201 86 853KB
English Pages 237 [238] Year 2017
Sarah-Jane Conrad, Klaus Petrus (Eds.) Meaning, Context and Methodology
Mouton Series in Pragmatics
Editor Istvan Kecskes Editorial Board Reinhard Blutner (Universiteit van Amsterdam) N.J. Enfield (Max-Planck-Institute for Psycholinguistics) Raymond W. Gibbs (University of California, Santa Cruz) Laurence R. Horn (Yale University) Boaz Keysar (University of Chicago) Ferenc Kiefer (Hungarian Academy of Sciences) Lluís Payrató (University of Barcelona) François Recanati (Institut Jean-Nicod) John Searle (University of California, Berkeley) Deirdre Wilson (University College London)
Volume 19
Meaning, Context and Methodology Edited by Sarah-Jane Conrad Klaus Petrus
ISBN 978-1-5015-1217-9 e-ISBN (PDF) 978-1-5015-0432-7 e-ISBN (EPUB) 978-1-5015-0423-5 ISSN 1864-6409 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. © 2017 Walter de Gruyter, Inc., Boston/Berlin Typesetting: Compuscript Ltd., Shannon, Ireland Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com
Contents Sarah-Jane Conrad, Klaus Petrus Introduction 1
Part 1: The contextual turn and the case for reference, truth-conditions and meaning Christopher Gauker Open texture and schematicity as arguments for non-referential semantics 13 Kepa Korta and John Perry Full but not saturated: The myth of mandatory primary pragmatic processes 31 Silvan Imhof How to get lost in context: Searle on context, content and literal meaning 51 Nikola Kompa Meaning and interpretation
75
Anne Bezuidenhout The role of context in semantics: A Relevance Theory perspective
91
Part 2: The contextual turn and the case for language use Stefano Predelli Boo semantics: Radical nonfactualism and non truth‐conditional meaning 117 Damon Horowitz Metaphor and mercurial content
141
vi
Contents
Part 3: The contextual turn and the case of analysis Dominik Aeschbacher Context, two‐dimensional semantics and conceptual analysis
171
Dan Zeman The use of the Binding Argument in the debate about location
191
Štefan Riegelnik Slices of meaning: Levels of analysis and the unity of understanding Contributors Index
229
227
213
Sarah-Jane Conrad, Klaus Petrus
Introduction
1 The context thesis and its methodological impact Philosophers of language and linguistics barely defend the view any longer that solely the linguistic meaning of a sentence and the syntactic order of the words it contains determine the truth-evaluable content of an utterance. The opinion that context-dependence is a common and quite prevailing phenomenon which is not limited to demonstratives and indexicals has gained considerable ground over the past few years (e.g. Bianchi 2004; Preyer and Peter 2005, 2007; Recanati, Stojanovic, and Villanueva 2010; Szabò 2005). The context thesis, i.e. the thesis that quite a few or even all linguistic expressions are highly context-sensitive, is thus widely accepted nowadays. Different conclusions have been drawn from this insight, however. Some take it to show that the basic concept of any theory of meaning – the semantic content or what is said by an utterance, which is often equated with the truth-evaluable and hence propositional content of an uttered sentence – is actually a pragmatic concept. They support this claim by pointing to specific features of the communicative context that need to be taken into account in order to specify what a speaker said by an uttered sentence. The concept of what is said is thus pragmatically infected, and specifying truth-conditions can no longer be seen as an exclusively semantic matter. On the contrary, pragmatics is necessary to make the concept of what is said available at all, and its content constantly changes relative to the communicative context. This very fact is taken to challenge the traditional project of a semantic theory, because pragmatic aspects affect not only what a speaker meant by an uttered sentence – i.e. what he or she communicates, possibly by means of an implicature (Grice 1989) – but also the proposition expressed by the literal use of a sentence. As a consequence, the semantics‐pragmatics distinction is not as clear‐cut as it may seem, and two domains, traditionally separated from both a methodological and a conceptual point of view, appear to be closely intertwined. Well-known names often cited in relation to this view are, for instance, John Searle (1978), Charles Travis (1985), Dan Sperber and Deirdre Wilson (1986), or Robyn Carston (2002). Others, by contrast, say that this reaction to the context thesis is unduly exaggerated. Although they readily acknowledge that context determines the content of a truth-evaluable utterance in numerous cases, they hold that the crucial question is how context adopts this role, and whether it is possible to account for the manifold contextual influences within a semantic framework. DOI 10.1515/9781501504327-001
2
Sarah-Jane Conrad, Klaus Petrus
Given that one can provide a systematic analysis of the role context plays and adapt the semantic theory accordingly in order to account for the different types of contextual influences, it is still possible, they maintain, to define a semantic concept of what is said by an uttered sentence. This way, the semantics-pragmatics distinction remains intact, and, more importantly, it is still possible to defend a formal semantics despite the changes the context thesis entails for our conception of semantics. Different authors such as Isidora Stojanovic (2008), John MacFarlane (2009), Jeffrey King and Jason Stanley (2005), Max Kölbel (2002), or François Recanati (2007) and many others defend this view. The bone of contention between these two main positions is the question of how to conceive of context, that is, how to account for contextual influences in determining what is said by an uttered sentence. On the one hand, we find the claim that the influences context exerts are manifold and genuinely pragmatic in nature; on the other hand, it is assumed that context and the role it plays in determining what is said can be modelled within a semantic framework. Still if the context thesis is taken on board and one acknowledges that context plays a much greater role than traditionally assumed, this raises the question as to what changes this implies with regard to the traditional view of language and communication as well as with regard to the methodology used to approach both areas. It is exactly to this very point that the authors contributing to this anthology set out to provide some answers. All papers collected in this volume take the context thesis on board, opting either for a more pragmatic or a more semantic reading of what is said, and evaluating the methodological consequences that follow from it for a theory of language and communication. It is important to stress the methodological perspective, because only too frequently the dispute over context has fed the suspicion that the whole quarrel is merely about terminological aspects, thus bearing no substantial weight (Borg 2007; Travis 2006). Although some conceptual incongruities indeed do blur the view, most disagreements are actually the result of certain methodological decisions. These are often not made transparent and are further obscured by quarrels about differing intuitions about extensively discussed examples like Tipper is ready or Fido is shaggy and their truth-evaluable content (see e.g. Bach 2001; Borg 2004; Cappelen and Lepore 2005; Carston 2008). At times, these discussions seem to dominate the debate as a whole, because so far no criterion has been available to decide the issue for good (Lötscher and Conrad 2015). This proves to be particularly problematic when such rather superficial disputes override or even conceal more systematically anchored questions. Being aware of this, it is all the more worth giving special attention to the methodological aspects of the debate.
Introduction
3
It goes without saying that the papers assembled in this volume do not come to a unanimous conclusion about the consequences one must draw from the context thesis. On the contrary, the anthology collects a variety of stances, which stand in tension with each other and thus indicate that different lessons can be learnt from the context thesis. Nevertheless, the papers cohere insofar as they all challenge and elaborate on established notions and methods that pertain to meaning, context and analysis. By doing so, they show that the context thesis provokes a fundamental shift and invites us to reconsider core concepts of philosophy of language as well as the methods applied by the theories of language and communication. It is therefore no exaggeration to say that we find ourselves confronted with a Contextual Turn, which, in the wake of the Pragmatic Turn, deeply changes our view of the philosophy of language. The different papers collected in this anthology suggest questioning three, ultimately connected assumptions of the philosophy of language. The first assumption relates to the status of referential semantics and, from a methodological point of view, to its power to explain truth-conditional meaning. This assumption has come under attack by the context thesis. Connected with the first assumption and thus owing to the predominant status of referential semantics, the second assumption gives priority to assertive sentences when language use is under consideration. The question it raises is whether the context thesis changes our understanding of language use from a methodological point of view, and what conclusions need to be drawn from this. According to the third assumption, philosophical analysis amounts to nothing more than semantic or conceptual analysis, granted by the metaphysical and epistemological foundations provided by referential semantics. Since the context thesis threatens to undermine the project of semantic or conceptual analysis, the question is whether this type of analysis can still be defended, and if so, how. As the following overview shows, most of the authors address more than just one of the above-mentioned assumptions. They generally have one specific focus, however, and this allows a grouping of the different articles on whether they draw on reference, truth and meaning, or whether they foreground language use or semantic analysis.
2 The Contextual Turn and the case for reference, truth-conditions and meaning The context thesis ultimately attacks referential semantics in its traditional form as the bedrock of semantic theories. Referential semantics has been the
4
Sarah-Jane Conrad, Klaus Petrus
paramount approach to clarifying the concept of meaning for a long time, although it has been seriously challenged from various sides since its heyday (see e.g. Brandom 1994; Chomsky 2000; Wittgenstein 1953). All the same, referential semantics is still central in many theories of meaning because of the explanatory role it plays with regard to the truth -conditions of an uttered sentence: referential semantics allows connecting the concepts of meaning and truth by holding that the propositional or truth-evaluable content of an uttered sentence is ultimately related to reference and predication. The content expressed by an uttered sentence can be equated with what is said. If the truth-evaluable content is exclusively related to reference and predication, prima facie no pragmatic or communicative information is needed in order to determine the truth-conditional content of an uttered sentence. As a consequence, what is said and what is meant by an uttered sentence can be distinguished from a conceptual point of view and classified as being either semantic or pragmatic. Both assumptions together smoothly open the door to formal semantics. Given the pervasive nature context plays not only regarding demonstratives and indexicals, but rather regarding lexical items in general, one might ask if referential semantics loses its explanatory power and if it has to be abandoned altogether or at least so massively improved that it becomes capable of capturing the manifold forms of context-influences. We also need to ask if it makes sense to defend the concept of semantic content any longer, and, if we do so, how it can be preserved. The context thesis ultimately raises some fundamental questions about how to conceive best of meaning and how to interpret the relation between meaning and truth. It is these very questions that lie at the centre of attention of the papers by Christopher Gauker, Kepa Korta and John Perry, Silvan Imhof, Nikola Kompa and Anne Bezuidenhout. Christopher Gauker identifies two phenomena, namely open texture and schematicity, as two embarrassing indeterminacies for referential semantics. As regards the phenomenon of open texture, the author claims that there are specific terms for which neither the objects referred to nor the past uses determine whether or not they can be applied to a given object. Hence, we are in some sense confronted with a blank space in the relationship between the word under consideration and the objects it can refer to. The second indeterminacy relates to specific terms that can be used in many different ways. This suggests interpreting their meaning as being merely schematic. Gauker takes both phenomena as a stepping-stone to arguing in favour of a non-referential semantics. The methodological lesson to be learnt from the Contextual Turn is therefore that referential semantics as an explanatory role model for semantics needs to be abandoned. The pressing question then, of course, is if a non-referential semantics still provides a basis for defining truth and validity, and if the project
Introduction
5
of a formal semantics can still be saved – a question the author answers in the affirmative. Kepa Korta and John Perry hold that the lesson to be learnt from the Contextual Turn is that a useful concept of what is said needs to capture everything that enters the process of utterance interpretation on the part of the hearer. The objects of the analysis are therefore not uttered sentences, but utterances. Despite this fundamental shift in perspective from language to communication and interpretation required by the context thesis, Korta and Perry emphasise that some specific worries turn out to be ill-founded, among others the worry that the concept of what is said might not be available without taking genuinely pragmatic processes into account. This worry is fed by the assumption that saturation, also known as “completion”, belongs to the mandatory processes of utterance interpretation. According to the two authors, different levels of analysis need to be distinguished. As a consequence, it is possible to identify at one level a truthevaluable proposition that is not in need of any saturation. Silvan Imhof takes a close look at John Searle’s Speech Act Theory, a pragmatic theory of language that includes a strong context thesis. According to Searle, context is necessary to determine the truth-conditions, or more generally, the conditions of satisfaction of a speech act. Despite his strong context thesis, Searle also defends the view that the meaning of every speech act can be captured by a sentence that has a semantic content. The semantic content of a sentence is related to reference and predication, and thus seemingly not affected by context. The assumption that the semantic content is invariable and independent from context is vital to Searle’s Speech Act Theory for methodological reasons: the autonomy of the semantic content allows Searle to claim that the analysis of speech acts amounts to the analysis of the meaning of the sentence uttered to perform a specific speech act. Imhof’s reconstruction of Searle’s framework shows, however, that the semantic content is dissipated by the strong context thesis, which is why Searle loses the methodological foundation on which he has built his Speech Act Theory. Nikola Kompa also addresses the question of how the Contextual Turn affects the methodological status of type meaning as an explanatory feature of successful communication. Type meaning describes the conventionalised core meaning of a sentence that directs its use. The context thesis suggests, however, that pragmatic derivations are in any case necessary. Therefore type meaning is not sufficient to explain successful communication. Since the necessary pragmatic derivations vary from one context to another, type meaning seems to lose its explanatory power as a basis for language mastery and successful communication altogether. Kompa discusses possible strategies that aim to defend the theory of type meaning. She presents alternative ways of conceiving of it such
6
Sarah-Jane Conrad, Klaus Petrus
as Augustín Rayo’s grab-bag-model, but finds all of them wanting for different reasons. Eventually she concludes that the context thesis shows that any semantic theory needs empirical input, and accordingly suggests a change in perspective by introducing elements of cognitive science into the philosophy of language. Anne Bezuidenhout, too, addresses the question of what framework is most suitable to account for the different types of contextual influences. She also believes that there is a need for a major shift in paradigm from the philosophy of language to cognitive science, because a theory of meaning can simply not account for the understanding of everyday language use. Instead, we had better avail ourselves of a framework from cognitive science. Championing a Relevance Theory stance, Bezuidenhout suggests interpreting context psychologically, so that it refers to the mutual cognitive environment of the persons involved in a conversation. This is a position that can be aligned with a Chomskyan internalist view of language. Bezuidenhout contrasts this type of context with the conception of a formal context as suggested by Stefano Predelli (2005), and she demonstrates how cognitive environments are constrained, and how such constraints can smoothly enter the analysis of indexicals.
3 The Contextual Turn and the case for language use As mentioned in the introductory part, the second assumption of philosophy of language relates to language use. Specific forms of language use are given priority within the referential framework, namely those that belong to the class of assertive sentences. If, however, referential semantics loses its overriding explanatory role within semantics, we must ask how this affects our analysis of language as such. Stefano Predelli and Damon Horowitz both address this question. Stefano Predelli studies the case of expressive devices. For this purpose he develops a language fragment as part of a radical nonfactualist language – i.e. a language fragment that consists of non-descriptive terms. Predelli’s contribution to the study of a non-truth-conditional language shows that context-dependence, regarded with fear and reluctance by proponents of traditional semantic frameworks, actually provides a convenient stepping-stone to a better understanding of natural languages in general. Once the phenomenon of context-dependence is properly understood and incorporated in our semantic theory, this knowledge can be used to expand the semantic framework and integrate non-descriptive elements of natural languages, for instance expressive devices. This is another
Introduction
7
lesson to be learnt from the Contextual Turn: it broadens our view of how language and its use work. Damon Horowitz challenges the basic assumption shared by the Literalists and the Contextualists. Both of the opposing camps claim that truth‐conditions are objectively fixed and determinate. Literalists explain this by relating the truth-conditional content to the conventional meaning of the sentence uttered, whereas Contextualists assume that the content of an utterance is sensitive to context and must accordingly be determined not only by semantics, but also by pragmatics. Neither Contextualists nor Literalists, however, are able to explain the truth-conditional content of metaphors systematically, but relate this phenomenon exclusively to pragmatics. Horowitz shows that after shifting the framework towards what he calls Hermeneutic Contextualism, the truth‐conditional content of metaphors can be systematically determined. According to his Hermeneutic Contextualism, the truth-conditional content of metaphors essentially depends on the interpreter of an utterance and is thus closely tied to the subject and her interpretation.
4 The Contextual Turn and the case for analysis As stated by the third assumption related to philosophy of language, the context thesis questions the status of conceptual and semantic analysis fundamentally. In accepting the strong context thesis, we need to ask what objects are actually available for analysis. This problem is addressed by Dominik Aeschbacher, Dan Zeman, and Štefan Riegelnik, who provide different answers. Dominik Aeschbacher deals with the question of whether conceptual analysis can be defended as a method for doing philosophy given the context thesis. Conceptual analysis is based on the so-called “golden triangle” which relates modality, apriority and meaning. As these three aspects are connected, conceptual analysis simply uncovers knowledge that is a priori and necessary in nature, and associated with the meanings of the particular expressions used. In the light of the context thesis, however, the semantic assumptions underlying conceptual analysis become questionable, and the different types of context-sensitivity threaten the whole endeavour. Aeschbacher shows that a modified version of a two-dimensional semantic framework in the spirit of David Chalmers and Frank Jackson is quite capable of countering possible problems related to the context thesis. Thus conceptual analysis can be defended against contextual challenges.
8
Sarah-Jane Conrad, Klaus Petrus
Dan Zeman puts the methodological value of the Binding Argument as an argumentation scheme to the test and subverts its power. According to defenders of the Binding Argument, context-dependent sentences like It is raining must be given an indexicalist reading, which means that we are required to assume that the sentence has variables in its logical form. In consequence, alternative analyses of the sentence, no matter whether they are semantic or pragmatic, have to be refuted. The Binding Argument thus seems to provide a methodological bedrock argument that makes it possible to decide what type of semantic theory to favour. Zeman shows that the Binding Argument fails to make good on this promise, because it can be blocked at different stages and alternative explanations for the phenomena at stake can be given. As a consequence, neither is the debate between truth-conditional semantics and truth-conditional pragmatics finally settled nor are we required to opt for a specific type of truth-conditional semantics. In the last contribution of this volume, Štefan Riegelnik deals with one of the most fundamental questions brought up by the Contextual Turn, namely with the question of where to draw the demarcation line between semantics and pragmatics for establishing two distinct domains of analysis. Riegelnik claims that there is no way of settling the issue in principle. He argues that this has to do with the fact that the distinction is problematic in itself. If there are semantic as well as pragmatic elements in an utterance, the presumption of a semantic-pragmaticdivide calls for an explanation of the unity of utterances. In other words, we need to answer the question of how semantic and pragmatic parts relate so as to form a unified utterance. Without an answer to this question, it is problematic to uphold the semantics-pragmatics distinction and hence mistaken to study semantic or pragmatic properties in isolation. According to Riegelnik, this is the methodological upshot of the Contextual Turn. The papers gathered in this volume all render the methodological assumptions transparent that are challenged by the context thesis, and they point out how this attack could be countered. By doing so, they identify a number of problems that are entailed from a methodological point of view, and they show that different lessons can be learnt from the Contextual Turn regarding both semantics and the (semantic) framework they consider the most suitable to handle the context thesis. Although the question of whether the philosophy of language can still make a stand in the face of the context thesis is not answered unanimously, all the authors address some systematic as well as methodological shortcomings and put them into sharp relief against the context thesis. The work for this anthology was supported by the Swiss National Science Foundation (project No. 114812). Special thanks go to Adrian Häfliger, Nathalie Lötscher, and Jonas Pfister as well as Christopher C. Pfisterer for their philosophical and Jonna Truniger for her linguistic support.
Introduction
9
References Bach, Kent. 2001. You don’t say? Synthese 128 (1–2). 15–44. Bianchi, Claudia (ed.). 2004. The semantics/pragmatics distinction. Stanford: CSLI Publications. Borg, Emma. 2004. Minimal semantics. Oxford: Clarendon Press. Borg, Emma. 2007. Minimalism versus contextualism in semantics. In Gerhard Preyer & Georg Peter (eds.), Context-sensitivity and semantic minimalism, 546–571. Oxford: Clarendon Press. Brandom, Robert. 1994. Making it explicit. Cambridge: Harvard University Press. Cappelen, Herman & Ernie Lepore. 2005. Insensitive semantics. A defense of semantic minimalism and speech act pluralism. Oxford: Blackwell Publishing. Carston, Robyn. 2002. Thoughts and utterances. Oxford: Blackwell Publishing. Carston, Robyn. 2008. Linguistic communication and the semantics/pragmatics distinction. Synthese 165 (3). 321–345. Chomsky, Noam. 2000. New horizons in the study of language and mind. Cambridge: Cambridge University Press. Grice, Herbert Paul. 1989. Studies in the way of words. Cambridge: Harvard University Press. King, Jeffrey & Jason Stanley. 2005. Semantics, pragmatics, and the role of semantic content. In Zoltan G. Szabò (ed.), Semantics versus pragmatics, 111–164. Oxford: Oxford University Press. Kölbel, Max. 2002 Truth without objectivity. London: Routledge. Lötscher, Nathalie & Sarah-Jane Conrad. 2015. Minimalismus, Kontextualismus, Relativismus. In Nikola Kompa (ed.), Handbuch Sprachphilosophie, 279–288. Stuttgart: Metzler. MacFarlane, John. 2009. Nonindexical Contextualism. Synthese 166 (2). 231–250. Preyer, Gerhard & Georg Peter (eds.). 2005. Contextualism in philosophy. Oxford: Clarendon Press. Preyer, Gerhard & Georg Peter (eds.). 2007. Context-sensitivity and semantic minimalism. Oxford: Clarendon Press. Recanati, François. 2007. Perspectival thought: A plea for (moderate) relativism. Oxford: Oxford University Press. Recanati, François, Isidora Stojanovic & Neftalí Villanueva (eds.). 2010. Context-dependence, Perspective and relativity. Berlin: De Gruyter Mouton. Searle, John. 1978. Literal meaning. Erkenntnis 13 (1). 207–224. Sperber, Dan & Deirdre Wilson. 1986. Relevance: Communication and cognition. Oxford: Blackwell Publishing. Stojanovic, Isidora. 2008. What is said: An inquiry into reference, meaning and content. Saarbrücken: VDM Verlag. Szabò, Zoltan G. (ed.), Semantics versus pragmatics. Oxford: Oxford University Press. Travis, Charles. 1985. On what is strictly speaking true. Canadian Journal of Philosophy 15 (2). 187–229. Travis, Charles. 2006. Insensitive semantics. Mind & Language 21 (1). 39–49. Wittgenstein, Ludwig. 1953. Philosophische Untersuchungen. Frankfurt am Main: Suhrkamp.
Part 1: The contextual turn and the case for reference, truth-conditions and meaning
Christopher Gauker
Open texture and schematicity as arguments for non-referential semantics Abstract: Many of the terms of our language, such as “jar”, are open-textured in the sense that their applicability to novel objects is not entirely determined by their past usage. Many others, such as the verbs “use” and “have”, are schematic in the sense that they have only a very general meaning although on any particular occasion of use they denote some more particular relation. The phenomena of open texture and schematicity constitute a sharp challenge to referential semantics, which assumes that every non-logical term has a definite extension. A different, non-referential approach to formal semantics defines truth as relative to a context and defines contexts as built up from exclusively linguistic entities. For any given utterance of a sentence, there will be one of these contexts that pertains to it. In this framework, open texture and schematicity can be understood as consequences of the complex nature of the pertaining relation between contexts and utterances.
1 Introduction According to a commonplace approach to semantic theory, every basic lexical item of a language has an extension. The extension of a singular term is an object. The extension of a one-place predicate is a function from objects to truth values. The extension of a two-place predicate is a function that takes an object as input and yields as output a function from objects to truth values. In other cases, the extension may be a function that takes some kind of function as its input and yields some kind of function as its output.1 The extensions of compound subsentential expressions, such as verb phrases, and the truth values of sentences, are somehow defined on the basis of the extensions of the basic lexical items. For instance, “Socrates is snub-nosed” is a true sentence if and only if, when the extension of “Socrates” (namely, the man Socrates) is the input to the function that is the extension of “is snub‐nosed”, the output is truth. Inasmuch as “reference” can refer to the relation between a predicate and its extension, as well
1 For an introduction to this way of thinking about semantic theory, see Heim and Kratzer’s textbook (1998). DOI 10.1515/9781501504327-002
14
Christopher Gauker
as to the relation between a name and its referent, it is commonplace to call this approach to semantics referential semantics. Although I said that the extension of a one-place predicate is a function from objects to truth values, I will, for simplicity, often conflate this function with the set of objects which, when the function is applied to them, yield the value true. Although I said that the extension of a two-place predicate is a function from objects to functions from objects to truth values, I will conflate this function with the set of pairs such that the first member yields a function that, when applied to the second member, yields the value truth. On this way of talking, “Socrates is snub-nosed” is true if and only if Socrates is a member of the set of snub-nosed things. An embarrassment for referential semantics is that a lot of terms do not seem to have any definite extension. In particular, there are two kinds of indeterminacy that should concern us. The first kind is what I will call open texture. To say that a term is open textured in my sense is to say that whether it applies to a thing is not always determined by the nature of that thing and past usage of the term. For example, the term “jar” is open textured because whether a given container belongs to the extension may be something that has not yet been decided. There may be a kind of glass vessel that could equally well be called a “jar” or a “bottle”. But if a soft drink manufacturer starts selling large quantities of a soft drink in such a vessel and in advertisements describes the beverage as packaged in a “jar”, then no one could object to the claim that “Choco‐cooler is sold in jars”.2 The second kind of embarrassing indeterminacy is what I will call schematicity. There are many verbs that seem to tell us very little but which we use a lot. Consider, for instance, the verb “to use”. The eater uses a knife, the light bulb uses electricity, the bird uses the air (to fly on), the viewer uses the television (for entertainment), the moon uses the sun (to reflect light). A true utterance of a sentence of the form “X uses Y” does not, in virtue of a meaning that that use of “use” shares with every other use, tell us much of anything about the relation between X and Y. And yet, we will usually take an utterance of a sentence of that form to be telling us something more. My first task in this paper will be to say something about the expectations that drive us toward referential semantics. Next, I will try harder to show that the kinds of indeterminacies that I have identified above really constitute a decisive reason to entirely abandon referential semantics. Finally, I will introduce my
2 Ludlow (2014) also appeals to phenomena of open texture to motivate some doubts about referential semantics.
Open texture and schematicity as arguments for non-referential semantics
15
alternative. The alternative will be a kind of non-referential semantics. The composition of semantic values will be handled by means of a recursive definition of truth in a context. For this purpose, contexts will be a kind of formal structure built up from linguistic entities. As I will explain, the job of grounding truth in the structure of the world will take the form of explaining the conditions under which a context pertains to a particular utterance. A variant on referential semantics takes the extensions of predicates to be not sets but properties and relations, conceived of as something other than sets. In this paper I will be concerned with the version of referential semantics that takes the extensions of predicates to be sets (or, more precisely, functions). The problems of open texture and schematicity are equally problems for the version that takes the extensions of predicates to be properties and relations, but I will not take the space to prove it. There are other aspects of natural language lexical semantics that we might struggle to incorporate into the framework of a referential semantics. These include: –– Indexicality: The reference of words like “this”, “you” and “now” may vary from situation to situation, and two or more occurrences of any one of them may refer to different things even in a single situation. –– Intensionality: In “Dottie is waiting for Santa Claus”, the extension of “is waiting for” cannot very well be a relation between existing objects. –– Polysemy: The extension of a word like “book” might shift from a set of concrete particulars to a set of abstract creations (Pustejovsky 1998: 28; Asher 2011: 89). –– Vagueness: There is no very sharp dividing line between a movie that counts as “long” and one that counts as “not very long”, or between one that counts as “exciting” and one that counts as “unexciting”. –– Variable types: In “She began to read a book”, the object of “begin” is an event, the reading, but in “She began a book”, the object is not an event (cf. Pustejovsky 1998: 115; Asher 2011 14–18). –– Permissible syntactic settings: We can say both “Mary likes reading books” and “Mary enjoys reading books”, but not both “Mary likes to read books” (which is fine) and “Mary enjoys to read books” (which is not fine) (cf. Pustejovsky 1998: 135; Asher 2011: 90). Each of these phenomena has been addressed in various ways in the literature while adhering to the basic paradigms of referential semantics. I do not assume that they have all been addressed successfully, but in this paper I will not attempt to discredit referential semantics by appeal to them.
16
Christopher Gauker
2 Desiderata for a semantic theory Among the many tasks for a semantic theory, there are two that a referential semantics seems especially well suited to perform. First, referential semantics seems to provide a means to define the conditions under which a sentence is true. Second, referential semantics seems to provide a means to define logical properties, such as the logical validity of arguments. As I said at the start, referential semantics aims to define the truth conditions of sentences of the language in terms of the extensions of the terms from which they are composed. Actually, this statement has to be modified to accommodate the context-variability of sentence truth. For instance, the truth of the sentence “Everyone is present” will vary with a contextually determined domain of discourse. The truth of “Dumbo is small” will vary with a contextually determined standard of size. The truth of “Tipper is ready” will vary with a contextually determined activity for which Tipper may or may not be ready. What we can expect of a sentence, then, is not that it be true simpliciter but only that it be true relative to a context. A context may be defined as a structure that provides values to parameters that we think of as variable across different situations. For example, domain of discourse and standard of size are two such parameters. The sentences of a natural language will generally have truth values only relative to a context in this sense. However, particular utterances or inscriptions of sentences may have truth values absolutely. There will usually be a single context that we can say pertains to the utterance. In terms of the pertaining relation, we can explicate the truth value of utterances in terms of the truth values of sentences relative to contexts, for the truth value of an utterance can be identified with the truth value of the sentence uttered relative to the context that pertains to it. Underlying the expectation that we can define truth relative to a context in the manner of referential semantics is a more general expectation that we can put in its place if need be. When we speak, we expect our words to satisfy certain norms – unless there is a reason to violate them. What we say should be relevant, informative, well-organized, concise, polite, etc. Among these expectations is that our words be adequately responsive to the way the world is. One kind of responsiveness to the way the world really is is the way that we might characterize as that of being true in the manner that referential semantics purports to define. But if it turns out that there are intractable problems with referential semantics, then there is a fallback position, namely, that our semantic theory should define somehow the way in which what we say should be adequately responsive to the way the world is.
Open texture and schematicity as arguments for non-referential semantics
17
The other desideratum that a referential semantics seems to satisfy is the need to define logical properties, such as the logical validity of arguments. From the starting point of referential semantics, we may arrive at definition of logical validity, by generalizing our definition of truth in a context to a definition of truth in a model and a context for that model. A model may be defined as a structure that includes a set of objects, the universe for the model, and an assignment of an extension to each nonlogical constant. An assignment may be conceived as an alternative reference relation. To accommodate contextrelativity, we might build into each model a set of contexts as well. Then we can say that an argument (a sequence of sentences, the last of which is the conclusion) is logically valid if and only if, for each model and each context in the set of contexts for the model, if the premises are true in that context in that model, then so is the conclusion.
3 Open texture In his 1945 contribution to a session of the Aristotelian Society, Friedrich Waismann introduced the term “open texture” into the philosophical lexicon (Waismann 1945). Waismann meant to criticize the idea that all empirically meaningful concepts can be defined in terms of sense data. He had a couple of objections, but one of them was that we can imagine situations in which we would not know whether to apply a term or not and that would challenge the definitions we had entertained up to that point. For example, we might think we could define “gold” in terms of the emission spectrum of gold, but we cannot rule out the possibility of a discovery that would force us to modify our definition (Waismann 1945: 122–123). Consequently, for many terms, even for a term of science such as “gold”, we cannot through lexical fiat decide in advance how we will use the term no matter what happens. This indeterminacy in the definition of a term is what Waismann called open texture. Referential semantics does not commit the error that Waismann was concerned to criticize. We can say that the extension of “chair” is the set of all chairs without committing ourselves to being able to precisely define “chair” using more basic terms. The referential semanticist is committed only to there being a definite set of things that is the extension of “chair”. There can be such a set though we have no means of specifying its membership and in no sense know which particular objects do in fact belong to it. Still, an analogous problem, which I will also call the problem of open texture, confronts referential semantics. I will say that a term F truly applies to an object o
18
Christopher Gauker
in a situation s if and only a sentence of the form “t is F” is true relative to the context that pertains to s, where t is used to refer to o. I will say that a term has open texture if and only if the true applicability of a term to an object is not always determined by the nature of the object together with the patterns that can be abstracted from our prior true applications of the term or from the past thoughts of those who have used the term. The problem of open texture is the fact that some terms do have open texture in this sense. That terms have open texture in this sense is also one of the lessons one might take away from Wittgenstein’s discussion of rule-following in the Investigations (1953). A different, more radical conclusion that one might also draw from that discussion of rule-following is that the applicability of a term is never determined by the patterns of use evident in past uses. (That appears to be how Kripke understands Wittgenstein in Kripke’s book on Wittgenstein (1982).) But here I am not committing myself to anything so strong. We may grant that past usage determines that the “+” sign denotes specifically the addition function. We may grant that the truth of an application of “cat” to a typical domestic cat and the falsehood of an application of it to a dandelion is strictly determined by past usage (and the nature of the cat and the dandelion). Still, in certain other cases, apart from some lexical fiat, there may be nothing in past usage that decides whether or not a term is truly applicable to a given object. A study by Malt, Sloman, Gennari, Shi and Yuan (1999) investigated the similarity of container classifications between speakers of American English, speakers of Argentinean Spanish and speakers of Chinese (in Shanghai). They found little correlation between the groupings imposed by the three languages. That is, the fact that two containers received the same label in one language did not make it likely that those same two containers would receive the same label in another language. Moreover, in each language the extensions of the container words did not well reflect the groupings that subjects produced when asked to sort containers on the basis of “physical”, “functional” or “overall” similarity. Most importantly for our purposes, when Malt et al. constructed a multidimensional scaling representing the subjective similarities between the various containers, it turned out that the regions occupied by the extensions of the several container words were not disjoint and the region occupied by the extension of a given container word was often not convex. Malt et al. speculate that the extension of a container word may be determined by a variety of processes that have little to do with the overall unity of the category. For example, in a process that Malt et al. call “chaining”, object a might be called a “juice box” because it is a cardboard carton containing fruit juice; a cone-shaped cardboard carton b might be called a “juice box” because, like a, it
Open texture and schematicity as arguments for non-referential semantics
19
is designed to contain a single serving of juice sipped through a straw; a coneshaped plastic container might be called a “juice box” because it is similar in function to c; and so on, to the point where a plastic bottle in the shape of Mickey Mouse is also called a “juice box”. For another example (my own, not theirs), our labeling practices may be driven by marketing considerations. Manufacturers of ice cream might prefer to sell their product in “cartons”, rather than “boxes”, just because we tend to think of what we call “boxes” as rather dirty. So the decision to label a certain container as a “jar” rather than as a “bottle” may be driven not by a difference in the meanings of the two words but rather by a similarity to particular exemplars, whether or not these are paradigmatic of the kind, and the selection of exemplars may depend on such things as marketing considerations. The container that holds the newly invented choco-cooler is similar enough to some things that have already been called “jar” that it too could be called a jar. It is also dissimilar enough from typical jars – the mouth is a bit too narrow compared to the body – that one could also call it a “bottle” instead of a “jar”. But if the manufacturer, in its advertising, describes the container as a “jar”, not as a “bottle”, and manages to get that usage accepted by consumers, then that’s what that container is – a jar. This same phenomenon of open texture affects terms for natural kinds such as “cat” as well as artifact kind terms such as “jar”. Consider, for instance, the (fictional) case of the Borneo weasel. Deep in the forests of Borneo, English- speaking explorers come upon a hitherto unknown creature. It looks very much like a weasel and in many ways acts like one. But genetic testing reveals that, actually, its genetic code is precisely that which had been taken to characterize the domestic cat (Felis catus). Its physical and behavioral features are in many ways weasel-like but in some ways cat-like. Its claws retract, but not as far as those of a typical cat. A difference in some neglected portion of a chromosome presumably accounts for the weasel‐like appearance. Thus, the explorers face this question: Shall we call it a “cat”? The Borneo weasel poses a challenge to lexicography, because we have various criteria for something’s being a cat that have never come into conflict in this way before. On the one hand, a thing is a cat if and only it looks and acts like a typical domestic cat. On the other hand, a thing is a cat if and only if it has the genetic constitution of a cat. By the first criterion, the Borneo weasel is not a cat. By the second, it is a cat. It is not reasonable to insist that the correct classification is determined by prior usage or by the thoughts of people who have used the word “cat”. In using the term “cat”, they may never have had any thoughts that would entail a decision about the Borneo weasel. If we assume that “cat” is a so-called natural kind term, then we will say that the genetic characterization has to take precedence over the phenomenological characterization. But we could
20
Christopher Gauker
modify the genetic characterization to exclude the Borneo weasel. And whether “cat” is a natural kind term in this sense is one of the questions that cannot be decisively answered on the basis of past usage or past thinking. Whether the Borneo weasel ends up being labeled a “cat” or not may depend on accidental features of the situation, by which I mean features of the situation that do not seem to provide an objective basis for judging the real extension of the word. If the discovery had first been reported in a popular magazine, the headline might read, “Bizarrely in Borneo: Cat looks like weasel!”, and in that case the label “cat” might stick. But if the discovery had first been reported in a scientific journal, a scientific classification might be introduced, adding perhaps a new subspecies to the genus Felis, and the term “cat” might simply not come up, in which case other terminology might take hold for describing this animal, leaving “cat” for application to members of the genus Felis that look more like a typical cat. Let us call these situations, where something happens that forces us to decide whether to apply a term to an object, though nothing in past usage or past thoughts decides the matter, a lexicographical choice point for the term. In the case of “jar” the question is which other things that have been called “jar” should serve as precedents and which similarities should count. In the case of “cat”, as I have imagined it, the question is which of two criteria that have always been satisfied jointly will guide our usage in a case where only one of the two criteria is satisfied. Terms with open texture are those that may face such lexicographical choice points. Open texture should not be confused with vagueness. As lexical choice points for a word are encountered and resolved one way or another, the questions they pose may be permanently decided and the meaning of a term is developed and enriched. By contrast, the vagueness of a term is a lasting feature of its meaning and does not indicate a similar place for enrichment. Likewise, open texture should not be confused with context-relativity as this is usually understood in referential semantics. The idea is not that the extension of a term with open texture may vary from one situation of utterance to another; rather, the decisions that are made at lexicographical choice points may govern all usage from that point forward. Can we plausibly maintain that, all along, even before the decision to call the choco-cooler container a “jar”, the term “jar” had an extension that either included the choco-cooler container, or that all along it had an extension that excluded the choco-cooler container (so that the decision to call it a “jar” was a decision to change the extension)? In the first of these cases, there would have had to be something about past usage and past thoughts that made it the case that the extension of “jar” included the choco-cooler container. In the second case, something about past usage and past thoughts made it the case that the
Open texture and schematicity as arguments for non-referential semantics
21
extension of “jar” excluded the choco-cooler extension. But on the contrary, in deciding whether to call the choco-cooler container a jar, we face a lexicographical choice point, not determined by past usage or thoughts. If we want to say that even before we encountered the Borneo weasel, the term “cat” had an extension that either excluded or included the Borneo weasel, then there are four options. The extension of “cat” consists of exactly those things that satisfy both criteria (appearance and genotype); exactly those things that satisfy either one or the other criterion; exactly those things that satisfy the genotype criterion; or exactly those things that satisfy the appearance criterion. For any of these options however, we should expect that there is something about past usage or the past thoughts of English speakers that makes it the case that exactly that class of things is the extension of “cat”. On the contrary, since nothing has ever been encountered that satisfies one of the criteria and not the other, and no one has ever thought about whether we should call something a “cat” if it satisfies one but not the other, there is no basis for deciding between these four options. Rather, we face a lexicographical choice point, at which we have to decide (deliberately or not) which of the two criteria our future usage will conform to. So we cannot maintain that prior to the discovery of the Borneo weasel, the extension of “cat” included it or that the extension of “cat” excluded it. Could we possibly deny my assumption that the extension of “cat” must somehow be determined by past usage or past thoughts of English speakers? Perhaps we could deny it by taking seriously the reality of future events. There has always been a fact of the matter about what would happen when English speakers first encountered the Borneo weasel even if, in some sense, something else could easily have happened. Suppose that in fact what would happen was that they would withhold the term “cat” from the Borneo weasel; then all along those animals have not belonged to the extension of “cat”. The reason the explorers took that decision was not that they had reason to think the extension of “cat” had all along excluded the Borneo weasel; they did not. They might just as well have decided to apply the term “cat” to the Borneo weasel. So yes, that decision was arbitrary in a certain sense. Their decision resolved what was a lexicographical choice point in the sense that past usage and past thoughts did not determine what they would or should say. Still, we can say that the extension of “cat” had always been that which it was taken to be at the point at which the arbitrary decision was made. This position cannot be sustained. On the one hand, it is supposed, the Borneo weasel does not, and never has, belonged to the extension of “cat”, because what we will in fact decide, when we first encounter it (deliberately or not) is to not call it a “cat”. Since it does not and never has belonged to the extension of “cat”, an utterance of “It’s a cat”, in reference to the Borneo
22
Christopher Gauker
weasel, would not be true. There may be cases where asserting that something is so makes it the case that it is so (e.g., that two people are married), or even cases where asserting that something is so makes it the case that it always has been so (although I cannot think of any examples). But surely it is not the case that asserting that the Borneo weasel is not a cat is what makes it the case that all along it has not been a cat. Could we perhaps say that until a lexicographical choice point for a term is surmounted, the extension of the term is, to some extent, indeterminate, and that resolving a lexicographical choice point makes the extension of the term to that extent more determinate? Before the explorers encountered the Borneo weasel, it was just indeterminate whether the Borneo weasel belonged to the extension of “cat”, and afterward it either determinately belonged or determinately did not belong. The problem with this approach is that it exacts a revenge. If it is indeterminate whether the Borneo weasel belongs to the extension of “cat”, then that means that the Borneo weasel does not belong to the extension of “cat”. But, as before, that is something we have no basis for saying. The story of the Borneo weasel is a made-up example, but I take for granted that it is representative of what has sometimes actually happened. More realistic examples might include fish, mass, water, mercury, slime mold, marriage and death. Liston (2010) poses a challenge to referential semantics similar to mine by means of the history of “rabbit”, which Old World settlers applied to New World hares. Even a word like “chair” could face a lexicographical choice point, if an ingenious furniture designer designed a sufficiently challenging case. To say that nothing in past usage or past thoughts decides which way to go with such terms is not to say that nothing important hangs on our decisions. Terms such as “person”, “marriage” and “death” can present lexicographical choice points that stand at the center of contentious social policy debates. But the moral questions decided by means of such debates do not necessitate any particular resolution of the lexicographical choice points. For instance, we could decide that a body kept alive in vegetative state does not really have to be dead before we can harvest its organs for transplants. The conclusion I draw is that referential semantics cannot accommodate the open-texture of many of the terms of our language. Referential semantics says that each non-logical term has a definite extension. But in many cases, there can be no possible basis for deciding whether a given object belongs to the extension or not. Even given the facts concerning the past usage of the term and the thoughts of the users of the term, no supposition about the object’s properties warrants any conclusion about whether it belongs. So referential semantics faces an intractable problem and is the wrong way to formulate semantic theories.
Open texture and schematicity as arguments for non-referential semantics
23
4 Schematicity The English language, as well as presumably every other natural language, contains many verbs of which we might like to say that their meaning is very schematic. An example is the verb “to use”. We may truly say that a person uses a knife, a commuter uses the subway, a light bulb uses electricity, a bird uses the air (to fly), the moon uses the sun (to reflect light), and so on. Other such schematic verbs of English include: “have”, “do”, “make” and “get”. (Compare “faire” in French, “machen” in German, “Κάνω” in Modern Greek, and “하다” in Korean.) Words in other categories, such as adjectives (e.g., “open”), may be schematic as well, but here I will focus just on verbs. The fact that the meanings of verbs may be schematic in this way poses a challenge to referential semantics. Apparently, all a referentialist can say about the extension of “to use” is just that it encompasses a great deal. It includes every pair of things such that the first in any way uses the second. (Of course, using is something that happens at a time. I will not worry just now about how to acknowledge that fact in our semantics.) But we cannot leave the matter there, because this answer leaves unexplained the fact that particular utterances of sentences containing “use” will usually communicate something more specific. If someone asks how John gets to work, and the answer is, “John uses the subway”, then we learn that John rides to a location near his workplace in the subway train. Or suppose we ask how John paid his restaurant bill and the answer is, “He used cash”. If “to use” just has the very broad extension, we would have to count this utterance as true even if what actually happened was that John handed the waiter his credit card wrapped in a dollar bill. Unless we can explain in what way a particular occurrence of “uses” refers to some more particular relation, referential semantics will not perform its appointed role in defining our obligation to make sure our words are adequately responsive to the way the world is, as I put it earlier. In answer to this challenge to explain what “uses” means on any particular occasion of use, one might reply that, while the meaning of “uses” is just very general, on any particular occasion of use the speaker will have some more specific relation in mind and the more particular relation that the speaker communicates by saying “uses” is only that which he or she happens to have in mind. If we say that someone used a knife, then typically we will have in mind that the person cut something with it. If we say that the commuter uses the subway, then we will have in mind that the commuter rides in the subway train. François Recanati has introduced the term modulation to describe, among other things, this process by which an interpreter identifies a more particular relation to take as the meaning of a schematic verb on a particular occasion of use (2004: 134). On his view, the
24
Christopher Gauker
correct interpretation is what it would be reasonable for the interpreter to take the speaker to have intended (Recanati 2013: 82). This answer in effect rescues referential semantics by telling us how, on any particular occasion of use, a particular extension attaches to a word. One objection to this answer is that it ignores the fact that the primary basis for our knowledge of what other people have in mind is what they tell us. If from the fact that the speaker says, “John uses the subway”, we learn that the speaker believes that John rides in the subway train, then that’s because there is something about his use of those words that allows us to make the inference. We can infer that the speaker believes that John rides the subway to work, because we interpret the speaker’s words as meaning that John rides the subway to work. So we need to be able to say what makes it the case that the speaker’s words have that meaning on the occasion of his or her uttering them other than that that is the meaning that we can reasonably assume the speaker to have had in mind. Another objection to this answer is that it assumes that thoughts cannot be schematic in just the way that spoken words can be. If mental representations can be schematic as well, though on the occasion of any particular tokening they have a more specific meaning, then obviously we cannot equate that more specific meaning of a thought with the meaning of some other thought that underlies it (Gauker 2012). A second sort of answer would be to say that the meaning of “uses” is context-dependent. Somehow, the meaning of “uses” and a context together determine the more specific relation, such as cuts or rides, that an utterance of “uses” will denote in any situation to which the context pertains. However, this appeal to context-relativity in the defense of referential semantics is selfdefeating. Context-relativity can legitimately play a role in referential semantics in perhaps two ways. First, it plays a role inasmuch as the truth value of an utterance is supposed to depend on something more than the reference of the referring terms and the meaning of the logical terms and particles (such as case markers) that compose the sentence uttered; it depends as well on the content of the pertinent context. For instance, the context may determine a domain of discourse or a standard of size. Second, context-relativity may play a role inasmuch as the reference of an indexical can be constrained in a rule-governed way by the context. For instance, there might be a defeasible rule according to which “now” has to be a period of time including the moment of speaking.3 If we now extend the notion of context-relativity to such an extent that even the extension of basic
3 In saying this, I am simply bowing to convention. In fact I am skeptical about this second purported role for context (cf. Gauker 2015).
Open texture and schematicity as arguments for non-referential semantics
25
nouns and verbs is relative to context, then it is not so clear that this is a way of defending referential semantics rather than a way of depriving it of all content. If for each token of a so-called referring expression, we now have to consider what it is about the circumstances of tokening that determines that the pertinent context assigns this extension rather than that one, then it is not clear that the appeal to extension serves an explanatory purpose. Rather, the truth value of an utterance will be explained somehow by appeal to these circumstances and their relation to the utterance. A third possibility (in addition to the idea that “uses” is just very general and the idea that “use” is context‐relative) is to say that a verb like “uses” is a kind of pro-verb. Much as a pronoun may pick up the reference of a prior proper noun or definite description, so too a pro-verb, such as “uses” may pick up the reference of a prior verb.4 One problem with this is just it does not seem to be the case that every understandable use of “uses” is preceded by the use of some other verb whose reference is picked up by that use of “uses”. When not, it might be said, “uses” has a deictic meaning, just as a pronoun without a prior anchor in overt speech might refer to an object deictically. The problem with this is that it is just not clear what it could mean to speak of deictic reference to a relation, as opposed to deictic reference to an object. A fourth possibility is to say that “use” is just ambiguous. The problem with this answer is that if “uses” is ambiguous, it is in at most finitely many ways ambiguous. We cannot write a recursive semantic theory for a language containing a term that is in infinitely many ways ambiguous. So for each of the different meanings of “uses”, we should be able to find, or invent, a synonym that has exactly that meaning and no other. For example, the many meanings of “uses” might include the meanings of “cuts”, “rides in”, “drives”, “is powered by”, and so on. But that just goes to show that positing ambiguities does not address the challenge, because each of these synonyms is schematic as well, though to a lesser degree (Searle 1983).
5 A non-referential semantics So far, I have been playing along with the assumption that contexts are structures that, so to speak, supplement models. A context supplies a domain of discourse, which is a set of objects, relative to which we can evaluate quantified sentences.
4 This answer was suggested to me by Reto Gubelmann.
26
Christopher Gauker
A context supplies a standard of size relative to which we can evaluate the adjective “small” as denoting the set of things that are smaller than that standard. And so on. But there is another way to think about contexts. A context, as I propose to define it, will contain a variety of structures built up entirely from linguistic entities. To illustrate the basic idea, let me start with a very simple example. Consider a language L having the usual syntax of the languages of first-order logic. Suppose it contains a finite number of one-place predicates, F, G, . . ., and a finite number of two-place predicates, R, S, . . ., a finite number of individual variables, x, y, . . ., and a finite number of singular terms a, b, . . . (We may find it helpful to think of these singular terms as subscripted demonstratives, such as “this1” and “this2”.) The logical operators of L are confined to the universal quantifier ∀, the negation symbol ¬, and the disjunction symbol ∨. For such a language we can define a context as a pair consisting of a base and a domain. For a context Γ, the base BΓ is a set of atomic sentences and negations of atomic sentences (literals). The base must be consistent in the sense that for no atomic sentence is it the case that both it and its negation is a member. But the base need not be maximal; there may be atomic sentences such that neither they nor their negations are members. (It can even be empty.) For a context Γ, the domain NΓ is a set of singular terms (not objects such as we might think of singular terms as denoting); it must be nonempty, and it must contain at least every singular term that occurs in any member of BΓ. Given these definitions of the language and set of contexts for the language, we can recursively define truth in a context as follows: (T0) If p ∈ BΓ, then p is true in Γ. (T¬) If p is false in Γ, then ¬p is true in Γ. (T∨) If p is true in Γ or q is true in Γ, then (p ∨ q) is true in Γ. (T∀) If, for every n ∈ NΓ, Φ[n/x] is true in Γ, then ∀xΦ is true in Γ. (TCl) No other sentence is true in Γ. (F0) If ¬p ∈ BΓ, then p is false in Γ. (F¬) If p is true in Γ, then ¬p is false in Γ. (F∨) If p is false in Γ and q is false in Γ, then (p ∨ q) is false in Γ. (F∀) If for some n, Φ[n/x] is false in Γ, then ∀xΦ is false in Γ. (FCl) No other sentence is false in Γ.
Here “Φ[n/x]” stands for the result of substituting the singular term n for the variable x wherever x occurs free in Φ. For example, suppose Γ = 〈BΓ, NΓ〉, where BΓ = {Fa, Fc, Gb, ¬Ka, Rab, ¬Hbc}, and NΓ = {a, b, c}. Then the following sentences will all be true in Γ: Fa, ¬Ka, ∀x(Fx ∨ Gx), ¬∀xHbx. Evidently, this
Open texture and schematicity as arguments for non-referential semantics
27
semantics is three‐valued. There will be sentences and contexts such that the sentence is neither true nor false in the context. Next, we may define logical validity very simply, as follows: An argument is logically valid if and only if for each context, if the premises are all true in the context, then so is the conclusion. The resulting logic is not classical. For example, sentences of the form (p ∨ ¬p) are not valid; they are not true in every context. The rule of universal instantiation (from ∀xΦ infer Φ[n/x]) is not valid. (That is, arguments having that form are not valid.) I have argued elsewhere (see especially my 2003) that these results are defensible as a representation of the logic of natural language. It is noteworthy that we can in this way get a defensible definition of logical validity merely by considering the values of the premises and the conclusion in every possible context. We do not need to employ any analogue of the set of models with all their various assignments of extensions to non-logical terms. In particular, we do not need to countenance one particular model that assigns to each non-logical term the extension it really has. In this respect, this kind of semantics can be characterized as non-referential.5 The basic idea behind this approach to semantics is that every time we add something of, so to speak, logical interest to the language, we accommodate it by adding structure to our definition of a context and adding clauses to our definition of truth/falsehood in a context. For example, if we wish to add a conditional operator having a logic that approximates to the logic of the natural language indicative conditional, we need to define contexts in such a way that contexts can, in a sense, have other contexts as members. Then we can reformulate the truth conditions for other sorts of sentences. For instance, an atomic sentence will be true in such a context if and only if it is true in every member of the context. A conditional, we can say, is true in such a context Γ if and only if for each context ∆ that is either a member of Γ or identical to Γ, if the antecedent is true in ∆, then so is the conclusion. (For details, see my 2005 monograph.) The definition of truth in a context, so understood, is, in one sense, quite uninformative. It simply defines a relation of truth between two abstract entities – sentences and contexts. However, it can be made to yield an account of the truth values of actual utterances when we add to it an account of the
5 This definition of logical validity in effect generalizes on the truth-value semantical definition explored by Hugues Leblanc (1976). An objection to this sort of semantics, with its substitutional interpretation of the quantifiers, is that it seems to validate the omega rule, which is an undesirable result. Leblanc shows that this objection can be addressed with modifications in the definition of logical validity, but there is more to say about this issue than I can take up here.
28
Christopher Gauker
ertaining relation between contexts and utterances. For as I said in section 2, p an utterance of a sentence is true (simpliciter) if and only if the sentence uttered is true relative to the context that pertains to the utterance. We can hope that our account of the pertaining relation will avoid invoking the reference relation that we are attempting to banish, but we cannot expect our account of it to be a tidy definition. My hope is that we can identify the context pertinent to a conversation along the following lines: Each interlocutor in a conversation will have his or her own take on the context. This we can think of as literally consisting of linguistic items written in the brain. A person’s take on the content pertinent to his or her conversation drives his or her behavior in cooperation with other interlocutors. The context objectively pertinent to a conversation can be identified in terms of the takes on a context that would be most effective in driving cooperation interaction. In light of this general conception, we can identify a variety of desiderata concerning the interlocutors and their environment that the context pertinent to a conversation ought to satisfy; the context that really does pertain to a conversation will be that which provides the best overall fit to these desiderata. (For further discussion, see my 2005.) As I have already explained, in these terms we can define logical validity, and so this kind of semantic theory satisfies our requirement that our semantics be able to do that. It also satisfies our requirement that it defines a way in which what we say should be adequately responsive to the way the world is. The definition of truth in a context does not do this, for as I said, that is just a relation between two abstract entities. But when we add the relation of pertaining between a context and an utterance, we get what we require: The normative expectation that we can define in terms of this semantics is that, other things being equal, the sentences we utter should be true relative to the context that pertains to the situation in which our discourse takes place.
6 Open texture and schematicity revisited A word has what I call open texture when its true applicability to an object in a given case is not always determined by the nature of the object, past uses of the term or the thoughts of speakers. When that is so, our use of the term may face a lexicographical choice point, at which point the question whether to apply the term to an object or withhold application can only be answered by a more or less arbitrary decision. That words have open texture in this sense is an embarrassment for referential semantics, because referential semantics demands that terms such as “jar” and “cat” have a definite extension.
Open texture and schematicity as arguments for non-referential semantics
29
The non-referential approach to semantics described in the previous section faces no such embarrassment. Because it posits no extensions, no question can arise about whether the choco-cooler container belongs to the extension of “jar”. Likewise, no question can arise about whether the Borneo weasel belongs to the extension of “cat”. In any situation in which the question arises whether to utter an atomic sentence of the form “x is a cat”, the question, “Does that sentence belong to the base of the context that pertains to our situation?” will still have to be answered (though not necessarily in just those terms). That, as I have already acknowledged, may be a messy, difficult question. In the “cat” case, part of what is involved will be thinking about similarities between the present objects and things we have applied “cat” to in the past. And part of what is involved will be thinking about what we expect to achieve by calling something a “cat”. But no part of the difficulty is deciding whether the referent of the singular terms belongs to the extension of “cat”. We can allow that a factor that may figure into the answer may be just the sort of arbitrary decision that I have called a lexicographical choice point. What I here call schematicity is a feature of verbs that seem to have only a very general meaning in the abstract, but which seem to denote more specific relations on any particular occasion of use. Schematicity is an embarrassment for referential semantics, because referential semantics does not seem to be able to offer any reasonable account of the relation between the general meaning of a schematic verb and the particular relation it denotes on any particular occasion of use. The non-referential approach to semantics faces no such embarrassment. Let us say that a context in which a sentence of the form “X uses Y” is true is a usescontext. Even on the non‐referential approach to semantics one can acknowledge that “uses” is schematic in the sense that the features of situations that make it the case that a uses-context pertains to a situation may vary widely. In one situation it may be the fact that something cuts something in that situation, in another that something rides in something in that situation, etc. That the pertinent feature of a situation that determines that a uses-context pertains to the situation may vary in this way from situation to situation is in no way contrary to expectations, given that, in general, the relation between a context and the situation that it pertains to may be a complex one, involving a competition of factors and amounting in the end to an optimality of fit.6
6 I thank audiences in Tartu, Zürich and Konstanz for stimulating questions that led to many improvements. Questions from Tobias Henschen, in Konstanz, and Matej Drobňák, during his visit to Salzburg, led to significant improvments.
30
Christopher Gauker
References Asher, Nicholas. 2011. Lexical meaning in context: A web of words. Cambridge: Cambridge University Press. Gauker, Christopher. 2003. Words without meaning. Cambridge, MA: MIT Press. Gauker, Christopher. 2005. Conditionals in context. Cambridge, MA: MIT Press. Gauker, Christopher. 2012. Inexplicit thoughts. In Laurence Goldstein (ed.), Brevity, 74–90. Oxford: Oxford University Press. Gauker, Christopher. 2015. How many bare demonstratives are there in English? Linguistics and Philosophy 37. 291–314. Heim, Irene & Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell. Kripke, Saul. 1982. Wittgenstein on rules and private language: An elementary exposition. Cambridge, MA: Harvard University Press. Leblanc, Hugues. 1976. Truth-value semantics. New York: North Holland Publishing Company. Liston, Michael. 2010. Rabbits astray and significance awandering: Review essay on Mark Wilson’s Wandering significance. Philosophy and Phenomenological Research 89. 809–817. Ludlow, Peter. 2014. Living words: Meaning underdetermination and the dynamic lexicon. Oxford: Oxford University Press. Malt, Barbara C., Steven A. Sloman, Silvia Gennari, Meiyi Shi & Yuan Wang. 1999. Knowing versus naming: Similarity and the linguistic categorization of artifacts. Journal of Memory and Language 40. 230–262. Pustejovsky, James. 1995. The generative lexicon. Cambridge, MA: MIT Press. Recanati, François. 2004. Literal meaning. Cambridge, MA: Cambridge University Press. Recanati, François. 2013. Reply to Gauker. Teorema 32. 81–84. Searle, John R. 1983. Intentionality: An essay in the philosophy of mind. Cambridge, MA: Cambridge University Press. Waismann, Friedrich. 1945. Symposium: Verifiability. Proceedings of the Aristotelian Society: Supplementary Volumes 19. 119–150. Wittgenstein, Ludwig. 1953. Philosophical investigations. Macmillan.
