi nition of solubility becomes circular 'for

determination of the liquid

because of the very structure of the argument, and so is Wilks' and Seuren's argument , whose structure we have faithfully reproduced. What Willes probably had in mind, and should have said, is that a defini tion of relevance in a context is insufficient if one claims, as we do, that the relevance-governed processin g of information does not take place in a flXed context but involves the search for an optimal context. This is why, in the book, which Wilks had not read when he grandly accused us of circularity, but which Seuren is supposed to be reviewing, we do not stop at the definition of relevance i n a context but go on, on the basis of this first definition, to define relevance for an individual, i.e. for a mechanism capable of accessing a variety of contexts. However, this takes place on pp. 1 42- 1 5 1 , and, by then , · less than halfway through the boo k, Seuren's interest has flagged. He dis misses the rest of the book in a single page. Here we will summarise the basic claims Seuren neglects, and contrast them with claims he attributes to us earlier in his review.

RELEVANCE AND ATfENTION For the individual, we claim, the effort required to process some newly-pre sented item of information consists of (a} the effort required to access a context, and (b) the effon required to derive the contextual effects of the new information in that context. The greater the contextual effects, the greater the relevance, and the smaller the processing effort required, the greater the relevance. Differe nt choices of context yield different contextual effects, and require different amounts of processing effort. We claim that humans tend to process each newly-p resented item as productively as possible: that is, to


'solubility ' : suppose a definition of 'solubility

155

R E LEVANCE AND COMMUNICATION

A communicator, by demanding an addressee's attention, creates a presump tion of relevance. She suggests, that is, that her utterance (or other act of inferential communication) is relevant enough to be worth the addressee's attention. But how relevant is that? Here, our claims are quite precise. To satisfy the presumption of relevance, an utterance must meet two conditions. First, it must achieve enough contextual effects, at a small·enough processing cost, to be worth the addressee's attention: that is, it must be more relevant than any other phenomenon in the addressee's cognitive environment. Sec ond, it must put the addressee to no unjustifiable processing effort: that is, it must be the most relevant utterance (or other act of i nferential communica tion) that the communicator could have used to achieve the intended contex tual effects. An utterance which satisfies these two conditions we call optimally rele vant. By demanding an addressee's attention, then, a communicator creates a presumption of optimal relevance. And the fact that she does so, we call the principle of relevance. Now the fact that an utterance creates a presumption of optimal relevance does not mean that it will actually be optimally relevant to the hearer. The hearer may be unable to find an i nterpretation on which it is optimally relevant to him. or one on which a rational speaker might have thought it would be so. When such an interpretation can be found, we say that the utterance, on this interpretation, is consistent with the principle of relevance. We show that every utterance has at most a single interpretation consistent with the principle of relevance.


obtain a s many contextual effects from i t as possible, given t h e available processing resources. Processing resources are scarce; more information is accessible at a given moment than anyone can hope to deal with. It is quite natural to assume that the human cognitive system has tended towards a maximally efficient use of the available processing resources, and hence, towards selection of a context in which the newly presented information can be most productively processed. Relevance theory thus provides a general answer to some fundamental cognitive questions: which phenomena should the individual attend to, which conceptual representations should he assign to those phenomena, and how should he process these representations? More specific answers depend on the individual's actual cognitive abilities (e.g. his perceptual abilities, the organi sation of his memory) and his p hysical environment: in our terms, on his cognitive environment. What makes communication possible is that humans, at least to some extent, share cognitive e nvironments, and are thus able, at least to some extent, to predict each other's allocation of processing resources. It is against this cognitive background that our theory of com mu nication is set.

1 56


Suppose, for example, that an utterance may be taken to express one of two propositions, P I and P2. Suppose, furthermore, that it is manifest in the shared cognitive environment of speaker and hearer that (a) PI is more accessible to the hearer than P2; (b) the hearer has access to a context C 1 in which PI has enough contextual effects to be worth his attention (call this set of effects E l ); (c) there is no more accessible context in which P I would have enough contextual effects to be worth the hearer's attention; and (d) there is no alternative utterance (more generally, act of inferential communication) which would have achieved the set of effects E 1 more economically. Then the overall interpretation which consists of selecting P I and processing it in context C 1 to obtain the set of effects E 1 is consistent with the principle of relevance - and is, moreover, the only interpretation consistent with the principle of relevance. For suppose the speaker, to whom by hypothesis all these facts are mani fest, had nonetheless intended to express p roposition P2 rather than P 1. The resulting interpretation - however relevant - would not be consistent with the principle of relevance, because it would put the hearer to the unjustifiable processing effort of, first, recovering and processing P 1 , next, recovering and processing P2, and then engaging in some further form of inference to decide between the two interpretations. All this the speaker could have spared him by simply rephrasing her utterance to eliminate P 1 entirely, or to make P2 more accessible than P l . What goes for disambiguation and reference assignment goes equally for determination of the intended context and contextual effects. Suppose it is manifest in the shared cognitive environment of speaker and hearer that (a) context C l is more accessible than context C2; (b) context C l will combine with the proposition expressed by the utterance to yield enough contextual effects, for a small enough processing effort, to be worth the hearer's atten tion ; and (c) no other utterance would have achieved these effects more economically. Then an interpretation based on context C2 and its associated contextual effects, however relevant, would not be consistent with the princi ple of relevance, because it would put the hearer to some unjustifiable processing effort. We show, more generally, that in every aspect of utterance i nterpretation involving the resolution of indeterminacies - of content, context or contex tual effects - the first accessible interpretation consistent with the principle of relevance is the only assumption consistent with the principle of relevance, and is the one a rational hearer should choose. In chapter 4, we illustrate the application of this criterion to every aspect of utterance interpretation. Let us compare the resulting theory with some comments made by Seuren in his review.

1 57 SEUREN'S IRRELEVANCIES

(5)

a.

Flag-seller: Would you like to buy a flag for the Royal National

b.

