LINGUIST List 10.657

Wed May 5 1999

Review: Ishikawa: Network Theory of Reference

Editor for this issue: Andrew Carnie <carnielinguistlist.org>



What follows is another discussion note contributed to our Book Discussion Forum. We expect these discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in. If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for discussion." (This means that the publisher has sent us a review copy.) Then contact Andrew Carnie at carnielinguistlist.org

Directory

  • Anne Reboul, Review of Ishikawa

    Message 1: Review of Ishikawa

     Date: Tue, 4 May 1999 14:22:35 +0200 (MET DST)
    From: Anne Reboul <Anne.Reboulloria.fr>
    Subject: Review of Ishikawa


    Ishikawa, Kiyoshi, (1998) A Network Theory of Reference, Indiana University Linguistics Club Publications, Bloomington, 139 pages,

    Reviewed by Anne Reboul, ISC-CNRS, France

    Ishikawa's book is a revised version of his Ph.D. dissertation at the University of Tokyo. It addresses the subject of reference and outlines a variety of Situation Semantics (i.e. a non-model-theoretic semantics; see Barwise & Perry 1983) which aims to account for Donnellan's distinction between attributive and referential uses of NPs (see Donnellan 1966, 1968), abandonning the view according to which this distinction amounts to the distinction between general and singular propositions and renouncing the strong realism of Situation Semantics in favour of an epistemic and "cognitive" approach to reference.

    Preface

    In defiance of model-theoretic semantics which accounts for natural language meaning in terms of links between utterances and external reality, the aim of Ishikawa's monograph is to outline and defend an alternative account in terms of change in a cognitive agent's information state. On this view, reference amounts to the ability of an agent to link characters through different frames of individuation.

    1. Introduction

    Model-theoretic semantics sees meaning as a bridge between language and reality, reality being conceived in set-theoretic terms as individuals, properties and relations between individuals and being expressed through a mathematically formulated model. Semantics in this perspective is essentially truth-conditional and everything which is not belongs to pragmatics. This view of meaning strongly implies a link between language use and real individuals in the world through reference seen as introducing these individuals into the propositions expressed. According to Ishikawa, however, this view of reference is flawed in as much as it has more to do with ontology (a philosophical and metaphysical concern) than with cognition, information states and the communicative potential of language which he views as central to reference. He proposes to outline an alternative, non model-theoretic approach to reference, centered on Donnellan's distinction between referential and attributive uses of NPs, which would also account, in a compositional way, for attitude reports. Ishilawa then proceeds with the examination of different variations on models, purporting to show that failure of this or that version of model-semantics is not enough to justify discarding the very notion of a model-theoretic approach to meaning. Rather one should show that the division of labor between semantics and pragmatics is inefficient and that model-theoretic interpretation is not relevant to natural language semantics. Finally, he argues that model-theoretic interpretation is not equivalent to meaning, resting his case on the interpretation of deictics, ambiguous proper names and concluding that meaning is not so much content as a way to reach content.

    2. The Referential/Attributive Distinction

    Ishikawa begins his discussion of the distinction by outlining the (different) distinction between general or quantified propositions (no individual in the proposition) and singular propositions (individual(s) in the proposition), based on the Russell-Strawson discussion of definite descriptions. He then describes Donnellan's distinction between attributive and referential uses as a challenge to the descriptions of definite descriptions of both Russell and Strawson, Russell's analysis failing to account for the referential use, while Strawson's fails to account for the attributive use. He notes that, though Donnellan is not entirely explicit on whether his distinction is semantic or pragmatic in nature, there seems to be a general agreement on equating the attributive use of definite descriptions with general propositions, while the referential use would correspond to singular propositions. Under these terms the attributive/referential distinction would amount to the fact that a given NP is used referentially if a) its descriptive content is not part of what is asserted, b) it introduces a real individual in the proposition ; it is used attributively otherwise. He then examines several attempts to derive the distinction through either pragmatic means (a gricean approach in terms of word meaning and speaker's meaning) or semantic means (a scopal approach which, however, leads to the postulation of a covert operator or a situation-semantics approach in terms of convergence or divergence between the described situation and the resource situation). All of these however do not detract from the description of the referential/attributive distinction in terms of singular/general propositions.