Kepa Korta and John Perry
Full but not saturated: The myth of mandatory primary pragmatic processes Abstract: On the issue of how much pragmatics has to do with what is said, philosophers and linguists divide into the minimalist and contextualist camps. Most members of both camps agree that in utterance comprehension, there are clear cases of “pragmatic intrusion.” The consensus is practically universal, when it comes to utterances containing indexicals, demonstratives and context-sensitive expressions in general. The basic idea is that without pragmatic provision of appropriate referents, no proposition is determined, so the hearer cannot very well understand what the speaker said (the proposition expressed or the explicature). Even “radical” minimalists like Cappelen and Lepore (2005) concede this. Recanati (2004) calls such pragmatic intrusion into the business of reference, “saturation”. Saturation is a mandatory primary pragmatic process. It is primary, in contrast with secondary processes of implicature inference. It is mandatory, in contrast with optional primary processes such as free enrichment. We will argue that the mandatory nature of saturation is a myth. Saturation is not needed to determine a truth-evaluable proposition. Indeed, at times it is not even required for an adequate understanding of what a speaker means by her utterance. We will offer several examples involving context‐sensitive expressions that make perfect sense of an unsaturated but truth-conditionally complete propositional content.
1 Introduction There are few assumptions in contemporary pragmatic theory as universal as what we will call “incompletism”. With this ugly word, we refer to the claim that, in utterance comprehension, in the absence of the operation of certain pragmatic processes, an utterance often fails to determine a complete, fully truth-evaluable proposition. It delivers only an incomplete proposition, something that in and by itself cannot have a truth-value. This may be identified as a subpropositional “logical form”, a “partial” or “gappy” proposition, a propositional “fragment”, “schema”, “radical”, “skeleton”, “template”, “matrix”, or “scaffolding.”1 All
1 We lack the scholarly instincts to track each term to its original author, but we think it is fair enough to say they collectively belong to Kent Bach (1994), Robyn Carston (2002), François Recanati (2004), Dan Sperber & Deirdre Wilson (1995), Ken Taylor (2001) and others. DOI 10.1515/9781501504327-003
32
Kepa Korta and John Perry
serve as propositional function, to use a more traditional term. Unless values are given for this entity’s variables or – gaps or slots – it will not have a truth-value, i.e., it will not be a (full) proposition.2 The ubiquity of incompletism is a matter of dispute. It depends on (i) the number of context-sensitive expressions admitted and (ii) the nature of context‐ sensitivity envisaged as well as (iii) what is taken to be a complete, fully truth- evaluable proposition. Cappelen & Lepore (2005) are a good example of a m inimalist position on the first two issues: the number of context-sensitive expressions is limited to what they call “the Basic Set”, namely, the pronouns “I”, “you”, “he”, “she”, “it”, “this”, and “that” in all their cases and number, the adverbs “here”, “there”, “now”, “today”, “yesterday”, “tomorrow” “ago”, “henceforth”, the adjectives “actual” and “present”, tense words and morphemes, nouns like “enemy”, “outsider”, “foreigner”, “alien”, “immigrant”, “friend”, and “native”, and adjectives like “foreign”, “local”, “domestic”, “national”, “imported”, and “exported”. Hence, the only kind of context-sensitivity admitted is indexicality (which alludes to the context-sensitivity proper of indexicals, demonstratives and contextuals). About the third issue, they are almost alone in holding that, for example, an utterance of “It’s raining” expresses a complete proposition even without the pragmatic provision of a location for the raining-event; and so does an utterance of “Tipper is ready” even without knowing what she is supposed to be ready for. When pressed, they are happy to answer in the following terms: (1) An utterance of “It is raining” expresses the proposition that it is raining, which is true if and only if it is raining. (2) An utterance of “Tipper is ready” expresses the proposition that Tipper is ready, which is true if and only if Tipper is ready. Most people disagree with Cappelen & Lepore, and think that locations must be provided for weather reports, and some other parameters are required by gradable adjectives, relational adjectives and a variety of expressions that show, at the same time, that there are more kinds of context-sensitivity besides indexicality. Among those who disagree are contextualists such as Charles Travis (1997), François Recanati (2004), Dan Sperber & Deirdre Wilson (1995) and Robyn Carston
2 Bach claims that there is no much sense in talking about incomplete propositions: “An incomplete proposition is no more a proposition than a sentence fragment is a sentence or a rubber duck is a duck” (Bach 2006: 441–2). That’s why he opts for the term “propositional radical”. Nevertheless, Bach himself talked about incomplete propositions in “Conversational Impliciture” (1994) and many other places.
Full but not saturated: The myth of mandatory primary pragmatic processes
33
(2002).3 But there is a point on which all, minimalists and contextualists alike, come to terms: the case of indexicals and demonstratives.4 Suppose that suddenly, without any given context, somewhat out of the blue, you hear the utterance (1) I am French. As long as you have a basic knowledge of English, and you assume that the speaker is talking literally, with the ordinary meaning of those words and their composition, and that she is making an assertion and not, for instance, just reading aloud a poem she just wrote, there is a sense in which you can rightly say that you understood the utterance. You understood the words uttered but you didn’t understand what the speaker said, in the philosophers’ and linguists’ usual favored sense of the verb “to say.” The fact that you have no clue about the context of the utterance that permits you to identify the speaker of (1) makes your understanding of (1) incomplete. You cannot assign a referent to the speaker’s use of “I”, so you don’t understand what the speaker said in uttering (1). Your understanding falls short of determining a complete proposition. Instead, you just get an incomplete proposition. Something like (2) x is French. Given this, pragmatic processes of provision of referents would be mandatory to obtain a fully truth-conditional proposition, or, what amounts to the same thing, to get a candidate for the proposition that counts as what is said by the utterance. Recanati (2004) calls the pragmatic processes that “intrude” in the determination of what is said “primary” pragmatic processes and claims that, among them, there is one kind, which he calls “saturation”, which is mandatory, unlike all the other pragmatic processes, which are optional. Not everybody agrees that there is a fundamental distinction to be made between primary and secondary pragmatic processes; Sperber & Wilson (1995, 2002), Carston (2007) and Curcó (2013) argue against Recanati on this point. But they all agree, minimalists and contextualists alike, that saturation, whatever you call it and however you characterize it, is a mandatory pragmatic process. And it is mandatory for the sole reason that, otherwise, the utterance would provide only an incomplete proposition.
3 Indexicalists such as Jason Stanley and Zoltan Szabo (2000) take the set of indexicals to be much larger than the Basic Set but they add a new kind of context-sensitivity: hidden i ndexicality. 4 There is no consensus about contextuals, and we won’t deal with them in this paper.
34
Kepa Korta and John Perry
In this paper, we argue against this consensus opinion. We argue that 1) there are no mandatory primary pragmatic processes; because 2) even without the provision of referents for referential expressions (or locations for weather reports) an utterance does determine a complete proposition. This proposition will typically not be “the proposition expressed” or “what is said” by the speaker. But it is a complete proposition that captures truth-conditions for the utterance. Thus, primary pragmatic processes are not needed to have such a proposition. If primary pragmatic processes are mandatory it is not due to incompletism. We’ll start by summing up the differences between primary and secondary pragmatic processes according to Recanati (2004) and the allegedly mandatory nature of some of the former. Then we’ll focus on the assumptions behind the main argument for that mandatory nature: together with incompletism, it depends on semantic underdeterminacy and propositionalism. We will show, in section 4, what’s wrong with these assumptions or, rather, what’s wrong with the use of these assumptions in arguing for incompletism. Their force ends when we recall that we are dealing, not with mere sentences, but with utterances of them. With that in mind, it is natural to consider a variety of truth‐ conditions or contents beginning with a level that is clearly complete but not saturated: the utterance‐bound or reflexive content. In section 5, we’ll consider various objections: (i) that the utterance-bound content is saturated after all (ii) that it’s not sufficient for understanding and (iii) that it should be ignored, since it plays no role in a theory of utterance comprehension. Needless to say, we’ll rebut all these charges. In section 6, we’ll show how our view fits perfectly with Grice’s view on what is said and, perhaps despite appearances, with the relevance-theoretic notion of explicature. We’ll end by drawing some general conclusions. We need to emphasize that we are adopting the hearer’s perspective on utterance contents and truth-conditions. That is, we are talking about the understanding rather than the production of utterances. When we take the speaker’s point of view and think about utterance planning and production, the debate does not make much sense. In all the relevant cases under discussion here the speaker does have a complete thought, i.e., a belief or another doxastic attitude with a complete truth-conditional content she intends to express via her utterance: there is no primary or secondary pragmatic process to be undertaken by the speaker. Typically she will know the (intended) referents for her referential expressions before they are uttered. The present debate, as many others in pragmatics, concerns utterance comprehension, but that shouldn’t hide the fact that pragmatics is also concerned with what the speaker means and does in uttering what she does.
Full but not saturated: The myth of mandatory primary pragmatic processes
35
2 Recanati on pragmatic processes Recanati (2004) has insisted on the distinction between primary and secondary pragmatic processes and its importance. Against this, some other authors, like Sperber & Wilson (1986, 2002), Carston (2007) and Curcó (2013), have argued that there is no significant psychological difference between them, even if some theoretical distinction may be justified. According to Recanati (2004), primary pragmatic processes (henceforth PPPs) are those involved, together with the semantic meaning of the sentence uttered, in determining the proposition expressed or what is said. These include disambiguation, reference fixing and, depending on your minimalist or contextualist allegiances, other processes of “enrichment”. In other words, PPPs belong to what Korta & Perry (2011b) called “near‐side pragmatics”, as they are prior to the determination of what is said. Secondary pragmatic processes (henceforth SPPs), belong to “far‐side pragmatics”, given that, of course, they go farther than what is said: In contrast, secondary pragmatic processes are ordinary inferential processes taking us from what is said, or rather from the speaker’s saying of what is said, to something that (under standard assumptions of rationality and cooperativeness) follows from the fact that the speaker has said what she has said. (Recanati 2004: 17)
Recanati makes a related distinction that concerns the level of operation of the pragmatic processes. Given that PPPs intervene on sentence meaning to jointly determine the proposition expressed, they can be said to work at the prepropositional level. SPPs, on the other hand, typically take the proposition expressed, what is said, as input; so, in that sense, they operate at the post‐ propositional level. A third difference has to do with the kind of cognitive system responsible for the process. Thus, PPPs typically are not available to consciousness, but rather belong to sub-personal cognitive processes. In contrast, the determination of implicatures, typically with what is said as input, occurs at the personal level, and so it is available to consciousness. This is related to the fact that SPPs are inferential processes, while PPPs would be blind, mechanical, i.e., non‐inferential and mostly associative. This is always the case according to Recanati (2004). Sperber & Wilson (1995), among others, argue that pragmatic processes are all inferential and guided by considerations of relevance. Given this, the distinction between PPPs and SPPs does not make much sense; it would concern only the level of utterance content to which these processes contribute: the level of what is said (or the explicature, in their terms) or the level of implicatures. However, practically all authors, including Sperber & Wilson, agree
36
Kepa Korta and John Perry
that there are some pragmatic processes that are mandatory, and that those are processes that contribute to what is said, that is, they are PPPs, if the distinction is to be kept.5 Tab. 1: Differences between PPPs and SPPs according to Recanati. 1.1 PPPs
1.2 SPPs
Involved in determining what is said
Take what is said as input
Work at the pre-propositional level
Work at the post-propositional level
Operate at the sub-personal level (they are not consciously available)
Operate at the personal level (they are consciously available)
Blind, mechanical
Inferential
Mandatory (saturation) or optional (enrichment)
Optional
Recanati, being a contextualist about what is said, admits that some PPPs such as enrichment are optional as all SPPs are. But he insists that some PPPs are mandatory, since they are required for the utterance to express a complete proposition. Recanati’s label for these PPPs is “saturation.” Under this label he includes not only reference assignment and disambiguation but also the provision of unarticulated constituents, when they are needed to get a full proposition: Saturation is the process whereby the meaning of the sentence is completed and made propositional through the assignment of semantic values to the constituents of the sentence whose interpretation is context-dependent (and possibly through the contextual provision of unarticulated constituents, if one assumes, as some philosophers do, that such constituents are sometimes needed to make the sentence fully propositional). (Recanati 2004: 7. Our emphasis).
The critical function of the process of saturation is, then, to complete what otherwise would be incomplete and pre-propositional. The key is then to provide the
5 The optionality of SPPs and PPPs other than saturation seems related to the cancelability of pragmatically determined aspects of content. Grice (1967a, 1967b) indicated that conversational implicatures are cancelable, and Sperber & Wilson (1986), Carston (2002), Recanati (2004) and many others extended it to all pragmatically determined elements. Now, if saturation is mandatory, it is inconsistent to claim that an element determined by saturation is cancelable, if we interpret being cancelable roughly as being eliminable, and not as being merely revisable (see Korta 1997, for discussion).
Full but not saturated: The myth of mandatory primary pragmatic processes
37
elements to get a complete proposition. And that’s the reason why saturation is a mandatory process: Whenever saturation is in order, appeal to the context is necessary for the utterance to express a complete proposition: from a semantic point of view, saturation is a mandatory contextual process. (Recanati 2004: 7. His emphasis)
Recanati’s “saturation” is very similar to Kent Bach’s “completion”: When a sentence is in this way semantically under-determinate, understanding its utterance requires a process of completion to produce a full proposition. (Bach 1994: 125)
Without completion (or saturation), no full proposition, ergo no understanding. That’s the claim.
3 Underdeterminacy, propositionalism and incompletism The main argument for the obligatory nature of saturation or completion is incompletism. But the route from incompletism to mandatory PPPs goes through two other widely held assumptions: semantic underdeterminacy and propositionalism. Semantic underdeterminacy has been much discussed in relation to the contextualist/minimalism debate, and it has been interpreted in related but significantly different ways. In particular, the following three somewhat different failures have been labeled “semantic underdeterminacy”: 1. the linguistic meaning of the sentence uttered fails to determine a complete proposition; or 2. the linguistic meaning of the sentence uttered fails to determine the literal truth-conditions of the utterance; or 3. the linguistic meaning of the sentence uttered fails to determine the i ntuitive truth-conditions of the utterance. There are various positions regarding underdeterminacy. If you are a contextualist, for instance, you probably think that all three varieties are pervasive in natural language; that the role of context in identifying propositions is not limited to indexicals, and that there is no coherent notion of literal meaning separable from the intuitive truth-conditions of the utterance, which linguistic meaning systematically underdetermines. But you don’t need to be a contextualist to embrace semantic underdeterminacy. And we are not going to take sides in the debate. We will deal with the non‐controversial part of semantic underdeterminacy,
38
Kepa Korta and John Perry
namely, 1) This is always true when we consider sentences containing indexical and demonstratives – the meaning of the sentence does not provide a complete proposition. Nobody denies that, nor do we. In fact, it sounds very much like a truism; the contextualist truism: sentences do not express propositions, utterances of sentences (that is, speakers uttering sentences) do. What we call “incompletism,” and deny, is version 1) of indeterminacy applied to utterances. The second leg of the argument for the mandatory nature of saturation is propositionalism, the idea that the performance (and understanding) of a full speech act like an assertion, a command or a promise involve the expression (and understanding) of a complete proposition. Of course, we are leaving aside wh‐interrogatives and other cases that may induce more or less controversy. We do not argue against it. So, we deny neither underdeterminacy nor propositionalism, but the assumption that they jointly lead to incompletism. The key to our objection requires attention to the distinction between sentence meaning and utterance content(s).6
4 Sentence versus utterance Consider again the sentence uttered in (1), namely, “I am French”. The sentence itself does not determine a complete proposition. If we adopt a token-reflexive account of indexicals and assume that the meaning of “I” is something like “the speaker of this utterance” then, at most we will have a “gappy” proposition with a slot to be filled in by a particular utterance of that sentence, something like: (3) The speaker of x is French. This is indeed a propositional function, that is, an incomplete proposition. That’s the most an indexical sentence can aspire to provide in propositional terms: a semi-gappy-incomplete propositional function-radical-template. Now, as soon as we consider not just sentences but utterances of sentences, that is, acts performed by a certain agent, the speaker, at a certain time, in a certain place, things are crucially different. To begin with, if we take again utterance (1), we don’t have just a propositional function like (2) or (3), but a full proposition like (4) The speaker of (1) is French.7
6 The reason for the parenthetical plural is that as a natural consequence of our discussion, we will argue that the utterance, not the sentence, has a variety of contents or truth-conditions. 7 We’ll follow the notational convention of Perry (2012) and use boldface to indicate that the propositional constituent is the referent itself and not any identifying condition on it. On the other hand, we will use small capitals for propositions, contents or truth-conditions.
Full but not saturated: The myth of mandatory primary pragmatic processes
39
Even if you heard (1) out of the blue, without any chance to identify the speaker, not even to guess her or his gender or age, this is available to you given your knowledge of the English language, the identification of the sentence uttered, and your perception of the utterance or, in some cases, your inference that there was an utterance. In face-to-face communication you are usually able to directly identify the speaker, the time and the place of the utterance you’re hearing. You may not know the speaker’s name, or what time is it, or where exactly is, but you have direct knowledge of those parameters, and they permit you to close the propositional function provided and have access to further contents. If you had the speaker of (1) in front of you, you would be able to grasp the following proposition: (5) The Person in front of me is French, and even, given your perceptual identification of the speaker, the following one: (6) This guy is French, with the speaker himself, not any identifying conditions on him, as a constituent. This is usually considered as the proposition expressed by the utterance, what the speaker said in uttering (1). We also consider it so – usually (see Korta & Perry 2007, 2011 and 2013). When not communicating orally or face‐to-face, often we do not perceive the utterance itself; we do not perceive the act of uttering a sentence, but only its product, what Perry (2012) calls a “token”. In that case, we might be unable to identify the speaker of the utterance, or the time, or the place, independently of the utterance – which may itself only be identifiable as “the production of this token”. Hence we talk of “utterance-bound” identification of the speaker, the time, the place, and in fact of the truth-conditions of the utterance. Suppose you find an anonymous hand-written note, apparently slid under your office door, which reads (n) You are Spanish. You don’t have any hint about its author, its addressee or addressees, the time it was written or placed there, or its purpose.8 The utterance-bound truth‐ conditions, or utterance-bound content, of (n), however, is available to you:
8 We should rather say its user, since being a token, can be used and re-used by other people other than its author; that’s why the relevant time is the time of sliding the note (or perhaps even the time the user expect the addressee to read the note; see Perry 2003).
40
Kepa Korta and John Perry
(7) The addressee(s) of n meet the conditions the author of n means by “Spanish” at the time of being said. Whether the referent of “you” is a sole person, all the occupants of the office or just some of them and who exactly he, she or they is/are is a matter of the speaker’s intention that goes further than the facts determining the minimal utterance-bound content. About the meaning of “Spanish” opinions can differ about whether it denotes a perfectly identifiable current European citizenship quite permanent in time, or it involves few things other than a passport which makes the denotation more intention-dependent and temporary. (7) stands for the latter option. Anyway, we know that the note has an author and has an issuing time, and that’s all we need to existentially close our propositional function and have a complete proposition. Coming back to (6), we call it the referential content or referential truthconditions of the utterance, as we take utterances to have a variety of truthconditions or contents; a variety of truth-conditions or contents that are set relative to various kinds of facts: –– the utterance-bound truth-conditions; set by the meaning of the sentence uttered plus the fact that an utterance has been produced. –– the speaker-bound truth-conditions; set by the above facts plus the identity of the speaker. –– the network-bound truth‐conditions; set by the above facts plus the notion‐ network supporting the use of a certain proper name. –– the referential truth-conditions; set by the above facts plus facts about the speaker’s intentions and contextual facts that set the values for context‐sensitive expressions as well as unarticulated constituents. –– and some other truth-conditions that can be distinguished as the product of all the various facts about conventions, intentions and circumstances involved. To our present purposes, however, it is the first level of truth-conditions which is critical: the level of utterance‐bound or reflexive truth-conditions, for they provide the proof that incompletism is false. It is false that without saturation we do not have a complete proposition. And thus it is false that saturation is mandatory to get a complete proposition.
5 Against utterance-bound or reflexive content Suppose you are quite convinced by our arguments, but still you wish to defend that saturation is mandatory. You might try to object as follows. The closure of
Full but not saturated: The myth of mandatory primary pragmatic processes
41
the incomplete parameters through the utterance parameters of speaker, time and place is just another way of saturating the incomplete proposition delivered by sentence meaning. If right, this would be a fatal objection to our argument, we would be just begging the question by an excessive narrow interpretation of “saturation”, and we wouldn’t have refuted incompletism. Remember that we deliberately choose to discuss only the non-controversial cases, indexicals and demonstratives. These are supposed to leave, in virtue of their context-sensitive meaning, a “gap”, “slot” or a free variable that is to be saturated to have a complete proposition. Saturation in this case amounts to providing the referents for the indexicals and demonstratives, their “semantic” values as they are often called (misleadingly, since their determination is not semantic, but pragmatic, as it involves intention recognition). Utterances containing indexical or demonstrative pronouns express singular propositions, with their referents as constituents, and saturation, as far as indexicals and demonstratives are concerned, is the process of fixing them. That is at least what Recanati (2004) clearly has in mind. Saturation provides the referents for indexicals and demonstratives. Without saturation we are left with gaps, slots or free variables. In other words, our utterance‐bound content, by definition, involves no saturation. Recanati (2004) himself acknowledges this when he takes the utterance-bound or reflexive proposition as the only coherent notion of minimal proposition: [T]he reflexive proposition is determined before the process of saturation takes place. The reflexive proposition can’t be determined unless the sentence is tokened, but no substantial knowledge of the context of utterance is required to determine it. Thus an utterance u of the sentence “I am French” expresses the reflexive proposition that the utterer of u is French. That it does not presuppose saturation is precisely what makes the reflexive proposition useful, since in most cases saturation proceeds by appeal to speaker’s meaning. (Recanati 2004: 65. Our emphasis.)
Saturation and existential quantification are two different ways of getting from sentence meaning to a complete proposition. It is contrary to Recanati’s understanding of his own term, and unhelpful, to use the term “saturation” with this portmanteau sense. Another possible objection could go along the following lines: “Ok, let’s accept that your utterance-bound or reflexive content is a complete proposition that is determined before saturation. Imcompletism is false. But the utterance-bound content does not amount to what is said, so understanding the utterance‐bound content does not count as understanding the utterance. So, even if saturation is not mandatory for reasons of incompletism, it is mandatory for reasons of full understanding of the utterance”.
42
Kepa Korta and John Perry
This is not an objection but a concession of our main claim: incompletism is false, so saturation is not mandatory for that reason. But it can be used as a preliminary move to a further point: that the utterance‐bound content has no role in a psychologically plausible account of utterance understanding.9 Concerning the first point. The notion of utterance‐bound content is not intended to capture “what is said” or “the proposition expressed”. Quite the opposite, we claimed that typically what is said corresponds to what we call the “referential content” of the utterance. Typically, but not always. For example with identity statement the referential content seldom captures what the speaker means to convey. Recanati himself does not take the reflexive proposition to be what is said: The reflexive proposition is admittedly distinct from that which the speaker asserts ... but why is this an objection? [The reflexive proposition] comes as close as one can get to capturing, in propositional format, the information provided by the utterance in virtue solely of the linguistic meaning of the sentence “I am French.” (Recanati 2004: 66)
However, we think that equating full understanding of an utterance with understanding of what is said is incorrect. Leaving aside that full understanding is most likely a chimera, we’d like to show, first, that understanding what is said is not necessary to adequately understand an utterance. Consider the following example:10 In New Orleans, after the Katrina, Davis McAlary checks regularly his ex‐ girlfriend’s mailbox since she moved to New York City. This time he sees the front door is open, it has been forced; he enters; there is a mess; he hesitates; he hears a noise upstairs; he utters “Who’s that? Who’s that? Anybody there? Hey! This is the Police! Anybody up there?”; he takes his cell phone from his pocket and holding it like a gun he shouts “I’m a cop with a gun, seriously!”; more noises upstairs; he runs out like crazy. Consider McAlary’s last utterance: (8) I’m a cop with a gun, seriously.
9 These two objections are the transposition of Recanati’s objections to the minimalist notion of what is said à la Cappelen & Lepore or Bach. Obviously, for a contextualist that notion does not correspond with an adequate notion of what is said; and, according to Recanati (and others like Sperber & Wilson or Carston) it plays no role, anyway, in a psychologically plausible explanation of utterance understanding. For some experimental work on the psychological role of (not so) minimal propositions, see Bezuidenhout & Cooper Cutting (2002). 10 From the TV series Treme, second season, third episode (2011), entitled “On your way down”.
Full but not saturated: The myth of mandatory primary pragmatic processes
43
Forget about “seriously”. Now, what is the relevant content? Perhaps, it’s a gappy proposition like (9) x is a cop with a gun.
We can discard that already. What he said by (8) is surely (10) Davis McAlary is a cop with a gun. But was saying this Davis’s goal? Whatever counts as identifying an individual in a singular proposition, it involves identifying the individual in other ways than as the referent of the uttered referential expression; that is, not simply in an utterance-bound way. Either directly, seeing him, for instance, or by a description like “the sweet but hopeless music lover who is an heterodox radio DJ and a frustrated musician.” Any of these would be fatal for McAlary’s purpose. He wants the hearer to think that there is an armed policeman downstairs, where the utterance clearly takes place. Anyone who sees him or has heard about him would definitely not be scared and would not fly from the scene. It seems clear that in this case the content that McAlary intends to convey to the hearer is rather (11) The speaker of (8) is a cop with a gun, i.e., the utterance-bound content. That’s what they are to grasp, and from which they are to infer that there is an armed policeman in the house, and it’s best to leave the house.11 The example shows that the objections are misguided if directed to utterance‐ bound content. It doesn’t amount to what is said – we never said that – but it explains what understanding consists in in many cases and, hence, it plays an important role in explaining utterance understanding. Thus, understanding what a speaker says in uttering a sentence is not necessary to understand the communicative act. And it is not sufficient either. This morning the first words Kepa told his friends were: (12) I am Basque. They looked at him with puzzled faces. His communicative plan was to exploit the widely recognized fact that Basques are punctual, and so to complain that
11 One can build alternative explanations like: “they just heard an utterance; they didn’t even understand the sentence; hearing a voice downstairs did all the work; Davis could have just uttered “Kennst du das Land, wo die Zitronen blühen?” and the robbers would have run all the same”. These are all possible stories, but not the story we are using as an example. Our story is possible and reasonable, and that’s all that matters.
44
Kepa Korta and John Perry
they were late and accuse them thereby for not being good Basques. It was probably too complicated early in the morning. Kepa failed. But no doubt they understood what he said. The problem is that they didn’t understand what he meant, that is, they didn’t understand why Kepa said what he said or, what amounts to the same thing, which the implicatures of his utterance were. This reveals a curious tendency in contemporary pragmatics: After 50 plus years of Gricean (and Austinian) pragmatics, the attention is centered on what is said (and its neighbor contents) rather than on implicatures. As we like to put it, the focus in pragmatics started on the far-side and then slowly slowly moved to the near‐ side. That is, it started by calling attention to phenomena and concepts that went far beyond what is said (illocutions, implicatures, presuppositions ...) and came to scrutinize the notion of what is said and facts in its vicinity. This is not good or bad in itself, research on both sides of pragmatics is surely important. But it could be the sign of a regress to the old code model of communication, a model that, as it is shown by Grice’s work, is essentially insufficient to account for human communication and demands to be supplemented (or substituted, depending how you interpret it) by a model that takes language as action. This emphasis on what is said as the measure of utterance understanding, forgetting about the importance of implicatures is, in our humble opinion, a remnant of the code model that we should discard.
6 True Neo-Griceans (or Neo-relevantists) Perhaps contrary to appearances, our view on utterance contents is well-rooted in Grice’s seminal work. It is also in substantial agreement with relevance theory, an important contemporary theory of utterance understanding that is rooted in Grice’s ideas. Our utterance‐bound content sounds as an echo of the following passage of “Logic and Conversation”, in which Grice is discussing what is said by an utterance of “He was in the grip of a vice”: Given a knowledge of the English language, but no knowledge of the circumstances of the utterance, one would know something about what the speaker had said, on the assumption that he was speaking Standard English, and speaking literally. One would know that he had said, about some particular male person or animal x, that at the time of utterance (whatever that was), either (1) x was unable to rid himself of a certain kind of bad character trait or (2) some part of x’s person was caught in a certain kind of tool or instrument (approximate account, of course). (Grice 1967a/1989: 25. Our emphasis.)
This is precisely how we characterize utterance‐bound content: the content determined by the knowledge of the language and the fact that an utterance has been made, nothing else. Referents for referential expressions are not assigned, the
Full but not saturated: The myth of mandatory primary pragmatic processes
45
time of the utterance is not determined and ambiguities are not resolved. It is not what is said, but it is looks like a complete content. So, leaving implicatures aside, Grice seems to explicitly admit at least two contents for the utterance: what we call the utterance‐bound and referential contents. The others are admitted implicitly since they derive naturally as the remaining information is “loaded”.12 As for relevance theory, the relevant comparison seems the one between our variety of contents and their “explicature”. The explicature has often been taken to be a proposition that results from processes of ambiguity and vagueness resolution, reference assignment and other pragmatics processes (roughly called enrichment processes) performed on the “logical form” of the sentence uttered. Or, following Recanati’s distinction, it is the result of applying the mandatory PPPs and, eventually, the optional PPPs to the initial – and incomplete – proposition that is the product of decoding. The “fully developed” proposition would constitute the input for the inference of implicatures. Cappelen & Lepore, to cite just one case, assume just that: We agree with her [Carston] that you need a contextually shaped content to generate implicatures in all of the cases she discusses. What’s needed in order to derive the implicature in these cases is a contextually shaped content, i.e., a contextually shaped what-is‐said. (Cappelen & Lepore 2005: 180)13
But we don’t think this is the right interpretation of the input of implicatures within relevance theory. Instead, as we understand the theory, it is assumed that both explicatures and implicatures are derived fast, on‐line and parallel, and the inferences are carried out following what they call the Relevance-theoretic comprehension strategy: (a) Follow a path of least effort in constructing an interpretation of the u tterance (and in particular in resolving ambiguities and referential indeterminacies, in going beyond linguistic meaning, in supplying contextual assumptions, computing implicatures, etc.). (b) Stop when your expectations of relevance are satisfied. (Wilson & Sperber 2012: 7. Emphasis added)
12 For a discussion of Grice’s notion of what is said see Korta (2013). 13 To be honest, in this passage they agree with Carston in “these cases”, the cases presented by contextualists in favor of an enriched notion of what is said. It’s not clear what Cappelen & Lepore think about other cases, but we think it is fair to say that many authors identify the concept of relevance-theoretic explicature with the fully enriched proposition. For instance, Yan Huang tells this in his entry on explicature: “An explicature corresponds roughly to the American philosopher Kent Bach’s notion of impliciture and the French philosopher François Recanati’s notion of pragmatically enriched said.” Huang 2012: 110–111.
46
Kepa Korta and John Perry
Interpreted according to our view, the hearer might well stop at the utterancebound content, without “going beyond linguistic meaning” and without “resolving referential indeterminacies”, because the utterance‐bound content is relevant enough – as in the McAlary example. Or she might go a bit further and stop at the speaker‐bound content, or the network-bound content, without going through the process required to fix the referents of referential expressions. This is all that is needed for the inference of implicatures (which can be on-line, and in parallel), and other perlocutionary effects the speaker intends to generate. There is nothing that demands the PPPs to operate at all, because the utterance-bound content might be relevant enough, and being complete, there is no necessity of any saturation process to work.
7 Conclusions We think it is now sufficiently clear that incompletism is false, and that, consequently, it does not justify the claim that saturation is mandatory. Reasons of “full” utterance understanding are also wanting. Understanding “what is said” is neither necessary nor sufficient to understand an utterance, and we believe that insisting on the contrary is the product of the old code model of human linguistic communication, according to which successful communication consists in transmitting a proposition from the speaker’s mind to the hearer’s mind. In our picture, what the speaker says, or, rather, the referential content of an utterance, is one content among others, and may or may not be the appropriate content to grasp in order to understand further contents of the utterance like implicatures and other perlocutionary contents. This seems to be much more in line with Gricean and Austinian views of language as action. Our pluralistic view on contents seems the only way to naturally explain actual communicative phenomena and overcome sterile debates caused by “monopropositionalism” and other remnants of the code model. It is true that abandoning incompletism even for the case of utterances containing indexicals and demonstratives would deprive both minimalism and contextualism of one of their few points of agreement. However, they would each gain on significant issues. The level of utterance-bound or reflexive content offers the minimalist a truly minimal, pragmatics-free kind of semantic content. Its determination does not require any appeal to context or to the speaker’s intention. Semantic content is immune to pragmatic intrusion. The price to pay is to admit that the minimal content does not amount to what the speaker says, but that’s a bullet most minimalist are ready to bite for their pragmatically-blended semantic content, anyway.
Full but not saturated: The myth of mandatory primary pragmatic processes
47
As for contextualism, the case for the pragmatic determinants of what is said remains strong. Anything beyond the utterance-bound content requires pragmatic processes (be they primary non-inferential or secondary inferential), so what is said by a speaker in uttering a sentence, our referential truth-conditions, would be highly context-dependent.14 What both minimalism and contextualism would lose together with incompletism is the mandatory nature of any pragmatic process. There is no mandatory PPP or SPP. At least not for reasons of incompletism. And if we consider reasons of good utterance understanding, PPPs are neither sufficient nor necessary. Acknowledgments: Earlier primitive versions of this paper were presented at “AMPRA. Pragmatics of the Americas Conference” (Charlotte, NC, USA), “IV Congreso Iberoamericano de Filosofía” (Santiago, Chile) and the “2nd ILCLIIFFs Workshop on Language, Cognition, and Logic” (Mexico City). We thank the audiences at these places and, in particular, Axel Barceló, Robyn Carston, Eros Corazza, Carmen Curcó, Wayne Davis, Justina Díaz Legaspe, Maite Ezcurdia, Eduardo García‐Ramírez, Larry Horn, Yan Huang, Nicolás Lo Guercio, Alfonso Losada, Eleonora Orlando, María Ponte and Deirdre Wilson for their comments, suggestions and criticisms. We benefited from grants by the Spanish Ministry of Economy and Competitivity (FFI2015-63719-P (MINECO/FEDER)), Basque Government (IT1032-16). The first author also benefited from a grant of the University of the Basque Country (UTR/PIU/UFI 11/18). The second author thanks the University of California, Riverside and Stanford University for their support. Sarah-Jane Conrad deserves special credit for her immense patience.
References Bach, Kent. 1994. Conversational impliciture. Mind and language 9. 124–162. Bach, Kent. 1994b. Semantic slack. What is said and more. In Savas Tsohatzidis (ed.), Foundations of speech act theory: Philosophical and linguistic perspectives, 267–291. London: Routledge. Bach, Kent. 2006. The excluded middle. Semantic minimalism without minimal propositions. Philosophy and Phenomenological Research LXXIII (2). 435–442. Bezuidenhout, Anne & Cooper J. Cutting. 2002. Literal meaning, minimal propositions, and pragmatic processing. Journal of Pragmatics 34. 433–456. Borg, Emma. 2006. Literal meaning. Review of François Recanati. 2004. Literal meaning. Mind 115 (458). 461–465.
14 For our position regarding the minimalism/contextualism debate, see Korta & Perry (2007b).
48
Kepa Korta and John Perry
Cappelen, Herman & Ernest Lepore. 2005. Insensitive semantics. A defence of semantic minimalism and speech act pluralism. Oxford: Blackwell. Carston, Robyn. 2002. Thoughts and utterances. The pragmatics of explicit communication. Oxford: Blackwell. Carston, Robyn. 2007. How many pragmatic systems are there? In María José Frápolli (ed.): Saying, meaning, and referring: Essays on François Recanati’s Philosophy of Language. Houndmills: Palgrave. Curcó, Carmen. 2013. From pragmatic processes to pragmatic processing: Consciousness, propositionality, and linguistic mandatoriness. Talk at the second ILCLI-IFF workshop on language, cognition and logic, Mexico, January 2013. Grice, H. Paul. 1975 [1967a]. Logic and conversation. In Donald Davidson & Gilbert Harman (eds.), The Logic of Grammar, 64–75. Encino: Dickenson. Also published in Cole Peter & Jerry L. Morgan (eds.), Syntax and semantics 3: Speech acts, 41–58. New York: Academic Press. Reprinted in Grice, 1989, 22–40. Grice, H. Paul. 1978 [1967b]. Further notes on logic and conversation, in P. Cole (ed.), Syntax and Semantics 9: Pragmatics. New York: Academic Press. Reprinted in Grice, H. Paul. 1989. Studies in the Way of Words, 41–57. Cambridge, MA: Harvard University Press. Huang, Yan. 2012. The Oxford dictionary of pragmatics. Oxford: Oxford University Press. Korta, Kepa. 1997. Implicitures: Cancelability and non-detachability. Report No. ILCLI-97-LICDonostia: ILCLI. Korta, Kepa. 2013. Grice’s requirements on what is said. In Penco Carlo & Filippo Domaneschi (eds.), What is said and what is not, 209–224. Stanford: CSLI Publications. Korta, Kepa & John Perry. 2007. How to say things with words. In Savas L. Tsohatzidis (ed.), John Searle’s philosophy of language: Force, meaning, and thought, 169–189. Cambridge: Cambridge University Press. Korta, Kepa & John Perry. 2007b. Radical minimalism, moderate contextualism. In Preyer Gerhard and Georg Peter (eds.), Context-Sensitivity and semantic minimalism. New essays on semantics and pragmatics, 94–111. Oxford: Oxford University Press. Korta, Kepa & John Perry. 2011. Critical pragmatics. An inquiry into reference and communication. Cambridge: Cambridge University Press. Korta, Kepa & John Perry. 2011b. Pragmatics. The Stanford encyclopedia of philosophy (Winter 2015 Edition), Edward N. Zalta (ed). http://plato.stanford.edu/entries/pragmatics Korta, Kepa & John Perry. 2013. Highlights of critical pragmatics: Reference and the contents of the utterance. Intercultural Pragmatics 10(1). 161–182. Perry, John. 2003. Predelli’s threatening note: Contexts, utterances, and tokens in the philosophy of language. Journal of Pragmatics 35. 373–387. Perry, John. 2012. Reference and reflexivity (2nd expanded and revised ed.). Stanford: CSLI Publications. Recanati, François. 2004. Literal meaning. Cambridge: Cambridge University Press. Sperber, Dan & Deirdre Wilson. 1995. Relevance: Communication and cognition (2nd edn.) Oxford: Blackwell. Sperber, Dan & Deirdre Wilson. 2002. Pragmatics, modularity and mind-reading. Mind and Language 17 (1/2). 3–23. Stanley, Jason & Zoltan G. Szabo. 2000. On quantifier domain restriction. Mind and Language 15. 219–61.
Full but not saturated: The myth of mandatory primary pragmatic processes
Taylor, Kenneth. 2001. Sex, breakfast, and descriptus interrruptus. Synthese 128. 45–61. Travis, Charles. 1997. Pragmatics. In Bob Hale & Crispin Wright (eds.), A companion to the philosophy of language, 87–107. Oxford: Blackwell. Wilson, Deirdre & Dan Sperber. 2012. Meaning and relevance. Cambridge: Cambridge University Press.
49
Silvan Imhof
How to get lost in context: Searle on context, content and literal meaning Abstract: Despite being a pragmatic theory of language, Searle’s theory of speech acts is methodologically based on semantic analysis. The properties of speech acts are supposed to be examined by analysing sentences whose literal utterance is sufficient to constitute a specific speech act by virtue of their semantic meaning. Searle’s method thus presupposes the concept of fixable, constant semantic content. Searle also advocates a very radical hypothesis about the dependence of the meaning of a speech act on the mental context of a speaker. The aim of the paper is to show that there is a serious clash between the methodological presupposition of constant semantic content and the thesis of the pervading influence of context on speech act meaning. In fact, if one takes up a contextualist stance as radical as Searle’s, semantic content cannot be taken as fixable and constant. Therefore, the preferred method of semantic analysis is by no means available.
1 Introduction My aim in this paper is to analyse a methodological problem of Searle’s theory of speech acts. The problem originates from a conflict between certain methodological assumptions and some fundamental theses of the theory. On the one hand, Searle assumes that the meaning of any speech act can be sufficiently determined by the meaning of a sentence whose literal utterance would constitute a speech act of the type in question. On this assumption, it is possible to characterise speech acts by a semantic analysis of suitable sentences. This methodological assumption is obviously based on the crucial presupposition that sentences have a constant semantic content in every literal utterance. On the other hand, the theory of speech acts contains some basic theses about how speech act meaning is constituted as well as about how the meaning of speech acts depends on the intentions of speakers, and about how context influences the meaning of speech acts. The problematic point is that the influence of context on the meaning of speech acts and the dependence of speech act meaning on the intentions of speakers undermine the aforementioned presupposition that sentences have a constant semantic content in every literal utterance. If, though, as a consequence of the theses of the theory of speech DOI 10.1515/9781501504327-004
52
Silvan Imhof
acts, no constant semantic content is available, there is no object available for a semantic analysis too. The opening sections 2 and 3 first specify the above-mentioned methodological assumptions and then expound on the three fundamental theses of the theory of speech acts. In sections 4 and 5, I shall explore some conceptual consequences of the fundamental theses concerning the notion of semantic content and semantic concepts in general. An examination of how content and context interact in section 6 will show that Searle’s notion of content is misguided. In section 7, I shall argue that against the theoretical background of the theory of speech acts no feasible concept of semantic content can be defended. The methodological consequences of my argument will finally be summed up in section 8: with no feasible concept of semantic content available, the theory of speech acts cannot proceed by semantic analysis. I should mention that my analysis of the problem refers exclusively to the theory of speech acts as it was put forward by John Searle. All the same, any theory of meaning that holds similar methodological assumptions and theoretical theses will presumably be affected by a similar problem.
2 The methodological primacy of semantic analysis in the theory of speech acts Searle’s theory of speech acts is a pragmatic theory of language. Language is conceived as consisting of rules that guide the use of linguistic items like words and sentences. Using such linguistic items according to these rules is to perform a kind of action called a speech act. From a methodological point of view, however, the theory of speech acts is a semantic project. The immediate objects of analysis are not utterance acts, but sentences whose literal utterances constitute the performance of certain speech acts in virtue of their meanings. Hence, although the theory of speech acts is a pragmatic theory of language, it is, from a methodological point of view, the result of a semantic analysis of sentences (see Searle 1969: 18; Vanderveken 1981, 1992; Vanderveken and Searle 1985: 7). This methodological view is grounded in two assumptions: First, if a sentence is uttered literally, “utterance meaning coincides with sentence meaning” (Searle 1979, 134); and if the sentence is sufficiently explicit, the meaning of the sentence is sufficient to determine the speech act performed in uttering the sentence. Therefore, the question as to what constitutes a speech act of a certain type can be answered by analysing the meaning of a sentence whose literal utterance
How to get lost in context: Searle on context, content and literal meaning
53
is the performance of a speech act of exactly that type.1 Accordingly, speech acts can be adequately characterised by analysing sentences whose literal utterance is sufficient to determine a speech act of a specific type. Second, semantic analysis can only lead to a full determination of the meaning of a speech act on the condition that, at least in principle, it is possible to find a sentence whose meaning is sufficient to determine exactly this particular speech act, if it is uttered literally. That is, it must, at least in principle, be possible to incorporate everything in a sentence that constitutes the meaning of the speech act. To this purpose, Searle argues for the Principle of Expressibility, according to which “whatever can be meant can be said” (Searle 1969: 19).2 It is this second assumption that allows Searle to state that “the study of the meaning of sentences is not in principle distinct from a study of speech acts” (Searle 1969: 18). The aforementioned assumptions about literal utterances of sentences and about the expressibility of what can be meant rest in turn on two crucial presuppositions: First, a sharp conceptual distinction needs to be drawn between sentences and utterances, and, correspondingly, between linguistic or word meaning and utterance meaning. As we have seen, if a sentence is uttered literally, linguistic meaning and utterance meaning coincide. But there are also non-literal uses of sentences like indirect speech acts, metaphors, etc. (see Searle 1979). In these cases, linguistic meaning and utterance meaning do not coincide, and therefore they are conceptually different. Second, while a speaker’s intentions determine whether an utterance of a certain sentence is to be taken literally or not, it only depends on the linguistic meaning of the sentence what kind of speech act is performed by its literal utterance. This means that linguistic expressions fulfil their function irrespective of what a speaker intends to express on a certain occasion. The semantic properties of linguistic expressions are therefore autonomous from the pragmatic properties of speech acts. It is this very autonomy that allows Searle methodologically to base the theory of speech acts on the analysis of the meanings of sentences in order to determine the rules governing the performance of speech acts.
1 “So what I shall do in my analysis of illocutionary acts is unpack what constitutes the understanding of a literal utterance in terms of (some of) the rules concerning the elements of the uttered sentence and in terms of the hearer’s recognition of the sentence as subject to those rules” (Searle 1969: 47–48; see also Searle 1969: 17–21, 45, 57, 94, 126). 2 A more explicit version of the Principle of Expressibility can be found in Searle (1969: 20). I shall not examine the validity of the principle here, but see Binkley (1979), Kannetzky (2002), and especially Recanati (2003), who argues that the principle is not compatible with Searle’s context hypothesis.
54
Silvan Imhof
In what follows, I shall try to show that these two presuppositions inherent in Searle’s underlying methodological assumptions are not compatible with the general framework of his theory of speech acts. Neither the conceptual difference between sentence meaning and utterance meaning nor the autonomy of linguistic meaning from speech acts and speakers can be maintained, if one holds, like Searle, both an intentionalist account of speech acts and a strong hypothesis about the influence of context.
3 The theoretical framework: three theses The methodological problem of the theory of speech acts arises when the methodological presuppositions presented in the preceding section are combined with some of the theses that characterise Searle’s broader theoretical framework. Thus, in this section I shall expound the three relevant theses: the thesis about speech act meaning (section 3.1), the thesis about the reduction of speech act meaning to a speaker’s intentions (section 3.2), and the hypothesis about the context dependence of speech act meaning (section 3.3).
3.1 Speech Act Meaning On the one hand, the thesis of Speech Act Meaning reflects the fact that speech acts are, in general, performed by using language, and that the meaning of a speech act depends, at least in part, on the meanings of the expressions used to perform the speech act.3 On the other hand, language does no work on its own. Language is essentially a means of performing speech acts, and it is operative only when speakers use it to perform speech acts. Hence, the meaning of speech acts depends, at least in part, on the intentional use of language by speakers. Speech Act Meaning thus amounts to the thesis that the meaning of a speech act generally depends in part on the speaker who uses language, and in part on the meanings of the linguistic expressions used by the speaker. Accepting Speech Act Meaning is to consider language and language users as two factors that each independently contribute to the determination of speech act meaning (Searle 1969: 17–19, 44–45, 48–49). In particular, language is supposed
3 Henceforth, I shall use Speech Act Meaning (capitalised) as a label for Searle’s thesis about how the meaning of speech acts is constituted, while speech act meaning (uncapitalised) is supposed to mean the meaning of a speech act.
How to get lost in context: Searle on context, content and literal meaning
55
to make an autonomous contribution to speech act meaning. As we have seen in section 2, the basic methodological assumption of speech act theory requires the autonomy of language from a speaker’s intentions. It has now become clear that this requirement is also incorporated in Speech Act Meaning. It is for this reason that it is crucial to the theory of speech acts from a methodological as well as from a theoretical point of view that the autonomy of language can be ensured. I shall show, however, that this is not possible, if one accepts the other two theses of the theory.
3.2 Intentionalistic reduction According to the thesis of Intentionalistic Reduction, the theory of speech acts is a branch of the philosophy of mind (Searle 1983: 160). It is the thesis that “speaker’s meaning should be entirely definable in terms of more primitive forms of Intentionality” (Searle 1983: 160). The thesis is based on a paradigmatic feature of speech acts: they are about something or, in other words, they are directed at something. Due to this feature speech acts can be said to represent something, i.e. to be intentional. However, intentionality is not primarily a feature of speech acts. In fact, their intentionality is only derived from the intentionality of mental states. Intentionality is thus imposed on speech acts by speakers using language in order to express their mental states. It is part of Intentionalistic Reduction that intentional or representational content can be shared by speech acts and intentional, mental states. The notion of content covers in both cases those aspects of intentionality that determine what things or state of affairs a speech act or a mental state is about or directed at.4 In the case of speech acts, intentional content is a part of speech act meaning. According to the thesis of Speech Act Meaning, the intentional content generally depends on the linguistic meaning of the sentences used to perform a speech act. In the case of a literal utterance of a sentence, the content of the sentence, i.e. the semantic content, is the same as the content of the intentional state the speaker intends to express by uttering the sentence. It is important to notice that Intentionalistic Reduction contradicts neither Speech Act Meaning nor the requirement of the autonomy of language from a speaker’s intentions. Although the reductionist thesis assumes that the
4 Thus, it covers the descriptive meaning or descriptive content. In the case of speech acts, the other aspect of intentionality is the illocutionary force; in the case of intentional, mental states it is the propositional attitude. I will confine the discussion to intentional content. See Searle (1983: 4–13, 26–29, 160–179).
56
Silvan Imhof
i ntentionality of speech acts or speaker’s meaning can be reduced to a speaker’s mental intentionality, it does not assume that linguistic meaning is reducible to a speaker’s mental intentionality. On the contrary: as Searle puts it, it is only possible that speakers impose intentionality on linguistic items by performing speech acts, if there are linguistic expressions that have meanings independent of a speaker’s intentions (Searle 1983: 167, 176).
3.3 The context hypothesis The third relevant element of the theoretical framework is Searle’s Context Hypothesis. I shall not go into the details of Searle’s description of context here nor shall I defend or criticise the Context Hypothesis.5 Rather, I shall take it as a given assumption and only point to its main characteristics. The Context Hypothesis applies to speech acts as well as to intentional, mental states. I shall first consider it with respect to the primary form of intentionality, i.e. intentional, mental states. As regards intentional, mental states, the hypothesis can be put like this: intentional states are constituted by two factors, namely intentional content and psychological attitude (Searle 1983: 5–6). The former defines what a state represents or what it is about. Basically, an intentional state represents its conditions of satisfaction; e.g., a belief represents what makes it true (Searle 1983: 10–13). However, the intentional content is not sufficient to determine specific conditions of satisfaction by itself. As a further determining factor, context needs to be taken into account as well. As Searle understands it, context is neither the external, situational, environmental or social context, nor the context of the utterance or communication. Rather, he takes it to be something mental. It is constituted (a) by the whole of an individual’s intentional states, the network, and (b) by some basic mental, nonintentional capacities and practices, the background.6 (a) Although an individual intentional state has conditions of satisfaction, it only has them within a network of other intentional states (Searle 1983: 19–21, 141–142). An intentional state presupposes further intentional states, and
5 For critical discussions see, e.g. Carston (1991), Recanati (1991), Stroud (1991), and Searle’s defences in Searle (1980, 1991, 1992). 6 I will use context as comprising both, background and network. Initially, Searle did not draw a distinction between network and background, and just spoke of contextual and background assumptions (see Searle 1979: 117–136). It was only in his later book Intentionality that he distinguished between background and network (Searle 1983: 141–142). Since The rediscovery of the mind, the network has been regarded as a part of the background (Searle 1992: 186–191).
How to get lost in context: Searle on context, content and literal meaning
57
the conditions of satisfaction of the state in question depend on what these further states are about. For example, my intention to become president of Switzerland presupposes some beliefs about the Swiss political system and involves further intentions like the intention to become a candidate for presidency or to overturn the government. Depending on what more I believe and intend, my intention to become president will have different conditions of satisfaction. So the general point is that the conditions of satisfaction of an individual intentional state are determined in a holistic way. (b) Intentionality or representation would not come about without a background of non-intentional capacities and practices relating to “how things are” and “how to do things” (Searle 1983: 143–144; see also Searle 1980: 227, 1995: 129, 1998: 109). In order to have the intention to drink a bottle of beer, for example, one must be acquainted with the behaviour of solid objects and liquids as well as with the handling of bottles. According to Searle, these non-intentional, mental capacities and practices provide “a set of enabling conditions that make it possible for particular forms of Intentionality to function” (Searle 1983: 157). Although background does not determine specific intentional states and their conditions of satisfaction, any intentional state is formed on the basis of a specific background and, therefore, has conditions of satisfaction only relative to this background. Thus, one can only have intentional states relative to a specific background. Because Intentionalistic Reduction holds, there is an important follow‐up: the Context Hypothesis directly applies to speech acts, because the intentionality of speech acts is derived from mental intentionality. Like mental states, speech acts have an intentional content and they represent their conditions of s atisfaction (Searle 1983: 10–11). Furthermore, speech acts share their intentional content with the mental states they are supposed to express. That means, however, that the content of a speech act, like the content of an intentional state, determines conditions of satisfaction only relative to a context. Thus, in order to determine the conditions of satisfaction of a speech act, the mental context needs to be taken into account as well.
4 The context hypothesis applied to semantic content After having established the methodological and theoretical framework, I shall now show step by step that within this framework it is not possible to uphold the
58
Silvan Imhof
two basic assumptions in a consistent way: the conceptual difference between sentence meaning and utterance meaning on the one hand, and the autonomy of linguistic meaning on the other hand. At this point it will emerge that the notion of semantic content is of special importance. For this reason I shall first sharpen Searle’s conception of semantic content before I present my argument. The best way to do so is to have a look at Searle’s initial application of the Context Hypothesis. The Context Hypothesis was first intended to criticise the traditional view of semantics holding that the meaning of a sentence is sufficient to determine the truth conditions, or, more generally, the conditions of satisfaction for every literal utterance of the sentence. According to the Determination View, the meaning of a sentence is a function of the meanings of its component expressions and its compositional structure (Searle 1979: 117–119, 1980: 223).7 If these elements and thus the semantic content of a sentence are known, also the meaning of any literal utterance of this sentence, i.e. its literal meaning, is known. In other words, the conditions of satisfaction the sentence has in every utterance in which it is uttered literally are known. Arguing against this view Searle claims that knowledge of the semantic content of a sentence will never be sufficient to determine the conditions of satisfaction for any literal utterance of that same sentence. A sentence or its semantic content determines the conditions of satisfaction only relative to a specific mental context. Therefore, without knowledge of the speaker’s mental context, it is not possible to know the conditions of satisfaction of a literal utterance of a sentence. Of course, the Determination View could still be saved, if there was a way of describing the relevant context, and, in doing so, making all the factors explicit that determine the conditions of satisfaction. However, examples provided by Searle indicate that context can never be fully known for two reasons. First, no exhaustive description of the relevant network and background of a specific utterance can be given. Second, there is no exhaustive specification of all the contexts in which a sentence can potentially determine different conditions of satisfaction, when it is uttered literally. As a consequence, no semantic description can capture all the relevant factors that determine the conditions of satisfaction, because one of these factors, i.e. context, is essentially pragmatic. Therefore, the Determination View is misguided. The argument reveals two crucial points about Searle’s conception of content. First, Searle attacks the Determination View because it does not acknowledge the influence of context, but he does not criticise the underlying notion of semantic
7 I take this label from Recanati (2003: 189–190). It covers not only truth conditional semantics, but also the extension of this account to conditions of satisfaction in general.
How to get lost in context: Searle on context, content and literal meaning
59
content. In fact, Searle agrees with the Determination View that semantic content is determined by the linguistic meanings of expressions and is thus independent of the speaker (Searle 1979: 120, 132–133, 1992: 184). Furthermore, the semantic content of a sentence is context-neutral, and it is constant and invariant in every utterance. This means that the contribution of the semantic content to the determination of the conditions of satisfaction is the same in every literal utterance (Searle 1979: 121, 125, 129, 1980: 224–227, 1983: 146, 1992: 178–179, 1995: 131–132). Hence the point of disagreement between Searle and the proponents of the Determination View is not so much about what semantic content is, but rather about whether it is sufficient to determine conditions of satisfaction or not. Second, there is an important difference between intentional states and sentences. Intentional states are intrinsically situated in a mental context, because not one single intentional state could exist without context. Sentences, by contrast, are not intrinsically connected to a context. Rather, they are supposed to be context-neutral. It is only when they are used to perform speech acts that they are related (or “applied”, as Searle says) to a context, namely the mental context of the speaker. Thus, in the case of literal utterances of sentences there are always two distinct factors involved: semantic content, which is determined by linguistic meaning, and mental context, which depends on the speaker. No similar distinction can be drawn in the case of intentional states, because both content and context belong to the same holistic mental intentionality of a speaker. The purpose of this section was to show that Searle regards semantic content not only as a factor that determines the conditions of satisfaction of a speech act, but also as a factor completely separable from and independent of the context. Obviously, this view is in perfect agreement with Searle’s presupposition of the autonomy of language that is incorporated in both the basic methodological assumption and the thesis of Speech Act Meaning. All the same, in the following sections I shall show that this traditional view of semantic content cannot figure in the theoretical framework of Searle’s theory of speech acts. If both Intentionalistic Reduction and the Context Hypothesis were true, then semantic content would simply vanish into context. If this is correct, the autonomy of language can no longer be upheld.
5 The roots of semantic concepts in mental intentionality My first point does not amount to an argument against Searle’s conception of semantic content. It is more like a reminder that one has to be careful about
60
Silvan Imhof
the conceptual dependencies of semantic concepts, if one adheres to Speech Act Meaning as well as Intentionalistic Reduction. I have already mentioned (see section 3.2) that Intentionalistic Reduction is compatible with Speech Act Meaning: although speaker’s meaning is reducible to a speaker’s mental intentionality, linguistic meaning is not reducible to a speaker’s intentionality. The autonomy of language is thus not directly threatened. Still, the question arises whether there is a conceptual connection between the semantic properties of sentences and mental intentionality for the following reason: Searle assumes that in a literal utterance of a sentence the meaning of the sentence determines the intentional content of the speech act, i.e. the content of the mental state the speaker intends to express. It might be suspected, however, that sentences can play this part only, because there is a close relation between basic semantic properties and those properties of mental states that make them intentional. Thus, if semantic properties actually do contribute to the determination of the intentional content of speech acts, they must in some sense be derived from the primary form of intentionality (Searle 1979: 130, 1995: 61, 99, 1998: 93). This means that semantic properties have to be understood in terms of mental intentionality. If, though, semantic properties can really be traced back to mental intentionality, the autonomy of language is no longer guaranteed. I shall now try to corroborate these suspicions. The semantic properties that contribute to content can be roughly reduced to the referential and the predicative meaning of expressions. Within the theory of speech acts, reference and predication are understood as the speech acts of reference and predication. They determine the content of the intentional state that is expressed in the performance of a full-blown speech act (Searle 1969: 23–24, 121–123). While in intentional, mental states there are no analogues for referential and predicative expressions or for referential and predicative speech acts, there must at least be functional analogues for reference and predication. Without such analogous functions, it would not be possible to say what it is for an intentional state to be about or to be directed at something, i.e. what it is for an intentional state to have intentional content. The functions of reference and predication are thus essentially required to determine intentional content. And this is why there is no fundamental difference between how the content of speech acts and the content of mental states are constituted, apart from the fact that the constitutive functions of reference and predication are performed in different ways. In this sense, the semantic concepts of referential and predicative meaning as well as the concept of semantic content can be reduced to the primary intentional concepts of reference, predication and intentional content. This conceptual reduction is not yet devastating for the assumption of the autonomy of language, but it shows that the autonomy of language cannot be taken for granted. This
How to get lost in context: Searle on context, content and literal meaning
61
means that an account has to be given that explains how language can contribute autonomously to the determination of content, although the basic semantic concepts are rooted in the primary concepts of mental intentionality. In order to tackle this task, it is necessary to rule out the possibility that the intentional properties of any particular speech act completely and directly depend on the mental intentionality of the performer of the speech act. If this possibility can be ruled out, it is permissible to assume that there really is an autonomous contribution of language to the intentionality of each particular speech act, although semantic properties in general can be reduced to the mental intentionality of speakers. Without a doubt, there are solutions for this task. In The construction of social reality Searle treats language as a part of institutional reality and in doing so takes collective intentionality to be a constitutive factor of language (Searle 1995: 13, 72–76, 1998: 153). If the semantic features of language are to be understood as the product of the collective intentions of a plurality of language users, the semantic properties of language are obviously rooted in their mental intentionality. Nevertheless, the contribution of language to the intentionality of a particular speech act does not depend on the intentionality of a particular speaker. Because the meaning of a sentence is based on collective intentionality, sentence meaning cannot be reduced to a particular speaker’s intentions. In this sense, language is independent of particular speakers. If backed by a solution like this, language could reasonably be said to be autonomous in the sense required by the methodological assumptions of speech act analysis.8 It seems that in spite of Intentionalistic Reduction the autonomy of language can be saved with reasonable theoretical effort. But what happens, if we take the Context Hypothesis into account as well? It may turn out that some of the relevant features of mental intentionality are essentially context-dependent. In this case, because of the conceptual dependence of semantic concepts on concepts of mental intentionality, it will be hard to show that context dependence can be excluded from the pertinent properties of language. If, however, these properties depend on context and the relevant context is the mental context of individual speakers, there will be an essential dependence of semantic properties on an individual speaker’s intentionality. In this case, the autonomy of language and the conception of a constant, context‐independent content will get into trouble.