Passer-by: No thanks, I always spend my holidays with my sister

Lifeboat Institution? in Birmingham. We point out that in order to recover the intended interpretation of (5b), the hearer must be able to access the assumption that someone who is not likely to need the services of a charity cannot be expected to subscribe to that charity - an assumption which is typically not manifest to everyone. Here, Seuren comments, 'SW cross metaborders unaware: they fail to distinguish, for a proposition P, between taking P as true, and taking P as believed to be true by an interlocutor. Yet P has a very different status in one case and in the other.' Indeed - or rather, almost. The distinction between assumptions that are actually true and those that are merely entertained as true has no place in a theory of cognitive psycho logy based on the assumption set out above: that the individual has no access to the semantic i nterpretations of h is beliefs, and is thus unable to distinguish between what he knows and what he believes (or more generally, assumes).


Take Seuren 's remark that 'there will be few utterances that fail to give rise to some inference together with some background knowledge. But most of these SW will want to rule out as irrelevant', as indeed we would. There are two mistakes here. First, it follows from our definition of relevance that an utterance is relevant to the individual if it has contextual effects in some context accessible to him. We discuss this issue at length (pp. 1 1 8- 1 23), and explain why intuitions may be hazy on this point. They may be hazy because where utterances are involved, people have intuitions not only about relevance but about consistency with the principle of relevance, and the two sets of intuitions do not coincide. An utterance may be relevant on a given interpretation without being consistent with the pri nciple of relevance on that interpretation; it may be consistent with the principle of relevance on a given interpretation without being relevant at all (for example, when you tell me something I already know). Seuren mistakenly assumes throughout his review that we regard the search for relevance as the key to recovery of intended contexts and contex tual effects. As shown above, what we argue is something quite different. In the interpretation of inferential com munication, and in particular in the recovery of intended contexts and contextual effects, it is the criterion of consistency with the principle of relevance which is crucial. As we have also shown, this criterion is powerful enough to exclude all but a single interp reta tion - a single intended context, a single set of contextual effects - for any utterance. Seuren does not even acknowledge its existence. Consider, in this light, Seuren's comments on our example (5):

1 58 He can, of course, distinguish between his own assumptions and those of others; or between assumptions that are manifest to him and those that are manifest to others. We claim that he can also distinguish between assump tions that are merely manifest to him and those that are mutually manifest to him and his interlocutor - or at least are treated as such . It is from the latter class of assumptions that intended contexts are drawn. Often, a speaker treats as mutually manifest an assumption that, before she spoke, was not manifest to her hearer at all. This is the source of what Seuren calls 'backward suppletion' (generally known as 'bridging implicature'), and other examples involving implicated contextual assumptions. Suppose Mary says to Peter: John entered the room. Both windows were open.

In assigning reference to the expression 'both windows' in (6}, Peter typically has to assume that the room John entered had two windows. This assumption - though easily accessible to him as a hypothesis via his encyclopaedic knowledge that rooms may have one or more windows - may not have been

I'TUlnijest to him before Mary spoke: that is, he may have had no more reason to believe that the room had two windows, than that it had one or three. By treating this assumption as mutually manifest - that is, by producing an utterance which is optimally relevant only in a context containing this assumption

-

Mary provides indirect evidence that she believes it; and if Peter

trusts her, he will believe it too. In just the same way, the passer-by in (5b) treats it as mutually manifest to himself and his hearer that someone who has no need of the services of a charity cannot be expected to subscribe to that charity. The structure of the interpretation process is the same for (5b) as for (6): the only difference is that an altruistic hearer may have difficulty actually accessing the required contextual assumption, and having seen that it i.r required, is most unlikely to accept it as true on the evidence of a passing stranger. Seuren's claim (and the earlier claim of Wilks 1986) that we are unable to make the subtle distinctions between speaker's and hearer's beliefs, and their beliefs about each other's beliefs, needed in pragmatic theory, are therefore gratuitous. We discuss these issues at length and provide a framework for solving them in terms of manifestness which, we claim, is more psychologi cally realistic than the usual 'mutual knowledge' framework.

'PERTINENT CRITICISMS'

We would like to end with three more general comments. The first is rather minor. In a number of places, Seuren refers to published criticisms of an early paper of ours, by Gazdar and Good, Moore, Wilks, and Clark, remarking that these 'pertinent criticisms . . . . do not seem to have been either heeded or


(6)

1 59

CONTEXT AND CONCERN

The second, more substantial, comment has to do with Seuren's own precon ceptions about the nature of relevance, which he takes us to share. Seuren regards it as uncontroversial that to see the point of an utterance is to see how it is relevant 'with respect to a concern'. Where we talk of relevance in a context, he feels free to replace our reference to contexts with reference to concerns. We strongly reject the identification of contexts with concerns. Contexts are sets of assumptions: we define relevance with respect to a context, describe how contexts are accessed, what it is to process an assump tion in a context, and what the effects of such processing might be. By con trast, we have no idea (and we doubt if Seuren has either) what a 'concern' might be, or how an utterance might be relevant with respect to a concern. Clearly, there would be little point in defining relevance if the definition itself made appeal to an unanalysed notion of 'concern'. One of the main recommendations of our approach to relevance is that it makes no appeal to such unanalysed notions. To the extent that systematic pretheoretical intui tions exist about relatedness, interests or concerns, we claim that they can be satisfactorily explicated in terms of our notion of relevance, eliminating the need for technical definitions of 'relatedness', 'interest' and 'concern' . Given the total lack of anything approaching a technical definition of these terms, this is a substantial claim, and one that should surely be welcomed.