    3. Scenarios for Definite Descriptions

    In this chapter, Ishikawa combines the description of Donnellan's distinction in terms of general/singular propositions with Russell's principle which is an epistemic requirement on singular propositions making the knowledge of which individual the proposition is about necessary to its understanding. His aim is to show through scenarios that this combination does not yield a convincing analysis of the referential/attributive distinction. To do this he describes four scenarios involving mainly the investigation by Holmes and Watson of Smith's murder, complete with the definite description "Smith's murderer", with revision of beliefs, etc. For reason of space, I refer the reader to Ishikawa's text. From these scenarios and their analysis, Ishikawa draws what he calls " lessons " : first, knowing who does not imply a referential use ; second, hearer's ignorance does not imply attributive use ; third, speaker's ignorance does not imply attributive use. Other lessons concern tests of the distinction : inclusion of " whoever it is " or of " must " do not imply attributive use.

    4. Further Scenarios

    Ishikawa proposes to reject the combination of the singular/general propositions distinction with Russell's principle and to keep only the singular/general distinction. He puts it to the test with further scenarios, implying not only definite descriptions but proper names. This leads him to the conclusion not only that the singular/general distinction fails to account for the referential/attributive distinction, but also that the direct reference account of proper names should be rejected in an analysis of natural language meaning.