8 Unfortunately, Searle does not really provide such an account. He rather relies on the concept of convention when he explains the status of public symbols like linguistic expressions (Searle 1992: 60–61, 72–76, 1998: 141, 153), and it is far from clear how conventions are related to mental intentionality.
62
Silvan Imhof
6 The interaction of context, content and conditions of satisfaction Searle shares the assumption of the Determination View that a sentence has a specific semantic content that contributes to the conditions of satisfaction and remains the same in every literal utterance of the sentence (section 4). This assumption has not been seriously challenged by my examination of the consequences of the thesis of Intentionalistic Reduction (section 5). What I shall show next is that, if the Context Hypothesis is accepted too, not only conditions of satisfaction essentially depend on context, but also the intentional content. In fact, it is not possible to separate the intentional content from context and to treat content and context as two independent determining factors of conditions of satisfaction. Intentional content in general and semantic content in particular depend on context and, therefore, on mental intentionality. As a consequence, semantic content is not autonomous from a speaker in the way that speech act analysis requires.
6.1 Searle’s view In order to argue for the Context Hypothesis, Searle gives different examples of the phenomenon that conditions of satisfaction vary between different literal utterances of the same sentence. I shall use a similar example in order to challenge Searle’s explanation of the phenomenon.9
Fig. 1: Cats on mats.
9 The scenario is a variation of examples in Searle (1979: 122–123).
How to get lost in context: Searle on context, content and literal meaning
63
Figure 1 illustrates the following situation: a person on earth who utters the sentence The cat is on the mat literally, says something true in the cat1-situation as well as in the cat2-situation. He uses is on in The cat is on the mat contextually relative to the direction of gravity on earth. In contrast, after a long intergalactic voyage an astronaut inside the spaceship has become accustomed to use is on contextually relative to the position of his spaceship, taking the tail fin for determining the up-direction. So the astronaut’s literal utterance of The cat is on the mat is true of cat1, but not of cat2. In the latter case, cat2 would be under mat2 relative to the astronaut’s context, and hence the utterance would be false. Thus, the respective truth values and truth conditions differ, although both speakers utter the same sentence literally.10 According to Searle, the example has to be interpreted like this: obviously, if the person on earth and the astronaut utter the sentence The cat is on the mat, their respective utterances have something in common (Searle 1979: 125, 128, 1980: 222–227, 1983: 145, 1992: 177–178, 1995: 130–131). Both speakers use the same sentence with the same expression meanings, and they both utter it literally. Accordingly, they both say the same about the same things (that, e.g., cat2 is on mat2), and thus utter the same sentence with the same semantic content. It is also obvious that truth values and truth conditions differ, although both speakers utter the same sentence with the same semantic content in the same situation. If, though, the semantic content is constant in all utterances, the difference between the truth values and truth conditions must be explained by something else. So, another factor needs to be taken into account, namely the speaker’s context. Searle’s interpretation thus leads to the following general conclusion: when uttered literally, the same sentence with the same semantic content can determine different conditions of satisfaction relative to different contexts.
6.2 Searle’s presupposition: the independence of content and context Searle’s explanation of the phenomenon is based on a crucial presupposition. He presupposes that it is possible to separate semantic content and speaker’s context, that is, that content and context are conceptually independent. While the
10 Note that the coincidence of the truth values of the utterances of both speakers in the cat1situation is only accidental. Just imagine the spaceship performing a roll of 180 degrees along its longitudinal axis.
64
Silvan Imhof
determination of truth conditions or, generally speaking, conditions of satisfaction requires context, content is supposed to be fixed exclusively by the semantic properties of the sentence, if the sentence is uttered literally. However, the possibility of the separation of content and context in turn implies the possibility of a clear separation of content and conditions of satisfaction, and, in fact, the conceptual independence of content from conditions of satisfaction.11 Only if the separation of content and conditions of satisfaction is possible can the semantic content of a sentence make the same contribution to speech act meaning in every literal utterance of the sentence, even though the conditions of satisfaction may vary. Finally, the conceptual independence of content from conditions of satisfaction in turn requires the possibility of the determination of content independently of conditions of satisfaction. In other words, the identity of conditions of satisfaction cannot be a criterion for the identity of content. The question then is if there is another criterion for the identity of content available?
6.3 The notion of content There is an obvious criterion for the identity of content: the same content is expressed, if the same is said about the same things, that is, if reference and predication are the same. As we have seen in section 5, the functions of reference and predication are constitutive of intentional content in general. In our example, the expressions the cat and the mat are used by both speakers to refer to the same cat (cat1 or cat2) and to the same mat (mat1 or mat2); and by means of is on the same relation (to be on) is ascribed to the cats and the mats in the utterances of both speakers, irrespective of their individual context. If both speakers use the same expressions to refer to the same things and to ascribe the same relation to these things, they both say the same about these things in virtue of the identical referential and predicative meanings of the expressions they both use. They both seem to perform the same acts of reference and predication, and thus to express the same content (namely that the cat is on the mat). Everything else being equal, only their different contexts can account for the difference in the conditions of satisfaction. Reference and predication are obviously suitable to identify content. Nevertheless, any specification of content by means of reference and predication is independent of context only if reference and predication themselves are independent
11 Note that the independence is not mutual, because conditions of satisfaction depend, among other things, on content.
How to get lost in context: Searle on context, content and literal meaning
65
of context. But they are not, as Searle explicitly remarks on reference: “Since linguistic reference is always dependent on or is a form of mental reference and since mental reference is always in virtue of Intentional content including Background and Network, proper names must in some way depend on Intentional content” (Searle 1983: 232). This means that even referring to something presupposes intentional states like beliefs about the intended object of reference (network) and basic abilities like the perception of objects (background). Although Searle makes no analogous statement about predication, it is plausible to assume that context is involved in predication too. Predicating something of something involves further intentional states like knowing the possible bearers of the property in question (network) and basic abilities like taking objects as bearers of properties (background). Since reference and predication are essentially contextual, there is no bare, context-free reference and no bare, context‐free predication; and since reference and predication are those intentional functions that constitute intentional content, there is no bare, context-free content too. The conclusion is that there is simply no intentional content without context. Intentional content is intrinsically related to a context, and it is constituted only relative to a context. Therefore, content cannot be separated from context. Furthermore, content cannot be separated from conditions of satisfaction either. If content is intrinsically related to a context, conditions of satisfaction are given as soon as some content is given. This conclusion indicates that Searle’s notion of content and, consequently, his interpretation of the context dependence of conditions of satisfaction need to be revised.
6.4 A revised view Starting off with the conclusion of the previous section, one gets a picture of the interaction of content, context and conditions of satisfaction that is quite different from the one Searle presents. (a) Different speakers have different intentional contents. In our example, both persons apply the concept of being on relative to their respective individual contexts, and, thereby, the content of their respective beliefs is fixed. If one knows, among other things, how they each apply the concept of being on, one also knows the content of their beliefs and how the content has to be taken in each case, i.e. that the concept of being on is applied relative to the direction of gravity on earth in one case, and relative to the position of the tail fin of the spaceship in the other. If, furthermore, one knows how the content of each utterance has to be taken as soon as reference and predication are established,
66
Silvan Imhof
one also knows the different truth conditions of the two different beliefs. That means, if one knows the content, one also knows the truth conditions, and vice versa. And if truth conditions vary, content must vary too, and vice versa. So the person on earth and the astronaut have different beliefs about the cats and the mats. (b) Different speakers express different contents. According to (a), it is not justifiable to assume that both speakers express the same content in their utterance of The cat is on the mat, only with different truth conditions depending on their respective contexts. Since the identity of content depends on the identity of reference and predication, and since both speakers perform at least different acts of predication owing to their different contexts, it follows that both speakers express different contents. The astronaut does not mean the same as the person on earth, when he utters the sentence The cat is on the mat, irrespective of the occasion he does so. But is the meaning of the sentence they use to express their beliefs still the same? Do both speakers say the same about the cats and mats in question, because the sentence they use has the same, constant semantic content? (c) Different speakers say something different. If linguistic reference and predication are derived from mental reference and predication as required by Intentionalistic Reduction, and if mental reference and predication essentially depend on context, then linguistic reference and predication depend on context too (Searle 1983: 232). And as the semantic content of a sentence is constituted by the referential and the predicative meanings of its constituent expressions, semantic content essentially depends on context as well. What has been said about the context dependence of intentional content in general also applies to semantic content. Most importantly, without context there cannot be any semantic content at all. According to the revised interpretation of the context phenomenon, the person on earth and the astronaut do not say the same, when they utter the sentence The cat is on the mat, because the conditions of satisfaction differ owing to their different contexts. The same sentence is used differently by both speakers, that is, it has different meanings depending on each speaker’s context. Therefore, the semantic content of the sentence cannot be regarded as being constant and context‐independent in every literal utterance of the sentence in whatever context. It might still be tenable to say that, in a certain sense, all utterances in our example are cases of literal utterances of the same sentence, because in every case what the speaker means corresponds to what the sentence means. This, however, is a void sense of literal meaning, since in every case what the sentence means does not only correspond to, but is completely determined by what the speaker
How to get lost in context: Searle on context, content and literal meaning
67
means. According to the corrected interpretation, which is based on the theses of Searle’s theoretical framework, sentences do not have a constant, contextindependent semantic content. Rather, there is an indefinite ambiguity resulting from the influence of context. For Searle, this kind of ambiguity is an “absurd result” (Searle 1983: 145).12 In this sense, which is the only sense available, the concepts of semantic content and literal meaning are certainly not very useful for semantic analysis.
7 The dissolution of semantic content in context 7.1 A possible way out Even if one accepts the theses of Searle’s theoretical framework and their consequences, one can still try to defend a conception of semantic content that is constant and neutral with regard to context. It may be true, one could say, that there is an ambiguity inherent in linguistic expressions. The sentence The cat is on the mat may determine different contents and different conditions of satisfaction relative to different contexts, but these contents are not entirely different and independent. Rather, they have something in common, since both speakers in our example, the person on earth and the astronaut, say something about the same objects (the two cats and mats) and ascribe the same relation to these objects (to be on). Hence, it should be possible to abstract the contextual features of literal utterances of a sentence that are responsible for the ambiguity, and thus to reach a core of meaning that is neutral with regard to context. This abstract core of meaning can plausibly be considered to be the invariant semantic content of the sentence.
12 Searle counters alternative interpretations of the phenomenon that ascribe the context sensitivity of conditions of satisfaction to semantic properties by making use of such notions as semantic ambiguity, variable functions, vagueness, incompleteness or hidden indexicality (Searle 1980: 224–225, 228–231, 1992: 182–186). It is worth mentioning that the alternative interpretation I suggest here does not rely on the notion of (semantic) ambiguity, but on the concepts of content, reference, predication and conditions of satisfaction as they figure in Searle’s theoretical framework. Ambiguity is not part of the theoretical explanation of the phenomenon, but rather an undesired result, similar to Kripke/Wittgenstein-style scepticism according to which one “would then be able to say anything and mean anything”, and “where meaning is concerned, anything goes” (Searle 1992: 184).
68
Silvan Imhof
Concerning this attempt to defend the concept of constant, context‐independent content I shall first give a general reply and then demonstrate what comes out, if the suggested procedure of abstracting contextual factors is applied to the situation illustrated in Fig. 1. Neither the general reply nor the demonstration based on the example will probably be decisive, but they will suffice, first, to cast serious doubt on the applicability of the procedure in general, and, second, to show that, even if the procedure could be applied, the result would not support a notion of semantic content that is useful from a methodological point of view.
7.2 A general reply The general reply is based on Searle’s claims about the range of the influence that context exerts. As long as one is confronted with only a few cases of literal utterances of the same sentence with different conditions of satisfaction, it may seem to be feasible to factor out the contextual differences in order to attain a common content. In our example, there is only one relevant contextual factor, namely the speakers’ different background regarding spatial orientation. If this difference is factored out, everything else remains the same for both speakers. It is, however, an important point of Searle’s argument in support of the Context Hypothesis that examples of deviant cases can be multiplied ad libitum. There is an indefinite number of possible contexts. This point is important, because it supports the claim of a general, pervading influence of context on conditions of satisfaction on the basis of a few arbitrary examples; and it also supports the argument against the Determination View holding that it is impossible to give an exhaustive description of all contextual factors that may be relevant to the determination of conditions of satisfaction (Searle 1979: 125–126, 128–130, 1980: 228, 1983: 148; 1995: 131). As a consequence, by applying the suggested procedure of abstraction, the contextual features that actually make a difference in a particular case can perhaps be sorted out. Nevertheless, this does not lead to constant, context- independent content, because there still remains an indefinite number of possible contextual features that may exert an influence on the determination of conditions of satisfaction, even though they make no difference in the case in hand. Thus, there is a conflict between the view that contextual factors cannot be specified exhaustively, which is part of the Context Hypothesis, and the suggestion that the procedure of abstracting from contextual factors can lead to constant, context-independent content. In the light of the argument presented in section 6, holding that without context there can be no content at all, the conflict appears to be even more striking. Therefore, the suggested procedure of
How to get lost in context: Searle on context, content and literal meaning
69
abstraction is no useful strategy for defending the concept of constant, context‐ independent content within Searle’s theoretical framework. Searle’s Context Hypothesis is radical and, therefore, undermines the concept of semantic content in a radical way.13 Let us now see what happens if one tries to apply the suggested procedure to the situation depicted in Fig. 1.
7.3 Applying the procedure Step 1. The only relevant contextual difference in the example is the different background regarding the spatial orientation of both speakers, which manifests itself in their use of is on. With everything else being equal, this difference must explain the different truth conditions. Thus, the determination of the spatial relation expressed by the literal use of is on relative to the centre of gravity of the earth or relative to the position of the tail fin of the spaceship is not part of the meaning of is on and has to be factored out. What remains as the semantic meaning of is on may be roughly described like this: is on is a two-place predicate which is (in a literal utterance) used to ascribe a certain spatial relation to physical objects. This description is rather unspecific because it only refers to a certain spatial relation without saying what kind of spatial relation it actually is. But if it only refers to an unspecified spatial relation, the description of the meaning of is on coincides with the descriptions of the meaning of any other expression the two speakers use to ascribe a spatial relation to physical objects, for instance is under, is beside of, etc. Strictly speaking, all these expressions have to be taken to be synonymous. It must be admitted therefore that the sentence The cat is on the mat has the same semantic content as The cat is under the mat, and that both sentences make the same contribution to the determination of truth conditions. Any use of is on counts as literal as long as it is used to ascribe whatever spatial relation to physical bodies, because the semantic content does not fix the specific differences in meaning between different expressions used for the ascription of spatial relations. The differences only occur when a particular speaker applies such expressions relative to his specific context.
13 Searle sometimes seems to acknowledge this point himself: “I wish to say that there is a radical underdetermination of what is said by the literal meaning of this sentence” (Searle 1995: 131; see also 1992: 181).
70
Silvan Imhof
Although this first step merely consists in abstracting just one contextual factor, the remaining content has already turned out to be quite unspecific and abstract. Yet further steps follow. Step 2. The defender of constant semantic content could remind us that the linguistic meaning of is on should not be considered in isolation. The expression should rather be seen as placed in a semantic system of expressions used to ascribe spatial relations. For example, if is on is used literally, it can only be used to express the opposite direction of is under for semantic reasons. There are semantic restrictions imposed on the use of is on which prevent its arbitrary use and which are part of its meaning. Expressions do not have a meaning in isolation, but only relative to whole systems of meaningful expressions like the system of expressions used to ascribe spatial relations. In our example, when the speakers use is on, they apply this whole semantic system. Although the system is applied relative to different contexts by both speakers, they both apply the same semantic system. The semantic system as a whole is invariant in the different cases. Still, what if the astronaut has completely adapted to relativistic space-time instead of our everyday earthly idea of space? In this case, although the astronaut still uses the same expressions for the ascription of spatial relations literally, he does not apply the same system of such expressions as the person on earth. He applies a different semantic system with different semantic restrictions on the use of spatial expressions. Not only are there different semantic restrictions in the different systems, but also the systems themselves are dependent on different contexts. They can only be applied relative to different beliefs and background assumptions concerning the properties of space. That means that a semantic system as a whole only works relative to a context. What seemed to be the constant semantic core of meaning after the first step of abstraction has now turned out to be context-dependent after all. Step 3. The defender of constant semantic content could now argue that we just have to dig deeper. Semantic content would then be even less substantial, but one could not exclude the possibility that after further steps we were finally to reach the semantic core that is invariant relative to all possible contexts. All the same, I shall not give a further reply, and stop my demonstration at this point. For my purpose, it suffices that I have raised the suspicion that the game could go on ad indefinitum. It is possible that whatever appears to be constant, context-independent semantic content at one point of the procedure can turn out to depend on context, because certain contextual factors have not yet been accounted for. No matter how deep one digs, one can never be sure of having reached semantic bedrock. And even if one reaches semantic bedrock at a certain point, the deeper it lies, the more abstract, unspecific and unsubstantial semantic content will be.
How to get lost in context: Searle on context, content and literal meaning
71
8 Conclusions Neither the general reply (section 7.2) nor the outcome of the attempt to apply the procedure of abstracting contextual features (section 7.3) is decisive in arguing against the assumption of constant, context-independent semantic content. There is probably no strict proof of the Context Hypothesis, and the defender of the concept of constant, context-independent content can still adhere to the suggested procedure of abstraction. All the same, my aim was to show that within Searle’s theoretical framework, the concept of constant, context‐independent semantic content as it is methodologically required for speech act analysis is not available. Although what has been said so far is not sufficient to dismiss the concept of constant, context-independent content altogether, it is sufficient to show that the concept, should it be available, cannot fulfil the methodological function it is intended to fulfil for the following reasons: (1) The more general and pervasive the influence of context is supposed to be, the less specific and substantial semantic content can be. Searle’s Context Hypothesis states a very general and radical influence of context. Right from the outset, it leaves not enough room for semantic content, if any, as my general reply indicated. (2) Even if the influence of context did not necessarily result in the complete dissolution of semantic content, and if it was possible to apply the suggested procedure of subtracting contextual factors from the total meaning, one could not know when one has reached constant semantic content. Since it is conceivable that there are indefinitely many possible contexts relative to which the same sentence can be uttered literally, one can never be sure that all possible contextual differences have actually been identified and factored out. So one can never be sure then, whether the procedure provides a suitable analysandum for semantic analysis. (3) All the same, it may be possible that, by applying the procedure, some abstract level can finally be reached where all contextual factors are eliminated. Semantic content might be given in terms of mere logical or formal or structural properties, and thus proper analysanda for speech act analysis would have been found. However, semantic content at this level would be very abstract and thin, and could therefore have almost no bearing on the meaning of speech acts. It is hard to see, then, how the analysis of semantic content of such an abstract kind could substantiate the claims of the theory of speech acts, which are quite concrete and specific about the nature of particular types of speech acts. (4) There is no guarantee, however, that there really is a final level where all contextual influence can be factored out. Even if one is left with nothing but
72
Silvan Imhof
logical, formal or structural properties to account for semantic content, one has still to expect the question whether the understanding of these properties does not presuppose some fundamental mental background abilities that enable speakers to apply logical, formal or structural operations. If this is the case, context is at work even at the most abstract levels. Lacking a criterion for deciding when content is free of context, there is no way of clearly discriminating content from context. This leads to the conclusion that the procedure of factoring out contextual differences in order to reach semantic content is neither applicable in practice nor in theory. In practice, no point can be determined where the procedure comes to a halt. In theory, it does not provide an operable conceptual distinction between contextual factors and neutral semantic content within the total meaning of a literal utterance of a sentence. (5) Finally, all this leads to the conclusion that, against the background of Searle’s theoretical framework, there are no proper analysanda for speech act analysis. Constant, context-independent semantic content is either too thin to fulfil the purposes of speech act analysis or it is not available at all. Therefore, the semantic analysis of sentences whose literal utterance constitutes the performance of certain speech acts in virtue of their meanings is, from a methodological point of view, a non-starter.
References Binkley, Timothy. 1979. The principle of expressibility. Philosophy and Phenomenological Research 39(3). 307–325. Carston, Robyn. 1991. Implicature, explicature, and truth-theoretic semantics. In Steven Davis (ed.), Pragmatics. A reader, 33–51. New York & Oxford: Oxford University Press. Kannetzky, Frank. 2002. Expressibility, explicability, and taxonomy. Some remarks on the principle of expressibility. In Günther Grewendorf & Georg Meggle (eds.), Speech acts, mind, and social reality. Discussions with John R. Searle, 65–82. Dordrecht, Boston & London: Kluwer. Recanati, François. 1991. The pragmatics of what is said. In Steven Davis (ed.), Pragmatics. A reader, 97–120. New York & Oxford: Oxford University Press. Recanati, François. 2003. The limits of expressibility. In Barry Smith (ed.), John Searle, 189–213. Cambridge: Cambridge University Press. Searle, John R. 1969. Speech acts. An essay in the philosophy of language. Cambridge: Cambridge University Press. Searle, John R. 1979. Expression and meaning. Studies in the theory of speech acts. Cambridge: Cambridge University Press.
How to get lost in context: Searle on context, content and literal meaning
73
Searle, John R. 1980. The background of meaning. In Manfred Bierwisch, Ferenc Kiefer & John R. Searle (eds.), Speech act theory and pragmatics, 221–232. Dordrecht, Boston & London: Reidel. Searle, John R. 1983. Intentionality. An essay in the philosophy of mind. Cambridge: Cambridge University Press. Searle, John R. 1991. Response: The background of intentionality and action. In Ernest Lepore & Robert Van Gulick (eds.), John Searle and his critics, 289–299. Cambridge: Blackwell. Searle, John R. 1992. The rediscovery of the mind. Cambridge, MA: MIT Press. Searle, John R. 1995. The construction of social reality. London: Allen Lane. Searle, John R. 1998. Mind, language and society. Philosophy in the real world. New York: Basic Books. Stroud, Barry. 1991. The background of thought. In Ernest Lepore & Robert Van Gulick (eds.), John Searle and his critics, 245–258. Cambridge: Blackwell. Vanderveken, Daniel. 1981. Pragmatique, Sémantique et Force Illocutoire. Philosophica 27(1). 107–126. Vanderveken, Daniel. 1992. On the unification of speech act theory and formal semantics. In Philip R. Cohen, Jerry Morgan & Martha E. Pollack (eds.), Intentions in communication, 2nd ed., 195–220. Cambridge, MA & London: MIT Press. Vanderveken, Daniel & John R. Searle. 1985. Foundations of illocutionary logic. Cambridge, New York & Melbourne: Cambridge University Press.
Nikola Kompa
Meaning and interpretation Abstract: My aim in the paper is to sketch an answer to the question what knowledge speaker and interpreter are required to have for communication to be successful. Obviously, I will not be able to provide a fully worked-out epistemology of language understanding. Nevertheless, I will try to indicate the direction in which an answer to this question may be found. A prima facie plausible answer to the question seems to be that speaker and interpreter ought to be in possession of a theory of (type) meaning for language L. Yet Davidson famously denied this assumption. In the second part of the paper I will address Davidson’s worry. Given that a smooth transition from type meaning to token meaning is thought to be hampered by the context sensitivity of linguistic interpretation, I will, in the third part of the paper, scrutinize the context sensitivity of linguistic interpretation more closely: How can context sensitivity be characterized, and how can it be resolved? In light of the (allegedly) massive context sensitivity of linguistic interpretation, some philosophers and linguists try to do without type meaning in anything like the traditional sense altogether. A fairly radical version of this idea can be found in Agustín Rayo’s grab-bag model of language-mastery. In the fourth part of the paper, the model will be discussed in more detail – and it will be found wanting, in certain respects at least. In the fifth part of the paper I will attempt to amend the grab-bag model. There is no doubt an Art in saying something when there is nothing to be said, but it is equally certain that there is an Art no less important of saying clearly what one wishes to say when there is an abundance of material; and conversation will seldom attain even the level of an intellectual pastime if adequate methods of Interpretation are not also available. C. K. Ogden & I.A. Richards, The Meaning of Meaning
1 Introduction Here is a question that I should like to see answered: What does an interpreter need to know in order to be able to successfully interpret a given utterance? As it stands, this is the topic of a research project rather than a question to be
DOI 10.1515/9781501504327-005
76
Nikola Kompa
addressed and answered in a single paper.1 I will, therefore, begin by trying to bring it in more tractable shape (§2); the aim of the paper is to thereby make a small contribution to the overall project. In the third part of the paper, the context sensitivity of linguistic interpretation will be investigated and the idea of successful interpretation will be elaborated on. In the fourth part of the paper, the grab-bag-model of language mastery will be discussed. The remainder of the paper (§5) will be devoted to the question of which methodology one ought to adopt if one’s aim is to develop an understanding of how linguistic interpretation is achieved.
2 Type meaning A prima facie plausible answer to the initial question seems to be that the interpreter needs to be in possession of a theory of meaning. Different things go by the name “theory of meaning”, though. According to Jeff Speaks, “theory of meaning” is ambiguous between the following two readings: The first sort of theory – a semantic theory – is a theory which assigns semantic contents to expressions of a language. [...] The second sort of theory – a foundational theory of meaning – is a theory which states the facts in virtue of which expressions have the semantic contents that they have. (Speaks 2010: 1)
The second kind of theory is clearly not what a hearer needs to possess in order to successfully interpret a speaker’s utterance. What about the first kind of theory? Theories of the first kind – semantic theories – assign semantic content to expressions. But, as Speaks points out, when philosophers try to conceive of a semantic theory, they are not so much interested in a semantic theory for English or French or any other particular language, but rather in what might be the right form for a semantic theory to take. They don’t answer questions like “What does run mean?” Rather, they ask, e.g., what kind of meaning or semantic value a semantic theory ought to assign to a particular expression. And they might differ over whether a theory ought to assign extensions or intensions, over whether the semantic values of sentences are structured or unstructured
1 I will be concerned mostly with what Guy Longworth calls “ability-understanding” and “achievement- understanding” (Longworth 2008: 51). I will not have much to say on the question of what the output – what Guy Longworth calls “state-understanding” (Longworth 2008: 51) – of a particular episode of coming to understand an utterance might be.
Meaning and interpretation
77
propositions, etc. Yet this is clearly not what hearer or speaker need to know for communication or interpretation to be successful either. Being told that, e.g., the meaning of a general term is a function from possible world to extensions does not help much when the question is what a speaker means to communicate by a particular utterance. If at all, speaker and interpreter of utterances of a particular language L need to be in possession of a semantic theory for language L–a theory that tells them what the particular expressions of language L conventionally mean; how competent speakers of L commonly use the expressions in question. They ought to be in possession of a theory of (type) meaning for L; or so it seems. Davidson famously denied that this is so. At the end of his paper A Nice Derangement of Epitaphs Davidson concludes “that there is no such thing as a language, not if a language is anything like what many philosophers and linguists have supposed.” (Davidson 1986: 483) In the paper, Davidson is concerned with malapropisms (We are all cremated equal). But he takes the point to generalize. So does Barry Stroud: The point is – and of course it does not depend on malapropism – that we often understand what people say even when the words they use standardly mean something different from what we take the person to mean in uttering them on a particular occasion. (Stroud 1998: 198)
Davidson finds fault with the idea that speaker and hearer have to share a system of rules or conventions for interpretation to succeed. According to Davidson the interpreter comes to a particular utterance equipped with what he calls “a prior theory”. He then adjusts this theory to the occasion; the result being what he calls “a passing theory”, a theory that is concerned with what the words the speaker uses mean on that occasion: The prior theory expresses how he is prepared in advance to interpret an utterance of the speaker, while the passing theory is how he does interpret the utterance. (Davidson 1986: 480)
All that must be shared for communication to succeed is, according to Davidson, a passing theory; a theory of meaning that is geared to the occasion. Prior theories, on the other hand, will usually not be shared, “and there is no reason why they should be.” (Davidson 1986: 480) Consequently, what speaker and interpreter commonly share is, so Davidson, not something they both learned, no set of behavioral regularities or rules: For we have discovered no learnable common core of consistent behavior, no shared grammar or rules, no portable interpreting machine set to grind out the meaning of an
78
Nikola Kompa
arbitrary utterance. We may say that the linguistic ability is the ability to converge on a passing theory from time to time [...]. (Davidson 1986: 482)2
On the assumption that what Davidson denies to have found is what is often called a theory of meaning of expression types (type meaning, for short), his point might be put thus: there is no substantive role for type meaning to play in interpretation. In order for communication to succeed an interpreter has to figure out the meaning of the expression as used on that occasion. She has to figure out the meaning of the expression token (token meaning, for short). One might question the validity of Davidson’s reasoning. One might claim, for example, that in order to recognize a malapropism as such an interpreter has to observe that a particular regularity in usage, a convention or rule, has been violated. Also, one might point out that a speaker cannot reasonably intend to communicate whatever he pleases. She can only intend to communicate what she can reasonably expect to get across. It seems to be, therefore, in her own interest to use expressions in the way they are commonly, i.e. conventionally, used in her linguistic community. So while type meaning and token meaning might, occasionally, come apart (as witnessed by the case of malapropism), this cannot be the normal case. In what follows I will not try to defend Davidson’s line of argument; I will not ask whether his considerations warrant the conclusion he draws. (For all I care, Davidson may be right, but for the wrong reason). My aim is to achieve a better understanding of the role type meaning plays in interpretation and linguistic understanding. Consequently, I will rephrase the opening question thus: Does an interpreter need to possess a theory of type meaning for L in order to successfully interpret utterances of L-sentences?
3 Context sensitivity According to commonly held opinion, a smooth transition from type meaning to token meaning is, occasionally, hampered by the context sensitivity of linguistic interpretation. To inquire into the relation between type meaning and token meaning would require, therefore, an investigation into the context sensitivity of linguistic interpretation: How much and what forms of context sensitivity does natural language exhibit; how can context sensitivity be characterized, and how can it be
2 And according to Stroud “no theory of meaning for a language can tell you what a person is going to do or how he is going to do it, and no such theory alone can tell you what a person is doing on a particular occasion, or how he is doing it.” (Stroud 1998: 201)
Meaning and interpretation
79
resolved? I will focus on the last two questions and start by explicating what I mean by context sensitivity. In a paper from (1949), Arthur Burks noted the following: Moreover, the meaning of a token of a non-indexical symbol is always the same as the meaning of the type to which it belongs. The case is different with an indexical symbol, however, for the spatiotemporal location of a given token of such a symbol is relevant to the meaning of that token: ‘now’ means two different things when it is uttered on two different days. Since the meaning of the type to which any symbol belongs (whether indexical or nonindexical) is always the same, it follows that the meaning of a token of an indexical symbol is different from the meaning of the type to which it belongs. (Burks 1949: 681)
Context sensitivity comes in various forms, though (for an overview cf., e.g., Kompa 2010). A well-rehearsed point in the literature by now is that context sensitivity is not limited to the case of indexical expressions. According to so-called contextualists (such as François Recanati, Robyn Carston, or Anne Bezuidenhout, to name just a few) context sensitivity is a ubiquitous feature of natural language. Others agree that there is more to it than just indexicality but insist that it is not that much more (Cappelen & Lepore 2005). How much context sensitivity natural languages exhibit and which form it might take is subject of heated debates; debates that I will try to stay clear of. Yet how context sensitivity is best to be characterized is controversial, too. Following Burks, I suggest to characterize it as follows: CS: An expression type is context-sensitive if and only if different token of it can have different extensions (a declarative sentence’s extension being its truth-value), owing to (systematic) difference in the respective contexts of utterance.
As a result, in order to successfully interpret a given utterance, any context sensitivity has to be resolved. Let us, therefore, ask how hearers interpret an expression that happens to be context-sensitive in the sense laid out above. (I hope that my examples are uncontroversial enough). Suppose a speaker, call him John, says Mary is rich
one time in the course of a conversation about Bill Gates and people of that ilk, the next time in the course of a conversation about middle‐class persons and their income and assets. What did John mean to communicate? In the first conversation it was, presumably, something to the effect that Mary is rich relative to a comparison class comprising Bill Gates and the likes of him. In the second conversation it was, presumably, something to the effect that Mary is rich compared to middle-class people. Someone is rich only compared to someone else, or relative to a certain standard or comparison class. (There is, pace Cappelen & Lepore, no point is asking whether Mary is rich simpliciter.) As a result, someone might reasonably count as being rich in one contextual setting but not in others as other contextual settings might require thinking of being rich quite differently. Mary, for instance, may well
80
Nikola Kompa
live up to the standard set by ordinary middle-class people but not to the standard set by people such as Bill Gates. Consequently, what John said might have been true in the context of the second conversation yet false in the context of the first conversation, even though he was talking about the same person, Mary. Someone who is party to the first conversation might reason as follows: John said, “Mary is rich”. Presumably, he tried to say something true. When would his utterance be true? Given that we have been talking about people such as Bill Gates, his utterance would be true if Mary were rich compared to these people. Accordingly, that is what he meant to communicate.
The hearer is thereby performing pragmatic (or abductive) inferences, drawing on whatever contextual clues she can get hold of. While traditionally it was thought that to interpret an utterance is to semantically interpret its syntactic structure, the result being a truth-apt content, context-sensitive interpretation seems to be a semanticpragmatic joint venture. Commonly, when hearers interpret an utterance, they interpret it in light of certain interests, against the background of certain assumptions, etc. (cf. Searle 1978, 1980). A pragmatic inference is, therefore, a type of practical, meansend reasoning as it is closely tied up with goals and purposes. It is an inference to the best interpretation. And it is a defeasible inference. More specifically, pragmatic inferences are defeasible or non-monotonic in the following sense: It may be the case that although one is justified in inferring conclusion C from a set P of premises, one is no longer justified in inferring C from a set M of premises, such that M ⊃ P. As a result, a pragmatic inference is liable to be defeated by new information. Yet there seems to be no limit to the amount of information that may turn out to be relevant (cf. Borg 2004: 79, 2007: 356; cf. also Recanati 2004: 54). If that is so, then there will be no algorithmic procedure for pragmatic interpretation as there is no guarantee that the task can be completed after finitely many steps.3
3 Inferences concerning an agent’s intentions are examples of pragmatic inferences, too. Here is an example of Emma Borg’s to illustrate the abductive nature of intentional interpretation: “So, imagine that you see Sally filling a glass of water from the tap. Then you might reason as follows: ‘Sally is getting a glass of water from the tap. The best explanation for this action is that Sally is thirsty and wants a drink; therefore Sally is thirsty and wants a drink’. Clearly this is a non-demonstrative piece of reasoning and it is susceptible to the influence of an open-ended range of contextual factors. For instance, say that you know that Sally has just come in with Sourav and that Sourav is wearing running gear and looks out of breath, then the best explanation for Sally’s action might be that Sourav is thirsty and Sally is getting a drink for him. Or imagine that Sally has just glanced at her potted plant, then the best explanation might be that she wants to water her plant.” (Borg 2004: 78–79).
Meaning and interpretation
81
What if no precise standard (for richness, for example) or comparison class has been mutually established so far? In that case, a speaker might try to establish one himself. Suppose a speaker, call him Tom, says: “John is tall. And Mary is rich” (not a very sensible utterance, admittedly; but let us put that aside for the moment) while no particular comparison class is contextually salient. He seems to be thereby suggesting to his audience that tall and rich be understood in their shared context in such a way that John falls under the term tall and Mary under the term rich. And if he gets away with it (as he will if what he suggests is compatible with what is jointly presupposed; cf. Stalnaker 1998, 1999b), then what he thereby said – given that they are all well-informed, i.e. not mistaken about John’s height and Mary’s wealth, competent and reasonable – will be true in their context simply due to the fact that they have agreed to take it to be true. Again, someone privy to the conversation might reason as follows: Tom said: “John is tall. And Mary is rich.” Presumably, he tried to say something true. What would it take for his utterance to come out true? The word tall would have to be interpreted in such a way as to make John a clear case of tallness and the word rich would have to be interpreted in such a way that Mary turns out to be a clear case of a rich person. So let us understand the expressions that way.
The idea, of course, is not new; David Lewis (1979) spoke of “accommodation”. As Mark Richard puts it: “My statement, that Mary is rich, is as much an invitation to look at things in a certain way, as it is a representation of how things are.” (Richard 2004: 226) Successful interpretation seems to be achieved via pragmatic inferences; and accommodation is a case in point. It requires that context sensitivity be resolved. That in turn requires that the interpreter knows, among other things, something about the purpose or point of the conversation, the participants’ concerns and background assumptions, and so on. To that end, she also needs to understand how people might interact with the world, to what purpose they do what we do, and so on. The meaning of an expression as used on a particular occasion seems to be a function of the ways people can reasonably and purposefully interact with the world. As Julius Moravcsik put it: There is one everyday meaning for ‘snow’. But it generates an indefinite variety of denotational uses (how dry? how light? How much of it in flakes?) These are the function of human interactions with nature. We cannot predict or give a qualitative exact specification of all past and future human interactions with nature that have or will affect our use of ‘snow’. (Moravcsik 1998: 43)
82
Nikola Kompa
4 The grab‐bag-model Some linguists and philosophers take the massive context sensitivity of natural language to show that the contextually relevant token meaning of a word or sentence is not a ready-made entity but has to be constructed, negotiated in context. According to Robyn Carston, for instance, often in linguistic interpretation “an ad hoc concept is constructed and functions as a constituent of what is explicitly communicated”. (Carston 2002: 357) Words provide information, which in turn is the starting point for pragmatic inferences that result in a contextually specified understanding or interpretation of the words as used on the occasion (cf. also, e.g., Recanati 2004: 146, Croft & Cruse 2004: 97). While this may still be compatible with the idea that the information is provided by the meaning of the word type, others try to do without type meaning in anything like the traditional sense altogether. A fairly radical version of this idea is what Agustín Rayo calls the grab-bagmodel of language-mastery. He asks what sort of semantic information someone has to associate with the items in his basic lexicon in order to master the language. According to the account he rejects, a subject has to be in possession of a significant amount of specifically linguistic information and her usage has to be guided by semantic rule. According to the account he favors, a grab bag of mental items is all it takes: With each expression of the basic lexicon, the subject associates a ‘grab bag’ of mental items: memories, mental images, pieces of encyclopedic information, pieces of anecdotal information, mental maps, and so forth. With the expression of ‘blue’, for example, a subject might associate two or three particular shades of blue, the information that a paradigmatic instance of ‘blue’ is the sky on a clear day, a memory of a blue sweater, and so forth. Different speakers might associate different grab bags with the same lexical item. A grab bag will typically not be enough to determine a range of application for the relevant lexical item independently of the subject’s general‐purpose abilities. But, by exercising sensitivity to context and common sense, the right kind of subject in the right kind of context might be in a position to use the grab bag to come to a sensible decision about what to treat as the expression’s range of application for the purpose at hand. (Rayo 2013: 648)
A defender of type meaning as a nonnegotiable explanatory ingredient in any theory of linguistic interpretation might pursue two strategies: 1) He might argue that the whole grab bag associated with the word F deserves to be called “F’s type meaning”; 2) Or he might argue that something important is missing from the grab bag – something that deserves to be called “type meaning”; that possessing a grab bag in Rayo’s sense is not sufficient for successful interpretation. The first line of reply meets with the following problem. If “type meaning” is meant to be just another name for the whole grab bag, then type meaning will
Meaning and interpretation
83
comprise mental images, encyclopaedic information, general‐purpose capacities, etc. It will thereby be drained off any specifically semantic content. The second line of defense is more promising. Here are some possible candidates for what might be missing from the grab bag. (a) The defender of type meaning might point out, for example, that an interpreter needs to be provided with a definition (that makes up the meaning of the expression type), understood as a set of singly necessary and jointly sufficient conditions of application. A triangle, e.g., is definable as a plane figure with three straight lines and three angels. But then, definitions are notoriously hard to come by. What is the definition of animal, or of object, or of justice, for that matter? And most terms fail to draw sharp boundaries anyway; they leave certain cases open. Most terms of our language seem to be partially defined at best, as suggested by Scott Soames (Soames 2010: 46). There are borderline cases of their application. That is the lesson to be learned from vagueness (cf., e.g., Williamson 1994; Keefe 2000; Shapiro 2006; Raffman 2014). And the classical paradigm of categorization by means of necessary and sufficient conditions has come under attack from another direction, too, with the advent of Prototype Theory in the 1970s. In a seminal paper (Rosch 1973), Eleanor Rosch showed that people judge some members of a category to be better members than others. As a consequence, category membership is now commonly taken to be a gradual matter, with some members being more typical members than others. When asked, e.g., to give an example of the category FURNITURE, chair is more frequently cited than, say, stool.4 (b) Alternatively (yet closely related), a defender of type meaning might claim that what is missing from the grab bag is a set of specifically semantic, i.e. meaning-constituting, beliefs. Some beliefs, he might claim, are part of the meaning of F, others are part of our F-theory, still others hold only
4 Unfortunately, prototype theory seems to be inflicted with problems of its own. What is a prototype, the prototype of the category ANIMAL, for instance, supposed to be? A pictorial representation will not do. And suppose the prototype of the category BIRD is a robin. Presumably, it is not a one-legged, limping robin that is old and featherless. It is, arguable, a paradigmatic robin. But that is the prototype of the Category ROBIN! Both categories have the same prototype, then! What needs to be answered is the question what features the prototype (of the category BIRD, e.g.) has to possess – a set of necessary features that is specific to this category will be hard to come by (“is able to fly, winged, two-legged”, e.g., is, on the one hand, too restrictive and on the other hand not restrictive enough); a set of sufficient conditions (“being a paradigmatic robin, being a paradigmatic sparrow”, e.g.) will not suffice to properly characterize the category. But then, prototype theory is no longer the only game in town; exemplar theory or statistical learning theory provide alternatives.
84
Nikola Kompa
contingently of (some) Fs. Yet the question of which beliefs are meaning constituting has also proven notoriously difficult to answer. Is the fact that cats purr part of the meaning of cat or it is part of our cat-theory? Of course, certain beliefs may be more central (in the sense of being widely shared and assumed to be shared, less open to negotiation, more entrenched, etc.) than others. My belief that grandma’s cat Toby is blind, for example, is hardly constitutive of the meaning of the word cat. But why think that the belief that cats are animals ought to be more dearly held than the belief that there are cats? Consequently, to concede that some beliefs are better entrenched than others is not to concede that the question which beliefs are meaning constituting and which are not is amenable to a definite answer. Moreover, one might point out that the idea of meaning-constituting beliefs will be of little use in explaining how speakers learn to competently use the words of their language in the first place. In order to learn to apply a general term F to varying objects children have to first learn to distinguish Fs from non-Fs: mama from nonmama, cups from non‐cups, milk from non‐milk, etc. Yet children will gradually also learn more about Fs; what they are good for, what purpose they serve; they also may learn about atypical Fs, etc. In many cases at least, what they have to learn in order to learn the ‘meaning’ of a term is first and foremost something about the things denoted by the term. One does not know the meaning of table, for example, unless one knows something about tables; what tables are for, what they have to be like to serve that purpose and so on. At first, maybe children only partially grasp the meaning of a word, but understanding improves as they learn more about the entity denoted. Concerning words such as table or chair, we will all, eventually, fare more or less equally well. But there are many other words where differences in knowledge between speakers are huge. Theoretical terms (such as electron) are the most obvious case in point. (Different contexts impose different requirements on us, though. It takes more to competently use the word electron in a discussion with scientists than in everyday contexts.) Yet expressions such as broker or philosopher will do, too. It is obvious that in order to know the meaning of these words, one needs to know something about brokers and philosophers, about what they do, what they are for, etc. And some know more than others. That is attested by the fact that these words hardly ever belong to a child’s first words. As Michael Tomasello puts it: For one thing, cognitive linguists highlight the importance of the various kinds of background cultural knowledge and categories that structure word meanings: a knuckle presupposes a finger, a pedestrian presupposes traffic, and a bachelor presupposes a culture in which marriage is expected. The meaning of many, if not most, words is thus not specifiable in isolation, but must be understood in the context of a larger set of cultural activities and entities. (Tomasello 2003: 53–54)
Meaning and interpretation
85
As we assemble knowledge about Fs our categorization becomes more robust, more reliable when it comes to hard cases. Different cases need to be distinguished, though. Take the case of nominal kind terms first (Schwartz 1977). These are more conventionally defined terms such as tool terms (drill), kinship terms (uncle), meal terms (dinner), moral act terms (cheat), etc. Children first tend to classify by characteristic features; later they move on to more theoretical, more defining features: For nominal kinds, characteristic features may predominate early on in concept acquisition but give way to defining features with increasing knowledge and conceptual sophistication. (Keil 1992: 60).
In the case of natural kind terms there is a shift from characteristic features to theory (Keil 1992: ch. 8); ontogeny recapitulates phylogeny. Once we have acquired more theoretical knowledge we may draw upon it in order to justify a particular classification or to decide hard cases. Usage will become more empirically and theoretically informed as one learns more about the world. Understanding improves as one gains knowledge. That also explains why – contrary to commonly held opinion – the greatest increase in vocabulary size happens between age 10 and 17: “[W]ord learning typically reaches its peak not at 18 month but somewhere between 10 and 17 years.” (Bloom 2000: 44) Consequently, correct and robust application and interpretation of an expression, the word table for example, (as used on a particular occasion) seems to require that we not only know that tables are pieces of furniture used to put thing on (note that this is not a definition as it fails to distinguish tables from sideboards, for example) but also what they typically look like, what people typically use them for, which kinds of things people tend to put on tables, what they are typically made of, or where they are typically found. To honor certain beliefs as meaning-constituting and to try to thereby extract a notion of type meaning seems to be a question of demarcation at best – at least if our goal is to understand and explain successful interpretation and competent usage. (c) Thirdly, a defender of type meaning might insist that we turn to meaning not only for information but also for guidance.5 We set out, he might remind us, to inquire into the knowledge hearers have to possess in order to successfully interpret a given utterance. Yet by phrasing the question the way we did (in terms of knowledge), we inadvertently invited a response along the lines suggested by Rayo. It might have seemed that by asking
5 There is a long‐standing debate about whether linguistic understanding consists in the possession of propositional knowledge or a practical ability (for an overview, cf. Dänzer 2015).
86
Nikola Kompa
what knowledge hearers have to possess we were asking what information they have to possess, thereby picturing meaning as something that provides information. What we missed, though, is that meaning is supposed to provide guidance for the application and interpretation of words. This has commonly put by saying that the meaning of a term is supposed to provide a rule or norm for its application and interpretation. Yet again, it has proven notoriously difficult to spell out the idea of following a rule or of meaning being normative (cf., e.g., Glüer/Wikforss 2015; Kusch 2006). But then, one might wonder whether information doesn’t also provide guidance. By furnishing a speaker with information about paradigmatic cases of the application of a particular term, for example, don’t we also provide guidance about how to apply (and interpret) that term? The problem is that speakers not only want to classify paradigmatic instances as such; they would also like to classify similar cases in like manner. What needs to be explained is how speakers and hearers manage to extrapolate from sample cases to new (and slightly different) cases.6 For purposes of illustration, consider the following example. We see, let us suppose, an object with a certain colour, a ripe lemon, for instance. The next day we see another object with a fairly similar colour, a ripe banana, let’s say. We observe that some things in the world have something in common, such as their colour. We sort like things alike by subsuming them under the same term (e.g. yellow). We thereby structure the manifold of experience; we sort things into categories. But then, no two situations or things that we encounter are exactly alike. By noticing similarities we also abstract from – for present purposes – irrelevant dissimilarities. Extrapolation requires abstraction, i.e. the omission of distinguishing yet irrelevant details.7 Yet which distinguishing details we omit and which similarities we deem relevant (and worthy of a name) depends on many factors. It is partly constrained by our cognitive architecture and our sensori-motor makeup as human beings. That we see certain shades of colours as being more similar than others depends, presumably, on our perceptual apparatus and our biological needs. In other cases, it
6 Not surprisingly, how classification is achieved is a highly contested issue in psychology and philosophy alike; cf., e.g., Margolis & Laurence 1999; Murphy 2002; Machery 2009; or Nimtz & Langkau 2010. 7 One might think that what is needed in order to understand and apply a particular term F is not a set of sample or paradigmatic cases but an abstract idea of F-ness that encapsulates what all cases of F have in common – an idea fervently advocated, e.g., by John Locke in his Essay Concerning Human Understanding. According to Locke, abstract ideas are the means by which our mind sorts things into classes. Unfortunately, Locke’s suggestion was subjected early on to harsh criticism and ridicule. Among his most prominent critics were David Hume and George Berkeley.
Meaning and interpretation
87
depends on culturally shaped interest and concerns. And it may depend on local, contextual interest as well. We divide men into friend and foe, books into those worth reading and those not worth the while, our colleagues into potential partners for cooperation and others, etc. Because these similarities and differences are important to us, we have words to mark them. And the ability to find and recognize patter seems to be a prerequisite for language acquisition in general and the acquisition of general terms as a means of classification in particular. So maybe (human) beings simply come equipped with a cognitive capacity to extrapolate, to find and recognize patterns. Maybe that is just a fact about human and other animals’ cognitive architecture. (The idea seems to be much en vogue in certain quarters of psychology and biology). The way we learn and employ general terms may, therefore, be the result of a biologically grounded capacity to detect patterns (statistical learning theory may help to provide a fuller account; cf., e.g., Romberg & Saffran 2010; Rebuschat & Williams 2012). Culture and context act as filters. Many patterns we are prone to find are cultural products and filtered by contextual interests and concerns. The biological mechanisms have to be supplemented by culturally transmitted and contextually relevant knowledge about ‘how it is done around here’. Yet without the ability to find and recognize patterns human beings would not be able to learn a language (cf., e.g., Tomasello 2003: 28).8
5 A plea for an empirically informed theory of linguistic interpretation What do we make of all this? One the one hand, one might think the grab‐bag‐ model slightly idiosyncratic, and the bag itself rather heterogeneous, even unmotivated. On the other hand, any attempt at spelling out the idea of type meaning meets with problems and seems to be lacking in explanatory value. In some cases of interpretation, speakers draw on (partial) definitions, in others on representations of paradigmatic cases of the application of the terms involved ( grounded, presumably, in an observation of certain regularities, certain patterns), in still others on encyclopedic knowledge or micro-theories, etc. The discussion in the last paragraph was, of course, a bit impressionistic. Yet even such a cursory
8 Consequently, each general term of our language has to pick out a recognizable pattern. Otherwise it could neither be learned nor taught: No pattern to recognize, no meaning to learn. It need not be a perceptually recognizable pattern, though. But the easier the pattern is to recognize that is picked out by a term F, the easier it will be to learn the term.
88
Nikola Kompa
reflection might lead one to conclude that someone who is to competently use and interpret natural language utterances has need (instead of type meaning) of at least the following: –– A culturally molded, yet also goal- and interest-directed capacity to find and recognize patterns, to see things as being similar in relevant respects (in light of certain interests and concerns, that is). –– The ability to perform pragmatic inferences, grounded in experiences of the ways in which classification and communication may be goal-directed together with the ability to recognize and share other people’s goals and purposes. (Being able to share goals and to recognize people’s purposes seems to be a prerequisite for language acquisition, too; cf. Kompa et al. 2013: 441). –– Beliefs about the entities denoted; the more beliefs one has acquired the more informed one’s usage turns out to be. Eventually, one will have acquired a (micro-)theory. The list is not meant to be exhaustive, of course. Yet it seems that we end up with a grab bag after all. And while a grab bag is a miscellaneous lot, it (thereby) draws attention to something important: the heterogeneity of the mechanisms, information, and (cognitive and social) capacities needed for linguistic i nterpretation. Moreover, the traditional attempts at explicating the notion of type meaning were found wanting. More specifically, they fail to take into account (and to help explain) how speakers actually achieve understanding and acquire interpretative skills in the first place. Consequently, we might want to open our mind to the idea that a (philosophical) theory of linguistic interpretation has to be at least compatible with, better still informed by, empirical findings regarding how speakers actually come to competently use a language, but also regarding the cognitive and social preconditions of linguistic interpretation. (A specifically philosophical task might then be to critically reflect on those findings and integrate those that stand up to scrutiny into a comprehensive account of language understanding and linguistic interpretation.) I will close, therefore, with a plea for a philosophically reflective, yet also empirically informed and methodologically more varied approach to language understanding and linguistic interpretation.
References Bloom, Paul. 2000. How children learn the meanings of words. Cambridge MA: MIT Press. Borg, Emma. 2004. Minimal semantics. Oxford: Oxford University Press. Borg, Emma. 2007. Minimalism versus contextualism in semantics. In Gerhard Preyer & Georg Peter (eds.), Context-Sensitivity and semantic minimalism: New essays on semantics and pragmatics, 339–359. Oxford: Oxford University Press.
Meaning and interpretation
89
Burks, Arthur W. 1949. Icon, index, and symbol. Philosophy and Phenomenological Research, Vol. ix. 673–689. Cappelen, Herman & Lepore, Ernie. 2005. Insensitive semantics: A defense of semantic minimalism and speech act pluralism. Oxford: Blackwell Publishing. Carston, Robyn. 2002. Thoughts and utterances: The pragmatics of explicit communication. Oxford: Blackwell Publishing. Croft, William & Cruse, D. Alan. 2004. Cognitive linguistics (Cambridge Textbooks in Linguistics). Cambridge: Cambridge University Press. Dänzer, Lars. 2015. Sprachverstehen. In Nikola Kompa (ed.), Handbuch Sprachphilosophie, 249–259. Stuttgart: Metzler. Davidson, Donald. 1986. A nice derangement of epitaphs. In Richard Grandy & Richard Warner (eds.): Philosophical grounds for rationality, 157–174. Oxford: Clarendon Press. Reprinted in Aloysius Martinich (ed.). 2001: The philosophy of language, 473–483. Oxford: Oxford University Press. (Quotations are from Martinich 2001.) Glüer, Kathrin & Wikforss, Åsa. 2015. The normativity of meaning and content. In The Stanford encyclopedia of philosophy (Summer 2015 Edition), Edward N. Zalta (ed.). http://plato. stanford.edu/archives/sum2015/entries/meaning-normativity/ Keefe, Rosanna. 2000. Theories of vagueness. Cambridge: Cambridge University Press. Keil, Frank C. 1992. Concepts, kinds, and cognitive development. Cambridge MA: MIT Press. Kompa, Nikola. 2010. Contextualism in the philosophy of language. In Klaus Petrus (ed.), Meaning and analysis: New essays on H. P. Grice, 288–309. Basingstoke: Palgrave Macmillan. Kompa, Nikola, Regine Eckardt, Henrike Moll & Susanne Grassmann. 2013. Sprache/Sprachverstehen/Sprachliche Bedeutung/Kontext. In Achim Stefan & Sven Walter (eds.), Handbuch Kognitionswissenschaft, 432–444. Stuttgart: J. B. Metzler. Kusch, Martin. 2006. A sceptical guide to meaning and rules. Defending Kripke’s Wittgenstein. Chesham: Acumen. Lewis, David. 1979. Scorekeeping in a language game. Journal of Philosophical Logic 8. 339–359. Longworth, Guy. 2008. Linguistic understanding and knowledge. Nous 42(1). 50–79. Moravcsik, Julius M. 1998. Meaning, creativity, and the partial inscrutabilty of the human mind. Stanford: CSLI Publications. Murphy, Gregory L. 2002. The big book of concepts. Cambridge MA: MIT Press. Machery, Erdouard. 2009. Doing without concepts. Oxford: Oxford University Press. Margolis, Eric & Laurence, Stephen (eds.). 1999. Concepts; core readings. Cambridge MA: MIT Press. Nimtz, Christian & Julia Langkau. 2010. Concepts in philosophy – A rough geography. Grazer Philosophische Studien 81. 1–11. Raffman, Diana. 2014. Unruly words – A study of vague language. Oxford: Oxford University Press. Rayo, Augustin. 2013. A plea for semantic localism. Noûs 47 (4). 647–679. Rebuschat, Patrick & John N. Williams (eds.). 2012. Statistical learning and language acquisition. Boston/Berlin: De Gruyter Mouton. Récanati, François. 2004. Literal meaning. Cambridge: Cambridge University Press. Richard, Mark. 2004. Contextualism and relativism. Philosophical Studies 119. 215–242. Romberg, R. Alexa & Jenny R. Saffran. 2010. Statistical learning and language acquisition. Wiley Interdisciplinary Reviews: Cognitive Science 1(6). 906–914.
90
Nikola Kompa
Rosch, Eleanor. 1973. Natural categories. Cognitive Psychology 4. 328–350. Schwartz, Stephen, P. (ed.). 1977. Meaning, necessity, and natural kinds. Ithaca, NY: Cornell University Press. Searle, John. 1978. Literal meaning. Erkenntnis 13. 207–224. Searle, John. 1980. Background of meaning. In John Searle, Ference Kiefer & Manfred Bierwisch (eds.), Speech Act Theory and pragmatics, 221–232. Dordrecht: Reidel Publishing Company. Shapiro, Stewart. 2006. Vagueness in context. Oxford: Clarendon Press. Soames, Scott. 2010. The possibility of partial definition. In Richard Dietz & Sebastiano Moruzzi (eds.), 2010: Cuts and clouds. Vagueness, its nature, and its logic, 46–62. Oxford: Oxford University Press. Speaks, Jeff. 2010. Theories of meaning. The Stanford encyclopedia of philosophy (Winter 2010 Edition), Edward N. Zalta (ed.). http://plato.stanford.edu/archives/win2010/entries/ meaning/ Stalnaker, Robert C. 1998. On the representation of context. Journal of Logic, Language, and Information 7. Reprinted in Robert Stalnaker, Context and content, 96–113. Oxford: Oxford University Press. Stalnaker, Robert C. 1999. Introduction. In Robert Stalnaker, Context and content, 1–28. Oxford: Oxford University Press. Stroud, Barry. 1998. The theory of meaning and the practice of communication. Crítica 30:88. Reprinted in Stroud, Meaning, understanding, and practice, 193–212. Oxford: Oxford University Press. (Quotations are from Stroud 2000). Tomasello, Michael. 2003. Constructing a language: A usage-based theory of language acquisition. Cambridge MA: Harvard University Press. Williamson, Timothy. 1994. Vagueness. London & New York: Routledge.