countered'. In fact, we made detailed replies at the time, which were publish ed in Smith ( 1 982) along with our original paper and the criticisms of Gazdar, Good, et.al. Seuren nowhere refers to these replies6• No doubt there will come a day when someone will say, with equal good faith, that we have neither heeded nor countered Seuren's 'pertinent criticisms' . Not only has Seuren failed to notice these earlier replies; he seems unaware too of the fact that the theory defended in Relevance is not the same as the one presented in our earlier papers (see note 1 1 , p. 26, where we draw attention to the difference between our original p rinciple of maximal relevance and our current principle of optimal relevance). If Seuren wishes to go on citing Gazdar, Good, et.al., he had better make sure that their objections apply to the theory we defend. This kind of sloppiness is displayed throughout the text. Seuren accuses us of failing to deal with objections which are fully dealt with, of failing to give definitions which are explicitly given, and of failing to discuss issues which are carefully discussed. To take j ust one example not already mentioned, Seuren wonders whether our account of inferential communication 'excludes questions, commands, and other non-assertive speech acts. SW are not explicit on this'. If he had finished the book (or even the table of contents}, he would have noticed that speech acts in general, and the semantics and pragmatics of non-declarative sentences, are explicitly dealt with in chapter 4, section 10.7

1 60 THE REVIEWER'S DUTY AND RIGHT

A reviewer has a duty and a right. The duty is to inform readers of the


contents of the book reviewed. The right is to express his own opinion of it. Seuren is much keener to exercise h is right than to perform his duty. While accusing us of 'inadequate treatment of the literature and uneven handling of topics' he confines himself to discussing only the first half of our book. Thus, all but four of his sixty quotations are from the first 140 pages of our 254-page text; thirty-five of these quotations are from the second (and shortest) chap ter. As we have shown, Seuren does not even give an adequate account of the few passages which he finds worth discussing, or rather denouncing, or explain how they fit in with the general design of the book. In fact, he utterly fails to represent the contents of Relevance. By contrast, Seuren makes his opinion of the book very clear. Given his theoretical choices, which are very different from ours, we would not have expected Seuren to be convinced by our arguments; we would have welcomed a reasoned expression of disagreement which, though it might not have convinced us, might have highlighted genuine problems and contrasted theoretical alternatives. But Seuren does not merely disagree, nor does he feel that we offer a conceivable alternative. Once in a while, a reviewer gets a book which does not meet basic standards of scholarship and which should not have been published in the first place. It then becomes not just the reviewer's right but his duty to publicise the fact and warn the community. Seuren felt that this situation had arisen with our book: he felt it incumbent upon himself to discuss it in detail 'mainly to warn against a creeping tendency, manifest in pragmatic writings these days, not to apply normal standards of precision ' and scholarship'. He then took an authoritarian stance, and felt free to rebuke us with scorn and sarcasm. Thus he talks of our 'loose and inconse quential way', our making 'a mess of things', his 'mounting amazement', our failing to maintain 'a minimum of coherence', our 'tortuous and longwinded path', our 'turgid text', our 'undeserving formal workmanship', our getting 'entangled in absurdities', our 'verbose text', our 'dodging', etc. Before taking such a stance and levelling such abuse at colleagues, one had better make sure that one is not misled by one's own theoretical biases, or by the fact that one's work is attacked (or, worse, ignored) in the book under review; or by personal considerations of any kind; one had better make sure that the case is incontrovertible. Seuren's case is non-existent. His verbal violence is totally unjustified. He claims that our calling the set of hypotheses we put forward a 'theory' is 'an imprope r appropriation of prestige' . We would like to know what Seuren considers to be worthy of the name 'theory' in pragmatics, if our work is not. In any case, prestige accrues to good theories; bad theories end up being an embarrassment to their authors. On the other hand, when Seuren appoints h imself our censor, and the defender of true scholarship against unworthy colleagues, he commits an utterly impro-

161 pe r appropriation of authority; h e is the one who ought t o b e embarrassed. There is one way though, in which Seuren's attack might cause one to doubt relevance theory. According to this theory, homo sapiens systemati cally tries to interpret new information as relevantly as possible. Seuren. has done his best to interpret our book as irrelevantly as possib le. Is he, then, a living counterexample to the theory? A more conservative explanation can be given: for Seuren, the assumption that our book would be totally irrelevant was more relevant than anything he expected to find in the book, and so, he tried to interpret our book so as to maximize overall relevance, and he did a great job of it. I:Xpart�nt oflinguistics

London WCJ£6BT UK and CNRS/ Univusltld� Paris X

NOTES I.

See, e.g., Sperber and Wilson ( 19 8 1 , 1 982a-d, 1986); Wilson and Sperber ( l 979, 1 98 1 , 1986a,

1986b, fonhcoming).

2

Our account of presuppositions (including, despite his denials, an account of the type of

examples that interest Seuren) is given in R�l�ance chapter 4, section S.

An earlier account

was

published in Wilson and Sperber ( 1 979).

3.

Seuren also seriously mistakes the role of ass u mption schemas in ovr framework. He regards

them as devices for inferring one assumption from another, saying: 'Ass u mptions are also taken to

be derivable from ass u mptions

in vinue of fixed schemata. Thus . . . when an assumption of

the form "If P then Q" is encoded for transmission, the hearer will weakly infer "If not-P then not-Q", and also "If Q, then Q because

P''. Such forma.:lon of assumptions is considered

standard'. What we actually say in the passage cited is thQt when available assumptions fit the

P' is, when entenaining a55umptions of a cenain form, humans

schema 'If P then Q', assumptions of the form 'If not-P then not-Q' and 'If Q, then Q because are standardly consid�r�d. That

standardly consider whether they have evidence for c;enain assumptions of related form. There is no talk of utterances, there is no talk of hearers, and there

i5 no talk of inference. We are merely

suggesting a possible procedure for hypothesis formation; the resulting hypotheses to be evaluated, as always, in the light of additional contextual assumptions. Seuren's 'counterexam ples' in (9) merely illustrate a variety of cases in which the standard hypotheses prove false.

4.

Here, Seuren may have been confused by his own unargued assumption that we are using a

standard first-order logic, supplemented in various ways. We nowhere make such a commit ment. Indeed, we envisage an entirely non-standard treatment of existential quantification. which readers who know something of the formal propenies of phrase-structure trees may be able to work out from the tree given on the last page of Seuren's review. S.