    5. The network theory

    Chapter 5, which occupies the main part of the book, is where Ishikawa introduces and describes his own theory of reference, the Network Theory. His point of departure is a divorce between a theory of linguistic meaning (his choice) and a theory of linguistic meaning as linked to non-linguistic factors (model-theoretic semantics). The main problem which he sees himself as having to account for is the fact that a given NP can be understood in different ways. Ishikawa begins by outlining a few basic tenets of his theory, among which the notion of an individual as a bundle of identifying properties in different frames. Referring is thus not reference to an outside individual but reference to an internal bundle of properties which are believed to be properties of the same outside individual. His system is based on the notions of information state (which he sees as roughly equivalent to the notion of belief state), facts (properties of individuals in different situations) and frame of individuation (a group of fact). In a given frame of individuation (or episode), the individuals individuated are roles (which are relative to episodes : one role-one episode). In his view, individuals are uniformities (bundles of roles) across episodes in a given information state. The modelization of an agent's information state advocated by Ishikawa is done in terms of Situation Theory, a given information state being considered as a model, implying the relativity of truth to information states. There are five classes of propositions depending on whether they describe facts, episodes, links, existence or quantification. Fact-describing propositions imply situations (s) supporting relations, roles and polarities (i.e. truth-values). Episode-describing propositions imply an episode (EP) grouping fact-describing propositions. Link-describing propositions imply a link (LINK) between roles, creating equivalence classes (or nodes) and depending on the agent's belief that the roles all pertain to the same individual. Existence-describing propositions imply predicating existence (EXISTS) of a set of roles (i.e. an individual existing in external reality, according to the agent's belief), this set (or node) being obtained through LINK. Quantification-describing propositions imply a quantificational relation (Q) and types whose situations and individuals are roles. They are subject to the usual constraints on quantification. Side by side with the formal description of his system, Ishikawa introduces a graphical representation (in standard boxological form, with links represented through nodes and lines between specific roles). Ishikawa then outlines some differences (not trivial ones) between his Network Theory and standard Situation semantics. The main one is that whereas Situation Semantics is realistic in that it implies real individuals, Network Theory merely allows roles, that is perceptions of individuals relative to episodes, i.e. cognitive entities. Given that Ishikawa's theory is based on the idea of information states, it is natural to analyze linguistic interpretation in terms of accommodation as a change of information states (i.e. a change of models), which, in combination with a version of HPSG, yields a compositional semantics. Ishikawa gives a few examples of both lexical entries and semantic interpretations for definite and indefinite NPs. He points out that given that roles are not equivalent to real individuals, the old familiarity/novelty distinction for indefinite/definite descriptions get a rather different analysis allowing to account both for the novelty of the information without necessarily implying the introduction of a new individual for indefinites, while definite descriptions, though their descriptive content must be satisfied by the role in the pre-update information state, can be linked to a role which does not satisfy it in the post-update information state. Ishikawa then turns to the referential/attributive distinction and notes that Network Theory shares with Mental Spaces Theory and Situation Semantics the advantage that the descriptive content of a definite description and the predicate which is applied to it need not be satisfied in the same episode (situation or mental space). The difference among scenarios (described in chapters 3 and 4) can be accounted for through the presence, absence or change of links between roles in different episodes, thus accounting for the change of beliefs informally described or implied by the scenarios, or through the addition of new information in a given episode, while the EXISTS predication, together with negation, can account for sentences such as " Smith was not murdered ". On the whole, this amounts to a redescription of the referential/attributive distinction in terms of the convergence or divergence between the descriptive content of the NP and the predicate : if they are predicated of the same role, then the definite description is used attributively ; otherwise, it is used referentially. Ishikawa finally deals with " knowing who " or the " particular individual " conception. He remarks that it can be accounted for through the network-theoretic notions of nodes and links combined with the (intuitive) notion of trustworthiness. He notes that this is near to the causal analysis of direct reference through a historical chain. He then turns to plugging what he calls " gaps " between network theory and conventional semantics, i.e. quantification and model-theoretic interpretation. He begins with quantification, insisting that his main concern is to prevent scopal interaction between indexicals, proper names and quantificational expressions. He gives a quantified example (" Every woman likes John "), as well as the lexical entry for "every". His analysis of this example requires that every woman role in an episode r has a link with a role who likes John in episode s. Regarding model-theoretic interpretations, Ishikawa remarks that some hard-core model-theoretic semantics (which he thinks would be philosophers of language rather than linguists) could see Network Theory as comparable to DRT. This concern with outward reality, according to Ishikawa, pertains more to metaphysics than to linguistics. In his own terms (in bold characters in his book), " Semantics is not metaphysics " (95). 6. Attitude Reports As Ishikawa points out in the introduction to this chapter, Network Theory, being constructed on " mental terms " (97), should give a straightforward account of attitude reports (restricted here to belief reports, Ishikawa's conviction being that his account of belief reports can be extended to other attitudes reports). According to him, belief reports are four ways ambiguous, between speaker's description and reported speaker's description and between attributive and referential readings, yielding inner referential (reported speaker's description used in a referential way), inner attributive (reported speaker's description used in an attributive way), outer referential (speaker's description used in a referential way) and outer attributive (speaker's description used in an attributive way). To deal with belief reports, Ishikawa introduces a "bel(a,k)" relation (in which a is a real individual and k a time point), which is recursive in that it can itself include another belief episode "bel(x,k)". Unsurprisingly, belief is treated as construction and revision of links between roles in different episodes. Ishikawa then turns to the Twin Problem, which is the opposite of Kripke's London problem (see Kripke 1979). Here the agent believes, not, as does Pierre in Kripke's puzzle, that one individual is two individuals but that two actually different individuals are one and the same. Again, this is treated in terms of the presence or absence of links between roles in episodes. Finally, Ishikawa examines the necessity vs. possibility of identity through two examples, " Aristotle might not have been the teacher of Alexander the Great " and " Aristotle might not have been Aristotle ", which have been analyzed, notably by Kripke (1980), as being respectively true and necessarily false. In Ishikawa's theory, they come out respectively as true and false (not necessarily false), given that the predicative relation can be revised and that the equative relation implies LINK and that LINK is reflexive. 7. Other Applications Ishikawa, in chapter 7, applies Network Theory to cleft and pseudo-cleft constructions, to indices, to the binding of indices and attitudes, and to the construction of a computational epistemology. This chapter is very short and does not give detailed analyses. Briefly, Ishikawa proposes to analyze clefts in terms of different roles for the descriptive content of the NP and the relation described in the VP, while indices (from binding theory) could be seen as roles, and binding indices and attitudes would depend on a sophisticated analysis of the attitudinal verb. On computational epistemology, Ishikawa suggests that roles linking could be language dependant. Network Theory, however, does not take into account these discrepancies and could be viewed as an epistemologic theory of cognitive agents sharing the same temporal and cultural spaces and thus building the same links. This could be extended to phonology, given that sounds could be analyzed as roles and nodes. 8. Concluding Remarks In this final chapter, Ishikawa defends his analysis against Perry's realism, against model-theoretic semantics, and against attitude-based semantics. His arguments against the first two have already been described. Against attitude-based semantics, he points out that roles in Network Theory are modes of identification in the sense of attitude-based semantics. He, rather surprisingly, ends his book with a (admittedly limited) defense of realism, pointing out that Network Theory is a theory of cognitive manipulation of information, but has no ontological implication whatsoever.