Anne Bezuidenhout
The role of context in semantics: A Relevance Theory perspective Abstract: The notion of context in Relevance Theory is a psychological one, and prima facie very different from the notion of context that is invoked by Kaplan and others working in the formal semantics tradition. In Relevance Theory, context is conceived of as the mutual cognitive environment of the partners in a conversation. This paper explores whether and how the Relevance Theory concept of context is related to the formal concept of context, which Predelli (2005) calls an index. I focus of the role of indexicals in language performance and mount evidence that the Relevance Theory notion of context is better calibrated with current work on language performance systems. The last two sections of the paper spell out some of the ways in which cognitive environments are constrained and how these constraints help to explain the use of indexicals and other referring expressions. I call the process of constraining cognitive environments “conversational tailoring”. One of the main concerns of this paper is to indicate some of the ways in which tailoring works when indexicals are used.
1 Introduction I am interested in exploring the relation between two conceptions of context, namely the psychologistic notion that belongs to the Relevance Theory framework and the formal notion that belongs to the traditional 20th-century philosophy of language framework. In section 2, I look at some of the ways in which pressure has been put on the formal notion of context because of problems in accounting for certain uses of indexicals. I will suggest that a psychologistic notion is better able to deal with these problematic cases of indexical use. In section 3, I address the question as to the nature of and the constraints on contexts in the Relevance Theory sense. In section 4, I return to the discussion of indexical use that was first raised in section 2, with a view to understanding how the Relevance Theory notion of context enters into an explanation of indexical use. This paper is not intended as an exhaustive examination of every notion of context that has ever been proposed. I have chosen to examine one psychologistic notion, the Relevance Theory notion of context as the mutual cognitive environment of a group of conversational partners, and to contrast this with one prominent DOI 10.1515/9781501504327-006
92
Anne Bezuidenhout
formal notion, the notion of context as what Predelli (2005) calls an index. There are other formal notions that are widely discussed in the literature, such as Stalnaker’s conception of context as common ground, where common ground is conceived of as a set of propositions, and where propositions in turn are treated as sets of possible worlds. The reason I have chosen the concept of an index as a foil for the Relevance Theory notion of context is because this paper is focused on rival treatments of indexicals and the notion of an index is the formal notion usually invoked in philosophical accounts of indexicals, as opposed to the notion of context as a set of possible worlds. Frege’s anti-psychologism in logic has been taken over largely unquestioned in the natural language semantics tradition in the philosophy of language. This is unsurprising, given that the tools and methods used in philosophical semantics are drawn from the Fregean logical tradition. Katz’s (1980) view of languages as abstract objects appeals to Fregean arguments to make the case for anti-psychologism in linguistics too. However, the view that has dominated in linguistics and the psychology of language over the last half century is the Chomskyan one, which treats languages as mental entities – systems of mental representations and rules for manipulating these representations. Relevance Theory accepts the Chomskyan internalist view of language as a system that is a part of the human mind-brain. Relevance Theory is often presented as a cognitive-psychological framework for addressing questions about language processing and language comprehension in particular, starting from the point at which the language faculty narrowly conceived has done its job and fed forward its output at what Chomsky calls the conceptual-intentional interface, the interface between the narrow language faculty and the conceptual systems that use language for communicative purposes. For an account of the language faculty narrowly conceived, see Hauser, Chomsky, and Fitch (2002: 1571). That is, Relevance Theory is presented as a cognitive-psychological account of the inferential phase of communication, which takes as input the representations that are the output of the decoding phase.1 This is not the place to elaborate on the Chomskyan program or to give a fullblown account of the inferential phase of verbal communication as understood
1 Although Relevance Theory accepts the Chomskyan picture of a language faculty narrowly conceived and its performance interfaces, it does not follow that Relevance Theory needs to accept Chomsky’s own suggestions about the grammar embodied by the language faculty. Alternative grammar formalisms, such as the dynamic syntax view defended by Kempson, Myer-Viol, and Gabbay (2000), may offer a better fit with the Relevance Theory view of the inferential phase of verbal communication.
The role of context in semantics: A Relevance Theory perspective
93
within the Relevance Theory framework. For present purposes, the main take-away point is that the Chomskyan program embraces psychologism and Relevance Theory embraces the Chomskyan picture of a language faculty narrowly conceived and its performance system interfaces. This puts Relevance Theory and philosophical semantics at odds with each other, given that the latter embraces Fregean anti-psychologism. I will try to argue that the psychologistic framework is needed for an account of indexical use and that the Relevance Theory notion of context is better aligned with current empirical work on language production and comprehension.
2 Contexts – cognitive environments or indexes? The notion of context that is used in Relevance Theory is a psychological one, and hence prima facie different from the formal notion of a context as an index i = , consisting of the values assigned to various contextual parameters i1, …, in on a particular occasion of use of an indexical or context‐sensitive expression. In Relevance Theory, a context is what is mutually manifest to the conversational participants in some ongoing conversational exchange. The information that is manifest to an individual at a time constitutes that person’s cognitive environment. Information is manifest if it is either explicitly represented in the individual’s working memory or could be so represented; i.e., if the individual’s attention were to be appropriately shifted, that information would become explicitly represented. This is an internalist conception of context that aligns it well with the Chomskyan view of language and with the dominant framework in the branches of experimental psychology that study language performance systems – that is, language production and comprehension. If our concern is to study the role of indexicals and context-sensitive expressions in language production and comprehension, the Relevance Theory notion of context seems to be a good fit, whereas the formal notion of context as an index seems out of place. Are these two notions of context in fact incompatible with one another? Which is to be preferred in psychological accounts of how context‐sensitive expressions are produced and understood? Can the notion of an index be “psychologized”? I will use notation as in (1) below to talk about the formal notion of a context as an index i consisting of the values for a set of contextual parameters, including minimally the speaker/agent a , the location l, and the time t of the context: (1) i = < a, l, t >
94
Anne Bezuidenhout
Other contextual parameters have been suggested, such as an addressee parameter to deal with deictic uses of you, a “demonstratum” parameter to deal with uses of simple demonstratives such as this, a world parameter to deal with uses of actual and actually, a parameter for the standards of precision operative in the context to deal with uses of vague expressions such as flat (Lewis 1979), and so on. I will suppress these other parameters, as they will not be relevant for the examples I will be discussing. This formal notion of context has been challenged because of problems in accounting for certain uses of indexicals. There are two kinds of uses of indexicals that have led to challenges to the notion of context as an index, which I discuss in the following two sub-sections.
2.1 Narrow versus wide context The first kind of challenging use of indexicals is illustrated by the different possible uses for sentences such as: (2) Colin is here. (3) We are healthier now than before. Depending on the assumptions that are mutually manifest to the conversational participants, the expression here in (2) might pick out the room, or the building, or the campus, or the city, or the country, etc. in which the participants are located. Similarly, now in (3) might refer to this week, this year, this generation, this millennium, etc. The first person plural pronoun we in (3) also exhibits this sort of flexibility. There are uses of we that include the addressee and uses that exclude the addressee and exactly who is included in the group along with the speaker can vary depending on assumptions operative in the context. See Nunberg (1993) for a discussion of the case of we. In all these cases apparently the interlocutors need to be sensitive to information beyond what is captured by an index i of the sort illustrated in (1). One suggestion for dealing with these cases is to distinguish between indexicals that have their references fixed automatically by applying the reference fixing rules that constitute their encoded meanings (their Kaplanian characters) in the operative context, and those indexicals that require an appeal to the speaker’s referential intentions in order to fix their reference. The claim would be that now, here and we in (2) and (3) fall into the latter category, whereas I and today would fall into the former category. See Perry (2001) for a suggestion along these lines. A similar suggestion is made by Recanati (2001), who calls I a pure indexical and
The role of context in semantics: A Relevance Theory perspective
95
now and here impure indexicals. The latter are in turn seen as a sub-variety of “intentionals”, which are expressions whose reference is merely constrained, but not determined, by their encoded meanings. This picture has also been associated with a distinction between narrow and wide context. Narrow context contains only the minimal information encoded in the character of an expression and is all that is required to fix the referents of the pure indexicals. Wide context, which includes information about a speaker’s referential intentions, is needed to fix the referents of the intentionals. This narrow/wide context distinction is one that Recanati (2001) embraces. However, in subsequent writings, he challenges the distinction as well as an associated contrast between the semantic content of a sentence-incontext, which is fixed in an automatic manner by a function that takes character and narrow context as arguments, and the pragmatic content of a speech act, which is fixed by appeal to speaker intentions, shared background information, and so on. Recanati (2005: 173) labels the view that embraces this distinction “Conventionalism”. Recanati (2004: 58) writes: We can maintain that the character of ‘here’ and ‘now’ is the rule that the expression refers to ‘the’ time or ‘the’ place of the context – a rule which automatically determines a content, given a (narrow) context in which the time and place parameters are given specific values; but then we have to let a pragmatic process take place to fix the values in question, that is, to determine which narrow context, among indefinitely many candidates compatible with the facts of the utterance, serves as argument to the character function. On the resulting view the (narrow) context with respect to which an utterance is interpreted is not given, it is not determined automatically by objective facts like where and when the utterance takes place, but is determined by the speaker’s intention and the wide context.
Recanati’s point seems to be that there really is only the pragmatic notion of context (wide context) and that no indexicals have their referents fixed automatically, because the context argument to the function that takes character and context to content (call this function f) is pragmatically determined. Predelli is equally critical of the narrow/wide context distinction, but reasons to a conclusion diametrically opposed to Recanati’s conclusion, even though he relies on a premise almost identical to Recanati’s, namely that pragmatics is needed to determine which narrow context (i.e., index in Predelli’s terms) serves as an argument to the function f. Predelli thinks that once we get clear about the relation between an index and a context in the wide, pragmatic sense (viz., one that includes information about the speaker’s intentions, shared background assumptions, etc.), we will see that all indexicals have their referents determined automatically. See Predelli (2005: 40–75).
96
Anne Bezuidenhout
Predelli argues for both a pre-semantic and post‐semantic role for p ragmatics. Post‐semantic pragmatics is just pragmatics as practiced by traditional twentieth-century philosophy of language, namely the study of Gricean conversational implicatures. The pre‐semantic role for pragmatics is to determine which pair best represents the speaker’s utterance. An index consists of the values for a set of contextual parameters, as illustrated in (1) above. A clause is not a sentence of a natural language but rather a theoretical representation of the syntactico-semantic structure of the uttered sentence. It will be something like a labeled tree or labeled bracket structure. The semantic machinery (what Predelli calls the Interpretive System or simply the System) takes such pairs as input and runs without any interference from pragmatics, yielding as output what Predelli calls a t‐distribution, namely a distribution of truth-values at points of evaluation. This is the theoretical counterpart of the intuitive, commonsense notion of an utterance’s truth-conditions. Thus the picture Predelli is suggesting can be illustrated as in Fig. 1 below: Utterance
→→
System
→→
t-distribution
Truth-conditions Fig. 1: Predelli’s “Interpretive System” (Predelli 2005: 4).
In other words, utterances are not assigned truth‐conditions directly, but only indirectly via some decision as to how best to represent the semantically relevant properties of the utterance – those properties to which the System is attuned. So there is pragmatic intrusion into the process overall, but this occurs at the pre-semantic, representational phase, and not directly into the compositional machinery for determining truth‐conditional content, as contextualists are wont to assume. This means that in the case of utterances of sentences containing indexicals, the pair selected on pragmatic grounds as best representing the uttered sentence-in‐context (i.e., the pair consisting of sentence’s syntactico‐ semantic structure and the values for the parameters of the index) will automatically, in a pragmatics-free way, determine the semantic values of the clausal elements that correspond to the indexicals in the represented sentence. These semantic values are determined as a part of the compositional process that determines the semantic content of the clause as a whole, which in turn, given a circumstance of
The role of context in semantics: A Relevance Theory perspective
97
evaluation, determines a truth-value. This will apply to all indexicals. There is no distinction between pure and impure indexicals, or between ones that have their referents automatically determined by the character‐context function f and those whose referents are merely constrained by f. To illustrate this, imagine the following context C, in which your British friend Colin is coming from out of town to visit you in your hometown of Miami. He promises to let you know when he arrives so you can make arrangements to meet for dinner. You unexpectedly have to travel to Seoul and are not able to check email for a couple of days. When you do check, on July 17, you find a message from Colin sent two days earlier in which he writes: (4) I arrived here today. On Predelli’s account, the pre-semantic representational phase of the process will determine something roughly like the pair as the pragmatically best representation of Colin’s (written) utterance, which the System will, without further pragmatic input, interpret as a yielding a certain t-distribution that corresponds roughly to the intuitive truth-conditions given in (7): (5) [S [NP The agent of i] [VP arrived in the location of i on the day of i]] (6) i = < Colin, Miami, July 15 > (7) Colin’s utterance of I arrived here today is true in context C if and only if Colin arrived in Miami on July 15. Of course (5) only roughly indicates the clause needed to represent (4), because the constituent structures of the NP and VP are suppressed. Such fine‐grained details are not needed to make my point. The pragmatic work that determines that the granularity for ‘here’ is (the greater metropolitan area of) the city of Miami and not, say, the bathtub in the Hilton Bentley Hotel in South Beach in which Colin is sitting while composing his email, is done pre-semantically. So we don’t have to think of here as an impure indexical. It functions semantically just like I and today. Predelli’s account is also able to deal with another problem that has been thought to challenge the notion of context as index, namely the “remote utterance” problem, which concerns uses of indexicals on voice recording machines, post-it notes, and other cases of signs created in one context but designed for use in other (possibly multiple other) contexts. Suppose that it is your friend Colin who is in Seoul and he faxes you a note on which he has written I will arrive today, which prints out on your fax machine date stamped July 15. However, when Colin writes the note in Seoul it is already July 16 in Seoul, the day after the intended day of arrival in Miami. Nevertheless, what Colin conveys is that he will arrive in Miami
98
Anne Bezuidenhout
on July 15. (I am assuming the 13‐hour time difference between South Korea and the eastern USA allows Colin to arrive in Miami on July 15). This example shows that it is not always the context of encoding that is operative in determining the values of the parameters of the index and example (4) in effect showed that neither is it always the context of decoding. Even more elaborate examples are possible that tinker with both the encoding and decoding ends simultaneously. See Predelli (1998) for examples and discussion. These examples apparently show that speaker intentions and wider background assumptions are needed to fix the referents of indexicals. Predelli of course would say that the wider background information comes into play at a presemantic stage, and hence these examples are no threat to the notion of context as an index. They only threaten the notion of an index coupled with the erroneous assumption that the values of the contextual parameters of the index will always be the agent, place and time of the utterance, as represented in (8) below: (8) i = < ua, ul, ut > Predelli (2005: 42) calls this the Simple-Minded view, and the sorts of examples we have been discussing show why. In a straightforward face-to-face conversation, the index that best represents the conversational context may indeed be (8). But we need an account that can deal with a much wider range of conversational settings. I will return to ask whether Predelli’s distinction between context and index can help with the questions about the relation between psychological and formal notions of context and about which notion best fits with current performance theories of language. Before I do, I discuss the second kind of use of indexicals that has led to challenges to the notion of context as an index.
2.2 Context-shifting uses The second kind of challenging use of indexicals is illustrated by possible uses for sentences such as (9) and (10), the former being an example of the historical present and the latter an example of free indirect discourse: (9) Now McEnroe flings his racket down and storms off the court. (10) Today, today she would confront him and get him to admit his guilt. The context in which (9) is uttered could be one in which someone, in 2104 in his own living room in Miami, is watching a video recording of a tennis match played back in the 1980s at Wimbledon and who is narrating the events as they unfold for the benefit of someone who cannot see the video. In this case, the now refers
The role of context in semantics: A Relevance Theory perspective
99
to an event time in the 1980s, not to the time of utterance. Similarly, the use of the present tense forms of the verbs to fling and to storm off does not signal that the reported events are co-occurring with the utterance event. The free indirect discourse example (10) is one in which today is used to refer to the day on which the woman’s internal ruminations occurred, not to the day of the report of those ruminations. This is also interesting because it mixes the use of the past tense would with the indexical today that is associated with the present in some sense (and which contrasts with yesterday and tomorrow, which are associated with the past and future respectively). Assuming the woman’s language of thought is English, her own thoughts would presumably use the present tense form will rather than the past tense form would. Her thought, if verbalized at the time of occurrence, would have been expressed by a sentence such as Today I will confront him. Examples such as (9) and (10) and many other sorts of “shifty” uses of indexicals and contextuals (viz., adjectival expressions like local and foreign and verbs like come and go that presuppose a perspective point) have been extensively discussed over the last decade or so and I cannot here rehearse all the arguments pro and con the idea that they involve context shifts – shifts from the context of utterance to an orienting context that fixes the time and place of the reported events, or to a perspective‐fixing context that represents a subjective point of view other than the speaker’s own. Moreover, there are arguments for thinking that some of the allegedly shifty uses of here and now are not after all pure indexical uses but rather demonstrative or anaphoric uses. Nevertheless, it seems clear that at the very least these uses are a challenge to the Simple-Minded view mentioned in section 2.1. Moreover, their interpretation does seem to presuppose a rich set of contextual assumptions, including information about a speaker’s intentions. However, this need not be a threat to the notion of a context as an index if we accept Predelli’s account. On this view, all the pragmatic work needed to fix the intended agent, location, and time of the context would be done pre-semantically, and it would be no challenge to the idea that the semantic interpretation function takes an index as one of its arguments. I therefore turn now to ask how Predelli’s notion of an index relates to the Relevance Theory notion of context as the mutual cognitive environment of the interlocutors in a conversation and whether Predelli’s account indeed shows that wide context plays no role in an account of the semantic/ interpretive processing of indexicals.
2.3 Contexts in language performance studies One might think that Predelli’s view shows us a way to incorporate both cognitive environments and indexes in a single theoretical framework for explaining the
100
Anne Bezuidenhout
use of indexicals in language production and understanding. The proposal would be to identify cognitive environments with the pragmatic contexts that Predelli argues are represented by indexes in the pre-semantic pragmatic phase of interpretation. Can we reconcile these two seemingly opposed conceptions of context in this way and see them as belonging to two separate phases of some overarching process? I think not, for several reasons. Firstly, it is far from clear that we should understand Predelli’s account as a model of language in use. It is true that he talks of an Interpretive System and even sometimes of an interpretive “module” (Predelli 2005: 11, n.2). His discussion is filled with talk of inputs to and outputs from this System, of syntactic processes of disambiguation, about the effects generated from such processes/ procedures, and so on. See Predelli (2005: 11–23). This certainly makes it seem as though his aim is to offer a model of the sorts of causal processes involved in utterance interpretation. However, I believe it would be uncharitable to construe his account as a psychological model of language processing, as it is pretty clear it would be hopeless so construed. If we think of the pre-semantic phase as one in which a hearer takes a speaker’s utterance as input and decides which bests represents the utterance’s encoded meaning and its intended context, it should be clear that most of the work in understanding the speaker’s utterance will already have been done and it would be completely unnecessary to assume a subsequent stage of semantic processing. This is most vivid in the case of elliptical or sub-sentential speech. If the representational phase includes the construction of a clause containing syntactico-semantic structure that was phonetically null in the input string, then the hearer has already understood what was implicit in the speaker’s message. Similarly, if the representational phase involves fixing the intended values of any contextual parameters, then the hearer has already understood what is being referred to. It would be unnecessary to have the Interpretive System run the pair to find out what the speaker intended to convey. Therefore, Predelli’s account should not be construed as a psychological model but rather as a vivid way of determining what belongs to the study of semantics proper. He shows us a way in principle to keep the boundary between semantics and pragmatics sharp. We do this by dividing pragmatics into pre-semantic and post-semantic phases, thereby avoiding the idea of “pragmatic intrusion” into truth-conditional content. Another way of construing Predelli’s distinction between pre-semantic pragmatics and semantics proper is to see the former as concerned with the epistemological question of how we come to know what indexicals refer to and the latter as concerned with the metaphysical question as to what the actual semantic values that constitute the context of interpretation (viz. index) are. Bianchi (2013: 133) suggests such division of labor between
The role of context in semantics: A Relevance Theory perspective
101
epistemology and metaphysics. Either way, Predelli’s account belongs to the philosophy of language and is at best very indirectly connected to psychology and to performance studies. Secondly, Predelli’s notion of an index is a representational one. Indexes represent pragmatic (wide) contexts. As Predelli (2005:40) says, indexes represent “the relevant contextual background in the austere format of a collection of co‐ ordinates”. However, cognitive environments in the Relevance Theory sense are themselves representational, and thus it would be a mistake to think that indexes represent such environments. While both cognitive environments and indexes are representational in nature, indexes are theoretical or abstract representations, whereas cognitive environments are mental representations, albeit posited as part of a psychological theory of language performance systems and so to that extent cognitive environments are also “theoretical” entities – entities that are not directly observable but that are posited as part of a theoretical explanation of language performance. Thirdly, if the contexts that indexes represent are understood commonsensically as physical situations that contain flesh and blood people with all their properties (including their mental ones), as well as material objects and events and all their properties, and also the physical sounds, hand movements or inscriptions that are the physical vehicles of acts of verbal communication, then cognitive environments are just as much representations of context as are indexes. And they are austere in their own ways. It is well known that what we mentally represent is partial, biased, contains inattentional blind spots, and so on. I conclude that we cannot accommodate both cognitive environments and indexes in a single overarching model of language performance or, more specifically, a model of indexical use. Thus we must choose between these two notions and it seems clear that cognitive environments, which mentally represent (wide) contexts, are of the right sort to play an explanatory role in an account of indexical use. Indexes are at best only indirectly connected to such empirical investigations. Thus Recanati’s argument (quoted above) wins the day over Predelli’s argument, at least if we are concerned to study language performance as opposed to offering a metaphysical theory of meaning à la Bianchi (2013).
3 What are cognitive environments and how are they constrained? In this section, I would like to briefly address the question as to the nature of and constraints on cognitive environments and then in the following section,
102
Anne Bezuidenhout
section 4, to sketch the way in which cognitive environments thus characterized enter into an account of indexical use. A person’s cognitive environment consists of all the information that is manifest to that individual at a time. Information is manifest if it is either currently explicitly represented in the individual’s working memory or could be so represented, under appropriate conditions. Sperber and Wilson (1986: 39) add that, at the time of representation, the person must accept the information as true or probably true. Sperber and Wilson (2014) say that information in a cognitive environment is more or less manifest depending on the degree of salience of that information and its degree of probability. A cognitive environment can thus be pictured as a field of represented or representable information where the manifestness of that information increases along two orthogonal dimensions, as illustrated in Fig. 2 below:
Probabilty
Salience
Fig. 2: Degrees of manifestness of information in a cognitive environment.
As noted above, information is manifest if it is either currently represented in working memory or could be so represented, under appropriate conditions. Almost anything could be manifest to you under some condition or another. You could explicitly represent arcane facts about quasars, if you were first to undergo graduate training in astrophysics. Clearly we do not want to allow conditions like this. What is manifest to you now is constrained by what you actually already know – by your current store of encyclopedic knowledge. Something that you already know in this sense could be activated in working memory and thus is a part of your current cognitive environment. We also want to allow that information in your current physical environment is manifest if, via the redirection of attentional resources and the (normal) operation of one or more of your sensory organs, you would become sensorily aware of this information. For example, the hum of the refrigerator in your apartment is manifest to you right now because even though you aren’t paying attention to that hum, you would become aware of it if you directed your hearing towards the part of the room where your refrigerator is located. In addition, information about your interlocutors’ informative and
The role of context in semantics: A Relevance Theory perspective
103
c ommunicative intentions, their preferences and abilities, as well as socio-cultural information about them (e.g., facts about their social status) will be manifest to you. Finally, interlocutors generally need to keep track of what has transpired so far in the conversation, and thus linguistic information at various levels (e.g., phonological, semantic, and pragmatic) will be mutually manifest. However, even this is still too broad, as we do not want to say that everything you know at a particular time is a part of the cognitive environment used to understand an utterance by one of your conversational partners at that time, say an utterance of Colin is here. In the 1970s, when Chomsky, Fodor and others popularized the idea of a modular mind, they argued that perceptual modules and the language module (Chomsky’s language faculty narrowly conceived, mentioned in section 1 above) are encapsulated, in the sense that the operations of such modules are sensitive only to domain specific information written in a proprietary code that is manipulated in accordance with module‐internal rules, but that the central processor is unencapsulated, in the sense that any piece of information could become relevant for its purposes. This modularist tradition treats pragmatic reasoning as a central cognitive process par excellence. Relevance Theory in its earliest iterations embraced the idea of pragmatics as part of the central processing system, and of pragmatic inferences as unencapsulated. Where Sperber and Wilson (1986) parted company with Fodor was that, while Fodor believed there could be no scientific account of central processes, Sperber and Wilson thought it was possible to give a scientific account of pragmatics, their Relevance Theory framework being just such an account. Sperber and Wilson resisted the idea of a pragmatics module, I believe, because accounts of pragmatics modules on offer in the 1980s appealed to ideas from then existing pragmatic theories (e.g., speech act theory, theories of politeness, and theories of turn-taking in conversation) and simply assumed that the pragmatic processing module (or its sub‐modules) operated according to internalized versions of these theories. See Kasher (1984). Relevance Theory on the other hand was intended to be a single overarching framework to explain all pragmatic processing and as an alternative to theories such as Grice’s theory of conversational implicature, Brown and Levinson’s politeness theory, and Searle’s speech act theory. According to Sperber and Wilson, pragmatic reasoning operates according to cognitive and communicative principles of relevance. These principles are conveniently summarized in Wilson and Sperber (2012: 6–7): Cognitive principle of relevance: Human cognition tends to be geared to the maximization of relevance. Communicative principle of relevance: Every act of overt communication conveys a presumption of its own optimal relevance.
104
Anne Bezuidenhout
Presumption of optimal relevance: (a) The utterance is relevant enough to be worth processing; (b) it is the most relevant one compatible with the communicator’s abilities and preferences.
The relevance of an ostensive stimulus – and utterances are paradigm examples of such stimuli – is a matter of degree. The greater are the cognitive effects of processing a stimulus and the smaller is the cognitive effort needed to process it, the more relevant will that stimulus be. (Wilson and Sperber 2012: 88). Cognitive effects are modifications of an individual’s cognitive environment. The information carried by the stimulus will either combine with existing assumptions to yield a contextual implication or will combine with existing information to alter the probability or saliency of that information or will interact so as to eliminate or cancel existing assumptions (Wilson and Sperber 2012: 176, 200). With respect to the effort factor, Wilson and Sperber mention two things that impact it, namely the form in which the stimulus is presented (its audibility, legibility, dialect, register, syntactic form, familiarity, etc.) and the degree to which it taxes the resources of memory and imagination. (Wilson and Sperber 2012: 176). As I said, initially Sperber and Wilson resisted the idea of a pragmatics module. In later iterations of Relevance Theory, however, they became more receptive to the idea of a pragmatics module, and suggested it is a specialized sub-module of the Theory of Mind module that has been proposed to explain the human capacity for “mind reading” – in the sense of being readily and automatically able to “read” a person’s purposive (ostensive) behavior and see it as a manifestation of that person’s underlying beliefs, intentions and desires (Wilson and Sperber 2012: 261–278). Since we want assumptions about interlocutors’ informative, communicative and referential intentions to be a part of the interlocutors’ cognitive environments and hence part of what is mutually manifest to them, we want the output from the Theory of Mind Module to interact with other information in a person’s cognitive environment. However, precisely because we need more than the output from the Theory of Mind Module to produce and understand verbal utterances, I would resist the idea of a pragmatics module as a sub-module of the Theory of Mind Module. Rather than accounting for the constraints on cognitive environments by arguing that pragmatics is modular and hence that the information that feeds into pragmatic processes is constrained in virtue of being domain specific or encapsulated, the key to understanding how cognitive environments are constrained is to understand the principles according to which they are modified in the course of conversational exchanges. We saw above that according to Relevance Theory, a speaker’s utterance, which is an ostensive stimulus, is intended to modify the cognitive environments of the conversational partners. Understanding the principles that govern such modifications is
The role of context in semantics: A Relevance Theory perspective
105
important for understanding cognitive environments and their role in utterance production and comprehension. The attempt to shape one’s interlocutors’ cognitive environments is what I call “conversational tailoring”. I have described some aspects of such tailoring in the context of a discussion of generalized conversational implicatures in Bezuidenhout (2015b) and in the context of a discussion of presuppositions in Bezuidenhout (2010, 2014). Not everything that is a part of your cognitive environment and hence that is now manifest to you is currently activated in the “spotlight of attention”. What determines which parts of one’s cognitive environment are active at any moment? This is just the old “frame problem” raised in early Artificial Intelligence research in the 1960s and 1970s. The issue initially arose in attempts to account for commonsense reasoning and conscious deliberation (including scientific theorizing), which modularists such as Fodor believed could never be explained scientifically. However, many of the same issues arise in the context of language performance studies, given the crucial role of pragmatic inferences in language performance. Such pragmatic processing is equally the target of those who believe that nonmodular processes are not scientifically tractable. Given how old the frame problem is, I cannot hope to summarize all the attempts to answer it here. For current purposes, I need merely to invoke the well‐entrenched notions of mental frames, scripts, or schemas, which are representations stored in long term memory and that contain default information about typical scenarios we have encountered in the past (e.g., we may have stored a “restaurant” script that contains information about the sorts of things that typically transpire in a restaurant). When one of these scripts is triggered by the current conversational situation, this will already to a large degree shape how you will interpret the actors, actions and events that transpire. Similarly, it has been argued that we operate with various sorts of mental schemas (or heuristics), such as a causal schema that accounts for our tendency to see cooccurring events as causally connected, or a purposive schema that accounts for our tendency to see events (even natural events) as the result of underlying agency or purpose. These ideas of frames, scripts and schemas certainly explain why not everything you know will be invoked in a particular conversational situation and how information in your cognitive environment might be limited to situation-specific information. However, they don’t explain what invokes a particular script or frame in the first place. It is on this issue that I think Relevance Theory has some valuable insights to offer. I will briefly describe two ways in which ideas from Relevance Theory can explain how particular frames are invoked in the course of a conversation and thus explain how interlocutors are able to shape one another’s cognitive environments.
106
Anne Bezuidenhout
Firstly, the very idea that the inferential phase of language comprehension is a relevance-driven process will partially explain why a particular frame is accessed. Relevance is a matter of balancing contextual effects and processing effort. Degree of effort is a factor in determining relevance. Thus the more effort it takes to retrieve a frame, the less relevant the stimulus that is interpreted relative to that frame would be (other things being equal). I assume that the processing effort involved in accessing a frame is affected by such factors as recency of use, frequency of use, ease of access (e.g., cases of self‐deception may involve blocking access to certain frames; a mother who refuses to read the signs of her son’s drug addiction is mentally blocking a way of framing the situation), and so on. And on the flip side, the more contextual effects that are yielded by invoking a particular frame, the more relevant the stimulus that is interpreted relative to that frame would be (other things being equal). Suppose you are dining in a restaurant with a married couple. As the meal draws to a close, the husband, gesturing towards his wife, suddenly comes out with an utterance of (11): (11) She will pay for it. You take him to have said that his wife will pay for the meal. After all, you are in a restaurant and maybe you have been thinking for a while about the touchy issue of who will pick up the tab for the meal. So the restaurant script will have been recently activated and be easily accessible. Moreover, the utterance interpreted relative to this frame answers a question that was on your mind and, you reasonably assume, on the minds of the other diners. That is, interpreted in this way, the husband’s utterance is highly relevant, and the relevance driven comprehension procedure is likely to halt at this point, without seeking alternative possible interpretations. Of course, it is always possible that the husband has been silently fuming all evening about his wife’s infidelity, which he found out about just before the start of the dinner when the private detective he had hired to follow his wife slipped him a set of photographs of his wife in flagrante delicto. Unable to contain himself anymore, he gives vent to his desire for revenge against his wife by uttering (11). In other words, in such a situation, interpreting (11) as the claim that the wife will pay for the meal involves a misunderstanding, although not one that can be blamed on the interpreters. This raises an interesting issue that is tangential to my main point, namely whether in this situation the husband’s intention to express his desire for revenge is the one that fixes the context for interpretation. I will not pursue this issue here. Suffice it to say that I follow Bianchi (2013), who argues that context‐fixing intentions must be ones that the speaker makes available to the hearer, in the
The role of context in semantics: A Relevance Theory perspective
107
sense that in the normal course of things a hearer could reasonably be expected to discern these intentions. Intentions that are hidden or opaque are not communicative intentions. The husband had no communicative intentions behind his utterance of (11). If the misunderstanding is detected, some conversational repair can be done and the correct frame for interpreting the husband’s utterance of (11) can be made accessible to the interlocutors. Or the husband might exploit the misunderstanding to cover up his embarrassment at showing his emotions in public and let the interpretation that his wife will pay for the dinner stand. The second Relevance Theory idea that can help to explain how and when frames, scripts or schemas will be invoked is the idea of procedural meaning. The idea is that some linguistic expressions encode procedures rather than concepts. There may also be expressions that encode both concepts and procedures. See Wilson and Sperber (1993, 2012). The notion of procedural meaning “arises from the observation that there are linguistic expressions which can help hearers to follow the [speaker’s] intended inferential path.” (Clark, 2013: 312). Expressions that encode procedures are called procedural markers. One important sub-class of such markers is the class of discourse connectives, such as but, however, nevertheless, inferential so, because, after all, moreover, and so on. Many other lexical items, linguistic constructions, or (para-)linguistic features have been said to encode procedural meanings. For example, prosody, syntactic structures (e.g., it‐clefts), interjections (e.g., wow and oh), illocutionary force indicators and hearsay particles have all been taken to encode procedures. For further discussion, see Andersen and Fretheim (2000) and Escandell‐Vidal, Leonetti, and Ahern (2011). Blakemore (1987, 2002) has discussed discourse markers from a Relevance Theory perspective, treating them as expressions that guide the interferential phase of interpretation. She argues that they encode rules that indicate the type of cognitive environment in which the utterances of which they are a part are to be processed. In this way they guide the hearer towards intended contextual effects and hence reduce the overall effort required to process the discourse. As Clark (2013) puts it, procedural expressions “guide the hearer by making one of the possible inferential connections more salient than others.” (Clark, 2013: 310). They “indicate the way in which the utterance might be relevant and so make particular ways of processing more salient” (Clark, 2013: 326). Consider for example a situation in which a speaker utters the sequence of sentences in (12) below: (12) He’s a heavy drinker. His wife left him. There are several possible ways in which these two sentences could hang together to create a coherent discourse segment. The speaker could intend to convey that
108
Anne Bezuidenhout
the wife’s leaving was a consequence of the man’s heavy drinking or that it was a reason for his heavy drinking. Or perhaps his heavy drinking and his wife’s leaving are just two of the ways in which the man’s life is falling apart. There are undoubtedly other possibilities too. Assumptions in the mutual cognitive environment (including the topic currently under discussion, the identity of the man, what his relation to the interlocutors is, and so on) may make it easy to figure out which of these possible relations between the two events is the one that the speaker intends to be talking about. However, there are ways for the speaker to make it easier on his listeners by using a discourse connective, such as in (13)–(15) below: (13) He’s a heavy drinker, so his wife left him. (14) He’s a heavy drinker because his wife left him. (15) He’s a heavy drinker. Moreover, his wife left him. In (13), one is constrained to interpret the man’s drinking as the reason for the wife’s leaving rather than the other way about. In (14), the inferential connection is just the reverse. In both these cases, certain types of scripts or frames are likely to be invoked. In the case of (13), a script about how alcoholic men behave in domestic situations might lead one to infer that the man physically abused his wife while under the influence of alcohol. In the case of (14), a script about how a man scorned by his love would act might lead one to infer that the man sought solace in his local pub night after night and became addicted to alcohol. In the case of (15), a script about people down and out on their luck might lead one to infer that the man has experienced further woes, such as having huge debts, and so on. There is a lot more that could be said about the notion of procedural meaning, about procedural markers, and about discourse markers in particular. There is an extensive literature devoted to all these issues. For current purposes, the main point I wish to emphasize is that procedural markers can help us understand how particular frames, scripts and schemas can be invoked in the course of an unfolding conversation. Procedural markers are used as a way to shape the cognitive environments of one’s interlocutors. They are devices for conversational tailoring.
4 Cognitive environments and the use of indexicals In section 3, I discussed the problem of how cognitive environments are constrained and argued that the use of expressions with procedural meanings is one of the ways in which interlocutors engage in conversational tailoring, shaping
The role of context in semantics: A Relevance Theory perspective
109
each other’s cognitive environments. This applies also to the case of indexicals and other referring expressions. According to Wilson and Sperber (1993), indexicals like I and now and pronouns like he and she encode procedural meanings that help hearers to infer the speaker’s intended referents. That is, their encoded Kaplanian characters are rules for guiding an interpreter’s search for the intended referent. The operative word here is “guide”. Recanati (2013: 162) calls the encoded meanings of indexicals “linguistic modes of presentation” and notes that they “may not be determinate enough to fix the reference” and often merely constrain the reference.2 The examples (2) and (3) using now, here and we that were discussed in section 2.1 illustrate how reference is merely constrained by encoded character. Given that the encoded character of here is something like the location of the context, this is not determinate enough to capture the speaker’s intended reference. Information that is mutually manifest to the conversational partners is needed in addition in order for the listener to determine the spatial extent intended in the particular context by the speaker’s use of here. The notion that certain referring expressions (e.g., definite descriptions, demonstratives, and pronouns) encode information that helps guide referential processing is well entrenched in linguistics and psychology. For example, Gundel, Hedberg, and Zacharski (1993) argue that referring expressions encode something about the degree of salience or “givenness” of their referents. When an entity is “in focus” it has the highest degree of salience, when it is “activated” is has a lower degree of salience, and so on. (Strictly speaking, it is mental representations of entities that are in focus, activated, etc.).3 This encoded information about degree of salience is an instruction about the type of process needed to retrieve or create an appropriate mental representation of the speaker’s intended referent and hence is a way that speakers have of shaping the cognitive environments of their interlocutors – of engaging in conversational tailoring. For example, pronouns (such as it, he and she in their deictic or anaphoric uses) signal that the intended referent is “in focus”, in the sense
2 Recanati (2013) goes on to argue that is a psychological mode of presentation that fixes reference. He conceives of such psychological modes of presentation as mental files that fix reference in a relational, as opposed to a “satisfactional”, manner. 3 Gundel and Mulkern (1998) have argued that resources beyond those of Relevance Theory are needed to account for the way this sort of procedural information helps constrain referential processing. However, Scott (2011) argues convincingly against this claim. Moreover, referring back to Fig. 2, it should be clear that providing procedural information about degrees of salience is one way to make information more manifest and hence to modify one’s interlocutor’s cognitive environment.
110
Anne Bezuidenhout
that the intended referent is something that is topical or has just been mentioned. Consider the use of it in (16) below: (16) There was a cockroach on the counter. It was very large. An interpreter of the second sentence in (16) is instructed to take her mental representation of the topical entity – the cockroach – and link it to her current linguistic representation, thereby ensuring that the mental dossier created when the cockroach was first mentioned (an action that in turn was directed by the speaker’s use of the indefinite expression a cockroach) will be updated with information about the cockroach’s large size. As mentioned above, Sperber and Wilson (1993) allow that expressions can encode both procedures and concepts. Pronouns do also encode conceptual information. Thus he encodes information about number, gender, and animacy in addition to signaling the “in focus” status of the intended referent. Uses of demonstratives such as this and that accompanied by some sort of ostensive gesture or the speaker’s direction of gaze, signal that the intended entity is “activated” (or “activate-able”) in the interlocutors’ mutual cognitive environment. The encoded information that guides processing would be something like: The referent is the entity that you currently perceive or are being directed to perceive or is the object of your mutual gaze with the speaker. Thus consider (17) below: (17) (Looking ostensibly at a wrench) Please hand me that. The interpreter is instructed to take his current perceptual representation of the wrench, which is being jointly attended to by speaker and listener, and to link this to his current linguistic representation of the speaker’s verbal instruction, thereby ensuring that this perceptual information will guide the requested actions involving the wrench, in this case the need for the listener to physically grasp the wrench and hand it to the speaker. Had the speaker used a complex demonstrative such as that wrench or that wrench on the hood of the car, the encoded conceptual information would provide additional constraints for the hearer to take into consideration in determining the speaker’s referent. In addition, this and that encode information about the proximal versus distal location of the intended referent respectively. A similar contrast exists for pairs of expressions such as these and those. Gundel and Mulkern (1998) discuss an example where the speaker’s use of the definite description the primitive reptiles, as opposed to the complex demonstrative these primitive reptiles, can make referential processing less smooth, because the definite description signals merely that the speaker’s intended referent is “identifiable”, whereas the complex demonstrative would signal that the referent is “activated”. In a context where more than one entity (or set of entities) is identifiable but only one is activated, processing is likely to be easier if the
The role of context in semantics: A Relevance Theory perspective
111
speaker uses a complex demonstrative. (Remember, these labels “activated” and “identifiable” strictly speaking say something about the status of mental representations of entities rather than the entities themselves). However, there are also cases where it seems to make little difference to referential processing whether the speaker uses a definite description versus a complex demonstrative. What would justify a speaker’s use of a complex demonstrative in such cases? Scott (2011) shows that in cases such as these, the encoded procedural information in these primitive reptiles may help hearers to recover implicated contents in addition to guiding the process of reference‐fixing. In such cases, had the speaker used a definite description such as the primitive reptiles instead, these implicated contents would not have been as readily accessible. Thus the use of the complex demonstrative yields an overall interpretation that has a greater degree of relevance (in the technical Relevance Theory sense of producing greater contextual effects for no extra gratuitous processing effort). A very similar point is made by Acton & Potts (2014), who argue that when the use of a demonstrative goes beyond what is strictly required for the speaker’s referential purposes, the speaker will be taken to communicate something in addition to the identity of the referent. So, for example, when Sarah Palin refers to ‘that Henry Kissinger’, the demonstrative is unnecessary, as the proper name by itself is a sufficiently strong referential device. What she communicates beyond the identity of the referent is that she and her interlocutors share a perspective on this man and his politics. Acton & Potts claim that the use of a demonstrative presupposes “perspectival alignment” and thus can foster a “sense of common ground and shared perspective” (2014: 4). The precise pragmatic and social effects that arise from the use of a demonstrative are “ultimately determined jointly by the discourse participants during interaction” (2014: 5). Sometimes the use of demonstratives can be deemed inappropriate, because people feel they do not share common ground with the speaker. This can arouse negative affective states in the audience. Thus Sarah Palin’s (over-)use of demonstratives, resonated with her supporters but alienated others. When I say that speakers use referring expressions with encoded procedural meanings to shape their interlocutors’ cognitive environments and to guide their referential interpretations, I do not mean to claim that this is a conscious process, either on the speaker’s or the interpreter’s end. On the contrary, it is a process that goes on unconsciously and automatically. However, there is ample evidence from psycholinguistic studies of language performance that supports the idea that people are sensitive to encoded information about the givenness status of (representations of) a speaker’s intended reference. On the comprehension end, there is evidence of a so-called Repeated Name Penalty. It has been found that there is a processing cost involved in
112
Anne Bezuidenhout
interpreting stretches of dialog when an entity’s full name or a definite description is repeated, as opposed cases where a pronoun or some other reduced form is used to make repeated references to that entity. See Gordon and Chan (1995), Almor (1999), and Almor and Nair (2007). This suggests that people expect full forms to refer to less salient entities rather than to ones that are already in focus or activated. On the production end, there is evidence that speakers prefer to use pronouns to refer to topical entities. See Rohde and Kehler (2013) for evidence in favor of this Topichood Hypothesis. A note of caution is in order; these effects have been teased out from a complicated set of experimental manipulations of multiple factors that might influence reference resolution and referring expression choice. These include factors about order-of-mention (first vs. last mention), grammatical role (subject vs. object), thematic role (agent vs. patient), information structural factors (topic vs. non-topic), semantic factors (implicit causality information encoded in verbs), and discourse factors (type of coherence relation holding between discourse segments). Much more needs to be said about the type of procedural information encoded by referring expressions of various types and how this constrains referential processing. Also, much more work is needed to explain how the experimental evidence that has accumulated about referential processing bears on these matters. However, my main point in this section was to indicate the way in which the Relevance Theory notions of context as what is mutually manifest and of procedural meaning fit into a psychological account of indexical use and to show that Relevance Theory aligns well with current empirical work on these topics.
5 Conclusions I have laid out some of the details of the psychological notion of context assumed within the Relevance Theory framework, namely the notion of context as the mutual cognitive environment of the conversational participants. I tried to show that the idea of a cognitive environment meshes well with current experimental work on language performance systems. I have also tried to show how the ideas of relevance-driven processing and of procedural meaning can help us to characterize cognitive environments and to explain the ways in which they are constrained in the course of language production and understanding. A slightly expanded treatment of the ways in which they are constrained can be found in Bezuidenhout (2015a). These ideas help us to see some of the ways in which interlocutors engage in a process of conversational tailoring. I also made some brief c omments
The role of context in semantics: A Relevance Theory perspective
113
about the role of cognitive environments in the production and understanding of indexicals and other referring expressions. This of course by no means exhausts the study of contexts as cognitive environments or of the role of context in language performance. For one thing, there is much more to language performance than the use of the handful of referring expressions I have discussed above. However, I hope this rather limited discussion succeeds in pointing to the ways in which Relevance Theory and its notions of a cognitive environment, of relevancedriven inferential processing, and of procedural meaning can contribute to current experimental work in language performance studies. The main take-away message of the paper is that if we turn our attention to an account of language performance, a different set of issues comes into focus than the ones typically of concern to philosophers of language. The one issue that I have paid the most attention to is the idea that interlocutors are engaged in a process of conversational tailoring. This idea itself is not new, and ideas from many different theoretical frameworks and from a diverse range of empirical investigations into language use could and should feed into a systematic account of such conversational tailoring. Such an overarching account is one that I will have to leave for another time and another venue.
References Acton, Eric & Potts, Christopher. 2014. That straight talk: Sarah Palin and the sociolinguistics of demonstratives. Journal of Sociolinguistics 18: 3–31. Almor, Amit. 1999. Noun-phrase anaphora and focus: The informational load hypothesis. Psychological Review 106: 748–765. Almor, Amit & Veena Nair. 2007. The form of referential expressions in discourse. Language and Linguistics Compass 1: 84–99. Andersen, Gisle & Thorstein Fretheim (eds.). 2000. Pragmatic markers and propositional attitude. Amsterdam: John Benjamins. Bezuidenhout, Anne. 2010. Grice on presupposition. In Klaus Petrus (ed.), Meaning and analysis: Themes from H. Paul Grice, 75–102. Basingstoke: Palgrave McMillan. Bezuidenhout, Anne. 2014. Presupposition failure and the assertive enterprise. Topoi online first. http://link.springer.com/article/10.1007/s11245-014-9265-4 (accessed 20 December 2015). Bezuidenhout, Anne. 2015a. Cognitive environments and conversational tailoring. Croatian Journal of Philosophy 15 (44): 151–162. Bezuidenhout, Anne. 2015b. The implicit dimension of meaning: Ways of “filling in” and “filling out” content. Erkenntnis 80(1): 89–109. Bianchi, Claudia. 2013. Illocutions in context. In Carlo Penco and Filippo Domaneschi (eds.), What is said and what is not, 131–139. Stanford: Center for the Study of Language and Information (CSLI) Publications. Blakemore, Dianne. 1987. Semantic constraints on relevance. Oxford: Blackwell.
114
Anne Bezuidenhout
Blakemore, Dianne. 2002. Relevance and linguistic meaning. Cambridge: Cambridge University Press. Clark, Billy. 2013. Relevance theory. Cambridge: Cambridge University Press. Escandell-Vidal, Victoria, Manuel Leonetti & Aoife Ahern (eds.). 2011. Procedural meaning: Problems and perspectives. Bingley, UK: Emerald Group Publishing. Gordon, Peter & Davina Chan. 1995. Pronouns, passives and discourse coherence. Journal of Memory and Language 34: 216–231. Gundel, Jeanette, Nancy Hedberg & Ron Zacharski. 1993. Cognitive status and the form of referring expressions in discourse. Language 69: 274–307. Gundel, Jeanette & Ann Mulkern. 1998. Quantity implicatures in reference understanding. Pragmatics and Cognition 6: 21–45. Hauser, Marc, Noam Chomsky & Tecumseh W. Fitch. 2002. The faculty of language: What is it, who has it, and how did it evolve? Science 298: 1569–1579. Kasher, Asa. 1984. Pragmatics and the modularity of mind. Journal of Pragmatics 8: 539–557. Kempson, Ruth, Wilfried Myer-Viol & Dov Gabbay. 2000. Dynamic syntax: The flow of language understanding. Oxford: Blackwell. Lewis, David. 1979. Scorekeeping in a language game. Journal of Philosophical Logic 8: 339–359. Nunberg, Geoffrey. 1993. Indexicality and deixis. Linguistics and Philosophy 16: 1–43. Perry, John. 2001. Reference and reflexivity. Stanford: CSLI Publications. Predelli, Stefano. 1998. Utterance, interpretation and the logic of indexicals. Mind and Language 13: 400–414. Predelli, Stefano. 2005. Contexts: Meaning, truth and the use of language. Oxford: Oxford University Press. Recanati, Francois. 2001. Are ‘here’ and ‘now’ indexicals? Texte 127(8): 115–127. Recanati, Francois. 2004. Literal meaning. Cambridge: Cambridge University Press. Recanati, Francois. 2005. Literalism and contextualism: Some varieties. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy: Knowledge, meaning, and truth, 171–196. Oxford: Oxford University Press. Recanati, Francois. 2013. Reference through mental files: Indexicals and definite descriptions. In Carlo Penco & Filippo Domaneschi (eds.), What is said and what is not, 159–173. Stanford: Center for the Study of Language and Information (CSLI) Publications. Rohde, Hannah & Andrew Kehler. 2014. Grammatical and information-structural influences on pronoun production. Language, Cognition and Neuroscience 29: 912–927. Scott. Karen. 2011. Beyond reference: Concepts, procedures and referring expressions. In Victoria Escandell-Vidal, Manuel Leonetti & Aoife Ahern (eds.), Procedural meaning: Problems and perspectives, 183–204. Bingley, UK: Emerald Group Publishing. Sperber, Dan & Deirdre Wilson. 1986. Relevance: Cognition and communication. Oxford: Blackwell. Sperber, Dan & Deirdre Wilson. 2014. What is the role of meaning in communication? Paper presented at the philosophy of language and linguistics conference, interuniversity center, University of Dubrovnik, Croatia, 8–12 September. Wilson, Deirdre & Dan Sperber. 1993. Linguistic form and relevance. Lingua 90: 1–25. Wilson, Deirdre & Dan Sperber. 2012. Meaning and relevance. Cambridge: Cambridge University Press.
Part 2: The contextual turn and the case for language use
Stefano Predelli
Boo semantics: Radical nonfactualism and non truth-conditional meaning Abstract: This essay discusses a semantics for Radical Nonfactualism, a non-cognitivist approach to normative expressions. Its scope is explicitly and exclusively semantic: my aim is not that of intervening in the (considerable) meta-ethical debate about the tenability and desirability of this or that version of non-cognitivism. I rather choose Radical Nonfactualism because of its characteristic semantic properties – which are, in turn, of philosophical interest primarily due to their ‘intermediate’ position between the truth-conditional and the ‘pragmatic’ dimensions. I start with a comparison with expressives (as in an old-fashioned ‘boo-hurray’ theory of normative predicate), and eventually develop Radical Nonfactualism on the model of (but with important differences from) the apparatus for non truth-conditional meaning I put forth in my 2013 book Meaning Without Truth.
This essay discusses a semantics for Radical Nonfactualism, a non-cognitivist approach to normative expressions. Its scope is explicitly semantic: I choose Radical Nonfactualism because of its commitment to a level of meaning somewhat intermediate between the classic sense of truth-conditional meaning, and meaning-external issues of presumably merely pragmatic interest. In particular, starting with a comparison with expressives, I eventually develop Radical Nonfactualism from the viewpoint of (an adaptation of) the philosophical framework for non truth‐conditional meaning I put forth in my 2013 book Meaning Without Truth (Predelli 2013). More precisely, the motivation for my project stems from the central tenet in the so- called nonfactualist and expressivist traditions in metaethics, namely the thesis that normative predicates such as “is wrong” or “is permissible” provide no descriptive contribution to the truth-conditions of the sentences in which they occur.1 As already hinted, my primary aim here is not that of evaluating this
1 For a small sample from the considerable literature on nonfactualism and expressivism, see Blackburn (1984), Blackburn (1993), Gibbard (2003), and Schroeder (2008a). ‘Radical’ in the Radical Nonfactualist approach which I take as an inspiration is supposed to contrast with ‘Hybrid’; for ‘hybrid’ versions of nonfactualism, namely what Schroeder (2008b) calls “adulterated positions”, see among many Barker (2000) and Ridge (2006). DOI 10.1515/9781501504327-007
118
Stefano Predelli
nonfactualist approach to normative predicates per se, but rather that of discussing its semantic consequences. In particular, if normative predicates do behave according to nonfactualist guidelines, their semantic analysis must be grounded on a nonfactualist view of meaning in general. This view must in turn inevitably abandon certain traditional assumptions about meaning, and must rather appeal to the semantic resources required for the study of the non truth-conditionally relevant aspects of an expression’s conventional profile. The main topic of this essay is thus the representation of this sort of nonfactualist meaning within a standard model-theoretic framework for formal languages.2 My conclusions are presented in the final section of this essay, where I proceed to the semantic analysis of a nonfactualist fragment LRN. The main body of my essay is devoted to an informal discussion of the resources needed for that analysis. In the first four sections, I take as my main inspiration certain insights from the philosophical literature on non truth-conditional meaning, starting with A. J. Ayer’s analogy between normative predicates and expressive interjections. I then proceed to an informal discussion of their interactions with apparently truth-conditionally relevant linguistic phenomena such as negation. On the basis of these considerations, I rehearse the methodological consequences of my approach in section four, where I discuss the relationships between the study of nonfactualist meaning and the traditional understanding of semantics as primarily concerned with “judgments of synonymy, entailment, contradiction, and so on” (Dowty et al. 1981: 2).3
2 My considerations bear important points of contact with the influential studies of non truthconditional meaning in Kaplan (1999) and Potts (2005). The peculiarities of my views are nevertheless worthy of mention. Kaplan (1999) approaches non truth-conditional meaning from the viewpoint of so-called propositional semantics, whereas Potts (2005) puts forth a type-theoretic semantic treatment. Although both Potts and Kaplan, not unlike myself, focus on issues of ‘context’, neither of them develops the sort of model-theoretic double-indexed framework which I find most suitable for the representation of non truth-conditional meaning. 3 The caveats at the beginning of this essay may be reinforced by noting that none of these projects are (and indeed should be) committed to the conceptual elucidation of the subject matter for the language under study: the uncontroversially fruitful application of the tools of intensional logic to the semantic analysis of expressions such as ‘possibly’ or ‘might’, to mention a celebrated and familiar example, hardly suffices as a clarification of the central concepts in the semanticists’ metalanguage, such as the very notion of a possible world and of related metaphysical ideas. In a similar vein, the fundamental concepts at work in my treatment of my normative fragments are fully defined within the metalanguage for the model-theoretic analysis in section five, and remain amenable to a relatively wide variety of (uncontroversially not equivalent) interpretations from the viewpoint of metaethics and the study of normativity.
Boo semantics: Radical nonfactualism and non truth-conditional meaning
119
1 Preliminaries: “Boo” and “Hurray” A. J. Ayer, a precursor of nonfactualism, insisted that normative predicates are similar to “thick exclamation marks” or to an appropriately stern tone of voice. Ayer’s analogy is not particularly illuminating, partly due to the fact that his terms of comparison, be they punctuation devices or intonational contours, hardly provide pedagogically transparent explanations.4 In this respect, the somewhat disrespectful moniker for Ayer-style non-cognitivism, the Boo/Hurray theory, strikes a more promising chord, especially in light of the growing interest towards expressions in the vicinity of “boo” and “hurray” – simple interjections such as “ouch” or “oops” and a few related constructions.5 Accordingly, what guides these preliminaries is the proposal to give due recognition to at least some similarities between normative sentences such as (1) stealing is wrong and more colourful modes of expression such as “stealing, boo!” or “boo for stealing!” Though more promising than Ayer’s menacing intonation, “boo” and its lot only provide partial similarities: surely, not every linguistically interesting property of “boo” is applicable to “is wrong” and predicates of that sort. Indeed, that “is wrong” is a predicate already indicates that the Ayer-inspired analogy may not be taken that seriously when it comes to syntax: on any decent account of grammatical acceptability, sentences such as (1) or “wrong actions will be punished” need to be recognised as well formed on the basis of the same regularities governing non-normative cases such as (2) stealing is difficult or “green cars will be impounded”, and independently of the (controversial) syntactic distribution of interjections.6
4 On exclamations and exclamation marks, though, see Nunberg (1990), Rett (2008), and Zanuttini and Portner (2003). The linguistic literature on intonation is of course extensive – in the semantic realm, especially because of its repercussions on matters of focus (for an overview, see Kadmon (2001); see also Bolinger (1989)). 5 On interjections, see Ameka (1992a) and Ameka (1992b), Cuenca (2000), Kaplan (1999), Norrick (2009), Padilla-Cruz (2009), Poggi (2009), Wharton (2009), Predelli (2013), Wierzbicka (1992), and Wilkins (1992). As indicative of the growing literature on slurs, see Anderson and Lepore (2001), Aoun and Choueri (2000), Hom (2008), Tirrell (1999), and Williamson (2009). For a pioneering collection of essays see Zwicky et al. (1971). 6 See (a)–(b) in section five for a reflection of this idea within the formal fragment L0. For a sample of the debate about the role of interjections within phrase structure “proper”, see McCawley (1998) and Potts (2005).
120
Stefano Predelli
The extent to which the analogy can be pursued is not syntactic (or, for that matter, phonological or pragmatic) but semantic: up to a certain point, the semantic behaviour of “is wrong” turns out to be closer to that of “boo” than to that of “is difficult” or “will be impounded”. Up to a certain point: as I explain later, even the semantic side of the analogy will have to be taken with a grain of salt. Yet, at least one property of interjections is bound to catch the Radical Nonfactualists’ attention from the outset: unlike a sob of despair, “boo” and “hurray” are meaningful expressions, which, unlike “is difficult”, fail to provide any descriptive contribution.7 Meaningful but non-descriptive: just the way Radical Nonfactualists would like to have it with “is wrong”. The idea of meaningful but non-descriptive expressions such as “boo” or “hurray” may then be represented in terms of constraints that address a level other than that of truth-conditions. Indeed, lone uses of “boo” or “hurray” are uncontroversially truth- conditionally idle (they are, in common parlance, not even truth-apt): the exclamation “boo!” may perhaps be chastised on a variety of grounds, but may presumably not be criticized for being false. Accordingly, these expressions’ occurrences in (at least superficially) sentential contexts, as in (3) hurray, stealing is difficult fail to affect the truth-conditions of the sentences they flank: (3) is intuitively true if and only if stealing is difficult, that is, if and only if “stealing is difficult” is true. As a result, the meaning appropriate for “hurray” or “boo” must have to do not with their effects on matters of truth, but, to use a deliberately neutral term, with their repercussions on questions of expressive adequacy (hereinafter e-adequacy) – as in the preliminary rough idea that, by virtue of its meaning, “boo” (or “hurray”) is expressively adequate in the mouth of an appropriately unsatisfied (or elated) speaker. Of course, much of interest remains to be said about what, at this stage, is merely a schematic allusion to a dedicated non truth-conditional dimension, namely e-adequacy. Still, lack of detail notwithstanding, one feature in this approach to “hurray” or “boo” is already worthy of attention: these expression’s e-adequacy conditions make reference to speakers. Utterly independently of questions of nonfactualism, the idea of a speaker has received a great deal of attention from the semantic viewpoint. In particular, speakers are uncontroversially elements of what are commonly called contexts – for instance, in the clause
7 This contrasts with Ayer’s occasional idea of verbal “ejaculations”, as in his eminently casual mention of “cries of pain” side by side with “words of command”: “people groan in all languages ... but say ‘ouch’ only in English” (Stevenson 1944: 39).