Moreover, suppose that vinually nothing were known about processing effon, but that it did

play a role in communication and cognition. The claim that it did so, and speculations about the role it played, would surely not 6.

be vacuous.

We elaborate on some of these replies in R�l�ance. For example, our description of

derivations ( I) and (2) above (which Scuren treats as a response to an objection fro m an 'imagi nary opponent') develops our original reply to Gazdar and Good. 7.

See also note 3. p. 262.

For an elaboration of these views, see Wilson and Sperber (fonhcoming).


Univuslty Co/kg� London Go.,."t!rSrrut

1 62 R EFERENCES Oark. H 1982: The relevance of common ground In: Smith (ed.). Cole, P., (ed.) 1 98 1 : Radical Pragmatics. Academic Press , New York. Dancy, J., Moravcsik, J. and Taylor, C. (eds.) to appear: LaltguDge and Value. Stanford Umver sity Press, Stanford. Gazdar, G. and Good, D., 1982: On a notion of relevance. In Smith (ed.) Myers. T., Brown. K. and McGonigle. B. (eds.). 1986: Reasoning and Discourse Processes. Academic Press , London. Oh, C.-K. and Dinneen, D., (eds.) 1979: Syntax and Seman11cs 1 1: Presuppositions. Academic Press, New York. Rips, L., 1983: Cognitive processes in propositional reasoning. Psychological Review 90. 1 :38-7 1 . Smith, N . (ed.), 1 982: Mutual knowltdge. Academic Press , London. Sperber, D. and Wilson, D., 1 98 1 : Irony and the use-mention distinction. In: Cole (ed.). Sperber, D. and Wilson, D., 1 982a: Mutual knowledge and relevance in theorin of comprehension. In: Smith (ed.). Sperber, D. and Wilson, D., 1 982b: Reply to Oark. In: Smith (ed.). Sperber, D. and Wilson, D., 1982c: Reply to Gazdar and Good. In: Smith (ed.). Sperber, D. and Wilson, D., 1 982d: Reply to Wilks. In: Smith (ed.). Sperber, D. and Wilson, D., 1986: Loose talk. Procudings of the Aristotelian Society. .•

Parassession on Grammar and Pragmatics.

Wilson, D. and Sperber, D., (forthcoming): Mood and the anal�is of non-declarative sentencn. To appear in: Dancy, Moravcsik and Taylor (eds.).


1985-6: 1 53-7 1 .

Travis, C . (eel.), 1986: Meaning and lnterpretallon. Oxford University Press , Oxford. Werth, P. (ed.), 1 98 1 : Convtrsation and Discount. Croom Helm, London. Wilks, Y., 1986: Relevance and beliefs. In: Myers, Brown and McGonigle (eds.). Wilson, D. and Sperber, D., 1979: Ordered entailments: an alternative to presuppositional theories. In: Oh and Dinneen (eds.). Wilson, D. and Sperber, D., 1 98 1 : On Grice's theory of conversation. In: Werth (ed.). Wilson, D. and Sperber, D., 1 986a: Inference and implicature. In: Travis. Wilson, D. and Sperber, D., 1986b: Pragmatics and modularity. Chicago Linguistic Society

COMMENTS ON MARVIN MINSKY: THE SOCIETY OF MIND Editorial Preface

_

•originally circulated as 'A framework for representing knowledge', Artifi cial Intelligence Memo no. 306, June 1 974, MIT, AI Laboratory.


Marvin Minsky has written a new book, "The Society of Mind", which is coming out one of these days at Simon and Schuster, New York. Style and presentation of the book are reminiscent of his seminal l 974 ' Frames' paper• : Minsky does not get too involved with a discussion o f details but tries t o get across and make plausible a particular way of thinking of mental phenome na, which , in one form or the other, has increasingly attracted cognitive scientists during the past years: a view of the mind as a 'society' of many small and highly specialized 'agents', each of them quite simple and not equipped with any form of intelligence, that interact with each other in order to produce what we perceive as manifestations of the human mind or intelli gence. In view of the relevance these developments have for the semantics of natural language, the Editors asked a number of semanticists for their comments on Minsky's book, or, to be precise, on an almost final version of the manuscript, which Marvin Minsky had kindly agreed to make available. We asked the commentators to focus, in particular, on those sections of the book that are specifically concerned with natural language semantics. In the following pages we print the comments we have received by Osten Dahl (Stoc kholm), John Marshall (Oxford), Keith Oatley (Glasgow), Ragnar Rommetveit (Oslo), and Tony Sanford (Glasgow). We are grateful for their and Marvin Minsky's kind cooperation. All references to Minsky's book are b y paragraph number. The Editor

Journal of s�nuuu ics 5: 1 65- 1 67

CONFLICT AND CONTROL AMONG MENTAL AGENTS

Keith Oatley

"The Society of Mind" is a terrific title. If shifts the discussion of mentality from concepts of processes and computational objects to concepts of agents.

organization of modules each of which is somewhat autonomous and only able to accomplish some simple thing, but is affected by the community of other agents with which it interacts. Two questions raised by such an approach are of how such units might cooperate, and of what happens when they conflict. I will concentrate on these more general issues, rather than discussing the specifically semantic proposals in the book. Just as earlier in this century social scientists started to ask what it is that coordinates individuals to create an organized society, Minsky now asks how the particles of mind, these agents which are each themselves mindless, interact in meaningful organizations which do have the properties of mind. Minsky is of course not the first to have drawn on a societal metaphor of mental organisation. John Hughlings Jackson (see e.g. 1959), the influential Victorian neurologist, was fond of it and used it to describe his theories of the hierarchical structure of mental and neurological organization, which come in some ways close to computational ideas. He was an early theorist of representations, and he described functions as higher and lower in a manner that has now become common. He advanced ideas of the evolution of more complex higher functions out of the lower ones to account for the way in which brain damage often destroyed apparently more complex and recently evolved mental processes, like voluntary language, while leaving lower, less highly evolved processes relatively unscathed. Here is a quotation from Jackson illustrating his societal theory of mind:

The higher nervous arrangements evolved out of the lower, keep down those lower, just a government evolved out of a nation controls as well as directs that nation. If this be the process of evolution then the reverse process of dissolution 1s not only a 'takmgotr of the higher, but it is at the very same time a 'letting go' of the lower. If the governing body of this nation were de11 troyed suddenly we should have two causes for lamentation: ( I ) the loss of services of eminent men; and (2) the anarchy of the now uncontrolled people. as


It implies a rewarding set of analogies between intelligent processes and human societies. It appeals to anyone who might wish to consider mind as an

1 66 The political sentiments expressed here are what one might expect from a successful Victorian doctor. They might warn us to be careful in projecting societal metaphors onto the mind. worthwhile their value must

In order to make such me taph ors

be described in the cognitive domain, not just left

as ideas which may seem recognizable i n society. Jackson was good a t proposing metaphors of evolution and hierarchy. W h a t cognitive scientists more recently have been good at describing are properties of computational hierarch ies. For instance a hierarchy might consist of an operating system as the top level, or a calling program that evokes lower level functions or procedures, with each of these procedures itself hierarchically arranged and able to invoke more pri m itive actions.

On reflection I am inclined

to think of the book less as an actual exploration, more a set ofsuggestions at the metaphorical level, which will stim ulate others to undertake explorations and instantiations. In the opening pages Minsky expounds the now familiar workings of hierarchical organizations. The agent Builder, that he describes, simply turns on and off subordinate agents such as Add, which itself can only act to invoke agents lower in the hierarchy, such as of its activities, however,

Builder

Find, Get and Put.

When seen in terms

becomes an agency. It accomplishes

something complex and seems to k n o w its job. The hierarchy is the chief orga nizational arrangement that we understand, or think we do, in society and in computation. But as Minsky hints, if agents have goals other than those set by a n agent higher in the hierarchy, all is not well. We need to discover principles of organization to supplement the hierarchical. If there is a single top level plan to build a tower of blocks, and if this

is

acco mplished hierarchically wi t h each lower element being switched on and off by the one above it, then organizational problems are simple. But as soon as agents are affected by agents other than those immediately above them in the hierarchy, the potential for more complex and intelligent forms of organization brings with i t the potential for conflicts. Minsky quite properly discusses such proble ms, and here we might expect to find the societal metaphor being a t its most helpful. In Section

3. 1 for instance, Minsky writes : "To settle arguments, nations

develop legal systems, corporations establish policies, individuals may argue, fight or compromise - or tum fo r help t o mediators outside themselves". The corresponding mental processes that Minsky proposes are that "conflicts between agents tend to migrate upwards to higher levels" and weaken their superior (Section 3. 1 ) , and that "The longer a conflict persists among an agent's subordinates, the weaker becomes that agent's status among its own

6.5 "No superior can know 1 6. 9, there i s a suggestion as to why children seem

competitors " . (Section 3 . 2). But in Section everything . . . " In Section

to swing violently between emotional states: "one agency attains control and


Minsky offers to explore with us forms of organization of agents which will extend our understanding beyond the hierarchy.

1 67

D�pr. of Psychology Univ�rsiry of Glasgow Glasgow. G ll BRT Scotland


forcibly suppresses the rest". Again a societal methapor, that of control and suppression is accompanied by a computational hint: that agencies are equipped with cross exclusion devices, which inhibit neighbouring agents . The implication of such suggestions is that whereas the hierarchy remains a uniform arrangement to enable coordination among agents, the problems of conflict are solved by a number of piecemeal methods. If I may adopt Minsky's style of argument: My intuition is that with such ideas we only scratch the surface of the problems that arise in organizations in which agencies are affected by several inputs or have any degree of autonomy. In society the hierarchy is by no means the only organization by which cooperative activities are accomplished. Minsky was the director of the AI lab in which was conceived the idea of exploring heterarchies, organizations with non-hierarchical structure. In his prologue he describes how his book is j ust such a heterarchical society of small ideas, with rather tangled webs of cross connections, reflecting the cross connectedness of the simple but complexly interconnected agents of mind. In society hierarchies seem to be favoured in organizations that have a single well defined goal. The mind is not like this. Each of us has multiple goals. One suggestion, made by Johnson-Laird and myself (Oatley and Johnson-Laird in press) is that, in modular cognitive systems, emotions have a role in setting up modes within which the activities of subsets of agents are compatible. So, for instance, in anxiety a pervasive signal interrupts ongoing activities, invokes agents to maintain vigilant watch on the environment, and others to check various matters to do with personal safety, while yet other agents t hat might initiate certain kinds of new plan are inhibited (cf also Gray's 1 982 arguments about this in the neuropsychological domain). Correspondingly among individuals anxiety spreads in a non specific way, and similarly induces wariness and carefulness. These pervasive signals, then , would be examples of non hierarchical communication. This is a relatively simple example. More complex examples involve verbal language. In society speech acts are capable of invoking a variety of forms organization. To put it crudely, the command is not the only speech act. Minsky himself has had a pervasive rather than a hierarchical influence on the society of cognitive scientists. His new book will stimulate us to think about the computational equivalents of the many and varied types of influence by which mentality may be organized.

1 68 REFERENCES

Gray, J., 1 982: The nturopsychology

of QIIXitty Oxford: Oxford Univcnity Preu. Hughlings-Jac'-son, J . , 1959: <cttd writings of John Hughlings-Jackson. Ed. J. Taylor. New York: Basic Books.

Oatley, K. and Johnson-Laird, P.N., (in press): Towards a cognitive t heory of emotions.

Cogrution and emotion.