    CRITICAL EVALUATION Ishikawa's book should certainly be read by anyone interested in both reference and attitudes. However, the putative reader should be warned that the book is difficult to follow for several reasons : first of all, it is rather badly organized: with some information given toward the middle of the book when which would be more useful towards the beginning; second, it assumes in the reader a rather thorough knowledge of both situation semantics, including its formal apparatus (no detailed information is given about Ishikawa's formalism and the reader is left to fend for him- or herself, returning to Barwise and Perry (1983) to get through the formalism - the same goes for the HPSG formalism); third, at the same time, the ideas defended are fairly simple (the notions of roles, linking, etc.) and could have been very clearly explained without resorting to the unnecessary obscurities which, from time to time, Ishikawa resorts to. Thus, the final impression is that this could have been a much better book had Ishikawa assimilated, for instance, the lesson of Situation Semantics (see Barwise & Perry 1983), which can be seen as a model of clarity and pedagogy, and had he relied less on his reader's complete and detailed knowledge of most of formal semantics for the past thirty to forty years. While I take it for granted that Ishikawa wanted to not only eschew model-theoretic semantics but also any type of semantics which has any commitment to reality outside of language, it is a puzzle that he chose as his anchor Situation Semantics (which, as he himself acknowledges, is a strongly realistic - though not model-theoretic - approach.) What's more, it is disturbing that his main hypothesis seems to be that language as a part of cognition should be studied entirely apart from reality: in fact, the most interesting question about cognition, language and reality is how it is that cognition and language are so powerful tools for dealing with reality, a question by the way which Barwise and Perry (1983) address in a sophisticated and detailed way and of which Ishikawa seems blessedly unaware. Finally, it is hard to see why a semantics based on mental representations (which seems to be what Ishikawa has in mind) could not also be used to relate linguistic use to reality through these mental representations, relying, for instance on Dretske's concept of information and on the notion of covariation(see Dretske 1981, 1995). There does not seem to be any contradiction in such a program, however difficult it may be, and it might be seen as a " more cognitive " approach to language use than Network Theory with its rejection of external reality as irrelevant to cognition. In other words, a cognitive approach to semantics can hardly deal only with linguistic phenomena.

    References Barwise, J. & Perry, J. (1983) : Situations and Attitudes, Cambridge, Mass., The MIT Press. Donnellan, K . (1966) : "Reference and definite descriptions", Philosophical Review 75, 281-304. Donnellan, K. (1968) : "Putting Humpty Dumpty together again", Philosophical Review 75, 203-215. Dretske, F. (1981), Knowledge and the flow of information, Cambridge, Mass., The MIT Press. Dretske, F. (1995), Naturalizing the Mind, Cambridge, Mass., The MIT Press. Kripke, S. (1979) : "A puzzle about belief", in A. Margalit (ed.) : Meaning and Use, Dordrecht, Reidel, 239-283. Kripke, S. (1980) : Naming and Necessity, Oxford, Basil Blackwell.

    Reviewer : Anne Reboul, Research Fellow at CNRS (National Center for Scientific Research), France. Ph.D. in Linguistics, Ph.D. in Philosophy, currently working at The Institute of Cognitive Sciences in Lyon, in the Linguistics team. Has written quite a few papers in French and English. Author of Rh\233torique et stylistique de la fiction (1992, Nancy, PUN), co-author of Dictionnaire Encyclop\233dique de Pragmatique (1994, Paris, Le Seuil. English translation in preparation for Basil Blackwell, Oxford), of La Pragmatique ajourd'hui (1998, Paris, Le Seuil) and of Pragmatique du discours (1998, Paris, Armand Colin). Has been working on reference for the past twelve years.

    Anne Reboul Institut des Sciences Cognitives-CNRS 67 bd Pinel 69675 Bron cedex France <reboulisc.cnrs.fr>