Boo semantics: Radical nonfactualism and non truth-conditional meaning
121
devoted to the truth-conditional interpretation of the first person pronoun “I”.8 The traditional formal understanding of a context as a sequence of parameters may then immediately be adapted within a first step in the elucidation of my hitherto schematic allusion to e-adequacy: (4) a context c is e-adequate for “boo” (“hurray”) if, in c, ca is unfavourably (favourably) disposed towards something, where, as usual, ca is the speaker (or “agent”, see Kaplan 1977) of c.9 Although little more than a quasi-formal rendering of the rough idea put forth above, this much highlights an intriguing parallelism: modulo the appeal to e-adequacy, (4) exemplifies a pattern familiar from the study of meaning-encoded contextual dependence, that is, of indexicality. As in, say, (5) a context c is t-adequate for “stealing is difficult now” if, in c, stealing is difficult at ct, where ct is the time of c, and “c is t-adequate for s” is a suggestive rephrasing of the familiar idea of a sentence s being true with respect to a context c.10 This parallelism yields a welcome payoff in section three. There, I proceed to the discussion of languages which combine descriptive and non-descriptive devices – as in (3) or, according to nonfactualism, in “if stealing is wrong then stealing is difficult”. Before I tackle this issue, in the next sections I present the Radical Nonfactualist approach to normative sentences in a format parallel to (4) and (5), and I discuss the semantic aspects of normative predicates that are not unveiled by their comparison with expressive interjections, including their behaviour in embedded environments.
8 My temporary identification of the “occupier” of the appropriate contextual role with “the speaker” is highly negotiable. For independently motivated discussions of the relationships between contextual parameters and the individuals involved in conversational exchanges, see for instance Kaplan (1977), Predelli (2005), and Predelli (2013); see also footnotes 9 and 12. 9 In a more appropriate account, (4) will need to be amended so as to mention an unfavourably/ favourably disposition towards some object or event “salient in c”. On the deictic profile of interjections, see Padilla-Cruz (2009) and Wilkins (1992). 10 Possibly not only “suggestive”: when t-adequacy is flanked by other adequacy dimensions, such as what I later call “n-adequacy”, the metasemantic issue of their relationships with “truth proper” may remain a topic of contention: see footnote 13. The locution “in c” is colloquial, but, for my purpose here, harmlessly so. The idea, familiar from the standard semantics for indexical languages, is that s is true with respect to c if, given the possible world w (or, more generally, point of evaluation) “determined” by c, the semantic value of s at c with respect to w is T; see Kaplan (1977) or Lewis (1980).
122
Stefano Predelli
2 Towards “is wrong” and “is permissible” The parallelism between (4) and (5) provides an intriguing template for nonfactualist approaches to normative predicates. In particular, (4) presents the meaning of “boo” in terms as manageable as those involved in straightforward truth-conditional clauses such as (5), with the notable exception of its appeal to a (here not further analysed) semantic dimension other than the notion of t-adequacy introduced in (5).11 Extending this model, mutatis mutandis, to “is wrong” is an offer Radical Nonfactualists cannot refuse. A Radical Nonfactualist treatment of normative sentences may then take the following general shape: (6) a context c is n-adequate for “stealing is wrong” if N(c) where n-adequacy is some non truth-conditional dimension or other, and where N(c) is a corresponding contextual constraint (as in: if context c has properties N). The details in this respect may well remain negotiable from the semantic viewpoint (though obviously not from the viewpoint of a full-fledged nonfactualist theory). But, once again, Radical Nonfactualists may do worse than to take (4) as inspiration, informally along the lines of (7) a context c is n-adequate for “stealing is wrong” if cj, the judge of c, disapproves of stealing in c, where cj may be identified with the speaker (as in individualist versions of nonfactualism), the speaker’s community (as in some form of collectivist nonfactualism), or, for that matter, some less immediately speaker-related parameter.12 My deliberately evasive attitude with respect to the make-up and nature of “the judge” extends to the other fundamental notion involved in (7): the very idea of
11 To reiterate the important background caveat from footnote 3: deliberately not further analyzed, for the point is not that of unveiling the sort of content contributed by expressive or normative expressions, but rather that of highlighting their role within a semantic apparatus also sensitive to questions of truth-conditions. 12 See Sinclair (2009) for an overview of some alternatives. Clearly, the “identifications” to which I allude pertain to the criteria appropriate for an application of this or that semantic framework to actual instances of language use – compare with the arguably not inevitable choice of the speaker as the occupier of the “agent” parameter suitable for the interpretation of particular acts of speaking (see footnote 9). The formalism in section five, clause (e), regiments judges as individuals, but this assumption is adopted merely for the sake of illustration: from the semantic viewpoint, “judge” represents whatever contextual coordinate is appropriate as a stand-in for the sort of conditions and/or objects which may independently be deemed to be worthy of attention.
Boo semantics: Radical nonfactualism and non truth-conditional meaning
123
n-adequacy. Indeed, in speaking of n-adequacy as “some non truth-conditional dimension”, I have remained silent with respect to its constitution (with the exception of its distinction from t-adequacy), and to the nature of the attitude with which it is associated, the disapproval mentioned in (7).13 Of course, as long as these (and, I suppose, a few other) dimensions remain at the purely schematic level, a fullfledged non-cognitivist analysis of normativity remains a distant prospect. But that is as it should be from the viewpoint of a semantic theory – for it would surely be perverse to expect that the study of normative expressions ought to suffice as a clarification of what these expressions are for. To demand that an analysis of moral talk yield insights in the very nature of normativity, after all, would be as bizarre as the expectation that, say, the analysis of interjections in (4) entail a theory of human emotions, or that the treatment of temporal adverbs in (5) yield an explanation of the metaphysical nature of time. This lack of extra-semantic commitments entails a similarly casual attitude when it comes to the distinction between the non truth-conditional dimension addressed by normative predicates, and that appropriate for the interjections from section one. Indeed, as far as Radical Nonfactualism is concerned, this distinction may either be almost non-existent, as in the crudely emotivist suggestion that “is wrong” expresses some generic displeasure, or fundamental, as in the view that what is at issue in (7) is a dedicated moral aversion. Of greater urgency, on the other hand, are more genuinely semantic contrasts between normative predicates on the one hand, and “boo” or “hurray” on the other – for it is these contrasts that highlight the semantic significance of the distinction between n-adequacy and e-adequacy, independently of the further aspects possibly involved in their intuitive interpretation. At least two points in this respect deserve to be mentioned before I come to compositionality, embeddings, and logic, in sections three and four. Note first that opposing interjections such as “boo” and “hurray” exemplify a different kind of contrast from that appropriate for incompatible normative expressions, such as the clash between (1), repeated here,
13 For metasemantic questions having to do with the relationships between t-adequacy (i.e., the sense of truth-conditions relevant for semantic inquiry), n-adequacy, and truth (in some appropriate sense of the term), see the discussion of the relationships between non-cognitivism and alternative theories of truth in Blackburn (1998), Divers and Miller (1994), Schroeder (2008a) chapter 11, and Smith (1994a). Since by definition Radical Nonfactualism refuses to classify normative sentences as true/false, in the sense in which this classification applies to descriptive scenarios, the tendency in the vernacular to label the former as “true” must anyway obviously not be taken as an insurmountable stumbling block for anyone interested in nonfactualism in the first place – independently of the philosophically important considerations about truth to which I have just alluded.
124
Stefano Predelli
(1) stealing is wrong and (8) stealing is permissible.14 In a nutshell: as the parenthetical remarks in (4) indicate, “boo” invokes some sort of unfavourable disposition, but the clause for “hurray” appeals to a gleeful stance not equivalent to the absence of hostility.15 “Is permissible”, on the other hand, covers a semantic space which includes mere tolerance, and which is thus larger than that corresponding to normative enthusiasm. Accordingly, the (informal equivalent of the) analysis of (8) should not require the judge’s approval of stealing, and should rather rest satisfied with mere lack of disfavour (momentarily disregarding the important complication discussed in the next paragraph).16 Continuing with the parallel with (4), what is needed must be, roughly, (9) a context c is n-adequate for “stealing is wrong” (“stealing is permissible”) if cj, the judge of c, does (does not) disapprove of stealing in c. The “important complication” to which I just alluded emerges from the possibility that, at least given some choices of what counts as a judge, judges are not, to put it colourfully, universally judgmental – say, the possibility that, when it comes to stealing, judges fail to take any stance (either because they have not considered the issue, or because they have not managed to reach a decision on that topic).17 Since indecision entails lack of disapproval, a clause such as (9) would incorrectly evaluate (8) as n-adequate on occasions where neither tolerance nor enthusiasm correctly describe the judge’s attitudes. If the judge’s inclinations guide the n-adequacy of normative predicates, then, inattention or agnosticism must be kept at bay, as in a definition explicitly sensitive to cases where one has simply not considered the question of stealing, or where one has anyway failed
14 For simplicity’s sake, I take “permissible” (and for that matter “wrong”) as exemplars of univocally normative predicates. I do not exclude that a full-fledged analysis of the uses of these expressions in English (in any tolerable sense of “English”) may unveil different meanings and connotations for these and cognate expressions. 15 Unsurprisingly, no interjection I can think of expresses the speaker’s indifference, though the juvenile “stealing, whatever …” may get close enough. 16 Of course, these remarks are of significance for the present project insofar as they highlight certain features of the mutual relationships between the parameters under discussion – further details in the attitudes mentioned within my informal explanation (“gleeful stances”, “approval”, etc.) remain tangential to the issue at hand. 17 For further comments and distinctions, see Dreier (2006); for a detailed discussion of these issues from the expressivist viewpoint, see Schroeder (2008a).
Boo semantics: Radical nonfactualism and non truth-conditional meaning
125
to reach any verdict about it. In turn, this naturally leads towards a three-valued approach, as in (still roughly and preliminarily) (10) “stealing is permissible” is (i) n-adequate with respect to c if cj is decided and does not disapprove of stealing in c; (ii) n-inadequate if cj is decided and disapproves of stealing in c; and (iii) n-indeterminate if cj is undecided where a judge is decided (with respect to stealing) if it approves, merely tolerates, or disapproves of stealing, and is undecided otherwise.18 Note that this predicament has an immediate analogue in the realm of t-adequacy, as long as the latter remains sensitive to presumed “factual indecisions” such as vagueness, future contingency, and the like. The consequences in the truth-conditional territory are familiar enough: the extension for, say, “is not bald” is distinct from the complement of the extension of “is bald”. As a result, approaches sensitive to these phenomena end up taking into consideration, side by side with t-adequacy (truth) and t-inadequacy (falsehood), some sort of t-indeterminacy, in turn formalisable within a non-bivalent framework. Since questions related to this choice are of independent interest, my approach in section five simply assumes, both for the normative and descriptive domains, one of the options currently on the market – any alternative remains easily obtainable given obvious adjustments. In my brief discussion of indeterminacy, I contrasted the normative predicate “is wrong” with its lexical counterpart, “is permissible”. Yet, when I shifted to the descriptive domain, I compared “is bald” with a compositionally richer construct, “is not bald”. This naturally raises the question of the treatment for a complex predicate arguably equivalent to “is permissible”, namely “is not wrong”. Which, in turn, nicely leads me to the discussion of negation and other embeddings, for factual and normative predicates alike.
3 Embeddings The behaviour of embedded sentences has traditionally been regarded as a stumbling block for speech-act oriented theories of meaning. To cite an obvious case in the descriptive domain, consider the idea that the meaning of (2), repeated here, (2) stealing is difficult
18 Bivalence, of course, is obtainable under the assumption of “universally decided judges”, for instance in systems where lack of prohibition entails permissibility.
126
Stefano Predelli
has to do with the fact that this sentence is used to assert that stealing is difficult. Since someone who utters (11) if stealing is difficult then pillaging is widespread does not thereby assert that stealing is difficult, it apparently follows that the occurrence of (2) in (11) does not have the meaning it has when occurring on its own. But this, so the story goes, disqualifies as equivocations obviously valid steps, such as that from (2) and (11) to “pillaging is widespread”.19 Although many details here deserve closer scrutiny, this particular type of embedding problem is clearly irrelevant for Radical Nonfactualism, a theory of the meaning of expressions independent of any commitment at the level of speechacts.20 Yet, a different embedding-related worry ought to be taken more seriously by anyone intrigued by the analogy with which I begun this essay, that between normative predicates and interjections. For, when it comes to embeddings, “boo” and its peers behave very differently from “is wrong” or “is permissible”: in a fashionable jargon, interjections are nondisplaceable, but normative predicates are not.21 For instance, switching for perspicuity’s sake from “boo” to “damn”, (12) if that damn Jones wins, I will resign
19 All of this is in close proximity of (at least some of) the (many) issues traditionally subsumed under the label of “the Frege-Geach problem”. See Geach (1965) and Searle (1962), and, for a sample of the contemporary debate, Sinclair (2011), Sinnott-Armstrong (2000), or Unwin (1999). 20 According to the standard framework for speech-act theory, speech-acts are representable in terms of pairs consisting of a content and a force. The semantic questions pertaining to content (and truth-conditions, entailment, etc.) are thus understood as “prior” to the specification of the conditions for nondefective speech acts – conditions which typically make explicitly reference to the very relationships between that content and a speaker (see among many Searle (1969), Searle (1975), and Searle and Vanderveken (1985)). That is not to say that the semantic considerations in this essay will have no interesting repercussions when it comes to the study of illocutionary force. In particular, the idea of content emerging from my analysis (see footnote 28) may well require either the introduction of “dedicated speech acts” (roughly along the lines of the informal idea of “expression”), or amendment of the customary conditions for familiar types of force, such as assertion. The study of the developments in speech act theory appropriate for Radical Nonfactualism is an important project, which nevertheless (uncontroversially, as far as I can tell) has no significant repercussions on the topic of this essay, the study of the semantic properties of sentences of a certain sort. 21 The term “nondisplaceable” comes from Potts (2005). The presumed exceptions occasionally put forth in the literature on this subject blatantly (and provably) appeal to metalinguistic or echoic scenarios in the sense of Horn (1989) or McCawley (1991) – tellingly, italics attempting to encode the appropriate intonational contour are pretty much a constant in these examples, see Barker (2000).
Boo semantics: Radical nonfactualism and non truth-conditional meaning
127
is unconditionally unsympathetic to Jones, regardless of the fact that the interjection is embedded within the conditional’s antecedent.22 But (13) if stealing is wrong then pillaging is widespread had better not be associated with unconditional disapproval of stealing, for (13) may well occur in the mouth of someone with no particular attitude towards stealing.23 Still, this much is, in itself, no cause for concern from a Radical Nonfactualist viewpoint: what served as an already partial analogy in sections one and two simply breaks down when it comes to embeddings. Just as any analysis of “boo” and “damn” presumably needs to take due care of their resistance to displacement, then, a Radical Nonfactualist treatment of normative predicates needs to take equal care of the fact that n-adequacy remains very much affected by “not”, “if”, “possibly”, and the like. It is at this juncture that the correspondences between t-adequacy clauses such as (5) and n-adequacy clauses such as (7), (9), or (10), yield an interesting payoff. After all, uncontroversially, “not”, “if”, etc. also have a bearing on the t-adequacy of factual sentences. Equally uncontroversially, the bearing in question entails straightforward effects of displaceability in the descriptive domain, as in the idea that, say, “stealing is not difficult” is t-adequate if “stealing is difficult” is not. What this suggests is that the semantics for the normative portions of the language may proceed with an approach to “not” or “if” parallel to that already at work in descriptive sentences – a highly desirable result, given that “not” and “if” are clearly non-ambiguously equally applicable to, say, (1) and (2). The idea, in a nutshell, is that the compositional clauses in the analysis of a descriptive and normative language be presented in terms insensitive to the distinctions between t-adequacy and n-adequacy – in contrast with the rules for a wider fragment that encompasses interjections, where the non-displacement of e-adequacy needs to be addressed on an ad hoc basis. Focusing on the contrast between descriptive and normative cases, let me then say that an atomic sentence s is adequate (inadequate, indeterminate) if either (i) s is factual and t-adequate (t-inadequate, t-indeterminate) or (ii) s is normative and n-adequate (n-inadequate, n-indeterminate – tedious and obvious mention of “with respect to c” often omitted in what follows). And, on the basis of this definition, let me
22 I switch to “damn” because it is an expressive with a more easily identifiable syntactic location: apparently, the occurrences of “damn” in (12) falls within the syntactic domain of “if”. 23 On this issue, see Anand (2007), Barker (2000), Geurts (2007), Hom (2008), Lasersohn (2007), Potts (2007), Ridge (2006), Sauerland (2007), and Schlenker (2007).
128
Stefano Predelli
proceed to the description of the effects of the aforementioned operators in terms of the adequacy of their arguments, as in section five, clauses (j)–(l). (The ultimate tenability of each definition, of course, will have to be assessed in detail, depending among other things on one’s favourite treatment of sentential operators in non-bivalent environments).24 A brief recap is in order. According to Radical Nonfactualism, the treatment of normative predicates appeals to clauses different from, but interestingly parallel to, those for descriptive expressions. Some aspects of these clauses may initially be highlighted by the analogy with interjections. In particular, (i) normative fragments bring to the foreground a semantic dimension distinct from t-adequacy, that is, they involve effects of meaning that are not recorded at the level of truthconditions; and (ii) these non truth-conditional aspects of meaning address a dedicated contextual parameter. Yet, the analogy with interjections ceases to be relevant when it comes to (iii) the contrast between normative predicates such as “is wrong” and “is permissible”, and (iv) their behaviour within embedded contexts. The trademark features of the resulting framework are: (v) an appeal to a non truth-conditional level of meaning, and, as a consequence, to a dedicated dimension of n-adequacy; (vi) an analysis of normative expressions in terms of a relation between appropriate contextual parameters (“judges”) and, say, stealing or pillaging; and (vii) a three-valued compositional apparatus sensitive to the general dimensions of adequacy, inadequacy, and indeterminacy.
4 Radical Nonfactualism Here, finally, is my presentation of a language that conforms to the dictates of Radical Nonfactualism. Starting with a syntactic issue that may be worth mentioning at the outset: in a Radical Nonfactualist treatment of any descriptiveand-normative language (English no less than the formal languages L0 and LRN I define in section five), the distinction between descriptive and normative predicates (and, accordingly, sentences) must be recorded at the lexical level. For instance, whether a predicate is descriptive or normative is encoded in L0 in terms of distinct variables such as “F” and “N”, and in English, presumably, in terms
24 To return to an already mentioned issue (footnote 10): the technical notion of adequacy may not only casually be voiced in the vernacular in terms of “truth”, but may even turn out to be interpretable as truth in some independently interesting (or, as the case may be, harmlessly uninteresting) sense. On these issues, inevitably, Radical Nonfactualism takes an explicitly neutral stance.
Boo semantics: Radical nonfactualism and non truth-conditional meaning
129
of dedicated lexical features (as in, say: “is difficult: intransitive, descriptive, ...”; “is wrong: intransitive, normative, ...”). This much is an immediate requirement of any nonfactualist approach, at least if conjoined with standard views about the syntax/semantics interface: it is a necessary condition for the development of nonfactualism that the distinction between descriptive and normative predicates be detectable at a syntactic level visible from the semantic viewpoint – in the common jargon, that it be recorded at the level of Logical Form. Descriptive predicates are then evaluated in the obvious (possibly modulo the details of non-bivalence) manner, that is, they are mapped to extensions and anti-extensions (as usual, with respect to independently needed parameters, say, possible worlds or times). For instance, the extension and anti-extension of “is difficult” are, respectively, the class of difficult and of non-difficult actions; and the extension and anti-extension of “is bald” are, respectively, the class of non-borderline hairy individuals, and the class of decidedly bald-headed people. Normative predicates are also assigned extensions and anti-extensions, but of a special kind: syntactically n-adic normative predicates/relations, such as the monadic “is wrong”, are semantically interpreted by appealing to (n+1)-adic relations involving a contextual parameter. As in: (14) the extension of “is wrong” in c is the class {t: Dec (cj, t) & Neg(cj, t)} the anti-extension of “is wrong” in c is the class {t: Dec (cj, t) & ¬ Neg(cj, t)} where cj is c’s judge parameter, and Dec and Neg are relations between individuals and action-types, presumably corresponding to the pre-theoretic ideas of “being decided” and being “unfavourably disposed” introduced in section three.25 Note that the semantic constraint on Neg relevant from the Radical Nonfactualist viewpoint consists exactly in its motivation for a trivalent framework: the complement of the extension of a normative predicate does not inevitably exhaust the remainder of the universe of discourse – that is, the union of extension and anti-extension is a possibly proper subclass of the universe. Non-cognitivists, of course, will be interested in a theory of moral psychology that explains Dec and Neg as the reflection of non-cognitive dispositions playing the appropriate roles within
25 Regarding “action-types”: throughout this essay, I followed the customary analysis of, say, “stealing” or “pillaging” (and, for that matter, “theft” or “murder”) as singular terms. Semantically, of course, the issue is worthy of greater attention – in particular, given the semanticist’s interest in reflecting the systematic structure in cases such as “not stealing is wrong” or “drinking while driving is wrong”. The options here range from radically clausal analyses of gerundive nominals (such as Lees’ contention that they result from sentences by transformation, Lees (1960)), to the nontransformational treatment pioneered by Schachter (1976); see also Pullum (1991) for a more updated approach.
130
Stefano Predelli
the judge’s decisional profile – the details in this respect, as usual, are only loosely constrained by the demands put forth by a Radical Nonfactualist semantics. The regularities for atomic instances may then proceed as customary for both descriptive and normative scenarios: (15) a sentence of the form p(t), p a descriptive (normative) predicate and t a clause, is t-adequate (n-adequate) with respect to a context c if the semantic value of t in c is a member of the extension of p in c; it is t-inadequate (n-inadequate) if it is a member of the anti-extension of p in c; and is t-indeterminate (-indeterminate) otherwise. Since the compositional processes generating non-atomic sentences inevitably yield sentences involving both descriptive and normative predicates, what now needs to be tracked at that level are both t-adequacy and n-adequacy results, along the general dimension of adequacy mentioned in section three. Given this notion, then, the outcomes are familiar and uncontroversial (again, modulo any independent controversy regarding the treatment of connectives in non-bivalent frameworks): say, (16) a sentence of the form not S is adequate (inadequate, indeterminate) in c if S is inadequate (adequate, indeterminate) in c, or (17) a sentence of the form S1 or S2 is adequate in c if either S1 or S2 are adequate in c, it is inadequate in c if both S1 and S2 are inadequate in c, and is indeterminate otherwise. So, for a negated sentence, we find that (18) stealing is not wrong is adequate in c if “stealing is wrong” is inadequate in c, that is, if stealing belongs in the anti-extension of “is wrong” – intuitively, if the relevant decided judge has failed to emit a verdict of condemnation. By the same token, in disjunctive cases, a sentence such as (19) stealing is wrong or grass is green turns out to be adequate if either “stealing is wrong” is n-adequate or “grass is green” is t-adequate, that is, if either grass is in the extension of “is green” or cj is appropriately related to stealing. The idea in (15) reflects one of the aspect of Radical Nonfactualism on which I insisted throughout this essay: by virtue of their meaning, normative predicates make a contribution at the level of n-adequacy, rather than at the level of t-adequacy. It is an immediate consequence of this approach that meaning may no longer be semantically regimented in terms of the customary, truth-oriented properties – given my focus on context-dependence throughout this essay, in
Boo semantics: Radical nonfactualism and non truth-conditional meaning
131
terms of what are commonly called characters, namely functions from contexts to intensions.26 Still, here as in the case of compositional clauses such as for instance (16) or (17), the modifications are obvious enough, and are achievable by replacing the descriptively focused idea of t-adequacy with the more general property of adequacy. This, as I explain in the remainder of this section, entails an equally obvious adaptation of the idea of “meaning guaranteed” inferential steps, and consequently of logical properties and relations. Since a descriptive expression’s meaning is fully formalizable in terms of character, the idea of meaning-guaranteed inferences immediately collapses into the notion of character-guarantee in the descriptive domain. So, a descriptive sentence such as “either grass is green or grass is not green” is judged to be true by virtue of character alone, in the sense that the truth-conditional contributions of its components (and, of course, of its syntactic structure) suffice for a conclusion of truth with respect to any context whatsoever. Similarly, a descriptive sentence such as “grass is green” character-guarantees the descriptive sentence “either grass is green or snow is white”, in the sense that, given any context with respect to which the former is true, the latter is true as well.27 When reformulated along model-theoretic lines, and given a choice of so-called “logical constants”, this much may be rephrased in the traditional logical terms of validity and entailment. For instance, “grass is green or grass is not green” is judged to be valid, in the sense that the characters of “grass is green” and “not” are such that “grass is green or grass is not green” ends up being true with respect to any context in any model. Similarly, “grass is green” is said to entail “either grass is green or snow is white” in the customary sense that the characters of “grass is green”, of “Snow is white”, and of “either … or” guarantee that whenever “grass is green” is true with respect to a model and a context, “either grass is green or snow is white” is true as well. For Radical Nonfactualists, an analysis of meaning-guaranteed relations in terms of character is insufficient when it comes to normative sentences, for the trivial reason that these are not the sort of expressions to which t-adequacy applies in the first place. Still, just as t-adequacy and n-adequacy are generalizable in terms of adequacy tout-court, a similarly neutral notion may be introduced, so as to extend the idea of character-guarantee consistently with the nonfactualist approach to the normative domain. At this point, the outcome should come as no surprise: since, for a Radical Nonfactualist, meaning encompasses not only character but also a dimension sensitive to n-adequacy, what must be at issue is a definition focused on adequacy in general, rather than on truth (t-adequacy) alone. As a preliminary
26 In the sense of Kaplan (1977). See in this respect the notion of bias in Predelli (2013). 27 Or, alternatively, never false; see the allusion to different three-valued logics in the text.
132
Stefano Predelli
stab to be sharpened along model theoretic lines in what follows: (i) a sentence s1 entailsRF (“entails in the logic for Radical Nonfactualism”) a sentence s2 if s2 is adequate with respect to all c such that s1 is adequate with respect to c; and (ii) a sentence s is validRF (“valid in the logic for Radical Nonfactualism”) if s is appropriate with respect to all c (with the usual modification given alternative choices of designated values in non-bivalent environments). So, for instance, the sentences (18) and (19), repeated here (18) stealing is not wrong (19) stealing is wrong or grass is green RF
entail
(20) grass is green because (i) “stealing is wrong or grass is green” is adequate in a context c if either “grass is green” is t-adequate in c or “stealing is wrong” is n-adequate in c, and (ii) “stealing is not wrong” is adequate in c if “stealing is wrong” is n-inadequate in c. It thus follows that for any such c, (20) must be t-adequate, and hence, being descriptive, adequate.28 Note that the point is not only that intuitive results may be obtained, both in the purely normative domain and in mixed cases such as (19). The point is also that the foregoing conclusions of adequacy are grounded on precisely the same regularities that explain the entailment from, say, (21) either pillaging is difficult or grass is green and (22) pillaging is not difficult to (20), that is, the regularities relevant for the preservation of t-adequacy. What this suggests is that the customary resources at work in the descriptive domain are
28 This is not to say that the distinction between strict entailment (in the sense of a character-based relationship, a fortiori one pertinent to t-adequacy) and the notion of entailmentRF introduced here may be obliterated. Suppose that the language under study contains expressions appropriate for whatever judge-involving relationship matters for n-adequacy – as in, say, “cj disapproves of murder”. Then, if c is adequate (n-adequate) for, say, “murder is wrong”, it follows that cj decidedly disapproves of murder, and hence that c is adequate (t-adequate) for “cj disapproves of murder”, so that the former sentence entailsRF the latter. Radical Nonfactualists would nevertheless (rightly, I suppose) resist the conclusion that the relationship at issue is describable in terms of strict entailment (see somewhat parallel considerations with respect to the “logic of interjections” in Potts (2005) and Predelli (2010) and (2013)).
Boo semantics: Radical nonfactualism and non truth-conditional meaning
133
straightforwardly applicable to the case of normative or mixed instances, modulo, of course, the focus on n-adequacy which any form of nonfactualism ought to envision ab initio.29 The ideas informally put forth in this section may be sharpened in terms of the framework developed in the next and final section of this essay. As already mentioned, the point there is not simply that of a formal exercise: my artificial fragments L0 and LRN are endowed with exactly those expressive features required by the genuinely semantic commitments of Radical Nonfactualism. Although they regiment normative languages along lines suitable for non-cognitivism, then, they also stress the extent to which a theory of meaning may afford to, and in fact ought to maintain a neutral attitude when it comes to issues arising within neighbouring but distinct quarters in the non-cognitivist programme. If Radical Nonfactualism, as encoded in L0 and LRN, provides an appropriate semantic theory, it is because it is in the position of yielding the right results about, say, the behaviour of normative sentences in embedded contexts, the interactions between the descriptive and normative sides of language, and their mutual logical relationships. If Radical Nonfactualism yields an appropriate semantic theory, on the other hand, it may rest satisfied with schematic indications of the role that some of its central notions are supposed to play, with no particular commitments to the actual meaning of particular expressions in this or that actual language. A formal model-theoretic apparatus, in which judges are represented merely as customized contextual parameters, and where their attitudes towards action- types are reflected by a non-constant dedicated relation, provides an ideal pedagogical device for this purpose.
5 L0 and LRN My formalisation of the general traits of Radical Nonfactualism proceeds by introducing two artificial fragments: a language L0 in which descriptive and normative predicates are suitably distinguished, but in which no normative predicate is treated as constant across models, and a language LRN built upon L0, but involving a constant one-place normative predicate.
29 The classic metasemantic notions close to many philosophers’ hearts are similarly immediately obtainable from that non-negotiable nonfactualist opening move, the rejection of t-adequacy as the sole semantic dimension. As in: the meaning of s is a function mapping contexts to contents*, where the content* of s with respect to c is a function mapping points of evaluation (chosen on the basis of the usual, independent considerations) to results of adequacy (or non-adequacy, or indecision).
134
Stefano Predelli
L0 Syntax L0 is characterized by a familiar syntax: predicates are coupled with the appropriate number of terms, and non-atomic sentences may be built by means of the usual connectives. More precisely, the following lexicon and formation rules exhaust L0’s syntax. Lexicon: singular terms, connectives and parentheses, a class of factual n-ary relation symbols, and a disjoint class of normative n-ary relation symbols. Formation rules: for all factual Fn, normative Nn, and terms t1 ... tn30 a. Fn(t1 ... tn) is an (atomic factual) sentence b. Nn(t1 ... tn) is an (atomic normative) sentence c. if s1 and s2 are sentences, so are ¬s1, s1∨s2, etc. L0 Models
Models are the usual affairs containing a universe of discourse U, a collection of contexts C, and the resources for the interpretation of the simple expressions in L0. In the case at hand, these resources consist of a pair of interpretation functions I+ and I−, intuitively related to the outcomes of extension and anti-extension appropriate for a non-bivalent framework. More precisely, then, an L0 model m is an n-tuple of the form m = , such that d. U is a class of individuals,31 e. C is a class of contexts (with C ⊆ U),32 and f. I+ and I− are functions such that, i. for any term t, I+(t) = I−(t) ∈ U ii. for any n-ary factual relation Fn, I+(Fn ) = {: i1 ... in ∈ U} and I−(Fn) = {: i1 ... in ∈ U}, such that I +(Fn) ∩ I−(Fn) = ∅ iii. for any n-ary normative relation Nn, I+(Nn) = {: j ∈ C and i1 ... in ∈ U}, and I−(Nn) = {: j ∈ C and i1 ... in ∈ U}, such that I+(Fn) ∩ I−(Fn) = ∅ 30 Expressions such as “murder”, “murdering”, etc., of course, are to be understood as terms merely for the purpose of the formalism. See also footnotes 24 and 32. 31 Superscripts hereinafter often omitted for legibility’s sake. 32 In the absence of indexical expressions, contexts are here envisaged only as providers of that parameter, and are accordingly formalised as themselves being individuals (judges), as in (e).
Boo semantics: Radical nonfactualism and non truth-conditional meaning
135
L0 Adequacy The definitions that follow are standard, modulo the appeals to adequacy, inadequacy, and indeterminacy discussed above. So, informally, factual atomic sentences are t-adequate if the objects assigned to the terms they contain belong in the predicate’s interpretation; and normative atomic sentences are n-adequate if the judge and the objects to which the terms refer stand in the relation contributed by the predicate’s interpretation. The usual compositional clauses take care of cases of negation, disjunction, and other connectives, and a definition introduces a familiar model-theoretic notion of entailment. g. for any atomic factual sentence S of the form Fn(t1 ... tn), t-adeq(S)m,c if ∈ I+(F) t-inad(S)m,c if ∈ I−(F) t-indet(S)m,c if neither t-adeq(S)m,c nor t-inad(S)m,c h. for any atomic normative sentence S of the form Nn(t1 ... tn), n-adeq(S)m,c if ∈ I+(Nn) n-inad(S)m,c if ∈ I−(Nn) n-indet(S)m,c if neither n-adeq(S)m,c nor n-inad(S)m,c i. for any atomic sentence S, adeq(S)m,c (inad, indet) =df S is factual and t-adeq(S)m,c (t-inad, t-indet) or S is normative and n-adeq(S)m,c (n-inad, n-indet) j. for any sentence S of the form ¬S1: (a) adeq(S)m,c if inad(S1)m,c and (b) inad(S)m,c if adeq(S1)m,c k. for any sentence S of the form S1∨S2: (a) adeq(S)m,c if either adeq(S1)m,c or adeq(S2)m,c, (b) inad(S)m,c if both inad(S1)m,c and inad(S2)m,c, and (c) indet(S)m,c otherwise l. etc. m. a sentence S1 entailsRF a sentence S2 =df for no m and c ∈ Cm, adeq(S1)m,c but not adeq(S2)m,c
As already mentioned, the genuinely normative dimension, and, a fortiori, the Radical Nonfactualist treatment of it, remain in the background in the treatment put forth thus far, with the notable exception of certain general differences between descriptive and normative predicates. This lacuna is corrected in the next fragment, where the peculiarity of certain normative predicates is brought to light by their treatment as constant expressions. LRN Syntax
In this fragment, the lexicon is now extended so as to include a constant normative predicate (“wrong”) and a dedicated 2-place relation (eventually aimed at
136
Stefano Predelli
encoding whatever agent-involving relationship nonfactualists may deem to be appropriate for “wrong”). Lexicon: logical symbols and terms as in L0. Descriptive predicates as in L0, together with the 2-place relation-symbol R. The one-place normative predicate “wrong”. Formation rules as in L0. LRN Models Models are as above, together with clauses pertaining to the interpretation of the constant normative predicate “wrong”: 2. A LRN model is of the form m = , with –– U and C as in (d)-(e) and –– I+ and I− as in (f), together with 3. I+(wrong) = { : j ∈ C and i ∈ U and ∈ I+(R)} 4. I−(wrong) = { : j ∈ C and i ∈ U and ∈ I−(R)}
So, for instance, applying to LRN the account of L0 adequacy in (g)–(h), the normative atomic sentence “wrong(t)” is n-adequate in m and c if ∈ I+(wrong), where i = I+(t). Hence by (p) n-adeqm,c(wrong(t)) if ∈ I+(R) – in quasi-English: “wrong(t)” is adequate as long as the “contextual judge” is disposed towards i as per R.33 Acknowledgments: Thanks to Ben Curtis, Neil Sinclair, and two anonymous referees for comments on early versions of this paper.
33 Clearly, the fragments studied in this section merely highlight the conclusions from the main body of this essay, but are not intended as a full-fledged analysis of normative talk. For one thing, the simple-minded treatment of expressions such as “murder” as singular terms would require at least additional resources for the analysis of “not murdering is wrong” – and, as a result, appropriate constraints on R. Consequently, although a formal counterpart for “permissible” may easily be incorporated in terms of meaning-postulates that appeal to “wrong” and negation, further clauses would be needed for an intuitively satisfactory rendering of, say “obligatory”. Yet, none of these independently interesting further details are of relevance when it comes to my proposal of a systematic, compositional treatment of fragments that contain both descriptive and normative predicates, and that approach the latter in the spirit of Radical Nonfactualism.
Boo semantics: Radical nonfactualism and non truth-conditional meaning
137
References Ameka, Felix. 1992a. Interjections: The universal yet neglected part of speech. Journal of Pragmatics 18. 101–118. Ameka, Felix. 1992b. The meaning of phatic and conative interjections. Journal of Pragmatics 19. 245–271. Anand, Pranav. 2007. Re-expressing judgment. Theoretical Linguistics 33. 199–208. Anderson, Luvell & Ernie Lepore. 2011. Slurring words. Nous 7. 25–48. Aoun, Joseph & Lina Choueiri. 2000. Epithets. Natural Language and Linguistic Theory 18. 1–39. Ayer, Alfred Julius. 1936. Language, truth, and logic. New York: Dover. Barker, Stephen. 2000. Is value content a component of conventional implicature? Analysis 60. 268–279. Blackburn, Simon. 1984. Spreading the word. Groundings in the philosophy of language. Oxford: Clarendon Press. Blackburn, Simon. 1993. Essays in quasi‐realism. Oxford: Oxford University Press. Blackburn, Simon. 1998. Wittgenstein, Wright, Rorty and minimalism. Mind 107. 157–181. Bolinger, Dwight. 1989. Intonation and its uses: Melody and grammar in discourse. Stanford: Stanford University Press. Copp, David. 2001. Realist expressivism: A neglected option for moral realism. Social Philosophy and Policy 18. 1–43. Cuenca, Maria Josep. 2000. Defining the indefinable? Interjections. Syntaxtis 3. 29–44. Dickey, Eleanor. 1997. Forms of address and terms of reference. Journal of Linguistics 33. 255–274. Divers, John & Alexander Miller. 1994. Why expressivists about value should not love minimalism about truth. Analysis 54. 12–19. Dowty, David, Robert Eugene Wall & Stanley Peters. 1981. Introduction to montague semantics. Dordrecht: Reidel Publishing Company Dreier, Jamie. 2006. Negation for expressivists: A collection of problems with a suggestion for their solution. Oxford Studies in Metaethics 1. 217–233. Geach, Peter. 1965. Assertion. Philosophical Review 74. 449–465. Geurts, Bart. 2007. Really fucking brilliant. Theoretical Linguistics 33. 209–214. Gibbard, Alan. 2003. Thinking how to live. Cambridge, MA: Harvard University Press. Harada, Shin‐Ichi. 1976. Honorifics. In Masayoshi Shibatani (ed.), Syntax and semantics vol. 5, Japanese generative grammar, 499–561. New York: Academic Press. Hom, Christopher. 2008. The semantics of racial epithets. Journal of Philosophy 105. 416–440. Horn, Laurence. 1989. A natural history of negation. Stanford: CSLI. Hornstein, Norbert. 1995. Logical form: From GB to minimalism. Oxford: Blackwell. Joyce, Richard. 2002. Expressivism and motivation internalism. Analysis 62. 336–344. Kadmon, Nirit. 2001. Formal pragmatics: Semantics, pragmatics, presupposition, and focus. Oxford: Blackwell. Kalderon, Mark. 2005. Moral fictionalism. Oxford: Oxford University Press. Kaplan, David. 1977. Demonstratives. Reprinted in Joseph Almog, John Perry & Howard Wettstein (eds.), Themes from Kaplan, 481–563. Oxford: Oxford University Press. Kaplan, David. 1989. Afterthoughts. In Joseph Almog, John Perry & Howard Wettstein (eds.), Themes from Kaplan. 565–614. Oxford: Oxford University Press.
138
Stefano Predelli
Kaplan, David. 1999. What is meaning? Explorations in the theory of meaning as use. Manuscript, UCLA. Lasersohn, Peter. 2007. Expressives, perspective and presupposition. Theoretical Linguistics 33. 223–230. Lees, Robert. 1966. The grammar of English nominalizations. Bloomington, IN: Indiana University Press. McCawley, John. 1988. The syntactic phenomena of English. Chicago: University of Chicago Press. McCawley, John. 1991. Contrastive negation and metalinguistic negation. Chicago Linguistic Society 27. 189–206. Norrick, Neal. 2009. Interjections as pragmatic markers. Journal of Pragmatics 41. 866–891. Nunberg, Geoffrey. 1990. The linguistics of punctuation. Stanford, CA: CSLI. Padilla Cruz, Manuel. 2009. Towards an alternative relevance‐Theoretic approach to interjections. International Review of Pragmatics 1. 182–206. Poggi, Isabella. 2009. The language of interjections. In Anna Esposito, Amir Hussain, Maria Marinaro & Raffaele Martone (eds.), Multimodal signals, 170–186. Berlin: Springer. Potts, Christopher. 2005. The logic of conventional implicatures. Oxford: Oxford University Press. Potts, Christopher. 2007. The expressive dimension. Theoretical Linguistics 33. 165–197. Predelli, Stefano. 2005. Context: Meaning, truth, and the use of language. Oxford: Oxford University Press. Predelli, Stefano. 2010. From the expressive to the derogatory. In Sarah Sawyer (ed.), New waves in philosophy of language, 164–185. London: Palgrave. Predelli, Stefano. 2013. Meaning without truth. Oxford: Oxford University Press. Pullum, Geoffrey. 1991. English nominal gerund phrases as noun phrases with verb-phrase heads. Linguistics 29. 763–799. Rett, Jessica. 2008. A degree account of exclamatives. In Tova Friedman & Satoshi Ito (eds.), Proceedings of the 18th conference on semantics and linguistic theory, 601–128. Ithaca, NY: CLC Publications. Ridge, Michael. 2006. Ecumenical expressivism: Finessing Frege. Ethics 116. 302–336. Sauerland, Uli. 2007. Beyond unpluggability. Theoretical Linguistics 33. 231–236. Schachter, Paul. 1976. A nontransformational account of gerundive nominals in English. Linguistic Inquiry 7. 205–241. Schlenker, Philippe. 2007. Expressive presuppositions. Theoretical Linguistics 33. 237–245. Schroeder, Mark. 2008a. Being for: Evaluating the semantic program of expressivism. Oxford: Oxford University Press. Schroeder, Mark. 2008b. Expression for expressivists. Philosophy and Phenomenological Research 76. 86–116. Searle, John. 1962. Meaning and speech acts. Philosophical Review 71. 423–432. Searle, John. 1969. Speech acts. An essay in the philosophy of language. Cambridge: Cambridge University Press. Searle, John. 1975. A taxonomy of illocutionary acts. In Keith Gunderson (ed.), Language, mind and knowledge, 344–369. Minneapolis: University of Minnesota Press. Searle, John & Daniel Vanderveken. 1985. Foundations of illocutionary logic. Cambridge: Cambridge University Press. Smith, Michael. 1994a. Why expressivists about value should love minimalism about truth. Analysis 54. 1–11.
Boo semantics: Radical nonfactualism and non truth-conditional meaning
139
Smith, Michael. 1994b. Minimalism, truth-aptitude, and belief. Analysis 54. 21–26. Sinclair, Neil. 2009. Recent work in expressivism. Analysis Review 69. 136–147. Sinclair, Neil. 2011. Moral expressivism and sentential negation. Philosophical Studies 152. 385–411. Sinnott-Armstrong, Walter.2000. Expressivism and embedding. Philosophy and Phenomenological Research 61. 677–693. Stevenson, Charles Leslie. 1944. Ethics and language. New Haven, CT: Yale University Press. Tirrell, Lynne. 1999. Derogatory terms: Racism, sexism, and the inferential role theory of meaning. In Christina Hendricks and Kelly Oliver (eds.), Language and liberation: Feminism, philosophy, and language, 41–79. Albany, NY: SUNY Press. Unwin, Nicholas. 1999. Quasi-realism, negation, and the Frege‐Geach problem. Philosophical Quarterly 49. 337–352. Van Roojen, Mark. 2009. Moral cognitivism vs. non‐cognitivism. In Edward Zalta (ed.), The Stanford encyclopedia of philosophy. Stanford, CA: CSLI. Wharton, Tim. 2003. Interjections, language, and the ‘Showing’/’Saying’ continuum. Pragmatics and Cognition 11. 39–91. Wierzbicka, Anna. 1992. The semantics of interjections. Journal of Pragmatics 18. 159–192. Wilkins, David. 1992. Interjections as deictics. Journal of Pragmatics 18. 119–158. Williamson, Timothy. 2009. Reference, inference and the semantics of pejoratives. In Joseph Almog & Paolo Leonardi (eds.), The philosophy of David Kaplan, 137–158. Oxford: Oxford University Press. Zanuttini, Raffaella & Paul Portner. 2003. Exclamative clauses at the syntax‐semantics interface. Language 79. 39–81. Zwicky, Arnold, Peter Salus & Robert Binnick (eds.). 1971. Studies out in left field: Defamatory essays presented to James D. McCawley. Edmonton: Linguistic Research Inc.
Damon Horowitz
Metaphor and mercurial content Abstract: What is it for an utterance to have truth-conditional content? While Literalists and Contextualists attempt to adjudicate the proper domains of semantics and pragmatics, they do not adequately answer this fundamental question. I propose that both camps’ concern with the determinants of meaning, and in particular with the crucial issue of indeterminacy thereof, might be better addressed by exploring a hermeneutic approach to utterance content, recognizing its inherently interpreter-sensitive character. I focus here on metaphor as an initial testbed for this proposal.
1 Utterances and their contents We commonly think of our ordinary indicative utterances as truth-evaluable – we can and do evaluate them as true or false. As language users, we have an intuitive (if often unarticulated) sense for what it would take for a given utterance to be true, i.e., for the truth-conditions of the utterance. When evaluating an utterance for truth, we are assessing whether those conditions are in fact met; we are evaluating whether “what is said” is in fact true. The idea that utterances have a basic truth-conditional content plays an important explanatory role in contemporary theorizing of language. This content is essential to an utterance’s intelligibility: it is what we fundamentally must grasp in order to understand an utterance; it is what we agree with or disagree with when we take such stands towards an utterance; etc. The basic truthconditional content is the message that the words of the utterance state or directly express – as distinguished from what is merely implied by the utterance, or other types of secondary contents, all of which arise from and depend upon a prior notion of “what is said”.1 Though expressing truth-conditional content does not exhaust all of the ways in which our utterances can be meaningful – e.g., it does not capture the non-propositional, experiential, or perlocutionary effects which utterances may have – it does play a central role in explaining the fundamental communicative potency of language.
1 Of course, the precise nature of “what is said” as a technical notion is a matter of some debate; the notion of content I take up and develop here is the ordinary one attached to the saying/ implying distinction (Grice 1989). DOI 10.1515/9781501504327-008
142
Damon Horowitz
But what exactly is the relationship between an utterance and its truthconditions? What makes it the case that an utterance has one particular basic content as opposed to another? Much of the disagreement in contemporary semantics/pragmatics debates arises from differing views that Contextualists and Literalists have about the determinants of meaning – and, as a consequence, about which elements of meaning “get into the proposition expressed”, about whether a candidate content is “what is said” by an utterance or if it is merely “what is implied” (or if it has some third intermediary status as an “impliciture” or the like), and so forth. In general, Literalists claim that the basic “what is said” level of utterance content is principally established by the semantic composition of words’ conventional literal meanings; while Contextualists claim that the determination of content, even at a basic level, is a context-sensitive and largely pragmatic affair. Yet both camps customarily take for granted the idea that a given utterance has a basic intrinsic content – established through some combination of literal meanings, speaker intentions, and other contextual factors. That is, they share the common assumption that there is a fact of the matter about “what is said” by an utterance: they agree that there is a thing to be determined, though they disagree about how the determination goes. What if this assumption is mistaken? What if an utterance’s basic truthconditional content is not, in fact, an objectively fixed and determinate entity? This possibility can be explored by considering alternative traditions of investigating meaning, methodologies developed outside of contemporary analytic philosophy of language. In particular, the hermeneutic tradition is one in which meaning is not seen as an objective and fixed entity, but rather something more mercurial – something partially created by interpreters through the act of finding significance in utterances. Hermeneutics explores the openness of interpretation, encouraging us to see meaning not as determined in any simple way by semantic conventions or speaker intentions, but rather, as constrained and developed by the diverse particulars of an interpretive context. Utterances are seen as entities which may be put into play and made meaningful through the act of interpretation as much as in their original speaking. But these somewhat impressionistic suggestions have usually struck analytic theorists as unsatisfying. The analytic tradition seeks a concrete notion of utterance content, one which can do substantial explanatory work – making sense of ordinary language reports and truth assessments of utterances, among other chores. It has not been obvious how an alternative hermeneutic conception of meaning could be relevant to this project. I believe that this methodological gap can be bridged: we can take inspiration from the hermeneutic tradition to stake out a new position in the analytic
Metaphor and mercurial content
143
debates. Specifically, I propose that the truth-conditional content of an utterance is interpreter-sensitive. There is no single objective fact of the matter about what an utterance means – about what it communicates, or even about the conditions under which it would be true – because it is in contexts of interpretation that utterance content is determined. Different interpreters in different contexts may correctly judge a single utterance to have different contents. I call this thesis Hermeneutic Contextualism. The claim that interpreters make determinations of utterance meaning may seem trivial. But the force of the Hermeneutic Contextualism thesis is that the content which interpreters reasonably judge utterances to have is the only content we need – at least for the explanatory tasks which drive the Contextualist debates.2 If we wish to give fair hearing to such a proposal, it can be useful to begin by considering apparently non-prototypical uses of language, cases for which our prior theoretical commitments are perhaps less firmly entrenched. With such a focus, we may be more attentive to the considerations which lead us to conclude that an utterance has a particular content, and thus, more open to reconsidering what exactly having a particular content consists in. In that spirit, the topic I investigate here is “what is said” by a metaphorical utterance: What is the basic truth-conditional content of a metaphor?
2 Do metaphors have truth-conditions? There are two main schools of thought about what a metaphorical utterance’s content consists in, corresponding to the above-noted general orientations in the semantics/pragmatics debates. Literalists (in the Gricean tradition) characterize metaphors’ significance as non- semantic: metaphors are analyzed as cases in which a speaker says one thing but means another.3 Specifically, Literalists take “what is said” by a metaphorical utterance to be a function of the conventional literal meanings expressed by its words. Literalists acknowledge that such literal interpretation may seem absurd,
2 It is uncontroversial that interpreters do judge utterances to have content; with the Hermeneutic Contextualism approach, I effectively take utterance content to be this judged content. And I argue that it is not clear that any other non- interpreter-sensitive notion of content is then needed. 3 I intend this broad classification to group together the diverse views of figures such as Searle (1979) and Davidson (1978: 31–47); contemporary advocates include Reimer (2001) and Camp (2006a: 280–309).
144
Damon Horowitz
rendering the utterance trivially false; and they grant that the anomalousness of a literal content may provoke an interpreter to search for further significance to the utterance, and even to hypothesize hidden metaphorical content which the speaker may have intended to convey. But if such metaphorical content is identified, Literalists claim that it is merely an implication of the utterance, it is not what the utterance has strictly-speaking stated; and thus, it is not relevant to the utterance’s basic truth-conditions. Indeed, some Literalists deny that a metaphorical utterance can properly be said to have a “metaphorical meaning” at all, instead characterizing metaphors’ apparent meaningfulness in terms of their brute force – e.g., that they induce an experiential effect of seeing one thing as another – not in terms of conveyance of truth-conditional content.4 Contextualists reject this picture.5 Contextualists claim that metaphorical utterances do have metaphorical truth-conditional contents. These contents involve pragmatically-determined “metaphorical meaning” of the sort that speakers ordinarily intend to communicate and that interpreters intuitively understand – they are not restricted to the “literal meanings” of the expressions employed. Contextualists suggest that “what is said” by the predicative metaphor “X is a Z” is simply that X has whatever it takes to count as a Z in the context of the metaphor: we evaluate the truth of such a metaphorical utterance by determining what it would take to be a Z in the context, assigning this metaphorical meaning to the predicate, and assessing whether X actually does have what it takes – whether X counts as a Z in a contextually-relevant way.6 Contextualists maintain that this procedure is not different from that involved in the context-dependent determination of truth-conditions for literal or “loose-talk” utterances. For each manner of speech, the content expressed by the utterance of a given expression
4 Davidson (1978). For such a Literalist, metaphorical utterances may seem meaningful because of the imagery they evoke or because of the implications we read into them; but there is no metaphorical truth (in the full-fledged semantic sense) to be found there. 5 What I am referring to here as the “Contextualist” view is as an updated version of the old “semantic” approach to metaphor as advanced by Black (1962) or Ricoeur (1978). Currently developed in different forms – not all of which accept all of the claims I mention here – by Hills (1997: 117–153), Bezuidenhout, (2001: 156–186), Recanati (2004), and Relevance Theorists such as Carston (2002). In addition, there are semantic theories of metaphor (i.e., holding that metaphors are truth-evaluable) which are not Contextualist theories; but these theories are not my focus here. 6 Throughout this discussion, when I speak of “a way to be a Z” or “what it takes to count as a Z”, I am referring to a way a subject might be in order to be called a “Z”. That is, I am talking about variability in what the expression “Z” can be used to refer to, not about variability in some postulated property of Z-ness; I am making a linguistic claim, not a further metaphysical one.
Metaphor and mercurial content
145
varies across contexts as a function of pragmatic factors – as now-familiar Contextualist examples of sense modulation and free enrichment have suggested.7 One compelling reason for siding with the Contextualists in this debate – i.e., for thinking of metaphorical utterances as expressing distinctively metaphorical truth-conditional contents – is that we find it natural to affirm or deny metaphorical utterances in conversation, using responses such as “That’s true!” or “That’s not so!”. In so doing, we appear to be agreeing or disagreeing with the truth of a metaphorical content that the utterance conveys. As Richard Moran has pointed out, such affirmation or denial is appropriate only as a manner of accepting or rejecting a propositional content that an utterance puts forward, not as a manner of merely indicating a positive or negative noncognitive response to the imagery of a metaphor or other experiential effects it may evoke.8 And as David Hills has pointed out, the fact that metaphorical content is available for a hearer to directly take up by employing the speaker’s very words suggests that this content is “lodged in the words” of a metaphorical utterance – it is “what is said” by the utterance, not merely “what is implied” by the utterance.9 To take an example: Imagine a context in which George and his friends are discussing whether Arnold – an ex-body-builder – is a good candidate for governor. If George utters approvingly… (1) Arnold is a bulldozer. …a hearer may reply with “That’s true!” or… (2) He certainly is! …as a way of agreeing with the proposition… (3) Arnold is powerful enough to dislodge any obstacle in his path.
7 In other words, the Contextualist advances a semantic continuity thesis about metaphorical and literal language. Metaphorical utterances directly express truths just as literal utterances do, using the same basic mechanisms of meaning, with the same potential illocutionary force. Metaphorical predications may differ from literal ones with respect to their style or to the experiential effects they provoke – and metaphors may rely upon less typical ways of satisfying a predicate (e.g., of counting as a Z) than literal language does – but there is no categorical difference between these manners of speech with respect to their status in forming truth-evaluable utterances. Early work on a continuity thesis is found in Rumelhart (1979); more recent elaboration in Wearing (2006: 310–332). For further discussion of the stylistic differences and semantic similarities between literal and metaphorical speech, see Horowitz (2006). 8 Moran (1989: 100); see also Wearing (2006: 313). 9 Hills (1997); see also Bezuidenhout (2001).
146
Damon Horowitz
…taking this proposition (3) to be communicated as the metaphorical content of George’s utterance (1) (“Arnold is a bulldozer.”) in this context. Or a hearer may say “That’s not so!” or… (4) No – Arnold is no bulldozer! …as a way of disagreeing with the proposition… (5) Arnold has the verbal forcefulness and aggressiveness needed to win support for difficult measures. …taking (5) to be the relevant metaphorically expressed content – for perhaps the hearer believes that Arnold is orally weak and easily resisted, not at all bulldozer-like.10 The Contextualist notes that in such examples, the respondent is affirming or denying metaphorical content directly, using the same expressions (e.g., “being a bulldozer”) with the same metaphorical meanings as in the speaker’s original utterance – in a way that is not possible with merely implied content (i.e., Gricean implicatures).11 George’s utterance (1) is taken to be true just
10 It may be tempting to analyze such cases as instances of metalinguistic negation, i.e., wherein it is not the truth-conditional content of the original utterance that is being denied, but rather some other property – perhaps the propriety of the word choice. (Recall the classic examples: “It isn’t warm in here; it is hot.”; “Those aren’t hippopotamuses; they are hippopotami.”; etc. See Horn (1989).) But this analysis does not quite fit the present examples. First, note that with metalinguistic negation, there is typically a second clause which involves contradicting the preceding denial of the offending material (e.g., “I haven’t deprived you of my cooking; I’ve spared you of it.”), pronounced with a characteristic contrastive intonation (as is consistent with the suggestion that the negating clause involves a mention rather than a use of the negated expression.) These features do not appear in the present cases of affirmations and denials, which use normal intonation, and which are typically followed by elaborating material, not contrasting clauses. Furthermore, the spirit of the present responses differs from what we find with complaints about stylistic or conversational impropriety: the interpreter who responds to “Arnold is a bulldozer.” with “No he isn’t!” would be expected to act upon this judgment (e.g., perhaps to vote accordingly) as upon a truth-conditional verdict, not as upon a merely metalinguistic criticism. Finally, note that applying the metalinguistic analysis here would create an undesirable asymmetry between the affirmation and the denial cases – for a “metalinguistic” (or an “echoic”) affirmation does count as endorsement of truth-conditional content. 11 For instance, a hearer cannot effectively reply to (1) by saying “That’s not so!” or… (i) No he isn’t! …as a way of disagreeing with the proposition that… (ii) George would vote for Arnold. …because this latter proposition (ii) is communicated merely as an implication of George’s utterance (1) in this context. The lesson from such further examples is that, unlike metaphorical content (e.g., (3)), merely implied content (e.g., (ii)) is not part of “what is said” by an utterance: the truth of George’s utterance turns on facts about Arnold’s political abilities, not on facts about George’s voting dispositions.
Metaphor and mercurial content
147
in case Arnold counts as a bulldozer, metaphorically: that is what George’s utterance says, the truth-conditional statement that it makes. When evaluating (1), we try to determine what it would take to be a bulldozer in the context, and then we assess whether Arnold is that way; we do not evaluate whether Arnold is “literally” a piece of construction equipment. In effect, we create a “customized” or ad hoc concept (e.g., a “nonce sense”) which represents a contextuallyrelevant way of being a bulldozer, and we treat the predicate as expressing this concept. For instance, the ad hoc concept we use in understanding (1) (“Arnold is a bulldozer.”) may involve the features attributed to Arnold in (3) or (5) (e.g., a certain power or forcefulness, here applied to body-building and politics, rather than construction sites) – these are the features that the truth of the utterance turns on. This type of affirmation/denial dialogue demonstrates why Contextualists believe that metaphorical utterances do express metaphorical truth-conditional contents, even though the pragmatically-determined context-sensitive metaphorical meanings which are “lodged in the words” of the utterance are not conventional literal meanings.12 For the Contextualist, these meanings are what a metaphor’s very intelligibility depends upon: to understand a metaphorical utterance is to grasp its basic metaphorical content, to recognize what it takes for the metaphor to come out true.