Jounral of s�mantics 5: 1 69- 1 7 3

CLOSE ENOUGH F O R AI? John C. MarsbaU


Two themes run through The Socitty of Mind ( 1 986). The first is that the mind-brain is not a unitary organ , but rather consists in a number of discrete agencies; the second is that these 'intelligent' agencies can each be fractionated into simpler agents which are individually 'mindless'. Neither claim is original; both are a commonplace of contemporary theories of cognition (Chomsky 1 980, Marr 1 982, Fodor 1983). The notion of multiple intelligences in something closely akin to its current form can be traced back to the work of Franz-Joseph Gall (Marshall 1980a); that wholes can achieve what parts cannot has been the stock-in-trade of engineering since the Ionian enlightenment (Marshall 1 977). Minsky's lack of originality in this respect is, of course, no vice if one believes, as I do, in the truth of both broad claims. But scientific virtue surely involves the effort to get a few of the details right. And that in tum is usually taken to involve empirical inquiry. The goal of discovering how many 'agencies' the mind-brain instanciates has been approached fro m a variety of angles. For example: Students of natural computation might inquire whether the formal structure of the rules exploited in decomposing a complex visual shape into its parts ( Hoffman and Richards 1 984) bears any similarity to the constraints exploited in parsing a sentence into its constituents (Berwick and Weinberg 1 984). Experimental psychologists have asked which pairs of high-level tasks can be accomplished simultaneously without performance decrement on either one (Allport, Antonis, and Reynolds 1972; Shallice, McLeod, and Lewis 1 985). With adults who have sustained relatively focal brain damage, neuropsychologists have investigated which faculties of mind may remain intact despite severe impairment to other cognitive domains (Marshall 1 984); likewise there has · been study of developmental disorders in which the growth of particular cognitive domains may be selectively impaired or preserved (Alexander, Ehrhardt and Money 1 966, Yamada 1 98 1 , Curtiss 1 98 1 , Bellugi, Sabo and Vaid 1 986). Such work is begi nning to answer the question : What is the structure of the (quasi- independent modules that serve human cognition? By contrast, Minsky's speculations about cognition are bizarre, to put it mildly. Are we intended, for instance, to take seriously the opinion that language acquisition consists in learning simple word-order frames and then slotting into them other little frames (26.8)? Evidence that would support such a view is not highly conspicuous in recent accounts of language

1 70


development (Wanner and Gleitman 1982, Pinker 1984, Berwick 1 985 , Hyams 1986). Can one see any but the most superficial relationships between the fact that two parts of a complex verb can be parsed together even when they are separated by other words and the fact that a plank can be recognized even when another is placed across it (26.9)? But this purports to be real science, for "in only a few decades we may be able to trace the ancestry of many such relationships, by examining the genes which generate the corresponding brain-structures" (26.9). Hope eternal may go down well at the Maharishi International University (6. 1 0), but more empirically-minded readers are likely to see red. Concerning which, Minsky remarks that "It's easy to design a machine to tell when there is something red" ( 1 1 . 1 ). The solution? "Start with sensors which respond to different hues of light and connect the ones most sensitive to red to a central 'red-agent', and so forth" ( 1 1 . 1 ). Minsky calls this device "a travesty" . . . . because it "would share no human notion of what colors come to mean to us" ( 1 1 . 1 ). Sadly, he does not indicate that the model is bull quite apart from what colours mean to us (or bulls). Unless Minsky knows of some fundamental objections to a whole line of investigation from Helmholtz' work on colour constancies to Land's retinex experiments, it would seem that the computations required to derive colour perception in V4 are far from simple (Zeki 1 980). Minsky's idea of what is easy to design can , however, be extended very simply to other domains: Start with sensors which respond to different faces and connect the ones most sensitive to grandmother to a central "grandmoth er-agent" . . . thus bearing out the truth of Minsky's profound aphorism : " The most efficient way to solve a problem is to already know how to solve it" (7.4; Minsky's italics). Yet the critical faculty is not entirely dormant in Society of Mind, for elsewhere 'photographic memory' is dismissed as either a myth or a fancy show put on by "professional magicians or charlatans" ( 1 5.3). It is not too clear what Minsky means by 'photographic memory', but if he is referring to eidetic imagery then my understanding of the literature is that the phenomenon can be found in some five percent of normal children (Gray and Gummerman 1 975). In the adult population, eidetikers are admittedly extremely rare; there is, however, at least one report (Stromeyer and Psotka, 1 970) that appears to document (with adequate controls) a quite astoundin� ability to re-visualize very complex images from memory. Have I missed a basic flaw in this research? When Minsky asserts that "there are many legends of persons having fabulous memories" ( 1 5 . 3), does he inted to include, say, Luria's The Mind of a Mnemonist ( 1 968) in the category of myth? If so, I'd appreciate hearing why. Minsky's own account of learning and memory is based on the notion ofK Iines: "Whenever you 'get a good idea', solve a problem, or have a memorable experience, you activate a K-line to 'represent' it. A K-line is a wire-like structure which attaches itself to whatever mental agents are active when you