3 Pragmatics and indeterminacy Literalists do not accept this Contextualist analysis. Literalists are anxious to maintain a privileged role for relatively fixed and relatively discrete conventional
12 Though there is not space to review the entire debate here, there are many other familiar reasons why a Contextualist continuity thesis is attractive. For instance: there is psycholinguistic evidence that metaphorical speech is processed in a similar manner to literal speech (Gibbs 1994), (Glucksberg, Gildea, and Bookin 1982: 85–98); for an opposing view, see Camp (2006b: 154–170); the phenomenology of metaphor comprehension is “direct”, “unreflective”, and has an “immediacy” and “transparency” common to literal speech (Bezuidenhout 2001), (Recanati 2004); (Guttenplan 2006: 333–359); and so forth.
148
Damon Horowitz
literal meanings in our understanding of utterance content.13 They believe that many linguistic phenomena demand a systematic explanation of the sort which can only be provided by postulating semantic mechanisms; they characterize Contextualists as relying too heavily on the overgeneralizing cauldron of pragmatics to explain communication. Specifically, Literalists argue that any approach which does not take conventional literal meanings as the fixed anchor for determining utterance content will be hopelessly unconstrained. This challenge comes in many forms. For instance, Josef Stern charges that pragmatic analyses generate an unmanageable proliferation of ad hoc concepts, leaving us unable to account for the fact that we often feel that the same concept has been referred to on two different occasions.14 Similarly, Jason Stanley charges that Contextualist accounts “overgenerate” in incorrectly predicting that a wide variety of combinations of predicate meanings would be acceptable to language users, when in fact many such ad hoc combinations are intuitively deemed infelicitous, and can be shown to be semantically ill-formed.15 But the deeper worry here goes beyond a concern about the efficacy and legitimacy of pragmatic explanations of felicity phenomena. The present disagreement concerns whether the very idea of metaphorical truth-conditional content is compatible with our understanding of what it is for an utterance
13 The Literalist explanation of linguistic communication rests on two foundational ideas: (a) that literal meanings are what we learn when we learn expressions in our language; and (b) that it is through the predictable compositional assembly of these meanings, via the coordination of syntax and semantics, that we are able to create and understand an infinite variety of novel sentences. The Contextualist suggestion is seen as a potential threat to this story. For if we admit that metaphors manifest the type of flexibility of meaning which Contextualists propose, it may be hard to resist generalizing this analysis to literal utterances with apparently context-sensitive meaning; and once we allow that the content of our ordinary literal utterances is largely shaped by pragmatic intrusions, then the celebrated achievements of compositional truth-conditional semantics may seem less explanatory of linguistic communication – for utterance content will turn out not to be simply a function of the algorithmic combination of words’ conventional literal meanings. 14 Stern (2006: 243–279) is concerned that with Contextualist accounts, “…we arrive at any number of different ad hoc concepts among which we cannot clearly distinguish one from another,” (p.256); for criticism, see Horowitz (2007); for a broader summary of challenges to the idea that words have discrete literal meanings, see Taylor (2012). There are echoes here of the objections raised by Semantic Minimalists that Contextualists cannot handle “samesaying” cases due to their unperspicuous individuations; see Cappelen and Lepore (2005); and for responses, Hawthorne (2007). 15 This general argument strategy against Contextualists is probably most familiar from the “binding argument”, as in Stanley (2005). It is also a more or less explicit premise in many related theories of hidden indexicals and the like.
Metaphor and mercurial content
149
to determinately express a proposition. Literalists suggest that metaphorical content is a fflicted with a special and unwelcome form of indeterminacy (akin to the familiar Quinian and Davidsonian varieties), which renders it unfit to play the role of “what is said” by an utterance. Metaphorical content cannot provide a firm basis for truth-evaluability, they argue, for its variability and open-endedness would yield unacceptable unclarity about the proper truth judgments of metaphorical utterances. This, then, is reason to resist the Contextualist suggestion that such content is directly expressed by an utterance in the traditional sense – i.e., the sense in which truth-conditional content determinately captures the conditions under which an utterance is or is not true – as opposed to being merely implied or otherwise communicated by the utterance. To see the worry here, recall the Contextualist claim that the basic truthconditions of a metaphorical utterance of the form “X is a Z” are constituted by whatever it takes for X to count as a Z in the present context: the utterance is true if X is a Z in a contextually-relevant way. Contextualists suggest that the “contextually-relevant way” to be a Z is “lodged in the words” of the utterance in the form of an ad hoc concept – it is not simply determined by the literal meaning of the expression “Z”. But here the indeterminacy problem looms: for there seem to be many different such pragmatically-shaped concepts we could conceivably form, each of which has some claim to being reasonable and relevant in a context. Which way to be a Z – which candidate truth-conditional content – is the one which the utterance’s truth actually turns on? For instance, consider again the alternative truth-assessments of George’s utterance (1) (“Arnold is a bulldozer.”): does the truth of this utterance turn on whether Arnold is sufficiently physically dominating (e.g., (3)) or sufficiently verbally aggressive (e.g., (5))? The Literalist observes that a number of different features associated with the metaphorical predicate may reasonably be proposed as “the relevant way” in which the utterance should be understood and evaluated. If two body-builders are within earshot of George, one of them might express agreement with (1) by taking up the proposition (3) and responding… (6) That’s true! He can lift huge weights that others cannot even budge. Whereas if two political pundits overhear George’s statement, one of them might express disagreement with (1) by taking up the proposition (5) and responding… (7) That’s not true! He is a timid and underwhelming public speaker. In ordinary conversation, either of these responses may seem acceptable as assessments of George’s metaphorical utterance. Even though they apparently
150
Damon Horowitz
have opposed verdicts, the respondents here might be in complete agreement about Arnold’s actual characteristics (e.g., physical strength paired with verbal weakness). However, they take the utterance (1) – picked out as the referent of “that” in each response – to express different contents, thus yielding different truth assessments. Or consider Romeo’s much-discussed utterance: (8) Juliet is the sun. Alternative possible interpretations of this metaphor have been articulated in a wide range of paraphrases: (a) Juliet is “warm, sustaining, comforting, perhaps awesome… something on which we are utterly dependent…” (Simon Blackburn); (b) Juliet is “dazzling… the center of (his) world…” (Lynne Tirrell); (c) Juliet is “glorious…(but) she may blind and burn you if you come too near her…” (Liz Camp). Each of these on its own may seem a reasonable approximation of the truthconditional content of Romeo’s metaphor. Thus, any of these might be taken up by Romeo’s interpreters in agreeing or disagreeing with his utterance. For instance, Mercutio may take up Romeo’s words and reply… (9) No she isn’t! …as a way of disagreeing with the proposition that Juliet is supremely beautiful – for perhaps Mercutio has a lower estimate of Juliet’s attractiveness. But alternatively, Benvolio might reply to Romeo by saying… (10) Yes, she sure is! …as a way of assenting to the proposition that Juliet is dangerous to pursue – after all, Benvolio is worried about Romeo slinking around in his enemy’s orchard in the middle of the night. The problem which the Literalist raises here is that such alternative interpretations reflect distinct propositions, which may not all be true simultaneously: being dazzling is not the same as being comforting, and being dangerous is not the same as being dependable – Juliet could be one without being the other. It is not difficult to imagine one interpreter vehemently disagreeing with another about any of the particular elements of these paraphrases, since they articulate potentially inconsistent claims. In the course of such disagreements, the disputants will accuse each other of not really understanding Romeo’s utterance – they have different ideas about what truth-conditional content that metaphor is expressing when uttered in that context.
Metaphor and mercurial content
151
The Literalist argues that this problem will inevitably arise from the Contextualist suggestion that we release the traditional assumption that words carry their standing conventional literal meanings – and not pragmaticallydetermined metaphorical meanings – across different contexts of utterance. Without conventional literal meanings as our authority, they worry, we will have lost our moorings entirely for adjudicating possible interpretations of an utterance.16 This way of developing the Literalist objection is potentially quite powerful, as it threatens to undermine the main consideration which motivates the Contextualist position to begin with. Recall (Section 2) that it is just such examples of affirmations and denials in ordinary dialogues which Contextualists take as evidence that metaphorical content is directly expressed by metaphorical utterances. But now the Literalist has drawn our attention to the fact that these forms of response may reflect a diversity of contents for a single metaphorical utterance; whereas if we stick with conventional literal meaning as the guide to content, no such unwelcome diversity arises. This observation, then, suggests that any alleged metaphorical content should be understood as merely an implication or other type of secondary message communicated by an utterance: the plurality of possible options disqualifies metaphorical content from playing the role of the basic truth-conditional content of the utterance; for that role, a single determinate content is required, if it is to be what utterance truth determinately turns on. The Literalist concludes that communicating metaphorical meaning is, if anything, a more impressionistic affair.
4 Speaker intentions and fallibility Contextualists do not find these Literalist challenges persuasive. Contextualists are prepared to discount the importance of literal meaning in determining utterance
16 Though here it is presented in the jargon of the Literalist/Contextualist debates, the difficulty of identifying a single satisfactory paraphrase of a metaphor has been a long-recurring theme of objections to the notion of “metaphorical meaning” – most famously in Davidson (1978), and defenses of this view such as Reimer (2001). The challenge which “the range of possible metaphorical interpretations” poses for attempts to provide a semantic treatment of metaphor is concisely articulated by Borg (2001: 227–248). Indeed, some Literalists have even proposed that this threat of unclarity is sufficient to substantiate a norm to prioritize literal speech over metaphorical speech – since metaphors inherently seem to lack a desired determinateness. See Camp (2006a); and criticism in Horowitz (2006).
152
Damon Horowitz
content because they have an alternative story available for how this determination takes place: they believe in the constraining power of pragmatics, and especially of speaker intentions, to settle the matter. Indeed, this belief that pragmatic factors do the lion’s share of the work in establishing communicatively-relevant meanings is at the root of Contextualists’ broader claim that an expression only has truthconditional content in a context of utterance.17 Thus, the Contextualist response to the Literalist challenge is simply to deny the allegation of indeterminacy. The Contextualist allows that the above examples (e.g., (9) vs. (10)) indicate how different interpreters may have different opinions about what proposition is expressed by a contested metaphorical utterance; but this only shows that some interpreters are mistaken about the utterance’s content, not that the content is actually indeterminate. For the Contextualist, pragmatic factors do fix the meanings of expressions in a context of utterance – there is nothing “unconstrained” about this common phenomenon. For example, in the above case of Romeo’s disputed utterance, Romeo himself might weigh in on the question. He may reject some of the paraphrases and accept others, perhaps recognizing them as part of what he intended – or even as something that he had not previously articulated, but is happy now to endorse after the fact. After all, Contextualists suggest, it seems right to grant that as the speaker of the utterance, Romeo has authority over the meaning he expresses, and the truth claim he makes. While there may be unclarity or openness about what a word could be used to mean, the actual meaning is fixed by a speaker in a context when the word is spoken; responses with alternative interpretations which do not respect this intention are simply incorrect.18 Thus, by taking speaker intentions as the decisive pragmatic factor in determining utterance content, the Literalist indeterminacy challenge could be deflected. However, Literalists are rightly reluctant to accept this response. While it seems to sidestep the threat of indeterminacy, it does so by requiring that we abandon our natural understanding of the above examples as acceptable responses to the original utterance – i.e., as manifesting different ways that interpreters may reasonably make sense of an utterance. The Contextualist explanation essentially resorts to an error theory of interpreters’ understandings of an utterance in order
17 This way of articulating Contextualism is most clearly sounded by Recanati – e.g., “Only in the context of a speech act does a sentence express a determinate content…”; Recanati (2004: 6) – but can be found across Contextualist theories. 18 For instance, Wearing (2006: 325) explicitly articulates this argument: she claims that the correct way of counting as a Z for a given context is settled by speaker intentions (within limits established by the conversation, the standing meanings of words, and such).
Metaphor and mercurial content
153
to explain away the problematic data; as such, it fails to explain our actual conversational practices. The Literalist reminds us that we should only take such a step as a last resort. In ordinary conversation, we do not always assume that a speaker has overriding authority in determining what his utterance means. Interpretation of a metaphor, or of any utterance, is not “restricted to an elaboration of what its author is taken to have understood by it,” as Moran has explained.19 A speaker cannot definitively vest his utterance with a particular content any more than he can definitively vest a stone he places on a trail with a particular significance (say, as a cairn): the speaker may have a content in mind, or he may intend to appeal to conventions for how his words (or his stone) may signify; but once he releases his utterance (or his stone), its actual significance on an occasion of interpretation will be what it is then taken to be. As the above examples show, hearers often can and do read more into utterances than speakers intend, evaluating utterances in consideration of how they imagine the utterance may be reasonably elucidated.20 The gulf between the truth-conditions which a speaker intends to convey and those which his utterance actually carries in an interpretive context can be considerable.21 The very idea of a speaker failing to say what he means depends upon this conceptual possibility. This is commonly acknowledged in the analytic literature in discussions of externalism (e.g., wherein content is seen as determined by the world in ways beyond a speaker’s intentions or intensions), and
19 Moran (1989: 107). For further discussion of “the degree of autonomy metaphorical meaning has from speaker intentions…” see Borg (2001: 232). 20 That is: what conversationalists are agreeing or disagreeing with when responding to an utterance need not be what the speaker hoped his words would mean when he uttered them. Hills (2004: 52,268) captures this idea nicely by describing the “oracular” quality that utterances may have: “Oracular utterances (are) utterances in which a speaker speaks to an effect she herself doesn’t yet fully understand, in the hope that what her utterance properly conveys is something she can and should mean, once she has discerned what that is…”. A more hermeneutic rendering of this point: “One should be able to understand an author better than he understood himself…The real meaning of a text, as it speaks to the interpreter, does not depend on the contingencies of the author and his original audience… the meaning of a text goes beyond its author… (or at least) it is enough to say that we understand in a different way, if we understand at all…”; Gadamer (1975: 297). 21 We should be wary of neo-Griceans’ too-hasty reduction of utterance meaning to speaker meaning here. An utterance is an independent entity unto itself, the product of an event of uttering – it is not the same entity as the intention of the speaker who speaks it, just as it is not the same entity as the sentence type it tokens. An utterance is a sentence spoken in a particular context and possibly interpreted in many others; its content may reflect influences from any of these sources. Thus, understanding the speaker’s intentions is not the same as understanding the utterance.
154
Damon Horowitz
also around cases of misdescription and malapropism (e.g., in which there is an explicit distinction between what a speaker intended to say and what was actually said by his utterance). For example, if George seeks to commend Arnold for his commitment to family values, George might malapropically proclaim… (11) Arnold understands the importance of bondage between a mother and child.22 …intending to express the proposition that Arnold understands the importance of a strong bond between a mother and child. But then a hearer, believing Arnold to be free of any bondage-endorsing beliefs, could reasonably respond… (12) That’s not true! Arnold does not advocate bondage in any form. In such a situation, we do not necessarily consider the hearer to be “mistaken” about the truth-conditional content of George’s utterance (11); rather, we may see this as a prototypical case of a speaker’s utterance not having the meaning he intended.23 In any event, deferring to speaker intentions would not help us resolve our indeterminacy woes in all cases – for speaker intentions themselves often have a kind of indeterminacy. To revisit our earlier example (8) (“Juliet is the sun.”): Romeo may reject the whole lot of proposed paraphrases of his metaphor, denying that he intended to convey any such determinate thoughts; the term “sun” may simply have come to mind for him and seemed fitting, free of any elaboration that might illustrate or justify this intuitive sense (e.g., “I just said that she is the sun. That’s all.”). Perhaps Romeo feels that it is up to each interpreter to figure out on
22 To borrow a phrase from Dan Quayle. 23 Malaprops and misdescriptions are typically analyzed as involving something screwy about a speaker’s communicative plan: e.g., a speaker (often inadvertently) intends to mean something by a term which that term is not conventionally used to mean, and fails. For instance, this is the analysis advanced by Reimer (2004). Whether or not we agree with Reimer (see Horowitz (2007) for a critique), or find the hearer’s response here unhelpfully uncooperative, the important point is that the very possibility of questioning what the utterance itself means in a context – as distinguished from what the speaker intends for it to mean – is sufficient to undermine the Contextualist’s present suggestion.
Metaphor and mercurial content
155
their own terms what it takes for Juliet to be the sun – that it is not a question we need to settle.24 Yet even in such cases when a speaker has no particular way of being a Z in mind when uttering “X is a Z”, conversationalists can and do project some such condition onto the utterance for the purpose of evaluating its truth – as the above examples show – regardless of whether the utterance is literal or metaphorical. Reflection on this phenomenon reminds us that we should not be too quick to take speaker intentions as uniquely dispositive in determining utterance content.
5 The dilemma But now we are left with a puzzle. The Contextualist’s natural explanation of the affirmation/denial dialogues suggests that metaphorical meaning really is lodged in the words of a metaphorical utterance, and that metaphors really do express truth-conditional content (Section 2). However, as the Literalist points out, there are many different reasonable candidates for what exactly this content is, reflected in the diversity of truth-evaluative responses which are possible – an indeterminacy that makes metaphorical meaning hardly seem “lodged” at all (Section 3). The Literalist accepts the indeterminacy examples at face value – as demonstrating the actual range of metaphorical interpretations which are possible for a given utterance – and concludes that so-called “metaphorical content” is essentially different in kind from basic utterance content. But in order to sustain
24 By analogy: in adjudicating what to make of an utterance’s content in cases of misdescription or malapropism, it is sometimes stipulated that the speaker’s higher-level intention to conform to the conventions of our language should guide us – e.g., using Kripke’s well-known identification of semantic reference with the speaker’s “general intention” to use a designator in a certain way. For instance, Reimer argues that malapropisms should be understood in this way (see footnote 23): they carry their conventional literal meanings, because those are the ones to which the speaker ordinarily intends to conform; Reimer (2004: 333). But if we accept this kind of stipulation, then we should also consider the possibility that we may as well defer to the speaker’s higher- level intention to participate in our ordinary practices of social negotiation and construction of meaning. That is, we may view an ordinary speaker as intending to have his hearers interpret his utterance in ways relevant to their contexts – including their unique beliefs, interests, purposes, and even idiolects – or, as hermeneuts would put it, intending to have his hearers truly understand his utterance. (A prototypical case here is the formal utterances of legislators who write law with the explicit intention that it will be interpreted in different future circumstances in a manner appropriate to the cases then at hand.) This is the basis of the proposal I elaborate in Section 8.
156
Damon Horowitz
this position, the Literalist must abandon our intuitive understanding of the affirmation/denial responses as direct reflections of the truth-conditions of an utterance. Disregarding these intuitions may seem implausible: do we really want to claim that the response (4) (“No – Arnold is no bulldozer!”) is not a denial of the basic proposition that the metaphorical utterance (1) (“Arnold is a bulldozer.”) puts forward? What business are truth-evaluative responses such as (7) (“That’s not true!”) in, exactly, if not that of responding to truth-conditional statements? The very form of the response, and the emphatic tone in which it may be delivered, weighs against accepting such a counter-intuitive claim. Indeed, such responses are central examples of the broader phenomenon of utterance truth-evaluability – of how conversationalists assess utterance truth – which our conception of utterance truth-conditional content should explain.25 By contrast, the Contextualist accepts the affirmation/denial examples at face value – as demonstrating that metaphorical truth-conditional content is directly expressed by a metaphorical utterance – and concludes that the diverse evaluative responses cited by the Literalist are misleading. But in order to sustain this position, the Contextualist must resort to an error theory about our ordinary understanding of the Literalist’s examples. We ordinarily find diverse responses (e.g., (6) or (7)) to a metaphorical utterance perfectly acceptable: neither respondent would be criticizable for their interpretation and response, even if they ignore or overrule the speaker’s intention. As argued in Section 4, an utterance is not like a flashcard – it does not have a single correct interpretation somehow inscribed on the flip side by the speaker’s intention.
25 Corroborating evidence for the truth-evaluability of metaphors can be found by observing the embeddability of semantically-relevant metaphorical content. In the domain of moral language, Geach (1965) famously argued that noncognitivist analyses of moral statements face difficulty in accounting for the evident truth-conditional contributions of such statements in embedded logical constructs. A similar argument challenges any non-Contextualist view which claims that metaphors lack directly expressed propositional contents. Such a view will have difficulty explaining metaphors’ evident contributions to the antecedents of conditional statements, e.g., (i) If music be the food of love, play on... …and to belief reports, e.g.,… (ii) Romeo believes that Juliet is the sun. The most natural explication of these examples is that the metaphorical phrase contributes a truth-conditional content to the larger utterance: (i) issues an imperative which should be followed if it is true that music counts as the food of love, metaphorically (in the relevant context); and (ii) attributes to Romeo a belief that is true if Juliet counts as the sun, metaphorically (in the relevant context). (See footnote 8.)
Metaphor and mercurial content
157
This, then, is our dilemma. Each of the existing camps in the debate feels forced to discount the significance of the linguistic evidence which the other advances. But we may be reluctant to resort to this tactic; we would like to develop a position which accommodates the intuitive understanding of all the examples. How might this be done?
6 Finding truth in indeterminacy: Mercurial content I suggest that the key to resolving this dilemma lies in questioning the assumption that the content of an utterance is determinately established in the context of utterance. Consider again how the dilemma arises. Contextualists claim that a metaphor’s truth turns on the primary subject having some features associated with the metaphorical predicate (e.g., there is some way that Juliet must be to count as the sun). When we agree or disagree with a metaphor, we are evaluating whether the primary subject actually is that way (e.g., we may disagree that Juliet really is so beautiful or comforting). And then the Literalist challenge becomes: Which way is actually the one that matters? However, there is something curious about a Literalist advancing such a challenge – for the same question may be raised for non-metaphorical utterances as well. Recall our example (1) (“Arnold is a bulldozer.”); but now suppose that George had saved himself a syllable and spoken literally, calling Arnold a “bully” instead of a “bulldozer”. The same indeterminacy we found above (e.g., (3) vs. (5)) would remain – e.g., is it Arnold’s intimidating physicality or his verbal aggressiveness that is at issue? – but presumably a Literalist would not take this as reason for questioning the truth-conditional status of literal content.26 Regardless, the question still stands: if one interpreter holds that we should only consider Arnold to be a bully if he really is physically dominating, while another interpreter believes that being verbally aggressive is sufficient qualification, which of these ways of being a bully matters to the truth of the utterance? Is Arnold really a bully or not?!
26 That is: While the Literalist claims that the “open-endedness” of metaphorical content renders it unable to function in the ordinary way we expect truth-conditional content to function, this claim is weakened when we observe that open-endedness is just as much a quality of literal meanings as metaphorical ones. See Bezuidenhout (2001: 172).
158
Damon Horowitz
But now the question starts to sound queer. Why must we assume that there is a single objective fact of the matter here? I think of this assumption as a relic from the Literalist tradition; it need not be part of an evolving Contextualist sensibility.27 Of course there may be straightforward facts about how on a specific occasion a specific interpreter does take an utterance and evaluate it for truth: in ordinary conversation, we do take different ways of being a bulldozer (or a bully) to matter; and these specific ways are articulated in our evaluative responses when we cite the features which qualify or disqualify the subject in our view (e.g., (5) “I disagree that Arnold is a bulldozer… he lacks the verbal aggressiveness needed to win support for difficult measures.”). But this only shows that interpreters do not treat utterances as hopelessly indeterminate or opaque; rather, interpreters assign to utterances whatever degrees and dimensions of determinate content are necessary for current purposes. The catch is that no particular content seems objectively inherent to the utterance – as the above examples show, different contents are assigned by different interpreters in different interpretive contexts. Given this explication of the phenomenon, it begins to look as though a diagnosis of indeterminacy is not quite accurate. Perhaps utterance content is not so much indeterminate as it is mercurial – an entity which changes faces on different occasions. The thought here is that the most constructive response to the lingering Literalist worry is not to deny the appearance of indeterminacy, but rather, to embrace it as essential to the phenomenon under study. Our task need not be to eliminate the intuition that a plurality of responses may be possible, but rather, to explicate further the kind of phenomenon we have encountered here and how it arises – and what it tells us about the nature of utterance content.
27 It seems to me that there is an inherent dissonance in simultaneously acknowledging ubiquitous context-sensitivity in meaning and stipulating a perspective-independent objective fact of the matter about content. Of course, it is theoretically possible to accommodate both of these ideas in a single theory – as Contextualists and Literalists both attempt to do, in various ways. But given an intuitive recognition of the prima facie tension between them, we might take occasion to question whether there is an alternative way to characterize utterance content.
Metaphor and mercurial content
159
7 The hermeneutic step: interpreter-sensitivity I wish to explore here the possibility that there is no perspective-independent fact of the matter about what it takes for a metaphor to be true – i.e., about which way the primary subject must be to count as satisfying the predicate – even though some such way of being does factor into the intuitive truth-conditions. Various hearers of a metaphorical utterance may each have in mind their own favored features of the predicate, which they take to be essential to the utterance’s truthconditional content. What I would like to suggest here is that we may ascribe to the utterance itself any one of these candidate contents which is appropriate in a given interpretive context. I propose that we move beyond standard Contextualism to consider an analysis of metaphor based on Hermeneutic Contextualism. Recall from Section 1 that Hermeneutic Contextualism holds the basic truth-conditional content of an utterance to be interpreter-sensitive: the sense in which an utterance “has truth-conditional content” is that an interpreter legitimately takes it to have such a content. There is no single perspective-independent fact of the matter about “what is said” by an utterance; different interpreters in different contexts may legitimately judge a single utterance – whether metaphorical or literal – to have different contents. Adopting the Hermeneutic Contextualism view allows us to resolve our dilemma. On this view, metaphorical meaning is indeed “lodged in the words” of an utterance – i.e., it is part of “what is said”, consistent with the Contextualist claim of direct expression – but it is so lodged only temporarily, on an occasion of interpretation; this allows for the plurality of possible evaluative responses which Literalists cite. (To clarify again: that a metaphorical meaning is indeed lodged is suggested by its availability for explicit affirmation or denial; and that a given metaphorical meaning is only lodged relative to an occasion of interpretation is suggested by the variation in reasonable but incompatible paraphrases that may be offered on different occasions.) For a given interpretative context, the utterance content takes on whatever degree and dimension of determinacy is required: a local code is effectively established for using the metaphorical expression in a fully truth-conditional manner but with a potentially novel contextuallyrelevant content. The important insight of the standard Contextualist approach is that pragmatic factors – i.e., as opposed to strictly linguistic knowledge of syntax, semantics, and conventional literal meanings – have a considerable effect on our understanding of utterances. These pragmatic factors determine what would be reasonable or unreasonable interpretations of an utterance in a given context.
160
Damon Horowitz
For instance, it would not seem acceptable for either the body-builders (in (6)) or the political pundits (in (7)) to take George’s utterance (1) as a statement about Arnold’s slow bulldozer-like speed of driving… (13) (?) No he isn’t! Arnold drives at the same speed as everyone else. …since driving speed is not particularly relevant to either the body-building or the political themes of their contexts. Such a response seems like a non-sequitur: it is infelicitous because it fails to pick up on a contextually-appropriate way to be a bulldozer – and it is pragmatic considerations, involving contextual relevance and general world knowledge, which make it so. But standard Contextualism fails to appreciate that it is in interpretive contexts that these factors are operative – in the interpretive acts of hearers – not in utterance contexts. By making this distinction, the Hermeneutic Contextualism approach illustrates why it should be that pragmatic factors are so important in the determination of utterance content: it is because the phenomenon of utterances bearing content is a product of interpreter activity, and interpreters themselves call upon pragmatic considerations in this activity, that pragmatics plays a large role in establishing truth-conditions.28 This approach for resolving the present dilemma has further motivation beyond the demands of the current examples. As explained in Section 1, the claim that utterance content is interpreter-sensitive is consonant with the broader conception of meaning developed in the hermeneutic tradition. Understanding an utterance involves understanding how it may be relevant or applicable to one’s particular circumstance and experiences; as these circumstances and
28 In other words: Hermeneutic Contextualism seeks to clarify the sense in which pragmatic factors “do work” in determining utterance content on any Contextualist account. From the Hermeneutic Contextualism perspective, this is a straightforward descriptive claim about the actual thoughts and actions of ordinary interpreters in making sense of an utterance – i.e., in determining the content of the utterance. In this light, the pragmatic factors which standard Contextualists cite are best understood as simply a subset of the more general range of hermeneutic pressures which bear on the actual interpretive life of an utterance.
Metaphor and mercurial content
161
e xperiences vary, there will be corresponding variations in the interpretations which are developed.29 From an analytic perspective, one way of understanding this claim is to see it as an answer to the question of what kind of a property “having truthconditions” might be. I am claiming that this property of an utterance is more like “being funny” or “having aesthetic appeal” than it is like “being loud” or “having a particular syntactic structure”. An utterance’s truth-conditional content is something of a secondary property, inherently observer- dependent – analogous to the appearance or value of an object, as opposed to the objective size or shape of an object. Both Literalism and standard Contextualism assume that utterance content must be a single discrete entity, fixed forever at the time of utterance; yet we are not tempted to make the analogous assumption that the humor or aesthetic appeal of an utterance are so fixed. In sum: With the Hermeneutic Contextualism approach, we can keep the Contextualist insight that the determination of utterance content is primarily a pragmatic affair, while not keeping the errant assumption that there is a perspective-independent fact of the matter about content determined for all time on the occasion of utterance.
8 Objections There are a number of objections which talk of Hermeneutic Contextualism tends to provoke. There is not room here for extensive discussion of these, but it may be useful to quickly sketch the types of responses which could be elaborated in defense of the view. First Objection: Isn’t there more to being true than being judged to be so? Even if an interpreter judges that an utterance is true (e.g., “Yes, Arnold certainly is a bulldozer!”), that does not show that the utterance really is true.
29 Of course, traditional methodological hermeneutics addresses the question of significance of entire works – e.g., works of art, or literature, or law. But the hermeneutic perspective on meaning can also provide important insights for analytic theorizing, if we apply its principles regarding full-scale works to the study of individual linguistic utterances. This conception of meaning can also be found in our ordinary folk understanding of the notion of significance as observer-sensitive: we take objects and events to be significant to someone; we do not suppose (upon reflection) that they have an absolute significance intrinsically. This is cause for entertaining the possibility that truth-conditional content itself may be similarly relativizable.
162
Damon Horowitz
Hermeneutic Contextualism is not a theory of Truth, i.e., a theory of the nature of truth, of what truth consists in; I am not suggesting a projectivist or relativist approach to truth here. Hermeneutic Contextualism is a theory about truthconditions, about one aspect of utterance significance: it claims that an utterance’s truth-conditional content is interpreter-sensitive. Given a truth-conditional content, there may be a perspective-independent fact of the matter about whether that proposition is actually true. That is, once we have determined a contextuallyrelevant way of being a Z, it may be an objective question whether X is in fact that way. Realists about truth can be Hermeneutic Contextualists; we can accept non-relativized objectivity about truth-values without accepting non-relativized objectivity about truth-conditions.30 Second Objection: Is Hermeneutic Contextualism claiming that it is simply up to the interpreter to choose which content to ascribe to an utterance? The worry here is that Hermeneutic Contextualism may be confusing the content that an utterance is judged to have by an interpreter with the content that it actually has. Does Hermeneutic Contextualism abandon the notion that an interpretation can be incorrect?
Hermeneutic Contextualism does claim that the content of an utterance is what we take it to be – but only when we are well-informed and well-intentioned hermeneutically-responsible interpreters. I am not suggesting that simply taking an utterance to have a content makes it so; some takings are legitimate, and others are not. Hermeneutic Contextualism acknowledges that interpreters are subject to normative pressures, of both hermeneutic and instrumental varieties, which constrain how they may correctly and effectively ascribe content to utterances. In one interpretive context, it may be important for an interpretation to respect the intentions of the speaker, in order to assess whether he spoke responsibly; but in another, it may be more important to adopt the most typical meaning of an expression, to best anticipate its public reception; and in a third, it may be more
30 It may be possible to develop an alternative explanation of the Literalist examples of diverse evaluative responses by taking up the approach of recent theories of truth-relativity; see MacFarlane (2005: 321–339); Egan, Hawthorne, and Weatherson (2005: 131–170). But many theorists find the relativization of truth itself implausible; see Horowitz (2007) for discussion of the tradeoffs between such views and Hermeneutic Contextualism.
Metaphor and mercurial content
163
important to bring out a less common meaning, one that captures the particular information relevant to immediately pressing circumstances.31 For example, we observed this type of hermeneutic variation earlier when considering alternative possible interpretations of Romeo’s utterance (8) (“Juliet is the sun”). Mercutio and Benvolio each interpret the utterance in a legitimate manner with respect to their concerns and interests in their respective contexts. Yet their interpretations ((9) and (10)) crucially differ: Mercutio is more attentive to the romantic impulse in Romeo’s actual thoughts, while Benvolio is more occupied with his own worries about the danger of the enterprise. Neither interpreter has made a linguistic mistake here: it is possible to be mistaken about the syntactic structure of an utterance, just as we may mishear a word or misgauge a pitch; but the utterance’s truth-conditions are not the type of property that can be misperceived in this way.32 Rather, the interpreters’ disagreement arises because they have different perspectives, and they have each performed reasonable and legitimate interpretations relative to their own perspective. In other words, we are not dealing with an unconstrained relativism or objectionable subjectivism here; hermeneutics is a principled business. Interpreters who violate hermeneutic pressures are criticizable: their interpretations seem mistaken, and we call them incorrect (recall (13)). But we should not reason from the fact that there are such pressures on interpretation to the conclusion that there is a single perspective-independent correct content for an utterance – that would be a fallacy. (That P is judged to be an incorrect interpretation of U does
31 We may grant that in some contexts, the legitimate way to take an utterance may in fact be to defer to speaker intentions (see footnote 34); or more precisely, to the speaker’s interpretation of his own utterance. But this only shows that Intentionalism may be reformulated as a special case of the hermeneutic view: it is the appropriate interpretive strategy for some contexts, but not for all contexts. In art theory, literary theory, legal interpretation, and any number of other disciplines, debates continue to play out regarding the proper roles for authorial intention and interpreter judgment in identifying and developing the significance of human works, as we attempt to find viable alternatives to the “intentional fallacy”. Hermeneutic Contextualism simply reconnects the study of utterance content to this broader enterprise. 32 See Horowitz (2006) for critique of the assumption that truth-conditions can be simply derived from semantics. If the notion of truth we are considering here is our ordinary notion – that which we believe in, argue about, care about – the kind of correctness that concerns us is a more humanistic affair.
164
Damon Horowitz
not entail that U has exactly one correct interpretation).33 Hermeneutic Contextualism holds that an interpretation of an utterance is correct when that interpretation satisfies the demands upon it in its particular hermeneutic situation. Thus, since the pressures on interpretation vary across contexts with our varying purposes, it may be correct to ascribe different contents to a single utterance on different occasions.34 Third Objection: We have strong intuitions that there is one single objective content for an utterance. If two people are genuinely disagreeing about the truth of an utterance, isn’t it counter-intuitive to claim that they can both be right?
Two people may disagree about whether an utterance is true for many reasons. They may share an understanding about what the truth-conditions of the utterance are – i.e., about the way things must be for the utterance to be true – and disagree that things are really that way. Or the source of their disagreement may be prior to this: they may have different views about what the truth-conditions of the utterance are in the first place. (Recall the above disagreements between Benvolio and Mercutio as interpreters of Romeo’s utterance.) This is still a “genuine” disagreement, often turning on important consequences of the opposing views – as we see in contemporary debates about whether the situation in Iraq counts as a “civil war”, whether a fetus counts as a “person”, and other similarly charged cases. Indeed, there are even genuine disagreements which originate at a level prior to the determination of content, concerning which hermeneutic pressures are appropriate to a given interpretive occasion – as exemplified, for instance, by debates between Intentionalists and Textualists in legal interpretation. It is true that it seems unintuitive to imagine that two parties can have different views about what an utterance’s content is and both be correct. But that
33 By analogy: the criticizability of a particular analysis of a work of art does not entail that there exists one correct objective analysis of the work’s meaning. Similarly in the case of individual utterances: we do not have good reason to believe that there is some single candidate for “the content” in ordinary cases that can satisfy all constraints simultaneously for all occasions. If the objector here continues to feel a lingering sense that there is something more to being the content of an utterance than being legitimately judged to be so, the burden is on the objector to say what that something more might be. 34 What the active interpretive pressures actually are in a given context is the subject matter for a methodological hermeneutics of single utterances. With the Hermeneutic Contextualism view (i.e., that “correctness” of interpretation is a matter of meeting whatever interpretive standards are in force in a given context), it may turn out that an Intentionalist or Textualist approach is the best account for a particular case. Developing principles to guide this determination is important work to do; but I am not proposing any particular principles on that front here.
Metaphor and mercurial content
165
restriction only holds when both parties are in the same hermeneutic situation – for then the same hermeneutic pressures will obtain, constraining what content may be correctly ascribed. Hermeneutic Contextualism does respect the intuition that a particular ascription of content may be the right one in a given interpretive context; in this sense, the appearance of determinacy may be upheld. But often we are not in the same hermeneutic situation with other interpreters, and differences in our contextual purposes, interests, and information may give us legitimately different views of what the content is. In these cases, we may look to a form of perspectivism to supply the desired normative constraint that goes wanting when we release our folk objectivity about utterance content.35
9 Conclusion Contextualists make a strong case for the claim that metaphors do have truth-conditions. When we agree or disagree with a metaphorical utterance, we are articulating our belief that it expresses particular conditions – via contextually-appropriate ad hoc meanings – and that these conditions are or are not satisfied. But once we accept a Contextualist analysis of the pragmatic factors shaping “what is said”, it may appear that there is considerable indeterminacy in precisely which truth-conditional content to ascribe to a metaphorical utterance – leading us to question what it is for an utterance to have such a content. I have proposed that the right response to this concern is not a rejection of the Contextualist insight, but rather a development of it, revealing its affinity with the hermeneutic tradition’s approach towards meaning – with broader methodological implications. The sense in which an utterance has a content is that interpreters legitimately take it to have that content; thus, it can be correct to ascribe
35 By analogy: think of two observers of an object reporting on its appearance from the same perspective (in which case it is odd to say they may disagree and both be correct) versus from different perspectives (in which case they may well disagree and both be correct). We may need to be tutored to realize that an utterance’s content is a relativistic property – perhaps by examining cases in which we naturally speak of an utterance meaning one thing to one person and another to another. But tutoring is also required in other domains when an “unobvious higher degree thesis” is advanced (e.g., we may need to be tutored to realize that an object’s motion is a relativistic property, dependent upon the perspective of the observer). Our untutored intuition that there is a single correct ascription of content to an utterance is mistaken – as we are often mistaken in believing that our own perspective provides the only right view – but our intuition that a particular content should be ascribed to the utterance given our context need not be abandoned.
166
Damon Horowitz
different contents to a single utterance on different interpretive occasions. Such is the Hermeneutic Contextualist thesis – challenging the standard Contextualist and Literalist assumption that “what is said” by an utterance is perspectiveindependent. By adopting a hermeneutic approach, Hermeneutic Contextualism presents a more thoroughgoing Contextualism,36 acknowledging the essential dependence of meaning upon the interpreter and the consequent mercuriality of utterance content. The arguments presented above focus on the case for Hermeneutic Contextualism as an account of metaphorical utterances’ content. But there are reasons to suspect that the considerations raised here in support of the view are also relevant to the analysis of literal utterances. As described in Section 2, Contextualists have developed independent arguments for the continuity of literal and metaphorical language; and the ubiquity of figurative and context-sensitive speech in ordinary language has been increasingly recognized in recent years. Further, the special type of mercuriality of content described here may be identified in literal and metaphorical speech alike – as we saw, for instance, in the parallel analyses offered for metaphorical “bulldozer” and literal “bully” predications in Section 6. In addition, there is reason to believe that a Hermeneutic Contextualist analysis is suitable for cases of prototypical literal speech, such as ordinary knowledge attributions: as has been argued elsewhere (see footnote 30), whether an epistemic subject counts as “knowing” something (literally) may depend upon elements of the evaluator’s context. Of course, work remains to be done to substantiate this continuity claim, and to respond to various objections to Hermeneutic Contextualism more thoroughly. But I hope this discussion is sufficient to at least recommend that path as a fruitful one to explore. Acknowledgments: Earlier versions of this paper benefited from comments received from audiences at the American Philosophical Association Eastern Division Meeting, the University of Maryland, and Stanford University. Thanks especially to Mark Crimmins, David Hills, Paul Pietroski, and Ken Taylor for helpful discussion.
36 Just as standard Contextualism claims that a single expression may legitimately have different meanings in different contexts of utterance, Hermeneutic Contextualism claims that a single utterance may legitimately have different contents in different contexts of interpretation. And it is in contexts of interpretation that utterance reports and truth-assessments (our explananda) are made.
Metaphor and mercurial content
167
References Bezuidenhout, Anne. 2001. Metaphor and what is said: A defense of a direct expression view of metaphor. Midwest Studies in Philosophy 25 (1). 156–186. Black, Max. 1962. Metaphor. In Models and metaphors. Ithaca: Cornell University Press. Borg, Emma. 2001. An expedition abroad: Metaphor, thought, and reporting. Midwest Studies in Philosophy 25 (1). 227–248. Camp, Elizabeth. 2006a. Contextualism, metaphor, and what is said. Mind and Language 21 (3). 280–309. Camp, Elizabeth. 2006b. Metaphor in the mind: The cognition of metaphor. Philosophy Compass 1 (2). 154–170. Cappelen, Herman & Ernest Lepore. 2005. Insensitive semantics. Oxford: Blackwell. Carston, Robyn. 2002. Thoughts and utterances: The pragmatics of explicit communication. Oxford: Blackwell. Davidson, Donald. 1978. What metaphors mean. Critical Inquiry 5 (1). 31–47. Egan, Andy, John Hawthorne & Brian Weatherson. 2005. Epistemic modals in context. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy, 131–169. Oxford: Oxford University Press. Gadamer, Hans-Georg. 1975. Truth and method. New York: The Seabury Press. Geach, Peter. 1965. Assertion. Philosophical Review, 74: 449–465. Gibbs, Raymond. 1994. The poetics of mind: Figurative thought, language, and understanding. Cambridge: Cambridge University Press. Glucksberg, Sam, Patricia Gildea & Howard Bookin. 1982. On understanding nonliteral speech: Can people ignore metaphors? Journal of Verbal Learning and Verbal Behavior 21. 85–98. Grice, Paul. 1989. Studies in the ways of words. Cambridge, MA: Harvard University Press. Guttenplan, Samuel. 2006. Contextualism, metaphor, and what is said. Mind and Language 21 (3). 280–309. Hawthorne, John. 2007. Testing for context-dependence. Philosophy and Phenomenological Research 73 (2). 443–450. Hills, David. 1997. Aptness and truth in verbal metaphor. Philosophical Topics 25 (1). 117–153. Hills, David. 2004. The pleasure of ulteriority: Four essays on verbal metaphor. Princeton: Princeton University dissertation. Horowitz, Damon. 2006. Must we be so literal? Paper presented at American Philosophical Association, New York, 27 December. Horowitz, Damon. 2007. Meaning and relativism in knowledge and metaphor. Stanford: Stanford University dissertation. Horn, Laurence. 1989. A natural history of negation. Chicago: University of Chicago Press. MacFarlane, John. 2005. Making sense of relative truth. Proceedings of the Aristotelian Society 105 (1). 305–323. Moran, Richard. 1989. Seeing and believing: Metaphor, image, and force. Critical Inquiry 16 (1). 87–112. Recanati, Francois. 2004. Literal meaning. Cambridge: Cambridge University Press. Reimer, Marga. 2001. Davidson on metaphor. Midwest Studies in Philosophy 25 (1). 142–155. Reimer, Marga. 2004. What malapropisms mean: A reply to Donald Davidson. Erkenntnis 60 (3). 317–334. Ricoeur, Paul. 1978. The rule of metaphor: Multi-disciplinary studies in the creation of meaning in language. Toronto: University of Toronto Press.
168
Damon Horowitz
Rumelhart, David E. 1979. Some problems with the notion of literal meanings. In Andrew Ortony (ed.), Metaphor and thought, 78–90. Cambridge: Cambridge University Press. Searle, John. 1979. Metaphor. In expression and meaning studies in the theory of speech acts. Cambridge: Cambridge University Press. 30–57. Stanley, Jason. 2005. Semantics in context. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy, 221–253. Oxford: Oxford University Press. Stern, Joseph. 2006. Metaphor, literal, literalism. Mind and Language 21 (3). 243–279. Taylor, John R. 2012. Contextual salience, domains, and active zones. In Hans-Jörg Schmid (ed.), Cognitive Pragmatics (Handbook of Pragmatics 4), 151–174. Berlin & Boston: de Gruyter Mouton. Wearing, Catherine. 2006. Metaphor and what is said. Mind and Language 21 (3). 310–332.
Part 3: The contextual turn and the case of analysis
Dominik Aeschbacher
Context, two-dimensional semantics and conceptual analysis Abstract: According to a traditional picture of philosophical practice, Philosophers engage in conceptual analysis, which results in knowledge that is a priori and necessary in its nature. This picture entails that there is an intimate connection between a priori knowledge, meaning and modality. Two-dimensional semantics in the spirit of David Chalmers shows how to make sense of these connections against specific worries brought in by semantic externalism. However, context affects natural language in many different ways. Tacit background knowledge, philosophical theories and vagueness have a huge influence on linguistic behaviour. It is therefore an open question whether two-dimensional semantics can provide an adequate theory of meaning that allows accommodating the various forms of context. In this paper, it is argued that if one accepts some further distinctions, the two-dimensional framework can do quite well with different kinds of context-dependence.
1 Introduction The search for accurate definitions of philosophically interesting notions has played a central role in philosophy for a long time and providing necessary and jointly sufficient conditions, known also as “conceptual analysis”, is well-known among philosophers in the analytic tradition. Conceptual analysis is taken to be a purely a priori matter and thus it is usually done from the armchair. When engaging in conceptual analysis, philosophers conceive of actual and counterfactual scenarios and intuitively apply or withhold a philosophically interesting notion. This in turn helps them in making explicit the conditions of application of the notion in question. The a priori character of this method is rather innocent, for it does not require a special faculty of rational insight, but merely a tacit knowledge of how the notions are usually applied. So in order to engage in conceptual analysis, philosophers merely have to be competent speakers. Let me call this picture of philosophical methodology “the traditional picture”.1
1 I call this picture “traditional” because the method that it describes can be found in many classical texts of analytic philosophy. Gettier (1963) is probably the most paradigmatic example. For similar descriptions of conceptual analysis, see Moore (1942), Grice (1989) and Jackson (1998). DOI 10.1515/9781501504327-009
172
Dominik Aeschbacher
Unfortunately, the traditional picture requires some strong assumptions that have been the target of many objections. One such assumption is that every competent speaker has an a priori access to the meanings of our linguistic expressions. Against this, proponents of semantic externalism such as Saul Kripke (1980), Hillary Putnam (1975) and Tyler Burge (1979) have argued that the meaning of a notion should be identified with its direct reference. In their semantic frameworks, competent speakers have therefore no a priori access to the meanings of their linguistic expressions. A second problem for the traditional picture arises from its assumption that all necessary truths are knowable a priori. Semantic externalists have argued, however, that there are necessary truths such as “water is H2O” which are not knowable a priori. Frank Jackson and David Chalmers (2001) have attempted to meet these challenges by developing the framework of two-dimensional semantics. Twodimensional semantics is similar to standard possible world semantics, but in contrast with the latter it assigns extensions and truth-values to two possible world parameters, rather than just one. This allows the two-dimensionalists to reconcile the semantic assumptions of the traditional picture with the arguments from semantic externalism. And yet, two-dimensional semantics encounters still other problems because there are many context influences on natural language and it is not clear whether two-dimensional semantics can save the traditional picture and conceptual analysis given these type of influences. In this paper, I will discuss the following three kinds of context influence that affect the traditional picture and two-dimensional semantics: (1) There is some evidence from experimental philosophy (Alexander and Weinberg 2007) and every day observations (Schroeter 2003) that ordinary people do not share the same tacit knowledge in order apply a notion. (2) The traditional picture presupposes that there are analytic truths, but background knowledge can have an influence whether a sentence is analytic or not; see e.g. Williamson (2007). (3) Competent speakers are affected by contextual factors when judging borderline cases. I will argue that with some minor distinctions, the two-dimensional framework can be saved despite the problems addressed above. In the first section, I will give a more detailed account of conceptual analysis and I will reconstruct the semantic and epistemological assumptions behind the traditional picture. In the following section, I will show how semantic externalism threatens the traditional picture and how two-dimensional semantics as developed by Chalmers (2006) can reconcile these objections with the traditional
Context, two-dimensional semantics and conceptual analysis
173
picture. In the three following sections, I will discuss three kinds of context influence and argue that none of them poses a problem for the traditional picture.
2 The traditional picture The traditional picture, as I understand it, can be characterized by at least two features2: (1) Philosophers engage in conceptual analysis. (2) There is an intimate relation between meaning, reason and modality. Let’s turn to (1). According to Frank Jackson (1998), conceptual analysis reveals an implicit folk theory that every competent speaker associates with a linguistic expression. The folk theory that is associated with a linguistic expression is the set of common sense beliefs that ordinary people have about a specific subject matter. These common sense beliefs relate to the properties that an entity must possess in order to be in the extension of the expression in question. More precisely, these properties specify a functional role that an entity must play in order to be the denotatum of the expression. For example, the functional role that we associate with the word “water” is something similar to a colorless liquid in oceans, lakes and rivers that is in normal circumstances drinkable and so on. For the sake of brevity, I will follow Chalmers (2006) and call the functional role of water “being watery stuff”. Whatever plays the functional role of watery stuff in our actual world counts as water according to our ordinary conception. The functional role that we associate with a linguistic expression determines its reference. For example, we succeed in referring to water (i.e. H2O in the actual world) when uttering “water”, because H2O plays the functional role that we associate with water. This picture of reference fixing implies a specific form of descriptivism. According to descriptivism, competent speakers grasp a reference-fixing criterion that is equivalent to some kind of descriptive content of the expression at hand. It is only in virtue of this descriptive content that a referential expression refers to something. So a proper name (or a natural kind term) e uttered by a speaker S at a time t refers to its referent R if and only if R uniquely exemplifies the set of properties that is associated with a descriptive content D.
2 A more subtle version of what I present here is also known as The Canberra Plan (BraddonMitchell and Nola 2009). I changed the label because I will deal with a simpler version that is not concerned with the technical aspects of the Canberra Plan. However, the arguments put forward in this paper can also be applied to the Canberra Plan.
174
Dominik Aeschbacher
For example, the proper name “Batman” might be associated with properties such as wearing a black costume, being a superhero and having no superpowers. “Batman” refers to Batman because he is the only person who exemplifies all these properties. Similarly, folk theories state that an entity must possess a particular set of properties and play a certain functional role in order to belong to the extension of a concept. A further important aspect of the traditional picture is that it is committed to the assumption that higher order facts are entailed a priori by lower order facts. Thus, truths about H2O state the same as truths about water, but they do so in a different vocabulary. It is the task of conceptual analysis to show how these different vocabularies are connected. For example, a conceptual analysis of “water” will reveal that the functional role associated with “water” is that of watery stuff. When you observe that H2O covers most of the Earth and that H2O is watery stuff, you know a priori that water covers most of the Earth. In other words, there is an a priori entailment of truths about water by truths about H2O. What is of note is that the traditional picture takes conceptual analysis to play a modest role (see e.g. Jackson 1998). Conceptual analysis does not reveal anything about the world itself, nor does it reduce philosophical problems to a confusion of language. It rather helps us to clarify what we mean with our linguistic expressions: it is nothing else than revealing our respective folk theory. But whether the folk theory is correct or not is a matter of empirical investigation. According to the assumption of the traditional picture mentioned in (2) there is an intimate relation between meaning, reason and modality. How are these notions connected? According to Jackson, conceptual analysis is equivalent to what he calls the method of possible cases (Jackson 1998: 31). By imagining counterfactual scenarios, we can observe whether we intuitively apply or withhold a linguistic expression e to a specific situation, which in turn helps us to specify what all the scenarios where e gets applied have in common. As soon as we know what they have in common, we know the conditions of application of e, which in turn are equivalent to the functional role associated with e. The semantic assumptions that underlie this method have been worked out by David Chalmers (see e.g. Chalmers 2006). According to him, the three notions “modality”, “meaning” and “reason” are interconnected and build what he calls a golden triangle. Immanuel Kant established the first link by showing that everything that is knowable a priori is a necessary truth. Thus Kant connected reason with modality. Secondly, Gottlob Frege (1986) established the link between reason and meaning by introducing a further aspect of meaning that he called a sense. While there are crucial differences between senses and Jackson’s folk theories,
Context, two-dimensional semantics and conceptual analysis
175
these differences do not matter for our current purpose. Like folk theories, senses provide some kind of descriptive content that fixes the reference of a linguistic expression. Introducing Fregean senses into a theory of language allows establishing a connection between reason and meaning. A statement “A↔B” (i.e. “A and B are equivalent”) is cognitively significant, especially when “A” and “B” are proper names and the equivalence relation is interpreted as the relation of identity. For example, the statement “Superman is identical with Clark Kent” is cognitively significant, but the statement “Superman is Superman” is not. The first statement is cognitively significant because “Superman” and “Clark Kent” have different senses and therefore we can merely know a posteriori that Superman and Clark Kent are identical. The second statement is knowable a priori because it involves twice the same name and the two occurrences of the name have, of course, identical senses. So the link between reason and meaning has been established by showing that the a priori nature of a statement can be explained by linguistic entities such as senses. Finally, Rudolf Carnap (1947) has established the connection between modality and meaning. The link is nowadays known under the label of “possible worlds semantics”. Possible world semantics allows us, very roughly, to individuate meanings as functions from possible worlds to extensions. These functions are called “intensions”. Two linguistic expressions have the same intension if and only if they are co-extensive across all possible worlds. On the other hand, if two linguistic expressions are not necessarily co-extensive, they have different intensions. Since Carnap’s intensions behave in the same way as Fregean senses do, I will follow Chalmers (2006) and speak of intensions instead of senses in the rest of the paper. With all links in the golden triangle established, it becomes clear how the three notions of modality, reason and meaning are interconnected: A statement “A↔B” can be known a priori, when “A” and “B” have the same intension and they have the same intension if and only if they are necessarily co-extensive. As a consequence, if “A↔B” is necessary, it can be known a priori (if “A” and “B” are proper names, the biconditional must be interpreted as identity). So the golden triangle guarantees that conceptual analysis can reveal knowledge that is a priori and necessary in its nature. Conceptual analysis consists in nothing more than revealing the conditions of application of a notion that can now be identified with its intension. Since the intension determines the extension of a notion in every possible world, it is possible to find out a priori what an expression means by applying Jackson’s method of possible cases.
176
Dominik Aeschbacher
3 Semantic externalism and two-dimensional semantics The traditional picture rests on a simple but solid semantic foundation. However, the arguments from semantic externalism might raise some doubt about its underlying semantic assumptions. Semantic externalists such as Saul Kripke (1980), Hilary Putnam (1975) and Tyler Burge (1979) argued convincingly that the reference of linguistic items is not mediated by a descriptive content, but given directly. Thus the meaning of a word such as “water” is not a description like watery stuff, but the chemical structure of H2O. Moreover, Kripke has argued that statements like “Water is H2O” are necessarily true. According to the traditional picture, such statements are merely contingently true for the statement is not a priori. Against this, Kripke holds that while it is not knowable a priori that water is H2O, it is nonetheless necessarily true. As a consequence of his criticism, the golden triangle shatters because the link between the necessary and the a priori has been cut. While it is not knowable a priori that water is H2O, it is nonetheless necessarily true. The arguments from semantic externalism threaten the traditional picture because they introduce a kind of context-sensitivity that makes conceptual analysis impossible. Conceptual analysis presupposes that competent speakers have an a priori access to the meanings of the linguistic expressions in their community. But if the meaning of “water” is not an a priori knowable functional role such as watery stuff, but merely the natural kind H2O, conceptual analysis turns out to be impossible. In an externalist framework, it is the external world that determines the meanings of our linguistic expressions and the only way to get to know them is through empirical investigation. It is even possible that speakers associate completely different properties with a linguistic expression and still succeed in communicating with each other. Fortunately, proponents of conceptual analysis have been busy looking for a solution to reconcile the traditional picture with semantic externalism. This has lead to what sometimes goes under the name “ambitious two-dimensional semantics” (Soames 2005), or just “two-dimensional semantics”. Two-dimensional semantics assumes that there is more to meaning than just an intension and an extension. It is thought to be ambitious because it provides a theory of representation and isolates an a priori aspect of meaning. In what follows, I will provide a short overview of the ambitious two-dimensional semantics without going into too many technical details. The two-dimensional framework developed by Chalmers (2006) and Jackson (1998) assumes that additionally to the classical intension that one may call a
Context, two-dimensional semantics and conceptual analysis
177
“2-intension”, there is another intension associated with a linguistic expression. I suggest to follow Chalmers in calling it a 1-intension. The 2-intension of a linguistic expression is the same that I mentioned above discussing the golden triangle. It delivers the extension of a linguistic expression when we consider other worlds as counterfactual. When considering worlds as counterfactual, you ask yourself what a linguistic expression would refer to in another possible world. For example, the expression “the inventor of the lambda calculus” refers in our world to Alonzo Church, but in another world considered as counterfactual it might refer to Alan Turing. It is worth noting that the 2-intension behaves a little bit differently when it comes to proper names or natural kind terms. The expression “water” refers to H2O in the actual world and since water is necessarily H2O, it has H2O as its extension in every possible world considered as counterfactual. So if worlds are considered as counterfactual, “water” cannot refer to another natural kind such as XYZ, which plays the functional role of watery stuff but has another chemical structure. The other intension, the 1-intension, behaves quite differently. The 1-intension picks out the extension of linguistic expressions when possible worlds are considered as actual. When you consider a possible world as actual, you do not think of it as a counterfactual world where things are different from how they actually are. Rather the world considered as actual is an epistemic possibility, i.e. a world that could turn out to be our actual world. Consider the following example. Suppose you wonder whether there is life on Titan. You do not know whether there is life on Titan, but it is possible. In this case, you have conceived a possible world as actual where life on Titan exists. It is possible that this world turns out to be the actual world. If you think of worlds in this way, the extension of our linguistic expressions is fixed differently. This point can be illustrated again by using the water example. Suppose that you do not know that water has the chemical structure H2O. If you consider worlds as actual, it is possible that “water” refers to a liquid that has the chemical structure of XYZ. However, it is also possible that it refers to stuff that has the chemical structure of H2O. Given these two possibilities, the sentence “Water is H2O” turns out to be merely contingently true. The set of all possible worlds that contain some kind of watery stuff can now be identified with the 1-intension of “water”. However, when you consider worlds as counterfactual and the actual world contains “H2O”, the statement “Water is H2O” is necessarily true. Correspondingly, there is a different set of possible worlds associated with “Water is H2O” which is its 2-intension. Why can 1-intensions behave differently than 2-intensions? It seems appropriate to interpret the 1-intension of a linguistic expression as its descriptive content, i.e. a functional role or a folk-theory that we associate with a linguistic expression.
178
Dominik Aeschbacher
It is something quite close to a Fregean sense. It picks out a reference across possible worlds when possible worlds are considered as actual. In the actual world, the functional role of watery stuff is played by H2O and therefore “water” refers to H2O in the actual world. However, if we consider a Twin-Earth as actual where everything is as on Earth, except that the watery liquid is XYZ, it refers to XYZ for XYZ plays the functional role of watery stuff on Twin-Earth. As a consequence, “water is H2O” is merely contingently true with regard to its 1-intension. Once the reference of “water” is fixed, you can consider worlds as counterfactual and ask yourself: Could water have been different from H2O? In that case you would ask yourself whether water is not water. Clearly, asking such a thing makes no sense. The distinction between 1-intensions and 2-intensions is quite intuitive for it turns on the fact that one can make a similar distinction between epistemic and metaphysical possibilities. 1-intensions can be considered as sets of epistemically possible worlds, while 2-intensions can be considered as metaphysically possible worlds. According to Chalmers and Jackson (2001), 1-intensions provide an interesting view on what it means to possess a concept. A competent speaker is supposed to be able to determine the extension of a linguistic expression across possible worlds considered as actual. So having a concept e of something implies to be able to determine the truth of a material conditional: If world w is actual, e refers to x.3 With this two-dimensional framework at hand the traditional picture can be saved from the criticism put forward by semantic externalism. First, it allows to say that 1-intensions take on the job of a functional role. Consequently, the aim of conceptual analysis is to make explicit our 1-intensions. It does not matter that the 2-intension of a linguistic expression can change with the external world, for conceptual analysis is merely targeted at whatever fixes the reference and this job can be done by the 1-intension. Second, the golden triangle can be re-established because the 1-intension behaves in the same way as Fregean senses do: A statement “A↔B” is therefore cognitively insignificant in case “A” and “B” share the same 1-intension; and if they share the same 1-intension, they have the same extension across all possible worlds considered as actual. Finally, it has been argued by Chalmers and Jackson (2001) that the twodimensional framework gives an interesting explanation for what is actually going on in thought experiments such as Putnam’s well-known Twin Earth experiment
3 My discussion of two-dimensionalism is quite simplified. However, it should be noted that neither Chalmers (2006) nor Jackson (1998) claim that 1-intensions behave like Kaplanian characters (see Kaplan 1989). Kaplanian characters will pick out a reference merely in contexts of utterance, while 1-intensions can pick out a reference in worlds considered as actual where there are no speakers. For a thorough discussion of these issues see Chalmers (2006).
Context, two-dimensional semantics and conceptual analysis
179
referred to above4. The thought experiment starts of by describing two subjects S1 and S2 that are mentally fully alike and hence associate the same descriptive contents with their linguistic expressions e1 and e2. Then you point to a difference in their environment, which shows that e1 and e2 do not have the same reference. This can only be done successfully, if you consider two worlds as actual. In Putnam’s Twin-Earth scenario, for instance, you consider first a H2O world as actual and you judge that H2O is the reference of “water”. Then you consider a XYZ world as actual and you judge that XYZ is the reference of “water”. In a last step, you conclude that e1 and e2 cannot have the same meaning because they have different referents. But how did Putnam determine the extension of “water” in the H2O world and in the XYZ world? The only intelligible answer is that the subjects S1 and S2 share the same descriptive content. Since their worlds are different, the same descriptive content picks out different referents. So it seems that Putnam has to presuppose that there is a descriptive content, namely a 1-intension that determines different referents across possible worlds. Hence, externalists like Putnam failed to refute descriptivism because they have to consider worlds as actual for their thought experiments to succeed in the first place. Despite this criticism they are right regarding what they say about usual intensions, namely 2-intensions. And this is exactly what ambitious two-dimensionalism amounts to when it suggests that there are two ways of thinking about necessity and possibility. Since ambitious two-dimensionalism states that 1-intensions are descriptive contents, and since externalists have to presuppose the two-dimensional framework for their thought experiments to succeed, externalists seem to presuppose descriptivism. At this point it is worth mentioning that the two-dimensional framework developed by Chalmers and Jackson does not reject the claims made by semantic externalists. To the contrary: It rather combines the traditionalist picture with an externalist semantics. I think that this fact is very important and all too often ignored by the critics. In fact the framework can account for both Fregean and Kripkean intuitions.
4 Bad context: Individual folk theories The two-dimensional framework has not been spared from criticism. One kind of criticism is that while it might be able to integrate externalism in the traditional
4 The same argument can also be applied to Burge’s famous Arthritis case.
180
Dominik Aeschbacher
picture, it fails to account for other kinds of context. More precisely, it has been objected that 1-intensions can be affected by context in such a way that conceptual analysis still is impossible. For example, proponents of experimental philosophy claim to have shown that intuitive judgments about counterfactual cases vary with contingent factors such as cultural background or priming (see e.g. Alexander and Weinberg 2007). Similarly, Laura Schroeter (2003) has pointed out that 1-intensions differ from one subject to the other because everyone determines different extensions when considering worlds as actual. Finally, David Chalmers (2006) acknowledges that whether a proposition can be known a priori varies with context insofar as that 1-intensions can differ from one individual to the next. For example, the sentence “If Neptune exists, it perturbs the orbit of Uranus” expresses a 1-intension that is knowable a priori by Leverrier because he defined the 1-intension of “Neptune” as whatever perturbs the orbit of Uranus. His wife, however, associates a different 1-intension with “Neptune” and therefore it is not a priori knowable for her that Neptune perturbs the orbit of Uranus. These examples point to a ubiquitous divergence in linguistic usage. It seems that despite the fact that competent speakers associate different folk theories with a linguistic expression they still succeed in communicating with each other. Hence, successful communication seems not to rely on shared 1-intensions. Unfortunately, this divergence in linguistic usage does not really support the traditional picture. While it does not challenge the connection between necessity and the a priori, it could still cause some troubles for conceptual analysis for the following reasons. Conceptual analysis is essentially a social endeavor. We use it in order to clarify specific subject matters. We also need it for scientific purposes such as the reduction of higher-order facts to lower-order facts. As I have shown in section 2, proponents of conceptual analysis like Jackson (1998) believe that conceptual analysis is committed to the a priori entailment of higher-order facts by lower-order facts5. However, if communication is possible without significant convergence in folk theories, it is not necessary to clarify the concepts, but one needs to look at the things themselves and this is achieved by doing empirical investigations and not through fancy thought experiments. I would like to add another problem in relation to the divergence in linguistic usage. It follows from it that it undermines the possibility of disagreement. If my 1-intension of a linguistic expression e is different from yours, you express a different 1-intension when you utter e. So when you utter e and I utter the negation ~e, I do not express the negation of the proposition that you utter and therefore, there is no disagreement. As a consequence, most philosophical disputes turn out to
5 For the connection between reductionism and conceptual analysis, see Jackson (1998).
Context, two-dimensional semantics and conceptual analysis
181
be pseudo-disagreements. For example, an internalist in epistemology believes that a subject knows a proposition solely in virtue of internal states or reasons of that subject, while an externalist believes that factors outside of a person can determine whether she has knowledge or not. But it seems that internalists and externalists do not have a genuine disagreement for both understand the word “knowledge” differently. As a consequence both of them talk past each other. The question, however, is if the traditional picture can be dismissed that easily. As far as I can see, it cannot, at least if you are a meaning pluralist. A meaning pluralist6 believes that a linguistic expression can express different semantic contents. It is obvious that proponents of two-dimensional semantics are meaning pluralists because they believe that a sentence expresses at least a 1-intension and a 2-intension. However, there is no reason to stop there. In order to explain the variety of folk theories and successful communication, I suggest that a speaker can also express two different kinds of 1-intension with one and the same utterance. The first kind of 1-intension is the intension that the majority of a community or its experts associate with a linguistic expression. In what follows, I will call this the public 1-intension. The public 1-intension of a specific expression accompanies the expression whenever it is used by a member of a specific language community. The second kind of intension is what a specific individual associates with a linguistic expression. In what follows I will call it the private 1-intension. A speaker counts as fully competent if and only if her private 1-intension converges with the public 1-intension. So a speaker expresses with an expression the private 1-intension she associates with it and at the same time, she expresses the public 1-intension that the community or the experts of it associate with it. In order to illustrate the point I will use this rather extreme example: Suppose I falsely believe that monkeys are a liquid that have an orange color and contain a lot of vitamins. My private 1-intension of the word “monkey” will include these properties and I will privately succeed in referring to orange juice when I think about monkeys. However, when uttering sentences such as “I had a glass of monkeys for breakfast”, my utterance expresses the public 1-intension that I had a certain kind of animal for breakfast. A speaker who utters a sentence such as “I had a glass of monkeys for breakfast” is completely incompetent with regard to the word “monkey”. Almost every property that he associates with the expression does not belong to the folk theory that is associated with the public 1-intension. However, there are other cases where a speaker’s private 1-intension differs merely in some minor aspects from
6 The expression “meaning pluralist” can be found in Chalmers (2006).
182
Dominik Aeschbacher
the public 1-intension. Should such a speaker be deemed to be incompetent? Perhaps, language competence is a matter of degree and a speaker can count as more or less competent with regard to a specific linguistic expression. I think that this suggestion makes sense for pragmatic reasons. If degrees of language competence are not allowed, many speakers would count as incompetent for there are many words (for example scientific expressions) whose public 1-intension is only known by a small group of experts. However, it seems that non-experts are quite capable to engage in conversations about scientific issues with the experts. If degrees of language competence are allowed, this fact can be explained very easily by pointing out that the ordinary people might have a partial understanding of scientific terms such as “electron” or “dark matter”. One important question that needs to be answered is what the nature of these public 1-intensions is. Various contextual factors determine what a public 1-intension of a particular linguistic expression is. In the simplest case, the public 1-intension corresponds to the functional role that the experts of a community have defined explicitly. According to Putnam’s (1975) metaphor of linguistic labor, the meaning of a linguistic expression is parasitic on the meaning that the experts have assigned to it. For example, if someone without expert knowledge uses a scientific term such as “electron” or “dark matter”, it has the meaning the experts have assigned to it. Adopting this metaphor, it seems plausible to say that the 1-intension that is expressed in a normal conversation usually corresponds to the public 1-intension that the experts have assigned to it. But things get more complicated with everyday terms where there are no experts for the subject matter in question. What could be the public intension of an everyday term like “bottle” or “table”? As far as I know, there are no experts for tables or bottles. I suggest to resolve this problem in the following way: If it is presupposed that the members of a linguistic community are rational and intend to adjust their own usage to the usage of their peers, the public 1-intension could be whatever the speakers would agree to while negotiating under ideal rational conditions. This process, of course, is not easy and may be affected by contingent social factors, but nevertheless it is prima facie available. One could say that the public 1-intension is some kind of ideal intension, which fits the purposes of the members of a community in the best way. It might be objected that the members of a community have quite different interests and are affected by different social factors when negotiating the public 1-intension. As far as I can see this is, however, unproblematic. It seems rather plausible to allow that a linguistic expression might have more than one public 1-intension assigned to it. A community is not a homogenous group of people, but might contain several groups as parts. The members of these groups might come to an agreement about the usage of a specific term. In that case there would
Context, two-dimensional semantics and conceptual analysis
183
be different groups in a community and each group would have their own public 1-intension. A linguistic community would therefore consist of different groups attributing different public 1-intensions to a specific expression, and yet they share a high degree of resemblance. At first sight, it might seem that introducing a public and a private 1-intension simply is an ad-hoc solution for saving conceptual analysis. However, the distinction can account for other phenomena, too and this strengthens the explanatory force of the distinction. For instance it allows to make sense of disagreement. Given the distinction between the public and the private 1-intensions, it is possible for two speakers to disagree even in case they associate different 1-intensions with a specific linguistic expressions. In such a case, one speaker expresses the public 1-intension of an utterance and the other its negation. This allows us to explain why two philosophers who disagree for example on the right analysis of “knowledge” do have a genuine disagreement even though they understand the word “knowledge” differently. Both are disagreeing on the public 1-intension of the word and not on their private 1-intensions. Furthermore, some pseudodisagreements can be solved if the two disagreeing parties make explicit what their private 1-intensions are. The pseudo-disagreement appears to be a genuine disagreement if one looks merely at the public 1-intension. However, if the speakers realize that there are too many differences between their private 1-intensions, they could solve the disagreement by adjusting their 1-intensions in such a way that they get closer to the public 1-intension. And this is the point where conceptual analysis takes over as it is meant to do: Conceptual analysis will help us to elucidate the character of a pseudo-disagreement and either show that we actually are confronted with a genuine disagreement or it will help to resolve the disagreement altogether. Furthermore, people who defend conceptual analysis can admit that irrelevant context factors do have an impact on our intuitions regarding the classification of counterfactual cases. However, they can reply to this that conceptual analysis targets the public 1-intension of philosophically interesting terms and not the individual private intension of a specific speaker. So while our intuitions might play an important role with regard to the individual private 1-intension, they are not sufficient for making explicit the public 1-intension. What is needed in addition is that philosophers negotiate about how to use a linguistic expression correctly, compare their individual intuitions with the ones others have and come up with an analysis that everyone can agree to. This exercise is of course less prone to error than the process where a speaker applies or withholds an expression without any serious reflections. From this, I conclude that even though experimental philosophers have addressed an important issue, they have nonetheless failed to make a point against the possibility of conceptual analysis.
184
Dominik Aeschbacher
5 Bad context: Experts have different folk theories The distinction between two kinds of 1-intension depends on the assumption that there is such a thing as a public 1-intension. Further above I have argued that one can make partially sense of this claim by adopting Putnam’s metaphor of linguistic labor, according to which the meaning of a linguistic expression behaves parasitic on the meaning that is given to it by the experts of a community. However, it is not obvious that the experts of a community share the same public 1-intension with one and the same linguistic expression. Timothy Williamson (2007) has pointed to this problem, though he introduced it for different reason. His criticism can, however, be applied to the problem at hand. In a series of counterexamples to analyticity, Williamson has argued that even experts do not associate the same conditions of application with a linguistic expression that belongs to their domain of expertise. As a consequence, Williamson doubts that one can make sense of the notion of analyticity. In order to give an overview of Williamson’s criticism, let’s turn to the notion of analyticity. Following Paul Boghossian (1997), analyticity is understood as an epistemic notion7. Let me define this epistemic understanding of analyticity as follows: Analyticity: A sentence e that expresses the proposition that P is analytic if and only if a subject S can know that P merely by grasping the meaning of e. Since I have introduced the two-dimensional framework, I take the relevant kind of meaning to be the 1-intension. Williamson has pointed out that if one interprets analyticity as an epistemic notion, one is forced to defend what he calls understanding-assent links. By this, he means that anyone who understands the sentence must also assent to it. This relationship is implied by the concept of epistemic analyticity because in order to know a proposition, one has to assent to it. According to Williamson (2007: 86), a proponent of analyticity has to accept the following understanding-assent link: UA1: Necessarily, whoever understands the sentence “Every vixen is a vixen” assents to it. However, Williamson has found different counterexamples to UA1 and these have at least two interesting features. First, the counterexamples all suppose that the
7 Epistemic notions of analyticity are opposed to metaphysical notions. Metaphysical notions of analyticity explain the phenomenon of analyticity by appealing to the metaphysical notion of meaning. Analytic sentences are taken to be true in virtue of their meaning. Epistemic notions of analyticity deny that analytic sentences are true in virtue of their meaning. Instead they try to characterize them through epistemic properties. For more information on this issue, see Boghossian (1997).
Context, two-dimensional semantics and conceptual analysis
185
speakers are competent and understand all the expressions used in UA1; and still the speakers do not assent to UA1. Second, the protagonists in Williamson’s examples are experts. One counterexample Williamson introduces reads as follows: Stephen is a philosopher who defends the philosophical thesis that borderline cases for vague terms such as “being bald” generate truth-value gaps; hence beside the value of being true and being false it is possible that a sentence is neither true nor false and must be attributed a third value “undefined”. For this reason, he endorses a three-valued semantics using Kleene’s three valued “strong tables”. According to this three valued system, an implication of the form “If p then p” is no longer valid. Suppose Stephen believes that some evolutionary ancestors of female foxes were borderline cases. Regarding these animals, the sentence “x is a vixen” is neither true nor false, but undefined. Since Stephen endorses Kleene’s three valued strong tables, the conditional “If x is a vixen then x is a vixen” is not true and one can conclude that Stephen does not assent to “Every vixen is a vixen”, even though he understands the sentence perfectly. One would not want to deny that Stephen is a competent speaker with regard to the words used in “Every vixen is a vixen”. He fully understands the word “vixen”, for he applies it correctly to all vixens and he can engage in every day conversations about female foxes. Moreover, Stephen is not incompetent with the word “every” and he not only engages in every day conversations by using the word “every” correctly, but he can also defend his view on truth-value gaps coherently. He defends a specific semantic theory about “every” that is possibly false, but doing so does not make him incompetent. Many philosophers and linguists defend semantic theories that are false, but that does not make them incompetent. Counterexamples like the one mentioned above provide lethal arguments against the assumption underlying the traditional picture for two reasons. First, it undermines the epistemic notion of analyticity that I characterized above. This is unfortunate for the traditionalist picture because conceptual analysis aims at delivering sentences that are analytic. Secondly, it shows that even experts seem to associate different conditions of application with a linguistic expression and still count as competent speakers within a specific linguistic community. This very fact speaks against adopting Putnam’s metaphor of linguistic behavior according to which the meanings of ordinary people behave parasitic on the agreed usage of the experts. However, I think that we must be careful what kind of conclusions we should draw from the counterexample above. Only because we defend a specific analysis of a linguistic expression, this does not mean the analysis is identical with our private 1-intension that determines our linguistic behavior in every day speech. For example, philosophers have believed for a long time that knowledge is justified true belief. But Edmund Gettier’s (1963) famous examination of the notion of
186
Dominik Aeschbacher
knowledge showed that this analysis is not accurate. Nonetheless, even though many philosophers believed that knowledge is equivalent to justified true belief, this analysis of knowledge did not determine their everyday use of the term. In view of this fact it seems necessary to introduce a further distinction. It is necessary to distinguish between the private 1-intension that can be equivalent with the public 1-intension, and what I would call a meta-theory regarding the public 1-intension. The meta-theory describes those believes that result from an adequate analysis of the public 1-intension. The meta-theory can be equivalent to our public or private 1-intension, without being necessarily equivalent. The justified-true-belief analysis of knowledge is a typical meta-theory of a public 1-intension of knowledge. It overlaps to a great extent with everybody’s public and private 1-intension without being identical with either of them because metatheories are often incomplete. All competent speakers are able to provide a meta-theory for a wide variety of linguistic expressions. For example, for the expression “bachelor” most people will provide the meta-theory that bachelors are unmarried men. However, this characterization of the meaning of the word “bachelor” is incomplete. Monks or the pope are unmarried men and yet they are not bachelors. This small example shows that it is difficult to provide a complete conceptual analysis, but easy to spell out some necessary conditions. In the former section, I pointed out that divergence about meaning can be explained by differences in the private and public 1-intension. But the distinction between 1-intensions and meta-theories provides a further explanation for this kind of divergence. Given the meta-theoretical level, it is possible that the private 1-intension of a speaker converges with the public 1-intension of the linguistic community a speaker is part of, while her meta-theory differs significantly from the meta-theories of other people. And yet, as long as speakers are capable of engaging in conceptual analysis and as long as they can agree on some core assumptions, they can count as competent speakers. Above, I have argued that language competence can be a matter of degree in so far as that a private 1-intension can diverge from the public 1-intension. However, language competence can be a matter of degree in yet another respect: a speaker whose meta-theory is closer to the public 1-intension can count as more competent. As a consequence, the terms “language competence” and “understanding” are somehow ambiguous between the two following readings: Understanding1: A speaker S is competent with a linguistic expression e if and only if a) her private 1-intension Pprivate associated with e converges with the public 1-intension Ppublic and b) S is able to provide an approximately correct meta-theory M of the public 1-intension Ppublic.
Context, two-dimensional semantics and conceptual analysis
187
The second kind of understanding can be characterized as a deep understanding that amounts to some kind of expert knowledge. This means that one has mastered Understanding1 of a specific expression. When engaging, for example, in conceptual analysis during a certain amount of time this will lead to a deeper understanding of the expression in question, especially in the area of philosophy. Thus the second way of understanding can be characterized as follows: Understanding2: A speaker S has a deep understanding of a linguistic expression e if and only if a) S is competent as spelled out in Understanding1, and b) S can provide a meta-theory M of e in such a way that M and e express the same public 1-intension (which would be identical to Ppublic). It is worth highlighting that Understanding2 can only be achieved when one inquires carefully in the sense of rational reflection undertaken under ideal conditions. For this reason, I will also speak of “understanding in a deep sense” when referring to understanding2. It is likely that most people do not have such a deep understanding of the expressions they use because they have not thought about the notions carefully enough. As my thesis regarding understanding as a matter of degree suggests, one can be more or less close to having an understanding of a word in this deeper sense. Conceptual analysis provides a deeper understanding of our words. However, it might be that some expressions have a 1-intension that is too complicated and so it is not possible to achieve a full understanding of them. The discipline of philosophy just shows how difficult it is to give a complete analysis of a term, and in some cases it might be even impossible. For this reason, conceptual analysis is not a trivial matter. This account of understanding has an interesting consequence with regard to the question what it means to assent to analytic truths. If understanding is such a demanding task then we should not assent blindly to analytic truths. Rather we have to think carefully about them and engage in rational discourse in order to get the most accurate meta-theories of our 1-intensions. So assenting to an analytic truth is not similar to a gut reaction where we blindly follow our intuitions or emotions. Rather it is based on rational reflection. A further consequence of this account is that it can show why analytic philosophers have a special expertise when it comes to philosophical notions such as “knowledge” or “causality”. Their meta-theories of the respective 1-intensions are more detailed than those of the ordinary people. Given the fact that they are experts in conceptual analysis, they know that some analyses are mistaken. For example, ordinary people might blindly assent to a justified-true-belief analysis of knowledge, while experts know that this analysis is mistaken.
188
Dominik Aeschbacher
Now that these distinctions have been made clear, it is possible to handle Williamson’s counterexample. I take analyticity and conceptual analysis to be a difficult endeavor that aims at understanding in the deep sense. If a speaker has achieved this kind of understanding, she is able to assent to analytic truths. It is obvious that the subjects in Williamson’s example do not have this kind of understanding. They hold controversial meta-theories about expressions that are not easy to analyze. So only because experts do not behave carefully and do sometimes deny analytic truths, this does not imply that conceptual analysis is in principle impossible. It still can be that everybody shares the same 1-intension with a linguistic expression, but that different people disagree with respect to its right analysis. Analysis is not easy, nor is assenting to analytic truths. Philosophy has always been a difficult task and it would be surprising if its methodology such as conceptual analysis is easy to realize.
6 Bad context: Vagueness At the end of this paper I want to address a last problem that the traditional picture must address, namely the phenomenon of vagueness. Different speakers seem to get to different results when applying a linguistic expression to borderline cases. For example, there are borderline cases between shades of orange and red. One and the same shade might be judged on one occasion to be red and on another occasion to be orange. If a speaker proceeds in a similar manner when considering philosophical thought experiments, this might be explained by the vagueness of the philosophical notion in question. Unfortunately, if conceptual analysis is supposed to deliver rigid characterizations of the conditions of application of philosophically interesting expressions, it does not allow this kind of context-sensitivity. The reason is that there would always be an unsolvable disagreement about the right analysis of a linguistic expression. I think the right answer to these worries is that one should not expect conceptual analysis to provide just a short list of necessary and sufficient conditions. Rather it has to deliver a complicated description of a folk theory that might also include the property of vagueness. When we find out that ordinary people have not made up their mind about philosophical notions such as “knowledge” or “causality” then this belongs to the very results that conceptual analysis must deliver. In such cases, philosophers must deliver more precise definitions. For example, it can be imagined that our folk theory regarding the phenomenon of free will is completely confused and vague. In that case it is the task of philosophers to come
Context, two-dimensional semantics and conceptual analysis
189
up with similar but more elaborated definition. Since this goes clearly beyond conceptual analysis and can be considered as some kind of engineering, I call this processes conceptual engineering. However, it is important that philosophers and scientists make explicit when they engage in this. Doing conceptual engineering and sell it as a conceptual analysis could lead not only to confusions, but also to unnecessary pseudo disagreements.
7 Conclusion According to what I call the traditional picture, conceptual analysis takes a central role in philosophical inquiry. In formulating two-dimensional semantics, Chalmers and Jackson have provided a sophisticated framework that can a) explain the a priori nature of conceptual analysis and b) incorporate both, descriptivism and semantic externalism in one coherent semantic framework. In this paper I have discussed three types of context-influence that might pose a problem for this framework. First I discussed the problem that subjects can be competent when using a linguistic expression without sharing the same tacit conditions of application. I distinguished between a public and a private intension and argued that this is not a real problem for conceptual analysis. Secondly, I considered the problem that background theories might affect whether a sentence is analytic or not. I have rejected the alleged counterexamples to analyticity by Williamson by arguing that the subjects in these thought experiments lack a deep understanding of the notions in question. Finally, I have considered the problem of vagueness and argued that vagueness is not a problem for conceptual analysis, but rather a feature that can be discovered through conceptual analysis. I conclude that there is no problem from context for conceptual analysis. Acknowledgments: I would like to thank Silvan Imhof and Fabrice Teroni for commenting on an early draft of this paper. I am very grateful to Sarah-Jane Conrad for her advice and extensive feedback.
References Alexander, Joshua & Jonathan Weinberg. 2007. Analytic philosophy and experimental philosophy. Philosophy Compass 2(1). 56–80. Boghossian, Paul. 1997. Analyticity. In Bob Hale & Crispin Wright (eds.), A companion to the philosophy of language, 331–368. Oxford: Blackwell Publishers.
190
Dominik Aeschbacher
Braddon-Mitchell, David & Robert Nola. (2009). Introducing the Canberra plan. In David Braddon‐Mitchell & Robert Nola (eds.), Conceptual analysis and philosophical naturalism, 1–20. Cambridge, MA: MIT Press. Burge, Taylor. 1979. Individualism and the mental. Midwest Studies in Philosophy 4(1). 73–121. Carnap, Rudolf. 1947. Meaning and necessity. Chicago: University of Chicago Press. Chalmers, David. 2006. The foundations of two‐dimensional semantics. In Manuel García-Carpintero and Josep Macià (eds.), Two-dimensional semantics, 55–140. Oxford: Oxford University Press. Chalmers, David and Frank Jackson. 2001. Conceptual analysis and reductive explanation. The Philosophical Review 110(39). 315– 360. Frege, Gottlob. 1986. Über Sinn und Bedeutung. In Günther Patzig (eds.), Funktion, Begriff, Bedeutung, 40–65. Göttingen: Vandenhoeck & Ruprecht. Gettier, Edmund. 1963. Is justified true belief knowledge? Analysis 23(6). 121–123. Grice, Paul. 1989. Postwar Oxford philosophy. In Paul Grice, Studies in the way of words, 171–180. Cambridge, MA: Harvard University Press. Jackson, Frank. 1998. From metaphysics to ethics: A defence of conceptual analysis. Oxford: Oxford University Press. Kaplan, David. 1989. Demonstratives: An essay on the semantics, logic, metaphysics and epistemology of demonstratives and other indexicals. In Joseph Almong, John Perry & Howard Wettstein (eds.), Themes from Kaplan, 481–563. Oxford: Oxford University Press. Kripke, Saul. 1980. Naming and necessity. Malden, MA: Blackwell Publishing. Moore, Edward G. 1942. A reply to my critics. In Paul Arthur Schlipp (Ed.), The philosophy of G. E. Moore, 535–677. Evanston, IL: Northwestern University Press. Putnam, Hilary. 1975. The meaning of “meaning”. Mind, Language and Reality 2. 215–271. Cambridge: Cambridge University Press. Schroeter, Laura. 2003. Gruesome diagonals. Philosophers’ Imprint 3(3). 1–23. Soames, Scott. 2005. Reference and description. Princeton: Princeton University Press. Williamson, Timothy. 2007. The philosophy of philosophy. Malden, MA: Blackwell Publishing.
Dan Zeman
The use of the Binding Argument in the debate about location Abstract: In this paper I inquire into the methodological status of one of the arguments that have figured prominently in contemporary debates about the semantics of a variety of expressions, the so-called “Binding Argument”. My inquiry is limited to the case of meteorological sentences like “It is raining”, but my conclusion can be extended to other types of sentences as well. Following Jason Stanley, I distinguish between three interpretations of the argument. My focus is on the third, weakest interpretation, according to which postulating variables for locations in the logical form of meteorological sentences is the best available explanation of the binding phenomena that such sentences give rise to. My aim is to show that even in this weak interpretation, the argument cannot be reasonably taken to hold. I accomplish this by showcasing several alternative ways to account for the binding data that have not been, as of yet, ruled out as flawed. In recent debates about the semantics of various types of expressions, one often encounters a particular argument purporting to show that certain sentences have a variable of a certain kind in their logical form.1 This is the “Binding A rgument”: it proceeds from the observation that complex sentences containing a quantificational phrase operating on simple sentences (the target sentences) have a bound reading, to the conclusion (obtained by employing prima facie plausible claims about the connection between semantics and syntax) that the target sentences have a certain variable in their logical form. The argument has been used against more than one view on the market and has been applied to a variety of expressions. Due to its widespread acceptance, from a methodological point of view the Binding Argument seemed to be a safe bet. In this paper I will be concerned with the use of the Binding Argument in the debate over the semantics of meteorological sentences like “It is raining”. To be more precise, I will be concerned with the methodological efficacy of the argument under one of its interpretations: namely, that according to which the postulation of certain variables in the logical form of the target sentences is the best
1 “Logical form” will be understood in this paper as referring to the level of syntactic representation free of ambiguities that constitutes the input to semantics. DOI 10.1515/9781501504327-010
192
Dan Zeman
explanation of the binding data proponents of the argument rely on.2 According to such an interpretation, the postulation of the relevant variables is the best explanation if i) explains the binding data; ii) there are no other accounts of the data that are better situated to do so. My aim is to show that, contrary to what its proponents claim, the Binding Argument under this interpretation is ineffective, for the simple reason that there is a number of alternative ways to account for the binding data that don’t postulate variables in the logical form of the target sentences and which have not been ruled out as faulty. I will thus present, in quite significant detail, several such alternative ways. The claim is not that such ways are better situated than the variables-postulation strategy to explain the binding data. Rather, it is that without a thorough examination of those ways and a careful comparison between them and the explanation consisting in postulation of variables in the logical form of the target sentences the Binding Argument (in the intended interpretation) cannot be reasonably taken to hold. The paper is structured as follows. I start with introducing the argument by means of an example taken from the debate between two mainstream views on the semantics of meteorological sentences and then offer a more general characterization of it as an argument-schema. In section 2 I present three interpretations of the argument and make clear which one I’m concerned with in this paper. In the same section I show that the third, weaker interpretation of the argument is consistent with criticisms leveled against the argument under the first two interpretations. Section 3 is then dedicated to the presentation of five alternative ways to account for the bound readings of complex sentences containing sentences like “It is raining” without postulating variables for locations in their logical form. In the last section I restate the main claims of the paper and their implications for the semantics of meteorological sentences.
1 The Binding Argument: an example and an argument‐schema To get a feel of how the Binding Argument is supposed to work, let’s start with a concrete example of how it has been used in one of the most vivid
2 This interpretation of the Binding Argument has been offered by one of the proponents of the argument, Stanley (2007, chapter 1). Other authors dealing with the argument take it to have such an interpretation (see, for example, Sennet (2008: 143)). Regarding the Binding Argument itself, its origin can be traced back to Chomsky (1957).
The use of the Binding Argument in the debate about location
193
debates in recent philosophy of language. When it comes to the semantics of meteorological sentences like ‘It is raining’, the two major views found in the literature can be subsumed under the labels truth-conditional semantics and truth-conditional pragmatics. The bone of contention between these two general views is whether such sentences have variables for locations in their logical form or not. Without going into much detail, truth-conditional semantics is the view (or, rather, a cluster of views) characterized by the claim that “any contextual effect on truth-conditions that is not traceable to an indexical, pronoun, or demonstrative (…) must be traceable to a structural position occupied by a variable” (Stanley 2007: 39). In contrast, truth-conditional pragmatics is the view (or, rather, a cluster of views) according to which the provision of elements in the content of utterances is done via optional pragmatic processes like free enrichment rather than by mandatory ones such as saturating a variable in the logical form (Recanati 2002, 2004 being the most prominent defender of the view). Now, the Binding Argument has been used to argue against t ruth-conditional pragmatics by adepts of a truth-conditional semantic view about meteorological sentences. Thus, Stanley (2007, chapter 1) notes that sentence (1) Every time John lights a cigarette, it rains has a reading according to which every time John lights a cigarette at time t, it rains at time t at the location in which John lights a cigarette. The location of rain in (1) is thus bound by the quantifier phrase “every time John lights a cigarette”. With this reading of (1) in mind, and using the regimentation of a similar argument given by Recanati, the instance of the Binding Argument against truthconditional pragmatics about locations could be put as follows: 1L. According to truth-conditional pragmatics, there is no variable for the location of rain in the logical form of the sentence ‘It is raining’. 2L. In (1), binding occurs: the location of rain varies with the values introduced by the quantifier ‘every time John lights a cigarette’. 3L. There is no binding without a bindable variable in the logical form. 4L. Therefore, there is a variable for the location of rain in the logical form of the sentence ‘It is raining’. 5L. Therefore, truth-conditional pragmatics is mistaken. (Adapted from Recanati 2002: 328–9). The same reasoning can be applied to argue against different views about the semantics of “It is raining”. Relativism about such sentences is a case in point. Relativism is the view (or, rather a cluster of views) according to which the location is contributed by the circumstances in which utterances of “It is raining”
194
Dan Zeman
are evaluated for truth.3 Assuming that a location cannot be provided both by a variable in the logical form of a sentence and by the circumstances with respect to which utterances of that sentence are evaluated for truth4, the Binding Argument can be used to refute relativism by showing that a location has to be provided by the former. On the other hand, there are many other types of sentences besides meteorological ones in connection to which various instances of the Binding Argument have been employed. Thus, drawing on the work of Barbara Partee and Kai von Fintel, Stanley (2007, chapters 1 and 6, for example) uses instances of the Binding Argument to argue against truth-conditional pragmatic views about a wide range of expressions, including gradable adjectives, quantifier phrases, relational expressions such as ‘local’, ‘enemy’, etc. Schaffer (2011) uses instances of the argument to argue both against relativism about predicates of personal taste and epistemic modals. This shows the argument’s versatility: it can be used to argue against several views about the semantics of various types of sentences. This, in turn, might explain the argument’s recent popularity in semantics. I mentioned above that my focus in this paper is on meteorological sentences, but the replies to the Binding Argument I will investigate below can easily be extended to other types of sentences. Because of this, it would be useful to have a more general rendering of the argument, one that abstracts away from particular instances found in the literature. Building on the regimentation provided by Recanati, the Binding Argument can be cast in the form of the following argument-schema:5 (1) According to position P, sentence s doesn’t have a variable for parameter π in its logical form. (2) Sentences of the form Q(s) have a bound reading where parameter π is bound by the quantifier phrase Q. (3) There is no binding without a bindable variable in the logical form. (4) Therefore, sentence s has a variable for parameter π in its logical form. (5) Therefore, position P is mistaken.
3 Relativism about meteorological sentences is not a very popular view, although see Lewis (1980) for an early relativist (in the sense used here) position. Kaplan (1989) also toyed with the idea of introducing locations in the circumstances of evaluation. For a contemporary defender of the view, see Recanati (2007). 4 A principle Recanati (2007: 34) dubs “Distribution”. For a discussion of this principle in the context of employing the Binding Argument against relativism, see Zeman (2015). 5 A similar schema, with roughly the same premises, can be extracted from Sennet’s (2008) discussion of the argument.
The use of the Binding Argument in the debate about location
195
One problem with this schema, and thus one that transfers to all its instances, is that, as it stands, it is invalid: the step from premises 1–3 to conclusion 4 is not mandated. Strictly speaking, what follows from the first three premises is the intermediary conclusion (IC): (IC) Therefore, sentence Q(s) has a variable for parameter π in its logical form. In order to get conclusion 4, we need a bridging principle relating sentences s and Q(s) in such a way as to make sure that the existence of a variable in the logical form of Q(s) guarantees the existence of a variable of the same type in the logical form of s. Recanati (2002) provides such a principle. Discussing the instance of the argument presented above, Recanati notes that it is missing an additional premise and claims that ignorance of that fact has lead Stanley to commit “the binding fallacy”. He claims that the following supplementary premise (SUP) is what the proponents of the relevant instance of the argument need in order for it to go through: (SUP) In (1), the sentence on which the quantifier ‘every time John lights a cigarette’ operates is the very sentence ‘It is raining’ which can also be uttered in isolation. (Adapted from Recanati 2002: 329) Recanati speaks about the same sentence here, but what he means is really a representation of the sentence at a deeper level than the linguistic surface. For my purposes here, I will take that deeper level to be the sentence’s logical form. Thus, extrapolating from Recanati’s suggestion, we arrive at the following bridging principle (BP): (BP) The logical form of sentence s is the same in Q(s) as when s appears in isolation.6 Using this principle and the intermediary conclusion above, we finally get the following valid argument-schema: (1) According to position P, sentence s doesn’t have a variable for parameter π in its logical form. (2) Sentences of the form Q(s) have a bound reading where parameter π is bound by the quantifier phrase Q. (3) There is no binding without a bindable variable in the logical form.
6 With a similar purpose in mind, Sennet offers what he calls “The Semantic Innocence Assumption” (2008: 143). The differences between (BP) and Sennet’s proposal are minimal.
196
Dan Zeman
(IC). Therefore, sentence Q(s) has a variable for parameter π in its logical form (BP). The logical form of sentence s is the same in Q(s) as when s appears in isolation. (4) Therefore, sentence s has a variable for parameter π in its logical form. (5) Therefore, position P is mistaken.
2 Three interpretations of the Binding Argument With the argument in good standing, we can now inquire into its methodological status. There are at least three interpretations of the argument, each bestowing upon it a different methodological strength. The three interpretations alluded to have been provided by Stanley himself, when answering various criticisms of the Binding Argument. In what follows I will briefly discuss each of them, as well as the main objection pertaining to each interpretation. According to the first interpretation, the argument “establishes the existence of covert structure, on pain of ungrammaticality due to vacuous binding” (Stanley 2007: 214). On the second, “a bound reading of a sentence is evidence for syntactic structure, since bound readings are the semantic effect of a syntactic process (…). [C]ertain kinds of semantic phenomena (for example, bound readings (…)) have ultimately a syntactic explanation.” (Stanley 2007: 214). Finally, “[o]n the third (and weakest) interpretation of the binding argument, it is an inference to the best explanation. By postulating a covert variable, one can account for the bound reading, and there is no other satisfactory way to account for it” (Stanley 2007: 214, my emphasis). While in its first and second interpretations the argument leads more or less directly to the postulation of variables (the bound readings establish and, respectively, are evidence for the presence of variables), this does not hold with its third. Under this interpretation the phenomena by themselves don’t lead directly to the postulation of variables; this result is rather achieved by accounting for the data and by showing that there are no alternative, equally plausible, explanations of it. In this sense the third interpretation of the argument is weaker than the first two. As mentioned from the outset, it is this third way of interpreting the Binding Argument that I will focus on in what follows. My aim is to show that, despite Stanley’s (2007, chapter 6) suggestion that for certain sentences the postulation of variables in logical form is the best explanation for bounds readings, this is not the case with meteorological sentences like ‘It is raining’. The reason the Binding Argument in its third interpretation doesn’t succeed in forcing one to postulate variables for locations in the logical form of such sentences is that there are many alternative
The use of the Binding Argument in the debate about location
197
ways to account for the bound readings. Some of those alternative ways have been criticized; others have not even been addressed. However, the point is that without showing that all the alternative ways of handling the data are implausible or less plausible than the postulation of variables for locations in logical form, the Binding Argument in the interpretation considered here cannot be used to that effect. Before moving to the detailed overview of some of the alternative ways of accounting for the data, let me first mention two common objections raised against the argument in its stronger (first and second) interpretations that will also help situate the present paper with respect to other criticisms of the Binding Argument. Thus, one early objection to the Binding Argument has been that it overgenerates massively. Perhaps the most striking illustration of this point is that offered by Cappelen and Lepore (2002), who, starting from the bound reading of the sentence (2) Everywhere I go, 2 + 2 = 4 and following the logic of the Binding Argument, arrive at the absurd conclusion that we would need to postulate a variable for locations in the logical form of sentences like “2 + 2 = 4”. More generally, given that virtually any sentence could be embedded in a more complex sentence that has a bound reading, we would end up postulating many more variables in the syntax than seems plausible. Now, Stanley (2007, chapter 6) has initially replied to this objection by distinguishing between the three interpretations of the Binding Argument mentioned above and by relegating Cappelen and Lepore’s worries to the first of these. I’m not sure that the point made by Cappelen and Lepore had anything essentially to do with ungrammaticality and thus whether Stanley’s move avoids their objection. However, it is also not clear that their objection is fatal. As both Stanley (2007, chapter 6) and Cohen and Rickless (2007) show, there are ways for the proponent of the Binding Argument to avoid the charge of overgeneration. Cohen and Rickless, for example, note a crucial disanalogy between Cappelen and Lepore’s example and (1), claiming that the crucial feature underlying binding phenomena (namely, that the interpretation of that embedded sentence should systematically be affected by the quantifier) is present in the latter but absent in the former. Be that as it may, the point that I want to stress here is that even if Cappelen and Lepore’s objection shows that the Binding Argument in its first (or, if Stanley’s classification of their objection is wrong, its second) interpretation is flawed, this does not mean that the argument in its third interpretation is flawed too. In other words, it is still possible that, despite overgeneration problems, the postulation of variables for locations in the logical form of sentences is the best available explanation for the phenomena. The main criticism of the Binding Argument in its second, strongest interpretation is that the prima facie plausible principle the proponent of the argument
198
Dan Zeman
uses to move from the existence of bound readings to the postulation of variables in the logical form of the target sentences is, in fact, not well supported. What I have in mind here is premise 3 in argument-schema provided above, “there is no binding without a bindable variable in the logical form”. This is a claim about the connection between semantic binding (which is what the intuitive bound readings put forward by the proponents of the Binding Argument are supposed to illustrate) and syntactic binding (which is a highly constrained, well regimented syntactic phenomenon).7 Many authors have denied premise 3 or similar claims (e.g., Carston 2002; Breheny 2003; Neale 2007; Collins 2007 – as well as most proponents of the answers to the Binding Argument that will be presented in the next section). Here, however, I will focus on one of the more recent and most substantial denials in the literature, the one found in Pupa and Troseth (2011). They object to what they call “coincidence”: the claim that there is a perfect overlap between semantic binding and syntactic binding.8 Stressing the point that the proponents of the Binding Argument incur a very heavy theoretical burden of a syntactic nature, they show that there are clear cases in which there is semantic binding without there being syntactic binding, for reasons having to do with the impossibility of moving for certain expressions or with lack of suitable expressions that should be c-commanded.9 Now, Pupa and Troseth’s criticism has focused on the case of domain restriction, but the same criticism might apply to the case of meteorological sentences as well. Indeed, they claim that “there’s little hope for [variables-postulating approaches] to similar phenomena such as the (…) spatial variability of weather verbs” (Pupa and Troseth 2011: 207). To be sure, no such evidence has been given. But, as before, what I want to point out here is that even if such evidence were to be given, this would still be compatible with the third, weaker interpretation of the Binding Argument. In other words, even if the binding data don’t provide conclusive evidence for the postulation of variables for locations in the logical form of meteorological sentences, it is still possible that such a postulation is the best available explanation for the phenomena. What matters according to the third interpretation of the argument is whether there are better ways to explain the phenomena than the variable-postulating approach. Pupa and Troseth have
7 To get the required reading, ‘binding’ in premise 3 has to be understood as semantic binding, while ‘bindable’ as syntactic binding. 8 Coincidence is assumed by Stanley (2007, chapter 1). 9 C-command is a crucial notion underlying the study of syntactic binding. Roughly, node N dominates node M if N is above M in the syntactic tree and there is a path from N to M that moves only downwards.
The use of the Binding Argument in the debate about location
199
shown that there are reasons to be dissatisfied with the argument in its second interpretation, but this does not mean that, in its third interpretation, the argument is useless. What does show its uselessness, however, is the following section.
3 Alternative accounts of the binding data Several replies have been explicitly given, and others could easily be extracted from the literature, to various instances of the Binding Argument – including the one I have focused on here. Perhaps the most radical answer consists in rejecting the argument by eschewing variables altogether. In variable-free frameworks such as that developed in Jacobson (1999), for example, the Binding Argument doesn’t make much sense. Challenging as they are, however, I will leave such frameworks aside in this paper. A more common answer is to accept variables, but to show that the data to be accounted for (such as the bound readings of sentences like (1)) can be handled in a framework that doesn’t require the postulation of variables (of the relevant kind) in the logical form of the target sentences. From the five ways to account for the data I have chosen to survey below, the first four rest on the idea of accounting for the bound readings of sentences like (1) by replacing quantification over locations with quantification over more encompassing entities, such as events, situations, contexts or indices. There are two ways to implement this idea. The first is to replace the variables for locations in the object language with variables for more encompassing entities in the object language itself. This way of implementing the basic idea is close to the spirit of Stanley’s proposal, if not to its letter, in holding on to the claim that binding has to be accounted for by the postulation of variables in the logical form – albeit not of the same kind as the ones the conclusion of the Binding Argument would compel us to postulate.10 To illustrate this way to account for the data, I will present a view that uses event variables instead of location variables, the “Event Analysis” proposed by Cappelen and Hawthorne (2007), and a similar view that could be
10 This answer seems to be compatible with the existence of instances of the Binding Argument that argue for the postulation of variables for those more encompassing entities themselves. However, I take it that the dialectical power of the Binding Argument resides in its capacity to argue against views that deny that we need variables for very specific parameters (such as locations) in the logical form of the target sentences. In any case, the existence of alternatives in which no variable of any kind needs to be postulated at the object level is enough to support my point.
200
Dan Zeman
extracted from Elbourne’s (2005, 2008) work that uses situation variables. For each view presented, I will show how exactly the derivation of the conclusion of the Binding Argument is blocked. The second way to implement the idea behind the type of approach mentioned is to replace the variables for locations in the object language with variables for the more encompassing entities in the metalanguage. Here, the departure from the proponents of the Binding Argument is clearer. To illustrate, I will present two versions of this way to account for the data: Pagin’s (2005) proposal to use quantification over contexts and Lasersohn’s (2008) quantification over indices view. As before, for each view presented, I will show how exactly the derivation of the conclusion of the Binding Argument is blocked. The fifth proposal that I will present is more peculiar. It differs from the others in granting all the premises of the argument, while at the same time blocking its conclusion. This view has been proposed by Recanati’s (2002, 2004) and consists in appeal to variadic operators, a device he introduced precisely with the aim of responding to the instance of the Binding Argument directed against a truthconditional pragmatic view of meteorological sentences. The five alternative ways to account for the binding data just mentioned are underpinned by different commitments and their proponents are following more or less determinate agendas. (For example, Recanati’s employment of variadic operators is directly connected with his general truth-conditional pragmatic view and he takes the avoidance of the conclusion of the Binding Argument to support that view.) For our purposes here, however, the differences between those approaches can be ignored. What is important is that under the views to be scrutinized the binding data are accounted for and that the instance of the Binding Argument presented in section 1 is blocked.
3.1 Quantifying over events The alternative to postulating variables for locations in the logical form of meteorological sentences that appeals to variables for events has been proposed by Cappelen and Hawthorne (2007). Their view, which they call the “Event A nalysis”, is meant to be an alternative to Stanley’s (2007, chapter 1) treatment of sentences such as (1), and it consists in adopting a Davidsonian event semantics (see for example Davidson (1967)) in combination with the idea that there are domain restrictions attached to certain phrases. In this case, the relevant phrases to attach domain restrictions to are verbs, which following the Davidsonian analysis should be conceived of as predicates of events. In this framework, binding phenomena are understood in terms of domain restrictions of events where the restrictor on
The use of the Binding Argument in the debate about location
201
events is bound by a higher quantifier. In the case of (1), the relevant type of higher quantifier is a temporal quantifier. Let me illustrate the view with an example given by the authors themselves. The bound reading of sentence (3) Every time Sam goes to the park, Nina is walking her dog, according to which every time Sam goes to the park Nina is walking her dog in the park where Sam goes to, is represented in the Event Analysis as (4) For all times t, if there is an event e1 that is a going to the park by Sam at t, there is an event e2 that is a walkingf(t) of a dog by Nina, where t is the time of event e1 (quantified over), while f(t) is a function from times t to the set of events that take place at t in the park where Sam goes. Walkingf(t) is thus the set of events arrived at by intersecting the set of events that take place at t in the park where Sam goes with the set of events of walking – that is, the set of events of walking that take place at t in the park where Sam goes. The problematic sentence (1) is represented in the event analysis as (5) For every time t, if there is an event e1 that is a lightning of a cigarette by John at t, then there is an event e2 that is a rainingf(t), where t is the time of the event e1 (quantified over), while f(t) is a function from times t to the set of events that take place at t in the location where John lights a cigarette. Rainingf(t) is thus the set of events obtained by intersecting the set of events that take place at t in the location where John lights a cigarette with the set of events of raining – that is, the set of events of raining that take place at t in the location where John lights a cigarette. How does this treatment help with blocking the Binding Argument? If (5) is the representation of (1) in the Event Analysis, then (IC) doesn’t hold – there is no variable for locations in (5). But if it doesn’t hold, then it means that either (IC) doesn’t follow from the premises, and thus the argument is not valid, or one of the premises needed for its derivation is rejected. So, which is the case? Actually, both hold. Premise 3, the principle that there is no binding without a bindable variable in the logical form, is ambiguous between a reading according to which there is no binding without a bindable variable in the logical form simpliciter, and a reading according to which there is no binding without a bindable variable of the same kind as the bound parameter (here, location) in the logical form. On the first of this readings premise 3 holds under the Event Analysis (since we have a variable for events in (5)), but (IC) doesn’t follow; on the second reading, the premise is rejected (since there is no location variable in (5)), but the location of rain is nevertheless bound. This means that in each situation the intermediary conclusion (IC) cannot be derived, and thus the Binding Argument is blocked.
202
Dan Zeman
3.2 Quantifying over situations A related way to account for the bound reading of (1) is one that replaces the postulation of variables for locations in the logical form of meteorological sentences with the postulation of variables for situations. Although not explicitly proposed for the case of locations, such a view could be extracted from Elbourne’s (2005) treatment of donkey sentences. In contrast to whole worlds, situations are spatially and temporally limited parts of worlds. However, in order to get around well-known objections to the traditional notion of situation11, a more fine-grained notion has been introduced: that of a minimal situation. A minimal situation s such that p is a situation that contains just enough individuals, relations and properties to make p true. Also, we need the part-of relation (symbolized ≤) to which situations are subject to: a situation s is part of a situation s’ if and only if s’ contains all the individuals, properties and relations that s does (and possibly some more). Now, in Elbourne’s framework, which uses minimal situations, sentence (6) If a farmer owns a donkey, he always beats the donkey and the priest beats the donkey too is represented as (7) λs1. for every minimal situation s2 such that s2 ≤ s1 and there is an individual x such that x is a farmer in s2 and there is an individual y such that y is a donkey in s2 and x owns y in s2, there is a situation s3 such that s3 ≤ s1 and s3 is a minimal situation such that s2 ≤ s3 and the unique farmer in s3 beats in s3 the unique donkey in s3 and there is an individual z such that z is a priest in s3 and the unique priest in s3 beats in s3 the unique donkey in s3, where s1 is the speech situation and s2 and s3 are minimal situations consisting in the individuals and relations specified. Elbourne is not concerned with sentences such as (1), but we could use (7) to get an idea of how (1) would be represented in his system. The representation of (1) would be something along the following lines: (8) λs1. for every minimal situation s2 such that s2 ≤ s1 and there is an individual x such that x is John in s2 and there is an individual y such that y is a cigarette in s2 and x smokes y in s2, there is a situation s3 such that s3 ≤ s1 and s3 is a minimal situation such that s2 ≤ s3 and it rains in s3,
11 See Kratzer (2014) for an overview.
The use of the Binding Argument in the debate about location
203
where s1 is the speech situation and s2 and s3 are minimal situations consisting in the individuals and relations specified. This treatment of (1) avoids the conclusion of the Binding Argument in much the same way in which Cappelen and Hawthorne’s analysis does. If (8) is the representation of (1) in the situation framework, then, again, (IC) doesn’t hold – there is no variable for locations in (8). But if it doesn’t hold, then it means that either (IC) doesn’t follow from the premises, and thus the argument is not valid, or one of the premises needed for its derivation is rejected. Again, both are the case. Premise 3 remains ambiguous between a reading according to which there is no binding without a bindable variable in the logical form simpliciter, and a reading according to which there is no binding without a bindable variable of the same kind as the bound parameter (here, location) in the logical form. On the first of these readings, premise 3 holds under the current analysis (since we have a variable for situations in (8)), but (IC) doesn’t follow; on the second reading, the premise is rejected (since there is no location variable in the logical form of (1)), but the location of rain is nevertheless bound. As before, this means that in neither interpretation the intermediary conclusion (IC) can be derived, and thus that the Binding Argument is blocked.
3.3 Quantifying over contexts A different way to account for the bound reading of (1) is to replace variables for locations in the object language with variables over more encompassing entities in the metalanguage. The first of the two views found in the literature is Pagin’s (2005), consisting in an appeal to quantification over contexts. Contexts, for Pagin, will be modeled as “sequences of context elements, or equivalently as assignments of values to a set of context parameters. For any context c, the associated model will contain as elements, to begin with, the speaker of c, if any, the addressee, if any, the (contextually salient) time of c, the location”, and, in general, “anything over and above the sentence (actually or potentially) uttered that is available and may be employed for achieving communicative success” (Pagin 2005: 328). Sentences will be true relative only to contexts, or indices, as in Montagovian semantics. Variables for contexts are variables in the metalanguage, and can be bound by quantifiers in the metalanguage. Such binding in the metalanguage is useful in order to handle quantifiers such as ‘every time’ that bind context elements – in this case, time. Using this idea, a sentence containing such a quantifier could be represented as (9) ‘Every time’^ s is true at c if for every context c, s is true at T(c), where ^ is the concatenation sign, s is a sentence, c the utterance context and T(c) is the time of c. But the original sentence ‘every time’^ s could be context s ensitive,
204
Dan Zeman
and thus will be itself evaluated relative to a context. To capture this situation, ‘every time’ will be construed as quantifying over contexts c’ which differ from c with respect to time but not in other respects. ‘Every time’^ s will thus be represented as (10) ‘Every time’^ s is true at c if for every context c’≈ c/t, s is true at c’, where ‘c’≈ c/t’ is to be read as “context c’ differs from context c at most with respect to time”. Now, when it comes to sentences such as (1), in order to account for the bound reading of that sentence we need first to let location vary together with time in the clause for ‘Every time’^ s. This is accomplished in (11): (11) ‘Every time’^ s is true at c if for every context c’≈ c/t + l, s is true at c’, where ‘c’≈ c/t + l’ is to be read as “context c’ differs from context c at most with respect to time and location”. Assuming that (1) is interpreted as a quantified conditional of the form (12) ‘Every time’^ ‘if John lights a cigarette, then it rains’ and substituting the new rendering of (1) for s in (11), we get (13) ‘Every time’^ ‘if John lights a cigarette, then it rains’ is true at c if for every context c’≈ c/t + l, ‘if John lights a cigarette, then it rains’ is true at c’. Now, if we assume that the following holds: (14) ‘if John lights a cigarette, then it rains’ is true at c’ if (if John lights a cigarette at T(c’) at L(c’), then it rains at T(c’) at L(c’)), where T(c’) is the time of c’ and L(c’) is the location of c’, we finally get (15) as the ultimate representation of (1) in Pagin’s framework: (15) ‘Every time’^ ‘if John lights a cigarette, then it rains’ is true in c if for every context c’≈ c/t + l, if John lights a cigarette at T(c’) at L(c’), then it rains at T(c’) at L(c’) which gives us the desired reading. In Pagin’s account, the conclusion of the Binding Argument is avoided by rejecting premise 3. We start by noting that, although we find a variable for locations in (15) – L(c’) – it is a variable belonging to the metalanguage and not to the object language. This makes (IC) fail. If we ask again why this happens, we are faced with the same options as before. And, again, we find that premise 3 is ambiguous between the two readings mentioned above. This time, however, premise 3 fails in both of its readings because according to Pagin’s account there simply is no variable of the relevant kind in (15). We have thus a more direct blocking of the Binding Argument by straightforwardly rejecting premise 3.
The use of the Binding Argument in the debate about location
205
3.4 Quantifying over indices The second way of accounting for the bound reading of (1) by appeal to quantification over entities in the metalanguage has been developed by Lasersohn (2008). His proposal is to “quantify directly over the [relevant] index, setting and resetting its value in tandem with the variable introduced by a quantifier” (2008: 324). Lasersohn is primarily interested in bound cases of predicates of personal taste, but he applies the same solution to locations as well. The first thing to mention about Lasersohn’s view is that it is a relativist view about locations. Thus, one main tenet of his view is that the circumstances of evaluation (the index, in Lasersohn’s Lewisian jargon) contain a location parameter. For Lasersohn, sentences will be true relative to a model, an assignment, a context and an index – index which comprises, besides possible worlds and other unorthodox parameters12, a location parameter as well, so that a sentence is true in a context if and only if the proposition expressed by the sentence in that context is true relative to the world of the context and the location of the context (or, more precisely, the location of the speaker in the context). Now, Lasersohn’s main claim is that quantifiers are able to bind both variables in the object language and parameters in the index (which are variables in the metalanguage). A certain quantifier, such as ‘every reporter’ (see example (18) below) could bind both a variable for individuals in the object language and a certain parameter in the index – say, the location parameter. For Lasersohn, quantifiers are sentence-abstract forming operators: when they bind variables in the object language, their effect could be described as in (16); when they bind variables in the metalanguage – specifically the location parameter – their effect could be described as in (17): (16) [[λnφ]]M,g,c,w,l = {x ∈ U | [[φ]]M,g[x/n],c,w,l, = 1}
(17) [[πnφ]]M,g,c,w,l = {x ∈ U | [[φ]]M,g[x/n],c,w,L(x) = 1},13
where φ is a sentence, M is a model, g an assignment function, c a context, w a possible world, l a location, g[x/n] is that sequence in which x is the n-th
12 Besides possible worlds, Lasersohn includes a parameter for judges in the index. And, to be more precise, in his view the location parameter is introduced in the index via another parameter he calls “the perspective point”, which is supposed to take care of all shiftable perspectival features of contexts related to the speaker, such as location and time. Since location is the only relevant parameter here, I present the view as if location were the only unorthodox parameter of the index. 13 I have slightly modified Lasersohn’s clauses to align with the simplification made (see previous footnote). For Lasersohn’s original clauses of the two operators, see Lasersohn (2008: 313 and 331, respectively).
206
Dan Zeman
element and which agrees with g in all other positions, L(x) is the location of x, U the set of individuals and λ and π are the two sentence-abstract forming operators. As (16) and (17) show, λ manipulates the assignment function (lambdaabstraction), whereas π manipulates the location parameter of the index, so that the truth of φ is dependent on the values taken by L(x). In order to deal with bound readings of sentences containing locational expressions without having to postulate variables for locations in the object language, Lasersohn allows quantifiers to bind parameters in the index. To use Lasersohn’s own example, sentence (18) Every reporter visited a local bar, has a reading according to which every reporter visited a bar that is local relative to the location of each reporter that visited the bar. This bound reading of (18) is usually accounted for by postulating a variable for locations in the logical form of ‘local’, an object language variable that could be free or bound. But this does not need to happen in Lasersohn’s account. (18) will be represented in Lasersohn’s system as (19) [[every reporter] π1 [[some bar that λ2 [pro2 is-local]] λ3 [pro1 visited pro3]]], where the quantifier ‘every reporter’ binds both a variable for individuals in the object- language (pro1) and the location parameter in the index. But the important point to note is that ‘every reporter’ doesn’t bind any object language variable for locations, since, as can be seen in (19), ‘local’ doesn’t have a variable for locations to be bound. Lasersohn doesn’t tell us how sentence (1) would be represented in his framework, but we get an idea from (19) how this could be done. Here is one attempt at representing (1) using Lasersohn’s apparatus: (20) [[every time] λ1 [John lights a cigarette at pro1]] π2 [it rains at pro1]]. In (20), the quantifier ‘every time’ binds a variable for times in the object language (pro1), whereas the more complex quantifier ‘every time John lights a cigarette’, containing ‘every time’ as its part, binds the location parameter in the index. But the important point to note is, again, that ‘every time John lights a cigarette’ doesn’t bind any object language variable for locations, since, as can be seen in (20), ‘rain’ doesn’t have a variable for locations to be bound. Thus, in Lasersohn’s account, as well as in Pagin’s, the conclusion of the Binding Argument is avoided by rejecting premise 3. We start by noting that there is a variable for location in the clause for the evaluation of (1) given in (17) – that is, L(x); however, it is a variable belonging to the metalanguage and not to the object language. This makes (IC) fail. If we ask again why this happens, we are
The use of the Binding Argument in the debate about location
207
faced with the same options as before. And, again, we come to the conclusion that premise 3 is ambiguous between the two by now familiar readings. This time, however, premise 3 fails in both of its readings because according to Lasersohn’s account there simply is no variable of the relevant kind in (20). Here we also have a more direct blocking of the Binding Argument by straightforwardly rejecting premise 3.
3.5 The variadic operators approach Finally, another alternative way to account for the bound readings of sentences like (1) without postulating variables for locations in the target sentences comes from Recanati, the chief opponent to the Binding Argument for locations. The main feature of Recanati’s view is his appeal to “variadic functions” – functions that take as input a predicate with a certain number of arguments and yield a new, related predicate with one more (or less) arguments. In the case of the “expansive” variadic function (the one that creates a new predicate with an additional argument), the new argument stands for what Recanati calls a “circumstance”: “a time, a location, a manner, or what not” (Recanati 2002: 319). Variadic functions are represented by variadic operators. The effect of the expansive variadic operator on a predicate can be captured by the following equivalence: (21) V (λx1 … λxn. φ (x1 ... xn)) ≡ λx1 … λxn. λy. φ*(x1 ... xn, y), where V is the expansive variadic operator, λx1 … λxn. φ (x1 ... xn) is the input predicate with n arguments (here represented as unsaturated), λx1 … λxn. λy. φ*(x1 ... xn, y) is the new predicate with n + 1 arguments and y is the newly introduced variable. Now, following other authors with similar proposals, Recanati makes the claim that several natural language expressions can be handled by using the variadic operator just defined.14 Here is what Recanati says about the category of modifiers: “modifiers are syntactically optional. They make a predicate out of a predicate. If we start with a simple predicate, say ‘rain’, we can make a different predicate out of it by adjoining an adverb such as ‘heavily’ or a prepositional phrase such as ‘in Paris’. (…) Semantically, I suggest that we construe the modifier as contributing a certain sort of [operator] which I call a variadic [operator].” (2002: 319). The contribution of the variadic operator won’t be the only one made by such expressions, however. In fact, they will be treated as having a twofold
14 Such as McConnell-Ginet (1982) and Keenan and Faltz (1985).
208
Dan Zeman
semantic effect: on one hand, they create a new, related predicate that differs from the input predicate by having an additional argument place occupied by a variable of a certain sort (the effect of the variadic operator); on the other hand, they provide a value for that variable. Let us apply this apparatus to the case at hand. With Vx: y capturing the effect of the modifier, subscript x indicating the type of variable the operator introduces and subscript y indicating the value given to that variable, sentence (22) It is raining in Paris will be represented as (23) Vlocation: Paris (rain), which is equivalent with (24) rain_in (Paris). We see that the locational expression ‘in Paris’ is treated as contributing both a variadic operator which transforms the zero-place predicate ‘rain’ into a new predicate with an additional argument place for locations (symbolized in (24) as ‘rain_in’) and the specific value for that extra argument place of the newly- created predicate (in this case, Paris). What about sentence (1), our main target? Recanati’s claim is that, as with prepositional phrases like ‘in Paris’, quantifier phrases like ‘every time John lights a cigarette’ also contribute a variadic operator. Simplifying greatly, (1) will be represented using the variadic operators apparatus as (25) ∀l (every time John lights a cigarette in l → Vlocation: l (rain)) which is equivalent with (26) ∀l (every time John lights a cigarette in l → rain_in (l)). We see that the quantifier phrase ‘every time John lights a cigarette’ is treated as both contributing a variadic operator which transforms the zero-place predicate ‘rain’ into a new predicate with an additional argument place for locations and binding the location variable of the newly-created predicate. The difference between prepositional phrases like ‘in Paris’ and quantifier phrases like ‘every time John lights a cigarette’ thus lays in the fact that the first provides specific value for the location variable, while the second simply binds that variable. How does this treatment avoid the conclusion of the Binding Argument? The main difference between this approach and the ones investigated before is that according to the present one, the first three premises, the intermediary conclusion (IC) and the bridging principle (BP) all hold, but conclusion 4 doesn’t
The use of the Binding Argument in the debate about location
209
follow. This is because, even if there is a variable for locations in Q(s), there is none in s taken in isolation. The reason for that is, of course, the presence of the variadic operator, which is responsible for the introduction of the location variable in Q(s). Now, Recanati has used the apparatus of variadic operators within a truthconditional pragmatic theory that makes essential appeal to optional pragmatic processes like free enrichment. In connection with (1), for example, Recanati claims that the variadic operator is contributed precisely by such a process. Many authors have expressed concern with respect to the nature of Recanati’s optional pragmatic processes, the most widespread worry being that they lack the systematicity and the constrained character of mandatory processes like saturation.15 But it should be clear that the variadic operators apparatus itself caries no commitment to optional (or of any kind, for that matter) pragmatic processes. Thus, one could claim that the process by which the variadic operator is contributed, although still pragmatic in nature, is more constrained than free enrichment. Alternatively, one could claim, as done here, that the variadic operator is part of the logical form of (1). Regardless of which of these alternatives is ultimately adopted, the result that the conclusion of the Binding Argument is avoided by giving (1) a variadic operator treatment remains.
4 Conclusion The Binding Argument has received its fair share of criticism in the literature. In this paper I focused on a weaker interpretation of the argument according to which the postulation of variables for locations in the logical form of meteorological sentences such as “It is raining” is the best available explanation of binding phenomena such sentences give rise to. Such an interpretation, it seems to me, is compatible with criticisms like those put forward by Cappelen and Lepore (2002) or by Pupa and Troseth (2011) that best apply to the other, stronger interpretations of the argument. However, this doesn’t mean that the Binding Argument in its third interpretation can be freely used: in fact, my aim was to show precisely the contrary. Thus, in section 3 I showcased a number of alternative ways to account for the binding data that have motivated the Binding Argument. The fact that such ways have
15 The most vivid challenge appears in Stanley (2007, chapter 6). For replies to such an allegation, see Elbourne (2008), Hall (2008) or Recanati (2010).
210
Dan Zeman
been proposed is, of course, not an indication of their correctness. However, what is important in this context is not so much whether they are ultimately correct, but whether they have been ruled out as incorrect by the proponents of the Binding Argument in its third interpretation. So far, this is something that hasn’t happened. To be fair, some of the views presented above have been criticized. For example, Stanley (2007, chapter 6) has argued that Pagin’s (2005) view is problematic with respect to certain quantified sentences. But Pagin’s view is just one alternative way to account for the binding data. And although Recanati’s truthconditional pragmatic view has been the focus of repeated criticisms by Stanley and other truth-conditional semanticists, the apparatus of variadic operators in itself has been rarely addressed – despite its importance in connection to the Binding Argument being acknowledged (Stanley 2007, chapter 7).16 This is not to say, of course, that the views that have not been engaged with in the literature are problem-free.17 But, for the Binding Argument in its third interpretation to work, what is needed is a more thorough investigation of all the alternatives on the market and a careful comparison of their respective virtues and shortcomings with those of the view postulating variables for locations in the logical form. It might be that at the end of such an enterprise the postulation of variables for locations in the logical form of sentences such as ‘It is raining’ is our best choice. However, as things stand now, no such claim can be reasonably made. Although I have mentioned some criticisms of the Binding Argument in its first and second interpretation, my focus in this paper has been on the third. But if one accepts that those criticisms are on the right track, the Binding Argument retains very little from its (once considered great) methodological power. Since the Binding Argument has mainly been employed in support of truth-conditional semantic views about “It is raining” and other types of sentences, it follows that the argument cannot be used by the proponents of such views. To be sure, this is not something that I have shown in this paper. What I did show, however, is that if one settles for the argument’s third interpretation, it cannot be used to support a truth-conditional semantic view about “It is raining” and its ilk.18
16 This is, of course, not to accuse Stanley of any omissions – most of the alternative views to his own have been put forward after his writings on the Binding Argument. 17 For some problems with all the views presented above, except the variadic operators a pproach, see Zeman (2011, chapter 3). 18 I’m grateful to Isidora Stojanovic for giving me the possibility of contributing to this volume and to Sarah-Jane Conrad for her support and feedback, as well as to an anonymous reviewer for this volume. Research for this paper has been financially supported by a MINECO Juan de la Cierva grant (JCI-2012-12974) and by the Semantic Content and Conversational Dynamics project (FFI2012-37658) at the University of Barcelona.
The use of the Binding Argument in the debate about location
211
References Breheny, Richard. 2003. On bindability. Amsterdam Colloquium 14. 81–87. Cappelen, Herman & Ernie Lepore. 2002. Indexicality, binding, anaphora and a priori truth. Analysis 62.4. 271–281. Cappelen, Herman & John Hawthorne. 2007. Locations and binding. Analysis 67(2). 95–105. Carston, Robyn. 2002. Thoughts and utterances: The pragmatics of explicit communication. Oxford: Blackwell Publishing. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Cohen, Jonathan & Samuel C. Rickless. 2007. Binding arguments and hidden variables. Analysis 67(1). 65–71. Collins, John. 2007. Syntax, more or less. Mind 116(464). 805–850. Davidson, Donald. 1967. The logical form of action sentences. In Nicholas Rescher (ed.), The logic of decision and action 81–95. Pittsburgh: University of Pittsburgh Press. Elbourne, Paul. 2005. Situations and Individuals. Cambridge MA: MIT Press. Elbourne, Paul. 2008. The argument from binding. Philosophical Perspectives 22(1). 89–110. Hall, Alison. 2008. Free enrichment or hidden indexicals? Mind & Language 23(4). 426–456. Jacobson, Pauline. 1999. Towards a variable-free semantics. Linguistics and Philosophy 22(2). 117–184. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard Wettstein (eds.), Themes from Kaplan, 481–563. Oxford: Oxford University Press. Keenan, Edward L. & Leonard M. Faltz. 1985. Boolean semantics for natural language. Dordrecht: D. Reidel. Kratzer, Angelika. 2014. Situations in natural language semantics. The Stanford encyclopedia of philosophy (Spring 2014 Edition), Edward N. Zalta (ed.). http://plato.stanford.edu/ archives/spr2014/entries/situations-semantics/_ (accessed 01.09.2015). Lasersohn, Peter. 2008. Quantification and perspective in relativist semantics. Philosophical Perspectives 22(1). 305–337. Lewis, David. 1980. Index, context, and content. In Stig Kanger & Sven Öhman (eds.), Philosophy and grammar, 79–100. Dordrecht: D. Reidel. McConnell-Ginet, Sally. 1982. Adverbs and logical form: A linguistically realistic theory. Language 58. 144–184. Neale, Stephen. 2007. On location. In Michael O’Rourke & Corey Washington (eds.), Situating semantics. Essays on the philosophy of John Perry, 251–394. Cambridge, MA: MIT Press. Pagin, Peter. 2005. Compositionality and context. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy: Knowledge, meaning, and truth, 303–348. Oxford: Clarendon Press. Recanati, François. 2002. Unarticulated constituents. Linguistics and Philosophy 25(3). 299–345. Recanati, François. 2004. Literal meaning. Cambridge: Cambridge University Press. Recanati, François. 2007. Perspectival thought: A plea for (moderate) relativism. Oxford University Press. Recanati, François. 2010. Truth-Conditional Pragmatics. Oxford: Clarendon Press. Pupa, Francesco & Erika Troseth. 2011. Syntax and Interpretation. Mind and Language 26(2). 185–209. Schaffer, Jonathan. 2011. Perspective in Taste Predicates and Epistemic Modals. In Andy Egan & Brian Weatherson (eds.), Epistemic Modality, 179–226. Oxford: Oxford University Press.
212
Dan Zeman
Sennet, Adam. 2008. The binding argument and pragmatic enrichment, or, why philosophers care even more than weathermen about ‘raining’. Philosophy Compass 3(1). 135–157. Stanley, Jason. 2007. Language in context. Selected essays. Oxford: Clarendon Press. Zeman, Dan. 2011. The Semantics of implicit content. University of Barcelona dissertation. http://www.tdx.cat/bitstream/handle/10803/22693/DZ_THESIS.pdf?sequence=1. Zeman, Dan. 2015. Relativism and bound predicates of personal taste: An answer to Schaffer’s argument from binding. Dialectica 69(4). 155–183.
Štefan Riegelnik
Slices of meaning: Levels of analysis and the unity of understanding Abstract: Although there is little agreement about where the demarcation line between semantics and pragmatics lies, there is strong consensus that those who ignore or deny the distinction are on a wrong track. In this paper I do not want to provide a further suggestion on how to make the distinction. Instead, I want to raise a problem for any theory of language based on a distinction between semantics and pragmatics. I argue that the idea of decomposing an utterance into a semantic and a pragmatic part precludes an account of the unity an utterance exhibits. For this reason, I conclude that approaches towards a theory of language based on a distinction between semantics and pragmatics should be reconsidered.
1 Introduction Philosophers and linguists of many persuasions appeal to the distinction between semantics and pragmatics. Their motives for introducing the distinction are legion (Morris 1938, Austin 1962, Levinson 1983, Stalnaker 1970, Grice 1989, Bach 2004) and so are the views on its nature. Broadly speaking, and from an uncritical perspective, the distinction is a way to differentiate elements and effects as constituents of an arbitrary utterance. I assume that an utterance is the primary unit of meaning and that meaning and interpretation are deeply linked.1 Granting these assumptions, the distinction between semantics and pragmatics might be understood as a way to decompose natural language utterances. By way of an example, imagine the utterance of “This is Ginger” where the speaker points to a specific person standing in the corner. With the help of the gesture an interpreter is able to recognise what the speaker intends to say by using the words “This”, “is” and “Ginger” and the pointing at the specific person. Decomposing the utterance into the words chosen by the speaker and the speaker’s gesture, the study of the meaning of the words, or their literal meaning, or what Grice identifies as the conventional meaning (Grice 1988: 25) falls within the scope of
1 Some readers might find this counterintuitive; to belie these worries, see Davidson 1967 and 1970. DOI 10.1515/9781501504327-011
214
Štefan Riegelnik
semantics. All further factors pertaining to the utterance, such as the gesture, belong to the study of pragmatics. The former is often considered to be the study of the context-invariant meaning; the latter is often considered to be the study of what is conveyed in a particular context. But wherever one draws the line between semantics and pragmatics, it appears to be plausible to understand the interpretation of the utterance as an interplay between elements and effects pertaining to both realms. For imagine the situation without the speaker’s gesture: then the interpreter might have taken the speaker to say something quite differently. For example, the speaker might be pointing out that the odd flavour of the wine is caused by ginger. Or, imagine that the speaker uses different words, say, “I”, “am” and “Ginger”. In this paper I am not concerned with the question of where philosophers and linguists should draw a line between semantics and pragmatics. Instead, I want to concentrate on the question of how these elements and effects are to be related so that an utterance emerges. For my purpose it is thus of secondary importance where or how the line is drawn, as long as there is a one. I am considering any theory of interpretation that asserts a semantics‐pragmatics distinction regardless of how it further articulates that distinction. Then, by assumption, any of these theories must provide some account of how pragmatic and semantic elements within a given utterance come together so that a unified whole emerges. Because it is in virtue of this unity that an utterance could be judged to be either true or false, or, appropriate or inappropriate. Consider again the “This is Ginger”-utterance: Both the gesture and the words are supposed to belong to the utterance and both are relevant for its interpretation. If so, then merely listing the constituting parts will not suffice to explain the interplay between semantics and pragmatics. For what needs to be accounted for is the unity the “This-is-Ginger”utterance exhibits. In what follows, I argue that it is precisely the requirement for the unity of an utterance that represents a stumbling block for theorists building frameworks predicated on the semantic-pragmatic distinction. In the context of the debate about the unity the way an utterance gets decomposed is of crucial importance. Thus even if one assumes that one can safely ignore the unity problem when it arises in the context of the debate about the semantics-pragmatics distinction, one can insist that the decomposition of an utterance results in elements which could be reunified in principle. This I take it to be a minimal condition for any theory of meaning.2 For, prima facie, if this condition is not met, one does not
2 I follow the truism that to understand an utterance is to know what it means. Thence a theory of meaning is on a par with a theory of interpretation.
Slices of meaning: Levels of analysis and the unity of understanding
215
know how linguistic decomposition relates to the utterance. As I argue, the idea of dividing an utterance into a semantic part and a pragmatic part conflicts with the minimal condition.3 The structure of my paper is as follows: In the subsequent section, I discuss the idea and some of the motives behind the distinction of the semantics and the pragmatics of an utterance. In section 3, I formulate by means of Russell’s analysis of propositions the problem of the unity of the utterance. As a point of reference, in section 4 I discuss briefly Davidson’s solution to this problem and his understanding of the semantics of a declarative sentence. In the last sections I focus on so-called near-side pragmatic effects in their alleged function to codetermine the truth conditions of an utterance. I take near-side pragmatic effects as an example because they display the required unity clearly. However, I think the scope of the problem is greater than the examples I discuss. In conclude that theories based on a distinction between semantics and pragmatics face a fatal problem when it comes to explaining the unity of an utterance.
2 In slices – ways to decompose an utterance Ever since the publication of Charles Morris’ Foundations of a Theory of Signs (1938) and the introduction of the trichotomy between syntax, semantics and pragmatics, philosophers of language and linguists have quarreled over the question of where to draw a line between these three realms. Traditionally, syntax and semantics deal with the context-invariant aspects of an utterance and pragmatics deals with the context-dependent ones. Accordingly, various utterances of the sentence-type “This is Ginger” share as the context-invariant part the sentence, irrespective of what the demonstrative “this” refers to. However, drawing the line in this way is now considered to be inadequate – for it is too simplistic (cf. Stanley 2007: 32). In the wake of H.P. Grice’s distinction between what is said and what is implicated by an utterance, the debate on what pertains to semantics and what to pragmatics has received a considerable boost.4 Attempting to develop a more sophisticated theory, Grice divides the interpretation of an utterances into two phases. The first one deals with truth conditions of an uttered sentence and results in what is said (Grice 1989: 25). This phase rests heavily on conventionally
3 See Hylton (1984) and Gaskin (2008) who stress the relevance of the problem of the unity. 4 Grice did not make use of the expressions “semantics” and “pragmatics”. He pointed out that his use of “say” and “what is said” or “implicate” or “what is implicated” deviate from ordinary usage (Grice 1988: 25).
216
Štefan Riegelnik
encoded sentence or word meaning and includes the assignment of denotations of the words (including indexical expressions) and their combination. The second phase takes as input that what is said and deals with everything else that is conveyed with an utterance. It runs along general conversational principles which enable the interpreter to infer what the speaker implicates. Although there is consensus that Grice’s model needs to be refined in several respects, his model still plays an important role in contemporary debates on the distinction between semantics and pragmatics. Most notably, contextualists of various manifestations argue that what is said with an utterance is influenced to a far larger extent by near-side pragmatic effects than Grice allows (Recanati 2001, 2004, 2010; Stanley 2007; Cappelen/Lepore 2005). Accordingly, contemporary contextualists’ theories of interpretation can be grouped by the degree of impact of pragmatic effects on truth conditions, i.e. by the number of assumed context-sensitive expressions. In contrast to the wide ranging and acrimonious discussions on the nature of the distinction, skeptical doubts concerning the viability of a distinction between semantics and pragmatics have been mostly neglected.5 The fundamental questions to be addressed are: Are natural languages suitable for such a distinction at all? Is it possible to draw a sharp line? Would it be better if philosophers and linguists gave up the quest for a demarcation line altogether? Kent Bach is among those philosophers and linguists for whom the answer to the first two questions is clearly “yes”. According to him, the reason for introducing the distinction is “to provide a framework for explaining the variety of ways in which what a speaker conveys can fail to be fully determined by the (conventional) linguistic meaning of the sentence he utters” (Bach 1999: 66). Bach cites indexicality, ambiguity, vagueness, semantic underdetermination, implicitness, implicature, and non-truth-conditional content as linguistic phenomena that require the introduction of a framework with different levels or slices. The framework he has in mind is supposed to cover the differences in the explanatory functions. Consider again an utterance of “This is Ginger”.6 Understanding the word meaning of “This”, “is”, and “Ginger” – usually considered to be the semantic part of an utterance and determined conventionally – would not suffice to understand what a speaker conveys by this utterance. Even if one consulted a myriad of dictionaries, one would not find an answer to the question of what the demonstrative “this”
5 A dissenting voice is Cappelen (2007). 6 In what follows, I focus on the demonstrative “this”, but the same line of reasoning would apply to other expressions taken to be context-sensitive.
Slices of meaning: Levels of analysis and the unity of understanding
217
used as part of a particular utterance refers to. So it appears to be the defining feature of utterances in which “this” is used that their meaning is different from context to context. This would make it plausible to think of the context of an utterance as what one ought to consider in order to understand an utterance containing “this”. To be more precise, since the reference of the word “this” vary with the context, it seems to be plausible to say that the truth conditions are co-determined by the context. Thus the appeal to the context of an utterance looks not only to be innocuous, but positively necessary in order to determine the meaning of an utterance of a sentence containing an indexical expression or a demonstrative. In contemporary debates on the role of an utterance’s context, minimalists (Cappelen/Lepore 2005, Borg 2004) claim that, basically, near-side pragmatic effects with regard to indexical expressions and demonstratives are relevant in order to determine what is semantically expressed. Moderate contextualists (Stanley 2007) agree with minimalists about this, but they also think that the class of expressions which are context-sensitive needs to be extended. Thus minimalists seek to confine near-side pragmatic effects to a very small class of expressions, whereas moderate contextualists are open to extending the class of contextsensitive expression. In contrast to the two aforementioned positions, radical contextualists (Recanati 2004, 2010, Bezuidenhout 2002) claim that near-side pragmatic effects are required for the emergence of a truth-evaluable entity even if the sentence does not contain any obvious context-sensitive expressions. They further claim that these effects cannot be confined to a specific class of expressions. The role of the semantical level of a framework is understood as that of a potentiality. For expressions are not considered to be meaningful independently of a speech act (Recanati 2004: 152). In any case, the boundary between semantics and pragmatics seems to be blurred. But in all three positions outlined above there is agreement that pragmatic effects determine to at least some degree the truth conditions of an utterance. In contrast to them, Bach insists that “semantics and pragmatics have distinct subject matters, sentences and utterances, respectively” (Bach 2004: 28). He maintains that we are able to draw a clear-cut line between these areas. For him “semantic properties are on a par with syntactic and phonological properties: they are linguistic properties. Pragmatic properties, on the other hand, belong to acts of uttering sentences in the course of communicating” (Bach 2004: 24). I quote some further philosophers and linguists to show that Bach is not alone in this view: The fact that a word or phrase has a certain meaning clearly belongs to semantics. On the other hand, a claim about the basis for ascribing a certain meaning to a word or phrase does not belong to semantics. (Almog/Wettstein/Kaplan 1989: 573) […] traditionally, […] semantics [is] to be the study of meaning, […] pragmatics is the study of language usage. (Levinson 1983: 5)
218
Štefan Riegelnik
A strong pragmatic effect on what is communicated is one in which context affects what is communicated, but not by affecting the referential contents of any lexical item in a sentence. (Stanley/King 2005: 140)
If one understands the distinction as clear-cut as stated explicitly by Bach, one might wonder why semantics and pragmatics are mentioned in the same breath at all. Put differently, if despite their “distinct subject matters” (Bach 2004: 28) both semantics and pragmatics belong somehow together, the assumed substantial relation between them needs to be accounted for. Or, if one admits that the distinction is as explicit and as clear-cut as suggested by Bach and also thinks that there is no substantial relation between semantics and pragmatics, then, admittedly, one does not need to provide an account of the relationship between these fields, i.e. the unity of the utterance. But then the motives for the debate about the distinction become even less comprehensible. For on Bach’s model, why should we understand pragmatics as being part of linguistics at all? In contrast to the philosophers and linguists I have just quoted, I hold that to account for the distinction between semantics and pragmatics requires one to understand these aspects as being mutually dependent. Intuitively, if one maintains that there is a distinction, one should be able to account for each of the aspects independently of the others. However, this seems to be deeply problematic as near-pragmatic effects show. So what I claim is that the identification of semantic properties is dependent on appealing to pragmatic aspects – and vice versa. Consequently, it is a mistake to study semantic or pragmatic elements and effects in isolation. This can be shown by the inability of advocates of such a distinction to meet the requirement I call the independency condition: the appeal to semantic or pragmatic properties has to make sense independently of other factors or levels of a framework built to explain the interpretation of an utterance. That this is the aim of proponents of a semantic-pragmatic distinction is shown by the above mentioned quotes where the possibility of the study of semantic and pragmatic entities and effects is assumed. However, as I argue in the remainder of this paper, the independent study of semantic and pragmatic elements and effects falls prey to the same problem that has plagued philosophers when it comes to account for the unity of the proposition.
3 The problem of the unity The problem of the unity of the propositions or the unity problem can be traced back to Plato and is addressed by philosophers in different areas. Nowadays, the problem is understood mostly as pertaining to the philosophy of language.
Slices of meaning: Levels of analysis and the unity of understanding
219
It emerges when we ask how singular terms are related to predicates so that one gets a unified sentence or proposition. I rephrase this question in the light of the debate on the distinction between semantics and pragmatics: how are elements and effects pertaining to either of these realms merged so that one gets a unified utterance? In order to demonstrate the pertinence of the unity problem for the debates on a distinction between semantics and pragmatics, I refer to Russell’s analysis of a proposition. I then show how, mutatis mutandis, the same problem besets the distinction between semantics and pragmatics. In dealing with the nature of the special unity of a proposition, Russell writes in the Principles of Mathematics: Consider, for example, the proposition “A differs from B”. The constituents of this proposition, if we analyse it, appear to be only A, difference, B. Yet these constituents, thus placed side by side, do not reconstitute the proposition. The difference which occurs in the proposition actually relates A and B, whereas the difference after analysis is a notion which has no connection with A and B. It may be said that we ought, in the analysis, to mention the relations which difference has to A and B, relations which are expressed by is and from when we say “A is different from B”. These relations consist in the fact that A is referent and B relatum with respect to difference. But “A, referent, difference, relatum, B” is still merely a list of terms, not a proposition. A proposition, in fact, is essentially a unity, and when analysis has destroyed the unity, no enumeration of constituents will restore the proposition. The verb, when used as a verb, embodies the unity of the proposition, and is thus distinguishable from the verb considered as a term, though I do not know how to give a clear account of the precise nature of the distinction. (Russell [1903] 2010: §54)
As Russell writes, under analysis a proposition loses its unity. Part of the analysis of a proposition is that we decompose it into its constituents. What, then, needs to be accounted for is the unity a proposition exhibits. One should not be puzzled by Russell’s realist view of propositions and what their constituents are, for at least in this context it is secondary what one takes to be the primary unit of meaning – Russellian propositions, sentences, utterances etc. Regardless which of them is considered to be the primary unit of interpretation, when analysing it, i.e. decomposing it into its constituents, one “destroys” its unity: one gets a mere list of constituents. However, an utterance or a proposition differs fundamentally from such a list. In contrast to a list, an utterance as the primary bearer of meaning can be, most strikingly, either true or false, or, appropriate or inappropriate. Only in virtue of a unity these predicates can be ascribed to an utterance.7
7 For contemporary contributions and the different presentations of the problem, see Davidson (2005), Gaskin (2008), contributions to Riegelnik (2010).
220
Štefan Riegelnik
If – as Russell anticipates – one tries to achieve the restoration of the unity by appeal to a relation, perhaps referred to by the copula, one will be disappointed. The so introduced relation is but a further element on the list of unjoined parts and there is still no unity between them in sight which could be true or false, appropriate or inappropriate. Further, for reasons of consistency, another relation needs to be introduced which is supposed to unite the original list of components with the initially introduced relation. But again, this relation is just a further item on the list of unjoined parts. And so forth, ad infinitum. This shows that the appeal to relations in order to restore the unity is a hopeless endeavour, for it ends in an inescapable infinite regress. It is not the nature of a relation that causes the regress – whatever entity or means one introduces as a form of “glue” in order to combine the components, one ends up in the regress just sketched.8 Russell discusses the problem of the unity of the proposition in the light of the debate on the semantic role of verbs or predicates: for him a theory of predication is supposed to explain the unity. He does not have the distinction between semantics and pragmatics in mind. However, the same problem emerges when we attempt to account for the relationship between semantic and pragmatic entities and effects. An utterance, taken as the primary unit of meaning or interpretation, is considered to comprise all elements and effects relevant for its constitution. An advocate of a distinction between the semantics and the pragmatics of an utterance might analyse the “This-is-Ginger”-utterance as follows: it consists, on the one hand, of the expressions used and, on the other hand, the part which is conveyed by pointing towards Ginger, and perhaps also further factors which need not concern us here. The choice of words and the way these words are uttered determine ideally in a decisive way the interpretation of the utterance. Counterfactually, the interpretation of the utterance would differ, if the speaker acted in a different way or if she used different words. Imagine if the speaker is sipping her wine, one can assume “ginger” in “This is ginger” refers to a spice. Conversely, she may say “This woman in the corner with the nice coat is Ginger” and, clearly, “this” refers to the woman named Ginger. But in virtue of what do the expressions and the gesture form a unity? As Russell’s analysis of a proposition shows, as soon as one needs to account for a combination of these parts, be it thought as the combination of the context-independent part with the context-dependent one, or the gesture and the words, the analysis is threatened by an infinite regress. For
8 Showing how prominent accounts fall into this trap is Davidson’s leitmotif in Truth and Predication (2005).
Slices of meaning: Levels of analysis and the unity of understanding
221
then one would need a further entity which unites the unifying entity with the original parts. So we are left with distinct entities.
4 Truth-conditional semantics and the unity of the sentence In contrast to contemporary truth conditional theories of meaning such as minimalism, moderate contextualism or radical contextualism, in Davidson’s account of the semantics of a declarative sentence one does not find a distinction between semantics and pragmatics. One could even sum up Davidson’s approach by emphasizing that any independent specification of truth conditions, or the supposed to be semantic meaning as independent from pragmatics, or the meaning of a predicate expression as independent from the meaning of the utterance etc. is utterly mistaken. For, by Davidson’s lights, such a distinction makes it impossible to understand how expressions are used as part of a sentence in such a way that one understands what somebody says (Davidson 1967: 17, 2005). Whether or not one holds Davidson’s approach towards a unified theory of meaning viable, it serves as a point of reference for the question of the unity of semantics and pragmatics. In Truth and Predication (2005) Davidson follows Russell in discussing the problem of the unity of a proposition in connection with the problem of predication. Davidson argues that the semantic role of a predicate cannot be explained unless we examine it in the context of a whole sentence. His solution to the unity problem is part of his comprehensive theory of meaning in which the principle of compositionality plays a key role. According to this principle, the meaning of complex meaningful expressions, i.e. sentences, is considered to be determined of the meaning of simpler expressions. The question now is: what concepts should one appeal to if one wants to explain the contribution of the parts of a sentence? In the case of a subject – usually a singular term – the concept philosophers appeal to is reference or denotation. But what concept one should appeal to if one wants to explain the contribution of the predicate? In order to determine the predicate’s contribution to the truth conditions, Davidson appeals to the concept of satisfaction: it pairs objects with so-called open sentences, which are expressions with free variables. An object satisfies an open sentence if the resulting sentence is true. Hence snow satisfies “x is white”, for “snow is white” is true and grass does not satisfy “x is white” for it is not the case that grass is white. The approach just sketched emerges from Davidson’s engagement with accounts that could not properly explain the unity of the utterance. Pursuant to this, Davidson’s own account is not subject to these deficiencies (cf. Davidson 2005: chapter 7). Most essentially,
222
Štefan Riegelnik
it does not generate an infinite regress and it links predicates as essential parts of sentences to the sentences’ truth conditions. “In a minimal but important respect”, Davidson states, “a theory of truth for a language does […] what we want, that is, give the meaning of all independently meaningful expressions on the basis of an analysis of their structure” (Davidson 1970: 55). The structure, as I have already pointed out, must not be understood as being independent of the utterance or as an additional part in the way Russell had thought of it in his multiple relation theory. In Davidson one does not find a distinction between semantics and pragmatics, or, generally speaking, the idea of different levels. Nor does one find a distinction between sentences and propositions, the latter expressed by the former relative to different contexts, influenced and shaped by pragmatic effects. This, of course, does not mean that Davidson denies that factors usually grouped under the label pragmatics are relevant for the interpretation of an utterance. It is rather the idea of a “framework with different levels” (see section 2) that stands in opposition to his idea of a sound theory of interpretation. Also, interestingly, many contextualists who advocate a distinction between semantics and pragmatics see themselves as being in opposition to Davidson – as stated explicitly by Bezuidenhout – “[i]f one embraces the contextualist perspective one must give up traditional Davidsonian style semantic theories” (Bezuidenhout 2002: 105) – then because their emphasis on the significance of pragmatics is a view which would be for Davidson a version of reductionism. This is so because according to Davidson, contextualists try to save a distinction between semantics and pragmatics by negating one part of the assumed relation. In Davidson’s model, ironically, defenders of the distinction end up trying to reduce one part of the distinction to the other.
5 Near-side pragmatic effects and the unity of semantics and pragmatics As I have already pointed out, in recent years debates on the semanticspragmatics distinction have centred on the question of the degree of the impact of pragmatic effects or contextual contributions on the truth conditions of an utterance. This has led to considerations about the interplay between semantics and pragmatics given that it is assumed that a sentence might express different propositions relative to different contexts.9 The thesis I am going to scrutinize is
9 An argument along similar lines can be formulated for theories understanding the process of interpretation in two or more phases – for what unites the phases?
Slices of meaning: Levels of analysis and the unity of understanding
223
that near-side pragmatic effects explain the divergence of sentence meaning and propositions expressed by sentences. In contrast to far-side pragmatic effects, near-side pragmatic effects are considered to be relevant for the determination of what is said (Grice 1989: 25) with the utterance of a sentence. This class of pragmatic effects – mostly contextual factors – comes into play when the knowledge of the sentence meaning is considered to be insufficient for the interpretation of an utterance.10 Since what is said with an utterance is usually associated with the truth-conditional meaning of an utterance, the governing assumption is that near-side pragmatic effects influence the determination of truth conditions of an utterance. Further it is assumed that these effects complement or modify the sentence meaning of the expressions or the semantic potentials used so that a truthevaluable entity – usually a proposition – results. Put this way, a proposition would be the result of a sentence enriched by near-side pragmatic effects. Intuitively, this seems to be correct, for one and the same sentence can be used to express different propositions. Again, the sentence “This is Ginger” uttered on different occasions can express different propositions, depending on the gesture, or more precisely, the person or the object “this” refers to. This requires that expressions composing a sentence have to be receptive to pragmatic effects: “this” must be understood as an expression being able to make varying contributions to the proposition expressed. If this were not the case, the proposition would be determined exclusively by the expressions used. Consequently, there would be no room for near-side pragmatic effects to determine or shape the proposition expressed by an utterance. However, the traditional way to cope with indexical expressions and demonstratives has on closer examination serious shortcomings. My main objection is that such an approach faces the same dilemma which I stated implicitly before: either near-side pragmatic effects are tied to what is considered to be the semantic level of an utterance in a way that there is no distinction between semantics and pragmatics, or, there is a distinction, but the grounds for understanding this interplay remain opaque. According to the first horn of the dilemma – near-side pragmatic effects must be tied to the semantic level, otherwise there is no mutual influence. And if there is no mutual influence, one can hardly speak of near-side pragmatic effects, for these effects are introduced precisely to influence the semantics of an utterance. For the sake of the argument, near-side pragmatic effects affect truth conditions via the expressions composing the sentence. They enrich or modulate the word meanings of the expressions composing the sentence. This entails that the expressions used have to be understood as incomplete. For if they were complete, near-side
10 This holds also for implicatures.
224
Štefan Riegelnik
pragmatic effects could not perform their function and enrich or modulate the word meanings. In other words, the incomplete part of the expression remains constant and gets completed or enriched from utterance to utterance or from context to context. This puts the semantic part of an utterance on a par with what Recanati calls “semantic potentials” (Recanati 2004: 152). Semantic potentials are taken to be the most basic unit. Above them, it is assumed that pragmatic effects are added at this higher level to supplement the semantics. Finally, the unity of these two levels is reflected by the utterance itself. However, this approach simply shifts the problem to the lower level. On the lower level, the same basic problem emerges. To wit, one still needs to account for how indexical expressions and demonstratives are open to pragmatic effects. In order to illustrate the point further, I return to Recanati’s notion of semantic potentials – assuming that the level of semantic potentials does contribute to the higher level. But for this, a third factor is required relative to which varying contributions are made by one and the same semantic potential. The third factor could be thought again as near-side pragmatic effects – except that in this case they are supposed to be decisive for the contribution a semantic potential actually makes to the determination of the expressions which determine the next level, i.e. the proposition. Further if one now considers the semantic potential as part of the framework, i.e. as a level of the framework, one encounters the same problem again: one part of the semantic potential remains constant across different utterances and one part makes varying contributions. But where should one now tie the near-side pragmatic effects relative to which a contribution to the expression is made? Again we might introduce one further level, a potential of the semantic potential, and so on, ad infinitum. One gets again caught in an infinite regress. In summation, I conclude that either we assume that the meaning of an expression as a contribution to the meaning of the utterance is complete or we end up in the regress I have just emphasized. This conclusion seems to be fatal for those who argue that truth conditions vary with contexts: near-side pragmatic effects, or contextual contributions, cannot “latch onto” the invariant part, or whatever is thought to constitute the semantic or lower level of a framework.11
11 What thereby is shown, too, is that drawing a distinction between semantics and pragmatics requires that the semantic level has to be regarded as significant independently from pragmatics. This is not so if pragmatic effects overrule qua pragmatic effects the contributions of the semantic level. By way of an example, even if one admits that it is possible that the sentence “The lion sleeps tonight” can be used to express the proposition that the woman in the corner is Ginger, for instance by way of an implicature, it cannot be a general principle. For if the role of the semantics is displaced, there is no relatum which corresponds to semantics as the counterpart of pragmatics. In other words, semantics and pragmatics collapse.
Slices of meaning: Levels of analysis and the unity of understanding
225
At the beginning of this section, I formulated a dilemma: either near-side pragmatic effects are tied to what represents the semantic level or the distinction represents an ill-founded postulation. As it is with dilemmas, the second horn shares with the first one its problems. If it is assumed that near-side pragmatic effects can be specified independently of any other level of a framework, then their contribution to the semantic level cannot be accounted for. In general, this means that once one of the levels involved in a framework based on a distinction between semantics and pragmatics is taken to be irrelevant, one should not assume a distinction anymore. As a consequence, it does not make sense to speak of a relation between semantics and pragmatics in such a way that it makes sense to identify the fields in question independently of each other. But then it seems that the idea of a theory of interpretation based on such a distinction makes no sense.
6 Conclusion Proponents of the distinction between semantics and pragmatics maintain that we can appeal to either aspect without any need to invoke the other. Indeed, this is exactly how one can disentangle pragmatic and semantic effects. This commitment to what I call the independency condition gives rise to a special version of the problem of the unity of an utterance. In comparing theories which draw a distinction between different levels with Davidson’s theory of interpretation, I have emphasized that the former cannot account for the unity. I further believe that this failure indicates the unintelligibility of understanding a sentence as distinct from the way we communicate. The idea that part of an utterance – a sentence as representative for semantics – remains constant and another part – the effects on the sentence – varies has proven to be unsuccessful. For either the varying factors are tied to the sentence’s level, so there is no essential distinction between semantics and pragmatics. Or, there is a distinction, but the grounds for being effective are opaque.
Acknowledgments I would like to thank Gabriele M. Mras and Charles Djordjevic for valuable comments on early versions of this paper.
226
Štefan Riegelnik
References Austin, John L. 1962. How to do things with words. Oxford: Clarendon Press. Bach, Kent. 1999. The semantics pragmatics distinction: What it is and why it matters. In Ken Turner, (ed.), The semantics-pragmatics interface from different points of view, 65–84. Oxford: Elsevier. Bach, Kent. 2004. Minding the gap. In Claudia Bianchi (ed.), The semantics/pragmatics distinction, 27–43. Stanford: CSLI. Bezuidenhout, Anne. 2002. Truth‐conditional pragmatics. Noûs 36. 105–134. Cappelen, Herman. 2007. Semantics and pragmatics: Some central issues. In Gerhard Preyer & Georg Peter (eds.), Context‐sensitivity and semantic minimalism, 3–23. Oxford: Oxford University Press. Cappelen, Herman & Ernie Lepore. 2005. Insensitive semantics: A defense of semantic minimalism and speech act pluralism. Oxford: Blackwell. Davidson, Donald. 1967. Truth and meaning. In Donald Davidson, Inquiries into truth and interpretation, 17–36. Oxford: Clarendon Press. Davidson, Donald. 1970. Semantics for natural languages. In Donald Davidson, Inquiries into truth and interpretation, 55–64. Oxford: Clarendon Press. Davidson, Donald. 2005. Truth and predication. Cambridge, MA: Belknap Press. Gaskin, Richard, 2008. The unity of the proposition. New York: Oxford University Press. Grice, Paul. 1989. Studies in the way of words. Cambridge, MA: Harvard University Press. Hylton, Peter. 1984. The nature of the proposition and the revolt against idealism. In Peter Hylton (ed.), Propositions, Functions, and Analysis. Selected essays on Russell’s Philosophy, 9–29. Oxford: Clarendon Press. Almog, Joseph, Perry, John, Wettstein, Howard & Kaplan, David (eds.). 1989. Themes from Kaplan. New York: Oxford University Press. Levinson, Stephen C. 1983. Pragmatics. Cambridge: Cambridge University Press. Morris, Charles. 1938. Foundations of a theory of signs. In Charles Morris (ed.), Writings on the general theory of signs, 17–72. The Hague and Paris: Mouton. Recanati, François. 2004. Literal meaning. Cambridge: Cambridge University Press. Recanati, François. 2001. What is said. Synthese 128(1/2). 75–91. Recanati, François. 2010. Truth-conditional pragmatics. Oxford: Clarendon Press. Riegelnik, Stefan (ed.). 2010. Predication and the unity of proposition. Special issue of Conceptus – Journal of Philosophy 96–97. Russell, Bertrand. [1903]/2010. Principles of mathematics. London & New York: Routledge Classics. Stalnaker, Robert C. 1970. Pragmatics. Synthese 22(1/2). 272–289. Stanley, Jason, King, Jeffrey C. 2007. Semantics, pragmatics, and the role of semantic content. In Stanley, Jason (ed.). Language in Context. Selected Essays, 133–181. New York: Oxford University Press. Stanley, Jason. 2007. Language in context. Selected essays. New York: Oxford University Press.
Contributors Christopher Gauker is the Professor of Theoretical Philosophy at the University of Salzburg, Austria. He works primarily in the philosophy of language, philosophy of mind and philosophical logic. He is the author of Words without Meaning (MIT 2003), Conditionals in Context (MIT 2005) and Words and Images: An Essay on the Origin of Ideas (Oxford University Press 2011). Kepa Korta is Associate Professor of Philosophy at the University of the Basque Country and researcher of the Institute for Logic, Cognition, Language and Information (ILCLI). He has authored and edited several books and numerous papers on philosophy of language, semantics, and pragmatics. Recently he has coauthored with John Perry the book Critical Pragmatics: An Inquiry into Reference and Communication (2011) and several articles in the journals Intercultural Pragmatics, Mind and Language, Philosophy and Phenomenological Research, Synthèse, in the Stanford Encyclopedia of Philosophy, and in various collections. John Perry is Emeritus Professor of Philosophy at Stanford University, where he taught from 1974 to 2008, and at the University of California, Riverside, where he taught from 2008 to 2014. He received his BA from Doane College in 1964 and his PhD from Cornell University in 1968. He has written books and articles on personal identity, the philosophy of language, consciousness, and other philosophical topics. He has received a number of awards for teaching and research, including honorary degrees from Doane College, the University of the Basque Country, and Bochum University. He presently conducts research at the Center for Language and Information at Stanford, does occasional teaching at Stanford and the University of California, Berkeley, and co-hosts the radio program “Philosophy Talk” (www.philosophytalk.org) with Ken Taylor. Silvan Imhof is working on the edition of Karl Leonhard Reinhold’s Gesammelte Schriften (University of Fribourg) and on the edition of Jeanne Hersch’s philosophical works (University of Zurich). He was a member of the research group meaning.ch at the University of Berne between 2004 and 2012. Nikola Kompa is a Professor of Theoretical Philosophy at the Institute of Philosophy, University of Osnabrueck/Germany. She works on a variety of topics within philosophy of language and epistemology. She is particularly interested in theories of language comprehension, the relation between language and cognition, metaphor and vagueness on the one hand and epistemic contextualism and the semantics of epistemic evaluation on the other. Anne Bezuidenhout is a professor of philosophy and linguistics at the University of South Carolina. She is interested in cognitive models of human natural language production and understanding. Her current research interests include: indexicals and context-shifting; generalized conversational implicatures; presupposition; foregrounding and backgrounding; and joint reference. Stefano Predelli completed his doctoral studies at UCLA in 1991, with a dissertation on indexicals supervised by David Kaplan. He then moved to Norway, where he taught for a few years at the University of Oslo. He is currently a professor at the University of Nottingham. His previous publications include Contexts: Meaning, Truth, and the Use of Language (OUP, 2005), and Meaning without Truth (OUP, 2013).
228
Contributors
Damon Horowitz holds a joint appointment with the Symbolic Systems and Structured Liberal Education programs at Stanford University. He has taught courses in philosophy, artificial intelligence, and cognitive science at Columbia, U. Penn., NYU, and San Quentin State Prison. He has served on the Board of Directors for several arts and humanities organizations, including the California Council for the Humanities, and has held senior advisory roles at major industry corporations. He earned his BA from Columbia, his MS in AI from the MIT Media Lab, and his PhD in Philosophy from Stanford. Dominik Aeschbacher studied at the University of Bern and is currently a PhD student at the University of Neuchâtel. In 2015, he received a Doc.CH grant for his project “The intelligibility of grounding” from the Swiss National Foundation. His research focuses on metaphysics and he takes a special interest in questions about essence, grounding and real definitions. Dan Zeman is a “Juan de la Cierva” Postdoctoral Fellow at the University of the Basque Country and member of the Basque Research Group for Theoretical Linguistics. His main research interest is in the various ways to formally capture the context-sensitivity of natural language expressions, with particular focus on perspectival expressions. In connection to the later, he is involved in the contemporary debate between contextualism and relativism and has written on various aspects of this debate (disagreement, retraction, the binding argument etc.). He has published papers in journals like Thought, Dialectica, Linguistics and Philosophy, as well as in several edited volumes. Štefan Riegelnik studied philosophy and economics in Vienna. He has taught courses in philosophy at the University of Zurich and the Vienna University of Economics. He was visiting professor at the University of Podgorica, Montenegro. Currently, he is the coordinator of the doctoral program “Philosophy: Language, Mind, and Practice” at the University of Zurich. His areas of specialization are philosophy of language and mind. He is working on questions concerning the semantics-pragmatics distinction and the unity of the proposition.
Index a priori 7, 171–72, 174–76, 175, 180 ambiguity 45, 67, 216 analysis 2–3, 5–8, 71, 118, 123–24, 126–28, 131, 136, 146–48, 154, 159, 164–66, 171, 186 analysis, conceptual 171–76, 178, 180, 183, 185–89 analysis, semantic 51–53, 61, 67, 71, 118, 148, 154, 159, 164–66, 171, 186 analysis, speech act 61–62, 71, 72 autonomy of language 55, 59–61 binding 191–210 Binding Argument, the 191–210 character 94–95, 97, 109, 131–32, 178 cognitive 6, 35, 84, 86–88, 91–93, 99–105, 107–13, 129, 145, 175, 178 content 1–2, 4, 7, 24–25, 28, 34–36, 38–47, 51–52, 55, 57–72, 80, 95, 100, 111, 122, 126, 133, 218, 141–66, 193, 216 content, descriptive 55, 173, 175–77, 179 content, intentional 55–57, 60, 62, 64–68 content, propositional 1, 31, 145, 156 content, refential 40, 42, 45–46, 218 content, semantic 1, 4–5, 46, 51–52, 55, 57–60, 62–64, 66–72, 76, 83, 95–96, 181, 210 content, truth-conditional 4, 7, 34, 100, 141, 143–51, 154–57, 159, 161–62, 165–66, 216 content, utterance 1, 7, 34–35, 38, 44, 141–43, 148, 152, 155, 157–66, 193 context 1–8, 13, 15–18, 24–29, 33, 37, 41, 46, 51–52, 54, 56–59, 61–72, 81–82, 84, 87, 91–94, 98–101, 105–06, 109–10, 112–13, 118, 120–21, 124, 128, 130–34, 142–66, 171, 179–80, 183–84, 188–89, 194, 100–200, 203–05, 210, 214–15, 217–19, 221–22, 224 context, mental 51, 57–59, 61 context, narrow 94–95 context, utterance 56, 99, 151–52, 157, 160, 203, 217 context, wide 94–95, 99, 101 context-dependent 8, 24, 35, 47, 61, 70, 144, 215, 219 DOI 10.1515/9781501504327-012
context-independent 68, 70–72, 220 context-sensitive 1, 31–32, 40–41, 79–80, 93, 142, 147–48, 166, 203, 216–17 context sensitivity 7, 32–33, 67, 75–76, 78–79, 81–82, 158, 176, 188 Context Hypothesis 56–59, 61–62, 68–69, 71 contextualism 7, 47, 143, 159–62, 165–66, 221 contextualist 7, 32–33, 35, 42, 45, 79, 96, 141–66, 216–18, 222 convention 5, 24, 38, 40, 61, 77–78, 85, 95, 118, 142–43, 147–48, 151, 153–55, 159, 215–15 conversation 6, 28, 44, 75, 79–81, 91, 98–99, 103, 106, 145, 149, 152–53, 155–56, 185 conversational 32, 36, 91, 93–94, 96, 98, 103–05, 107–09, 112–13, 121, 146, 143, 155–56, 210, 216 descriptive 6, 55, 117, 120–36, 160, 173–79 enrichment 20, 31, 35–36, 45, 145, 193, 209 entailment 118, 126, 131–32, 135, 174, 180 externalism 53, 171–72, 176, 178–79, 189 externalist 172, 176, 179, 181 free enrichment 37, 145, 193, 209 grab-bag 6, 75–76, 82–83, 87–88 hermeneutic 141–42, 153, 159, 160–66 hermeneutic contextualism 7, 143, 159–66 implicature 1, 31, 35–36, 44–46, 96, 103, 105, 146, 216, 223–24 incompletism 31–32, 34, 37–38, 40–42, 46–47 indeterminacy 4, 14, 17, 38, 125, 128, 135, 141, 147, 149, 152, 154–55, 157–58, 165 indexical 1, 4, 6, 8, 24, 31–33, 37–38, 41, 46, 79, 91–102, 108–09, 112–13, 121, 134, 148, 193, 216–17, 223–24 indexicality 15, 32–33, 67, 79, 121, 216
230
Index
intention 40–41, 46, 51–66, 80, 92, 94–95, 98–99, 103–04, 106–07, 142, 151–56, 162–64 internalist 6, 92–93, 181 interpreter 7, 23–24, 75–78, 81, 83, 106, 109–11, 141–44, 146, 150, 152–54, 157–66, 213–14, 216 literal 1, 7, 26, 28, 33, 37, 44, 51–53, 55, 58–60, 62–64, 66–72, 141–45, 147–53, 155–59, 161–62, 166, 213 malapropism 77–78, 154–55 meaning 3–5, 7, 13–14, 20, 23–25, 29, 33, 35, 38, 40–41, 51, 58, 60–61, 66–67, 69–72, 76–77, 84, 86, 107–08, 112–13, 117–18, 120, 122, 125–26, 128, 130–31, 133, 141–42, 152, 154–55, 159–60, 162–63, 165–66, 171–76, 179, 181–82, 184, 196, 213–14, 216–17, 219–24 meaning, conventional 7, 142–43, 151, 159, 213 meaning, encoded 94–95, 100, 109, 121 meaning, linguistic 1, 37, 42, 45–46, 53–54, 56, 58–60, 70, 216 meaning, literal 37, 51, 58, 66–67, 69, 142–44, 147–48, 151, 155, 157, 159, 213 meaning, metaphorical 144, 146–47, 151, 153, 155, 159 meaning, semantic 35, 51, 69, 221 meaning, sentence 35, 38, 41, 52, 54, 58, 61, 223 meaning, speech act 51, 54–55, 59, 64 meaning, theory of 1, 4, 6, 52, 75–77, 101, 125, 133, 171, 214, 221 meaning, type 5, 75–78, 82–83, 85, 87–88 meaning, word 16, 53, 84, 216, 223–24 meaning, utterance 52–54, 58, 143, 153, 165, 224 mental representations 24, 92, 101, 109–11 metaphor 7, 53, 141, 143–48, 150–51, 153–55, 157, 159, 165, 182, 184–85 metaphorical 143–49, 151–52, 155, 157, 159, 165–66 network 40, 46, 56, 58, 65
open-texture 4, 13–15, 17–20, 22, 28 pragmatic 1–5, 8, 31, 33, 35–36, 41, 45–47, 51–53, 58, 80–82, 95–105, 111 pragmatics 1–2, 7–8, 31, 34–35, 44–47, 95–96, 100, 103–04, 141, 147–48, 152, 160, 193, 213–25 predication 4–5, 60, 64–67, 145, 166, 220–21 proposition 1, 5, 31–43, 45–46, 77, 92, 142, 145–46, 148–50, 152, 154, 156, 162, 180–81, 184, 205, 215, 218–24 propositional 1, 4, 31–32, 35–36, 38–40, 42, 55, 85, 118, 141, 145, 156 propositionalism 34, 37–38, 46 reference 3–5, 13, 15, 17, 21, 24–25, 28, 31, 35–36, 45, 60, 64–66, 94–95, 103–04, 109, 111–12, 120, 126, 155, 172–73, 175–76, 178–79, 215, 217, 221 referential 6, 13–14, 17, 23, 25, 27, 29, 34, 40, 43–47, 60, 64, 66, 109–11, 173 relativism 163, 193–94 relevance 34–35, 45, 103–04, 106, 111–13, 160 rule 18, 24, 27, 29, 52–53, 77–78, 82, 86, 92, 94–94, 103, 107, 109, 127, 134, 136 saturation 5, 31, 33, 36–38, 40–42, 46, 209 schematicity 4, 13–15, 23, 28–29 semantic 1–8, 13, 15–16, 22, 25, 28, 34–35, 37, 41, 46, 51–53, 57–63, 66–72, 76–77, 80–83, 95–100, 103, 112, 117–18, 120–24, 128–30, 133, 142–43, 148, 171–72, 174, 176, 178–79, 181, 185, 189, 196, 198, 208, 210, 213–18, 220–25 semantics 2, 4, 6–8, 14–15, 23–24, 27–29, 56, 58, 91–93, 100, 117–18, 121, 127, 129–30, 141, 159, 163, 172, 175–76, 179, 185, 191–94, 200, 203, 213–25 semantics, formal 2, 4–5, 13, 91 semantics, referential 3–4, 6, 13–17, 20, 22–25, 28–29 semantics, truth-conditional 8, 148, 193, 221 semantics, two-dimensional 7, 171–72, 176, 178–79, 181, 184, 189 semantics-pragmatics 1, 2, 8, 142–43, 214
Index
speech act 5, 38, 51–57, 59–62, 71–72, 95, 125–26, 217 syntactic 1, 15, 80, 96, 100, 104, 107, 119–20, 128–29, 131, 161, 163, 191, 196, 198, 207, 217 theory 2, 5, 34, 44–45, 51–52, 72, 76–78, 82–83, 85, 87, 101, 103–04, 117, 119, 122–23, 126, 129, 152, 156, 158, 162–63, 173–77, 181, 185–88 theory, folk 173–74, 177, 181, 188 theory, Relevance 6, 44–45, 91–93, 99, 101, 103–05, 107, 109, 111–13 theory, semantic 1, 2, 6, 8, 13, 16, 25, 28, 76–77, 123, 133, 185 theory, pragmatic 5, 31, 51–52, 209 theory, speech act 5, 51–55, 59–60, 71, 103, 126 truth 3–4, 13–18, 27–28, 31, 58, 80, 117–18, 121, 123, 125, 130–31, 141–42, 144–47, 149, 150–52, 155, 157–58, 162–64, 166, 174, 178, 187–88, 194, 206, 221–22 truth conditions 1, 3–5, 7, 16, 27, 34, 37–40, 47, 58, 63–64, 66, 69, 96–97, 120, 128, 141–44, 149, 153, 156, 159–65, 193, 215–17, 221–24
231
truth-conditional 3, 4, 6–8, 31, 33–34, 117–18, 120–23, 125, 128, 131, 141–66, 193–94, 200, 210, 221 truth-evaluable 1–2, 4–5, 31–32, 141, 144–45, 217, 223 truth value 13–14, 16, 24–25, 27, 31–32, 63, 79, 96–97, 162, 172, 185 unity 8, 18, 213–25 utterance 1, 5, 7–8, 13–16, 20–21, 23–25, 27–28, 31–34, 36–47, 52–53, 58–59, 63, 65–66, 75–81, 85, 88, 95–100, 103–07, 141–66, 181, 183, 193–94, 213–25 utterance, literal 51–53, 55, 58–60, 62–64, 66–69, 72, 145, 148, 166 vagueness 15, 20, 45, 67, 83, 125, 171, 188–89, 216 validity 4, 16–17, 27–28, 53, 78, 131 variables 8, 26, 32, 41, 67, 128, 191–210, 221 what is said 1–2, 4–5, 31, 33–36, 41–47, 69, 141–45, 149, 159, 165–66, 215–16, 223