171


solve a problem or have a good idea" (8. 1 ). I suspect that this is a sloppy paraphrase of some ideas in Hebb ( 1 949) about the construction of cell assemblies, but it's difficult to be sure. In any case, it should be pretty obvious that the activation of K-lines is a statement of the problem of association and learning, not the solution to that problem . Hebb himself, and many later scholars (e.g. Hinton and Anderson 1 98 1 . Finkel and Edelman 1 985, Rumelhart and McClelland 1 986) have tried to work out how associative memories could be formed from systems of modifiable synapses. Minsky's device, however, seems to work not by modifiable synapses but rather by exclamation marks: "When you activate that K-line, later, this arouses the agents attached to it, putting you into a 'mental state' much like the one you were in when you solved the problem or got that idea. This should make it relatively easy for you to solve the same or similar problems!" (8. 1 ). I should be so lucky! ! Many of the terminals in the "frames of mind" (24.2) that K-lines arouse are assigned by default, we are told. Thus "As soon as you hear a word like 'person', 'frog' , or 'chair', you assume the details of some 'typical' sort of person, frog, or chair" (24. 2). Again, no evidence is provided on the basis of which we could evaluate the claim. Try as I may, I have the utmost difficulty in imagining what a 'typical' person could be. Perhaps I ' m trying too hard and my unconscious knows all along? Thus, "when someone says 'John threw a ball' you probably, unconsciously, assume some certain set of features and qualities of the ball, like color, size and weight" (24.4). Yet no priming experiments (for example) are quoted in support of these unconscious assumptions. We know that ball will (temporarily) activate bat and dance (Seidenberg and Tanenhaus 1 986). Does Minsky know of results where ball also primes, for example, red, eighteen inches in diameter, and half a pound'! Minsky is also holding out on us with respect to his latest findings in neuroanatomy. He thus appears to believe that dividing the hemispheres at the corpus callosum is merely one of "many other ways to draw i maginary boundaries through brains" (28.8). Many of us have criticised the dichotomania of work on cerebral laterality, but never I fancy in quite such cutting terms (Marshall 1 98 1 ). If I were unfortunate enough to suffer from intractable epilepsy I might consider asking a reputable neurosurgeon to perform a commissurotomy; but I cannot i mmediately think of any circumstances in which I would request him or her to divide my brain on a line running from the right frontal lobe to the left occipital. Elsewhere ( 1 1 .8), Minsky writes: "My own theory" - I think he means Lenneberg's ( 1 967) theory - "of what happens when the cross-connections between these brain halves are destroyed is that, in early life, we start with mostly similar agencies on either side. Later, as we grow more complex, combinations of genetic and circumstantial effects lead one of each pair to take control of both". Again , my own reading of the literature (Marshall 1980b) is that the basic anatomical asymmetries are present at birth and that a large number of functional

1 72

N�ropsychology Unit N�rosci�nce Group

"J"M Radcliff� Infirmary Wood.Jrock Road Oxford OX2 6HE England

REFERENCES Alexander. D. . Ehrhardt. A.A. and Money, J 1 966: Defective figure drawing, geometnc and human. in Turner's syndrome. 1M Journal of Nuvous and M�nral Drs�a.u 1 42: 1 6 1 - 1 67. Allpon. D.A .. Antonis, B. and Reynolds. P . 1 972: On the division of attention: A disproof oft he stngle channel hypothesis. Quarrerlq Journal of Expuim�nral Psychology 24: 225-235. .•

.


asymmetries can be found at as early an age as we have reliable techniques to measure them. Such an interpretation on my part is not dramat-ically idiosyncratic (see Curtiss 1 985). We may, of course, be wrong, but it would be nice to be presented with the evidence. I tried to find such evidence by reading "between the lines" for the "more technical version" or "second level" (P- 1 ) of The Society ofMind. The best I could manage was confirmation of Minsky's "hope that the second level does not show" (P- I ) . But I now see t hat Minsky never intended to write "a text of scientific scholarship" (P- I); rather we are recommended to approach the book "as an adventure story for the imagination" (P- I). What can one say? Only that Society of Mind does not quite attain the logic, intellectual clarity, and scientific depth of Raidl!rs of the Lost Ark. When drafting the book, Minsky apparently spent a lot of time "deleting every reference to scientific evidence" (P- 1 ). The only positive suggestion one can make is that he spend a little time putting them all back in. A good place to start would be the discussion of autism, a condition where, as Minsky writes, the affected children "do not establish effective communication with other people, al though they may acquire some competence at dealing with physical things" (29.6). Thanks to the outstanding work of Baron-Cohen, Leslie and Frith ( I985) we can now give some serious content to such clinical observations. Autistic children, it would seem, are unable to impute to others beliefs which differ from their own; a considerable proportion of the autistic child's social i ncompetence may arise from this underlying deficit, a deficit which is furthermore not found in children with Down's syndrome whose mental age (on standard tests) is lower than that of the autistic population studied by Baron-Cohen, Leslie and Frith. A careful reading of their fine paper should help show Minsky's non-specialist (P- I ) what the difference is between controlled experimentation relevant to an explicit theoretical claim and making it up as you go along. Upon experiencing some difficulty in tuning his guitar, Bob Dylan was once heard to remark "Close enough for folk music" . Perhaps The Society of Mind is close enough for Artificial Intelligence?

1 73 Baron-Cohen, S . , Leshe, A.�. and Frith, U., mmd'?

1 985: Does the autistic child have

a 't heory of

Cognition 2 1 : 37-46.

Bellugi , U., Sabo, H. and Vaid. J . .

1986: Spatial deficits in children with Williams syndrome. I n : Spatial Cognition: Brain Ba.su and

J . Stiles-Dav1s, M. Kritchevsky, a n d U. Bel l ug1 (eds.).

D�elopment.

Ch1cago U m versny Press. Chicago.

1985: The Acqui.mion of Syntactic Knowledge. M I T Press, Cambndge, Mass. BeJ'Wick, R.C. and Weinberg, A .. 1 984: The Grammatical Basis of Linguistic Performance. MIT

Berwick. R.C.,

Press. Cambridge, Mass.

1980: Rules and Representations. Columbia Umversny Press, New York. 1 98 1 : Dissociat1ons between language and cognition. Journal of Aumm and Developmental Disorders I I : 1 5-30. CurtiSS, S . , 1 985: The development of human cerebral lateralization. In: D. F. Benson and E.

Chomsky, N . , Curtiss. S . .

Za1del (eds . ) ,

The Dual Braur: Hemispheric Specialization in Humans.

The G u 1lford P�s.

New York.

1985: Interaction of synaptiC modification rules within USA 82: 1 29 1 - 1 295.

populations of neurones. Procudings ofthe National Acadtmy ofSctencu.

1 983: The Modulamy of Mind. MIT Press, Cambridge. Mass. 1 975: The enigmatiC eidetic image: A cntical examination of methods, data. and theones. Psychological Bullettn 82: 383-407. Hebb, D.O., 1 949: The Organizatton of &haviour. Wiley, New York. Hinton, G . E. and Anderson, J .A. (eds.), 1 98 1 : Parallel Models ofAssoctative Memory. Erlbaum,

Fodor, J .A . ,

G ray, C . R . and Gummerman. K.,

Hillsdale, N e w Jersey.

1984: Paru of recogmtion. Cognillon 1 8: 65-96. 1 986: Language Acquuition and The Theory of Parameters. D. Reidel. Dordrecht. Lenneberg, E. H . . 1 967: Biologtcal Foundations of Language. Wiley, New York.

Hoffman, D.O. and R1chards. W.A., Hyams. N . M . ,

Luna, A . R . ,

1968: The Mind of a Mnemonist.

Bas1c Books. New York.

1982: Vision. Freeman. San Francilco. Marshall. J.C., 1 977: Minds, machmes and metaphors. Soctal Studies of Sctence 7: 475-488. Marshal l , J . C . , 1 980a: The new organology. The Behavioral and Bratn Sciences 3: 23-25. Marshal l , J.C., 1 980b: On the biology of language acquisition. In: D. Caplan (ed.), Btological M arr, D . ,

Studtes of Mental Processes.

�IT Press , Cambridge, Mass.

1 98 1 : Hemisphenc specialization: what, how and why. The Behavioral and Bratn Sciences 4: 72-73. Marshall, J.C., 1 984: Multiple perspecuves on modularity. Cognition 1 7: 209-242. Minsky, M . , 1 986: The Soctety of Mind. MS. Pinker. S., 1 984: Language Ltanrability and Language D�e/opment. Harvard University Press,

Marshall, J.C.,

Cambridge, Mass .

1 986: Parallel Distributed Processing: Explorations in the

R u melhart, D . E . and McClelland. J . L . ,

Microstructure of Cognition.

Volume I:

Se1denberg. M.S. and Tanenhaus. M . K . ,

McGill Studies in Cognitive Scienct.

Foundations.

M I T Press, Cambridge, Mass.

1 986: Modularity and lexical access. In: i . Gopnik (ed.),

Ablex, Norwood, New Jersey.

Shallice, T., McLeod, P . and Lew1s, K.,

1 985: Isolating cognit1ve modules with the dual-task

paradigm: Are speech percept1on and production separate modu les?

Quattrly of Jounral of

Experimental Psychology 37A: 507-532. Stromeyer, C . F . and Psotka. J . ,

1970: The detailed texture of eidetic images. Nature 225:

346-349. Wanner, E. and Gleitman, L.R. (eds.),

1 982: Language A cquuition: The State of the Art.

Cambridge University Press. Cambridge.

198 1 for the independence of language and cognition: case study of a l:CU Working Papers in Cognmve Ltngumics 3 : ll l - 1 60. Zeki, S .. 1 980: The representation of colours in the cerebral cortex. Nature 284: 4 1 2-4 1 8 . Yamada, J . , Evidence

'hyperlingUisllc' adolescent.


Finkel, L . H . and Edelman. G . M . ,

loUJ711Jl of Semantics 5: 1 75- 1 76

SOME COMMENTS ON 'THE SOCIETY OF MIND' A.J. Sanford


This is a fantasia on how a distributed processor might support behaviour as complex as that of the human mind. It contains few references to the vast body of research on which it stands, be that computational, psychological, or neurO'logical. The basic scheme is that mentality is the result of subtle assemblages of astronomically numerous small, relatively dumb agents which deal with only some aspect of the world or of other agents. It is exciting, as I assume the author intends it to be. It is also to me somewhat irksome, since many parts of it are written in a thoroughly disjointed way. I wonder whether the form in which it is written is representative of some theory of how communication at some level might normally take place. The result is a selection of cameo questions, sometimes linked only loosely to each other, and sometimes only weakly dependent on the main theme of the mind as a set of distributed agents. The idea, traceable in part to some of Minsky's earlier work, is already being taken seriously by many cognitive scientists. Distributed processing, largely in parallel, does seem to be the way the brain must work. Here Minsky relates this to aspects of the nature of mind, and in so doing, will doubtless introduce the interested layman (and maybe not a few psychologists) to the idea that mind consists in the activities of numerous largely independent agencies. The core of the book is the theory of how the independent agents come to communicate with each other, how they exercise control over one another, and how they come to be grouped into various families. This is a major issue in the understanding of any distributed system, and Minsky freely mixes psychological observations with computational ones in pain ting h is picture. The sections on language vary greatly in detail, and do not address the issues which interest the psycholinguist in any systematic way. One thing which is loud and clear however is that Minsky sees language as directly influencing the mental states of the listener/reader. Language input is assumed to activate numerous agents at numerous levels. Part of the discussion brings in his earlier ideas about frames, now common currency amongst those who will admit that to understand how language is understood, even to the level of syntactic disambiguation , almost certainly requires accessing situation-specific knowledge. The criticisms of knowledge based accounts of language understanding, particularly when frames are invoked, are well-known. How can you ever demonstrate that someone is

1 76

D�pt. of Psychology Univ�rsity of Glasgow Glasgow. G I 2 8RT Scotland


using a frame, script, or other kind of knowledge structure? More to the point is that in his treatments of language understanding, Minsky supposes that in principle very many ways of looking at the significance of an utterance can be active at the same time. For instance, in the context of Charniak's well known story, the sentence Mary gives Jack the kite could be interpreted as a physical transfer, a transfer of possession, or a realisation of a gift-giving role, etc. Minsky suggests that all could co-exist. Once a realm has been activated, it continues (autonomously ?) to activate various agencies specific to itself. Quite clearly, this pulls the rug from beneath arguments often seen in the psycholinguistic literature that so-

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents