Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Linguistic understanding and semantic theory
(USC Thesis Other)
Linguistic understanding and semantic theory
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
LINGUISTIC UNDERSTANDING AND SEMANTIC THEORY by Brian Bowman _______________________________________________________________________ A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (PHILOSOPHY) August 2012 Copyright 2012 Brian Bowman In loving memory of my grandmother, Elsyee Langstroth. ii Acknowledgments Special thanks to my committee members: First, Scott Soames, whose seemingly boundless dedication, enthusiasm, and insight was more than anyone could hope for in an advisor. This dissertation would have been much poorer without his numerous contributions. George Wilson, who offered sound guidance and encouragement, especially when my ideas were first germinating in a rather inchoate form. Barry Schein, who challenged me with penetrating and difficult questions that helped sharpen the view. This brief paragraph hardly does justice to, or constitutes proper thanks for, the contributions made by my committee members, but it will have to do. Thanks also to the many individuals who facilitated my intellectual development and, directly or indirectly, contributed to this work (in no particular order): Jeff King, James Higginbotham, Robert Cummins, Michael Jubien, Karen Neander, Michael Glanzberg, Tomis Kapitan, Geoff Georgi, and Lewis Powell. This is by no means a complete list and I've probably forgotten some that deserve mention. iii Table of Contents Dedication ii Acknowledgments iii Abstract vi Preface xi Chapter One: Summary of the Project 1 Introduction 1 Clarifying the Negative Part of the Project 1 Semantic Theory and Understanding 4 Understanding and Competent Use 18 The Basic Picture 25 The Variable Standard of Understanding 27 An Alternative Semantic Framework 35 Conclusion 40 Chapter Two: Why Semantic Theory Can't Be a Theory of Understanding I: Names 41 Introduction 41 Preliminary Considerations 41 The Central Argument 48 Conclusion 64 Chapter Three: Referential, Recognitional, and Inferential Competence 65 Introduction 65 Referential Competence 66 Recognitional Competence 79 Inferential Competence 84 Conclusion 88 Chapter Four: Why Semantic Theory Can't Be a Theory of Understanding II: Predicates 89 Introduction 89 Preliminary Considerations 90 The Central Argument 98 Conclusion 121 iv Chapter Five: The Limits of Mere Referential Competence 123 Introduction 123 Thesis I 124 Theses II and III 130 Conclusion 143 Chapter Six: How Recognitional and Inferential Competence Contribute to Understanding 144 Introduction 144 Understanding Names 144 Understanding, Discourse, and Predicates 149 Conclusion 162 References 163 v Abstract This book defends a theory of linguistic understanding, or what is alternatively called linguistic competence or semantic competence; it also defends a view about how linguistic understanding is related to semantic theory. The central target for a theory of understanding is to answer the question, “What is it to understand and competently use a language?” It is natural to think that understanding a language is a matter of knowing what various expressions in the language mean. Since semantic theory is a theory of what various expressions in a language mean, it is a natural and widely held view that linguistic understanding is simply a matter of knowing something like a semantic theory. According to this view, we essentially arrive at the best theory of understanding by pointing to the best semantic theory and simply saying that to understand and be competent with a language is to know that. Thus, the theory of understanding has been thought by many to be a trivial offspring of semantic theory, and the theory of understanding is generally passed over as an independent topic of inquiry. I argue, however, that linguistic understanding cannot simply be a matter of knowing a semantic theory. I develop a theory of understanding distinct from semantic theory, explain how the two are related, and why both are essential. Four major threads run through the chapters. First is a proposal about how to understand competent use. A theory of understanding must begin with a characterization of the linguistic abilities we are trying to account for. Competence consists in at least three abilities: referential competence, recognitional competence, and inferential vi competence. Referential competence is the ability to use a word with its meaning – using it in various sentences to make assertions, commands or promises, to ask questions, and form linguistically mediated thoughts (I will henceforth only mention the case of assertion). For example, a speaker is referentially competent with the name ‘Felicity’ if she can use it to refer to Felicity and thus, among many other things, use the sentence ‘Felicity is a horse’ to assert that Felicity is a horse. It is widely accepted in the literature that we can successfully use certain words, such as names and some natural kind terms (i.e. ‘elm’), without knowing much about their reference or meaning. I argue that the relevant sense of successful use in these cases is referential competence. I extend this idea by arguing that mere referential competence is possible for a much greater variety of expressions, including some that carry a higher standard of understanding (‘red’, ‘bachelor’, ‘horse’ etc.). Such words do carry a higher standard of understanding, but this doesn’t mean mere referential competence isn’t possible, only that there is a larger gap between mere referential competence and a “normal” level of understanding. The gap is filled in my view by recognitional and/or inferential competence. A speaker is recognitionally competent with a word to the extent that she can recognize whether or not it applies in a given circumstance – i.e. to the extent that she can correctly determine whether or not ‘red’ applies to various objects. A speaker is inferentially competent with a word to the extent that she can see how its meaning relates to the meanings of other words in the language. Knowing to infer ‘Felicity is a horse’ from ‘Felicity is a mare’ is an instance of inferential competence with ‘mare’ (and ‘horse’). A key point about the abilities is that they suggest a highly graded picture where vii understanding can range from minimal to robust. Mere referential competence, accompanied by no significant measure of the other abilities, is the most minimal level of genuinely linguistic understanding possible. Robust understanding comes with the addition of a significant measure of recognitional and/or inferential competence. The second thread is negative, arguing that semantic theory alone can’t provide a theory of understanding. I argue for this primarily on the grounds that the notion of “knowing” a semantic theory isn’t suitably graded to account for competent use. One speaker can be more competent with a given expression than another speaker. But if we ask semantic theory to give us a theory of understanding, it can only give us the same claim for both speakers – namely, that each knows the meaning of the expression, and this clearly gives us no way to explain how they differ in competence. The third thread connects competence to the theory of speech acts. There are two questions about the extent to which mere referential competence is possible. First, are there classes of words for which mere referential competence leads to serious breakdown? I argue that this is so – for example, if a speaker doesn’t understand what inferences are licensed by a logical term such as ‘or’, her attempts to use the term will be defective. Second, how many words can a speaker be merely referentially competent with – in a sentence or an entire language – before breakdown ensues? I argue that mere referential competence is only secure against a rich background of understanding for most other words. A speaker typically will not use more than one or two minimally understood words in a sentence, and can, in many cases at least, do so successfully. But attempts to use much more than one or two words lead quickly to defective speech acts. viii Even in these cases, however, I argue that the defect isn’t always a complete breakdown of the speech act, so that it doesn’t count as an assertion at all. I think the defect can instead be a violation of a necessary condition on fully rational action, and that the speech act may count as an assertion even though it is irrational in a certain sense. Defending this view involves a graded theory of speech acts – since understanding is graded, the rationality of speech acts is too, and similarly admits of more or less, and different kinds or aspects. The last thread is a defense of the claim that my proposed theory of understanding is a requisite part of an overall theory of language. The strategy here is to show how the theory of understanding is required in order to explain conversational communication. This involves a shift from individual assertions to the broader context of conversations. Since conversational competence is graded, this helps explain why linguistic competence is too. A conversation is a set of assertions coordinated with one another, and with the non-linguistic activities of its participants. So in order for a speaker to fully understand a conversation, she must not only be able make assertions, but must understand how the assertions in a conversation relate – both to each other (secured by inferential competence), and to things that agents are able to recognize in the world (secured by recognitional competence). I argue that individual words are normatively associated with information about the word’s content (i.e. ‘horse’ is normatively associated with information about horses). This information is part of what must be known in order to fully understand a word, and supplies pragmatic presuppositions in a conversational context that help speakers arrive at a common understanding of the two kinds of ix conversational relation they must track. Hence, what it is to understand and competently use a language, or even know the meaning of a word, involve a heavily pragmatic dimension. The overall view I defend is thus one according to which the theory of understanding belongs on the pragmatic side of the picture. The theory of understanding works in tandem with semantic theory, via a story about the semantics-pragmatics interface, to provide an overall account of linguistic communication. x Preface Two theoretical projects have traditionally claimed to offer theories of linguistic meaning. The first is semantic theory. Guided by the thought that language is a medium of representation carrying information about the world, it aims to discover how language does this. On this approach, the meaning of a linguistic expression consists in its association with a representational content. The second is a theory of linguistic understanding. Guided by the thought that shared knowledge underlies a speaker’s ability to use language competently or felicitously as others do, it aims to determine the nature of this knowledge. On this approach, the meaning of an expression can include whatever knowledge is necessary for its competent use. 1 A popular view holds that these two projects come to the same end. According to this view, knowledge of an expression’s representational content suffices for, and explains, its competent use. 2 Those denying this view typically marginalize one project, 1 For ease of exposition, I am using ‘semantic theory’ to designate representational theories of content. ‘Semantic theory’ is often used to designate theories of meaning broadly construed, and thus doesn’t mark the distinction between representational theories and theories of competent use. 2 Asher (1988, p. 1) writes, “A ‘received view’ underlying much work in the philosophy of language is that a semantic theory must also count as a theory of semantic competence, a systematic account of what competent speakers of a language must know to use the language correctly.” A version of this thesis is held by Davidson (1967) and his contemporary followers, such as Higginbotham (1992). Here the claim is that knowing what a theory of truth for a language says about an expression suffices for understanding the expression. xi by outright rejection or simple neglect. 3 This is a rough characterization of a complex literature, but the overall consensus is that if semantic theory can’t serve as a theory of understanding, then so much the worse for one or the other. In my view this consensus is unwarranted. A central project running throughout the chapters of this book is a defense of the following two theses: (I) Semantic theory can’t serve as a theory of understanding. (II) Semantic theory and a theory of understanding, though distinct, are not only complementary, but mutually vital to an overall theory of language. In defending (I) and (II), I primarily focus on proper names and simple predicates, and assume a semantic theory according to which they are directly referential. On this view, the meaning of an expression is identified with its worldly reference – the meaning of a name is the object it is used to designate, and the meaning of a predicate is the property instantiated by the kind of thing it is used to talk about. In my view, this is the best semantic framework going for expressions of this kind, because it gives the best account 3 This view is rejected by Foster (1976) and Soames (1989) (among others). Soames extends his negative thesis to include semantic theory conceived in the framework of structured propositions, arguing that such a theory does not account for semantic competence. Those who reject the view in favor of a theory of understanding include Dummett (1996a; 1996b), who argues that a theory of truth – and a fortiori, a theory incorporating direct reference – cannot serve as a theory of understanding, and must be replaced by a theory capable of explaining the practical abilities involved in language use (in his view, a theory of verification and falsification – see esp. Dummett (1996b, p. 74)). Those who reject the view in favor of semantic theory include Katz (1990; 1996) and Soames (1984). They argue that the target of linguistic theory, including semantic theory, should be the description of linguistic facts, such as the fact that an expression stands for its meaning or content, but that such facts are not directly facts about individual speakers (i.e. their abilities or psychological states). Soames differs from Katz in that Katz claims semantic theory isn’t even an empirical discipline, while for Soames it is. On Soames’ view, semantic theory is grounded in facts about individual speakers, but is a considerable abstraction away from this, and should not be literally construed as the content of a speakers’ psychological state that underlies and explains her competence. For criticism of Katz and Soames (1984), see Higginbotham (1983) and Antony (2003). The view I defend here is compatible with that of Soames, since I agree that semantic theory cannot serve as a theory of competence, though I offer different reasons for this. Compared to Soames, I am merely taking the next crucial step – namely, sketching the outlines of a theory of semantic competence that is vitally needed as a supplement to semantic theory in an overall account of linguistic communication. xii of their representational content, but it isn’t my aim to defend this claim. The only sense in which I defend this framework is by showing how it can be supplemented by a plausible theory of understanding to provide an appealing overall account of linguistic competence and communication. I don’t take thesis (I) to be a point against the framework. I am skeptical that thesis (I) can be denied for alternative frameworks. So I am skeptical that we can adjudicate between competing semantic frameworks by the extent to which they can serve as theories of understanding. Instead, I think we can adjudicate between competing frameworks by the extent to which they are elegantly compatible with a plausible supplementary theory of understanding. While I discuss alternatives briefly, it isn’t a central aim of this book to defend direct reference against competing frameworks. Again, my aim is to show how the theory of direct reference is an elegant companion to a plausible theory of understanding. For the most part, I leave it up to others to show how their favored brand of semantics can either serve as a plausible theory of understanding, or – what I think will be the only viable defense – show how their favored brand of semantics is a friendly companion to a plausible theory of understanding. Another central project running throughout the chapters of this book is to defend a plausible conception of competent use. A theory of understanding must begin with a characterization of the linguistic abilities we are trying to account for. Competence in my view consists in at least three abilities: referential competence, recognitional competence, and inferential competence. Referential competence is the ability to use a word with its meaning – using it in various sentences to make assertions, commands or promises, to ask xiii questions, and form linguistically mediated thoughts. For example, a speaker is referentially competent with the name ‘Felicity’ if she can use it to refer to Felicity and thus, among many other things, use the sentence ‘Felicity is a horse’ to assert that Felicity is a horse. A speaker is recognitionally competent with a word to the extent that she can recognize whether or not it applies in a given circumstance – i.e. to the extent that she can correctly determine whether or not ‘red’ applies to various objects. A speaker is inferentially competent with a word to the extent that she can see how its meaning relates to the meanings of other words in the language. Knowing to infer ‘Felicity is a horse’ from ‘Felicity is a mare’ is an instance of inferential competence with ‘mare’ (and ‘horse’). While I discuss these abilities throughout the book, they take center stage in chapter three. A key point about the abilities is that they suggest a highly graded picture where understanding can range from minimal to robust. Mere referential competence, accompanied by no significant measure of the other abilities, is the most minimal level of genuinely linguistic understanding possible. Robust understanding comes with the addition of a significant measure of recognitional and/or inferential competence. We are already familiar with the idea that, for names and some natural kind terms, it is easy to pick up the word conversationally and in some sense understand or be a competent user of it (this point was made most famously by Kripke (1980) and Putnam (1996)). I argue that the relevant sense of competence for these cases is referential competence. I then extend this idea by arguing that referential competence can be had cheaply for a much greater variety of expressions, including those that have traditionally been thought to xiv carry a higher standard of understanding (such as ‘red’, ‘sister’, ‘bachelor’, etc.). Such words do carry a higher standard of understanding, but this does not mean that mere referential competence isn’t possible, only that there is a larger gap between mere referential competence and what counts as a “complete” or “normal” level of understanding. A speaker can be merely referentially competent with ‘bachelor’, it’s just that this constitutes a much more impoverished level of competence than does mere referential competence with ‘elm’. The contrast between ‘bachelor’ and ‘elm’ shows that there is a variable standard of understanding for different words. I don’t think this variable standard is an explicit linguistic convention, but is rooted in culturally shared interests. I argue for (I) primarily on the grounds that the notion of “knowing” a semantic theory isn’t suitably graded to account for competent use. One speaker can be more competent with a given expression than another speaker. But if we ask semantic theory to give us a theory of understanding, it can only give us the same claim for both speakers – namely, that each knows the meaning of the expression, and this clearly gives us no way to explain how they differ in competence. In detail, of course, the argument is more complex than this – it begins in chapter one, and takes center stage in chapters two and four (devoted to a discussion of names and predicates respectively). Chapter one is an introductory overview of the whole project, but here I lay the groundwork for defending (II) by forging a plausible link between semantics and understanding. I show how the theory of direct reference naturally motivates a distinction between (i) knowing the semantic rule associating a content with an expression, and (ii) xv knowing things about the content that aren’t part of understanding the rule, but are commonly associated with the word, exploited in communication, and thus properly conceived as pragmatic presuppositions of a communicative context. This distinction paves the way to explaining the graded nature of linguistic competence. One speaker may be more competent with an expression than another speaker because she knows more about its content. Knowing more about a given content goes along with having a richer perspective on it. But semantic theory itself has nothing so say about perspectives on content, so we must look to a theory of understanding for this. I then forge the connection to linguistic competence through the plausible idea that having a higher level of recognitional or inferential competence with an expression is a manifestation of having a richer perspective on its content, or knowing more about its content. Thus, the theory of understanding can be understood to supply, in companion to the theory of reference supplied by semantics, a graded, pragmatic theory of sense – where a theory of sense is traditionally understood as a theory about the “modes of presentation” by which speakers grasp or think about reference. Thinking in terms of the abilities raises a question about the extent to which mere referential competence is possible, an issue I tackle in chapter five. In actual practice, a speaker will typically have mere referential competence for a relatively small set of words in her repertoire, and will typically not use more than one or two of these words in any sentence she utters. So in the normal case, a minimally understood word will be standing against a rich background of understanding. It is plausible to expect that there are limits to how many words a speaker can be merely referentially competent with – in a xvi sentence or an entire language – before some kind of breakdown ensues. Also, there are some words for which the consequences of mere referential competence are particularly severe; in these cases the lack of recognitional and inferential competence is apt to lead rather swiftly to breakdown, even when the word stands against a rich background of understanding. Is it possible, for example, to be referentially competent with a logical term such as ‘or’ without understanding the inferences it licenses? I argue that breakdown does in fact ensue in such extreme cases, though the breakdown isn’t always a complete breakdown of the utterance (so that it doesn’t count as an assertion at all). In some cases the breakdown is a violation of a necessary condition on rational utterance, and the utterance counts as an assertion even though it is defective in a certain sense. Defending this view, leads, in the chapter, to an account of what it is for linguistic utterances to count as fully rational acts. Due to the fact that understanding is graded, the rationality of an utterance is also graded, and similarly admits of a distinction between different kinds or aspects. Finally, chapter six concludes the defense of (II) by showing how the proposed theory of understanding is a requisite part of an overall theory of language. The strategy here is to show, through a series of examples, how the theory of understanding is required in order to explain conversational communication. This involves a shift from individual assertions to the broader context of conversations. Since conversational competence is graded, this helps explain why linguistic competence is too. A conversation is a set of assertions coordinated with one another, and with the non-linguistic activities of its participants. So in order for a speaker to fully understand a conversation, she must not xvii only be able make assertions, but must understand how the assertions in a conversation relate – both to each other (secured by inferential competence), and to things that agents are able to recognize in the world (secured by recognitional competence). I argue that individual words are normatively associated with information about the word’s content (i.e. ‘horse’ is normatively associated with information about horses). This information is part of what must be known in order to fully understand a word, and supplies pragmatic presuppositions in a conversational context that help speakers arrive at a common understanding of the two kinds of conversational relation they must track. Hence, linguistic competence and what it is to know the meaning of a word involve a pragmatic dimension. The resulting conception of the relation between semantics and pragmatics is one according to which pragmatics makes an important contribution to linguistic competence and knowledge of meaning, and explains the truth of what might otherwise seem paradoxical – understanding an expression requires much more than knowing the semantic rule that determines its meaning. xviii Chapter One Summary of the Project Introduction The purpose of this chapter is to give a summary overview of the project. This will hopefully prevent us from losing sight of the forest as we begin to inspect the trees in later chapters. In next section, I clarify the negative part of the project, and in the following sections I sketch the positive view. Clarifying the Negative Part of the Project One of the central aims of this book is to defend the claim that semantic theory can’t serves as a theory of understanding (i.e. thesis (I) from the preface). This is the negative part of the project. The aim in this section is to clarify the sense in which I deny that semantic theory can serve as a theory of understanding. The argument for this will partially unfold in subsequent sections, and culminate in chapters two and four. Semantic theory and linguistic understanding are commonly tied together by the thought that understanding is knowledge of meaning. The thought is that competent language use is secured by knowing the meaning of a language. Since semantic theory states the facts about meaning, semantic theory can provide an account of understanding by describing the content of the knowledge which is thought to explain competent language use. Call this popular conception of semantic theory and its relation to understanding (PC). 1 (PC) A semantic theory can serve as a theory of linguistic understanding. The motivating thought behind (PC) is quite plausible, but I’m going to challenge this popular conception. I do not deny that competent speakers have knowledge of meaning, or that a semantic theory can provide at least a partial description of this knowledge. Let’s consider some theses which clarify the view stated by (PC). I accept (PC1), which I take to be the weakest possible understanding of (PC). (PC1) Some of the facts about meaning stated by an adequate semantic theory for language L are among the facts known to competent speakers about the meaning of L. However, if all that is meant by (PC) is (PC1), then I don’t think the view is particularly interesting. I deny any stronger understanding of (PC). Most importantly, I deny (PC′) A semantic theory can be adequate, explaining everything we want a semantic theory to explain, while also serving as an adequate theory of linguistic understanding, and thus also explaining everything we want a theory of linguistic understanding to explain. As I shall argue, the explanatory constraints on semantic theory are distinct from the explanatory constraints on a theory of understanding, and it is simply not possible to have a semantic theory which satisfies both sets of constraints. While (PC′) is the primary target, there are other ways of understanding (PC) which I also deny. First, it is possible to have a perfectly adequate semantic theory which has more information about meaning than some competent speakers know, and hence I deny 2 (PC2) All of the facts about meaning stated by any adequate semantic theory for language L are among the facts known to all competent speakers about the meaning of L. Second, there are things which competent speakers know about meaning, which are required for their competence with the language, but which may not, and in some cases should not, be stated by semantic theory. Hence, I also deny (PC3) All of the facts known to competent speakers about the meaning of language L are among the facts about meaning stated by any adequate semantic theory for L. Finally, bringing in the idea that the knowledge of meaning is supposed to explain competent language use, I deny (PC4) All of the facts stated by an adequate semantic theory are facts which serve to explain linguistic competence. I contend that, in some cases at least, the fact that a speaker knows what semantic theory states about the meaning of a word, and that she is minimally competent with it, are coeval facts about the speaker each having a similar explanatory basis, so that one fact isn’t explained in terms of the other. This common explanatory basis, I shall argue, involves, among other things, propositional knowledge which is more minimal and schematic than what is stated by semantic theory. So there is a sense in which semantic theory, as a theory of understanding – i.e. as a theory of the knowledge underlying and explaining competence – actually attributes too much knowledge to speakers, though it does describe knowledge which competent speakers have (modulo the point made above with (PC2)). 3 While I will give some reasons for doubting (PC) in this chapter, the full case against (PC′) and (PC2) – (PC4) will be taken up in chapters two and four. As we shall see, the point that a semantic theory can have in one sense more information than competent speakers know and in another sense less information is due to the point that linguistic understanding is graded, and can range from very minimal to robust. It will turn out that speakers who have the most minimal sort of understanding know less than what is stated by an adequate semantic theory, while speakers who have more robust understanding know more than what should be stated by semantic theory. Since competent speakers can know both less and more than what is appropriate for semantic theory to state, semantic theory cannot serve as a theory of understanding. Semantic Theory and Understanding Another central aim of this book is to defend the claim that semantic theory and a theory of understanding can be made perfectly complementary (i.e. thesis (II) from the preface). For this it is required that we forge a link between the two, which will explain how the two are intimately related. Linguistic understanding is graded, but we don’t get anything appropriately graded out of semantic theory to explain it, so we stand in need of a graded notion to link the two. This will come by way of the distinction I motivate between (i) knowing the semantic rule associating a content with an expression, and (ii) knowing things about the content. Although the notion of knowing a semantic rule associating content and expression isn’t graded, knowing things about the content is, and this will 4 provide the requisite link. So the goal for this section is to explain how the notion of knowing things about content is related to understanding. In my view, the best semantic framework should incorporate the notion that some names directly refer to objects and some predicates directly express properties. One advantage of this framework is that it nicely accounts for the point famously made by Kripke in Naming and Necessity that some names “rigidly designate” objects (call these proper names). In the case of predicates, it elegantly captures the attractive idea that some predicates directly stand for a certain kind of thing without expressing structurally complex meaning (call these simple predicates). 1 I will assume this framework here, only defending it by showing how it can be supplemented by a plausible theory of understanding to provide an appealing account of linguistic competence and communication. 2 1 See Kripke (1980). An expression rigidly designates its (actual) extension just in case it has the same extension relative to every possible world (or circumstance of evaluation) in which it has an extension. In the case of a name, its extension is its referent, so a rigid name has the same referent relative to every possible world in which its (actual) referent exists. Taking the extension of a predicate to be the property it stands for, predicates are also rigid on this definition, but for my purposes this point isn’t interesting (though it’s generally interesting – see Soames (2006), which is a reply to critics of Soames (2002)). To be clear, a predicate having complex meaning is not simply a matter of it contributing a property that might be understood to have parts (in the sense that bachelorhood might be thought to be composed out of the property of being unmarried and the property of being male). For comparison, Bill Clinton occurs as a simple in the unit set {Bill Clinton}. Bill Clinton is a complex object in the sense that he has parts – arms, legs, etc. – but these parts are not parts of the unit set. Similarly, ‘bachelor’ expresses a property that might be complex in the sense that it might be understood to be composed of other properties, but it doesn’t follow from this that ‘bachelor’ contributes complex meaning to the propositions expressed by sentences in which it occurs, since the properties that might compose bachelorhood are not necessarily component parts of the proposition. It is true in my view, however, that a speaker who is fully competent with ‘bachelor’ needs to cognitively articulate the component parts of bachelorhood in some fashion, even though ‘bachelor’ need not be understood as having complex meaning. 2 This kind of semantic theory has for some been the most prominent target of pessimism for any reconciliation with a theory of understanding. Dummett (1996a, pp. 23 – 24), for example, remarks that this kind of theory is “unintelligible” as a theory of understanding. The view I sketch here shows how such a theory can be made intelligible as part of a theory of understanding, though I agree with Dummett insofar as I think that such a theory by itself cannot serve as a theory of understanding. 5 A semantic theory of this kind has axioms such as: (1) ‘Felicity’ refers to Felicity. (2) ‘horse’ expresses H (where H is the property of being a horse). It also includes recursive rules assigning descriptions of structured propositions to sentence types, allowing the derivation of theorem (3) on the basis of (1), (2) and (4). (3) ‘Felicity is a horse’ expresses the proposition <Felicity, H>. (4) For any proper name n and simple predicate F, the sentence ⎡n is F⎤ expresses the proposition <o, P>, where o is the referent of n and P is the property expressed by F. 3 ‘<o, P>’ is a description of the structured proposition that says, of o, and of P, that the former has the latter (i.e. ‘<Felicity, H>’ describes a proposition that says, of Felicity, and of the property of being a horse, that she has it). This is a theory of the representational content of proper names and simple predicates. Theories of representational content are essential to accounts of how speakers are able to use language to assert, think, and communicate. If ‘Felicity is a horse’ is primarily used to assert, believe or communicate <Felicity, H>, it’s plausible that knowing (3) provides most of what one needs to know for this. According to this line of reasoning, semantic theory can essentially serve as a theory of understanding since what it tells us about the representational function of language can also serve as a theory of the knowledge enabling competent use. At best, however, this only shows that semantics can give a partial account of understanding. We haven’t yet said what abilities are involved in competent language 3 For simplicity, (4) doesn’t accommodate contextually sensitive expressions or variables. 6 use, and, despite its initial plausibility, the tenability of the thesis that semantic theory can explain understanding ultimately hinges on whether an agent who knows the semantic theory will have these abilities. I will argue that the relevant abilities extend well beyond the capacity to use a sentence to assert, believe, or communicate the proposition it semantically expresses. What abilities will an agent have on the basis of knowing (1) – (3)? The claim that S knows (1) – (3) entails (1′) – (3′), which attribute equivalent knowledge to S. (1′) ∃x(x = Felicity & S knows that ‘Felicity’ refers to x). (2′) ∃p(p = H & S knows that ‘horse’ expresses p). (3′) ∃x∃p(x = Felicity & p = H & S knows that ‘Felicity is a horse’ expresses the proposition <x, p>). So the claim that S knows (1) – (3) is a claim that S stands in the knowledge relation to propositions which are “singular” in the sense that they are directly “about” an object or property (or both). It’s widely noted that singular knowledge attributions are poor predictors of behavior. The claim that S knows (2) only tells us that S knows, of the property of being a horse, that ‘horse’ expresses it. The claim gives no further information about how S thinks of the property. So the claim is a poor basis for predicting S’s behavior, and it doesn’t help to add that S knows (1) and (3) as well. Furthermore, what a speaker knows about content involves different non- transparent perspectives on the content. A speaker can have two or more perspectives on an object, property or proposition, each of which involves different information about it. A speaker can have two different ways of thinking about Felicity, one of which involves 7 knowing, in this way, that she is a race horse, while the second way involves withholding belief that she is a race horse, and hence not knowing it in this other way. 4 The notion that we understand the world via perspectival modes of apprehension has played an important role in accounts of language and thought going back (at least) to Frege. But what hasn’t been adequately addressed so far is that these perspectives are graded, ranging from minimal to robust depending on how much information they include. What we need from a theory of understanding is an account of how gradations in perspective are connected to gradations in competence. Semantic theory has nothing to say about this. Another widely accepted point about singular propositions is that they can be cheaply entertained. I will assume here that a causal connection between an agent and an object or (instances of) a property, in conjunction with a few other minimal requirements, suffices to entertain singular thoughts about it. 5 But the causal connection can be quite distant, and need not grant the agent much knowledge about the target of her thought. For example, if you are walking along a beach and see a barely discernable object way off in the distance, appearing as nothing more than a dark speck in your visual field, you can nonetheless point to it and ask, “What is that?” Despite your lack of information about the object and meager visual representation of it, you have singled it out in thought, 4 Knowledge and belief are inherently perspectival. This view has been explicitly defended by Salmon (1989). Salmon’s view is that the ordinary binary notion of belief is an existential generalization over a ternary belief relation (designated “BEL”) involving a perspective or way of grasping the proposition believed. The relation between the two is governed by the following principle: A believes p iff ∃x(A is familiar with p by means of x & BEL(A, p, x) (even if A withholds belief in p relative to another perspective so that ∃y(A is familiar with p by means of y & ¬BEL(A, p, y))). Since on this view, the ordinary notions of knowledge and belief are binary, ordinary singular attributions do not give any information about the perspective involved in the agent’s knowledge or belief, though there will always be such a perspective. 5 This kind of view has been defended by Recanati (1993) and Bach (1987) (among others). 8 directly referred to it by means of your demonstration, and put forth a direct query about it. The explanation of your direct cognitive relation to the object in this case will include the point that you have a causal link to it. The ease by which singular thought can be had explains how some expressions, such as names and natural kind terms (i.e. ‘elm’), can be easily acquired in conversation and deferentially used with considerable ignorance of their meaning or reference. 6 It has largely gone unnoticed, however, that this phenomenon extends to other classes of words, including some that carry a higher standard of understanding (‘red’, ‘bachelor’, ‘horse’, etc.). Deferential use is possible for such words, but there is a larger gap between this minimal level of competence and a “normal” level of understanding. For example, suppose that a speaker S is party to a conversation in which Jones assertively utters the sentence ‘Felicity is a horse’, thereby sincerely asserting that Felicity is a horse. This is the first time S has encountered ‘Felicity’ and (by freak linguistic chance) ‘horse’. Further, suppose S only learns from the conversation so far that ‘horse’ is a count noun standing for a kind of animal and that ‘Felicity’ refers to an individual said by Jones to be of this kind. Despite her minimal understanding, S can carry on from here to use ‘Felicity’ and ‘horse’ for herself with their respective meanings. If S is doubtful of Jones’ assertion, she can turn to another speaker and successfully ask, “Is it true that Felicity is a horse?” Or she can successfully report on what Jones has said and what he believes by uttering, ‘Jones said that Felicity is a horse’ and ‘Jones believes that Felicity is a horse’. If S chooses to accept Jones’ testimony, she can report a belief 6 See Kripke (1980) and Putnam (1996). 9 she and Jones share by uttering ‘Jones and I believe that Felicity is a horse’. Or she can simply report her belief more directly by uttering, ‘Felicity is a horse’. Furthermore, her deferential use isn’t limited to parroting Jones. She can use ‘Felicity’ and ‘horse’ in a range of sentences. She can point to some animal and successfully ask, ‘Is that Felicity?’ or ‘Is that a horse?’ Similarly, she can assert that horses are animals by uttering ‘Horses are animals’. 7 Given her minimal level of understanding, it perhaps seems prima facie doubtful that S knows the semantic rules for ‘Felicity’ and ‘horse’. Still, she is using these expressions with their respective meanings, and it would be odd if she could do so without knowing their semantic rules in some sense. And there is good reason to think that she does know their semantic rules. S knows that ‘Felicity’ refers to whatever individual her interlocutors are using it to refer to, and that ‘horse’ expresses whatever property her interlocutors are using it to express. The attribution in (5) will be true (dropping the reference to her interlocutors). 8 7 This does not mean, however, that relative to certain communicative contexts, S’s speech acts involving ‘Felicity’ or ‘horse’ will be entirely non-defective, and it can turn out that communication between her and other speakers with richer understanding can break down in significant ways. I illustrate some of these ways in chapter six. 8 Presumably S knows these terms are governed by standards of use that apply broadly to the community, not just to her interlocutors, so the reference to her interlocutors may not be needed. Even if it is, this doesn’t affect anything I say (so long as her interlocutors are in fact using these terms in accord with the community standards). 10 (5) S knows that ∃x∃p(‘Felicity’ refers to x & ‘horse’ expresses p & ‘Felicity is a horse’ expresses the proposition <x, p>). 9 Furthermore, S is part of causal chains of communication connecting ‘Felicity’ and ‘horse’ back to Felicity and (instances of) the property of being a horse. 10 She is also using these expressions against a rich general background of understanding and communicative experience, including common “animal talk” using similar names and predicates in the same kind of construction (i.e. ‘Rex is a dog’). Given this, we have sufficient reason to export the restricted quantifiers in (5), yielding (5′). 11 (5′) ∃x∃p(‘Felicity’ refers to x & ‘horse’ expresses p & S knows that (‘Felicity’ refers to x & ‘horse’ expresses p & ‘Felicity is a horse’ expresses the proposition <x, p>)). (5′) entails that S knows (1) – (3), so S knows everything semantic theory says about ‘Felicity’, ‘horse’ and ‘Felicity is a horse’. 12 The semantic rules can be minimally learned because the rules are minimal. The rule for ‘horse’ only says, of the property of being a horse, that ‘horse’ expresses it, and it’s easy to learn this much conversationally as S does. To illustrate how startling this 9 The linguistic practice of the community (using ‘Felicity’ to refer to Felicity, ‘horse’ to stand for horses, and ‘Felicity is a horse’ to say of Felicity that she is a horse) makes it the case that ∃x∃p(‘Felicity’ refers to x & ‘horse’ expresses p & ‘Felicity is a horse’ expresses the proposition <x, p>). In the conversation, S gains reliable (though not infallible) evidence of this fact, and this suffices for her to know it. 10 Here I am obviously alluding to the causal theory of names sketched by Kripke (1980), and suggesting a similar story for predicates. 11 While I don’t advocate completely liberal exportation of quantifiers outside the scope of attitudinal verbs, I think the conditions on exportation are minimal, and similar to the minimal conditions sufficient for singular thought. 12 For the purpose of discussion, I’m making the simplifying assumption that any given name or predicate only stands for one object or property. This assumption avoids issues, having to do with ambiguity and the individuation of expressions, tangential to the issues raised here. 11 result is, it’s helpful to consider an example involving unfamiliar words from a first person perspective. Suppose I tell you that Tanya is a fossa by assertively uttering, ‘Tanya is a fossa’ (and that ‘fossa’ and (my use of) ‘Tanya’ are unfamiliar to you). Assuming I am a reliable user of the relevant words, you learn, through deference to my usage, that ‘Tanya’ is a name for a certain individual, and ‘fossa’ is a count noun standing for a certain kind. You can go on from here to deferentially use these words with their meanings. If you assertively utter ‘Tanya is a fossa’, you assert the same proposition I do, and are held to the same attendant standards of truth and falsity I am. If it turns out that Tanya isn’t a fossa, we are equally culpable for having spoken falsely. You obviously haven’t learned much about the relevant expressions, but you know what semantic theory says about them. You know that ‘Tanya’ refers to Tanya, that ‘fossa’ expresses the property of being a fossa, and that ‘Tanya is a fossa’ says of Tanya that she is a fossa. It’s natural to object at this point that you don’t know what ‘fossa’ means. However, if knowing what a word means is a matter of understanding it and being able to use it competently, this isn’t an objection to the view. According to the view I’m defending, semantic theory doesn’t give a complete account of what it is to know the meaning of a word in this sense, precisely because one can know what semantic theory says about it yet fall short of understanding it. The felt sense in which you don’t know what ‘fossa’ means is, I think, due to the fact that you don’t know what fossas are (i.e. 12 you don’t know what the kind in question is). 13 But it’s possible for (6a) to be true while (6b) is false. (6a) You know that ‘fossa’ expresses F (where F is the property of being a fossa). (6b) You know what fossas are. For comparison, suppose you and I are put in a drug induced slumber, and later wake in a windowless room we have never seen before. I could truthfully say to you, “You know that you are here.” This is true because you have singular knowledge of your present location – you know, of the room you are in, that you are there. But I couldn’t truthfully say to you, “You know where you are.” Intuitively, to know where you are, you need to know something about how your present location relates to other places in ways that might facilitate getting from one to the other. Something similar can be said of (6a) – (6b). You can know, of the property of being a fossa, that ‘fossa’ stands for it, without knowing what fossas are. To know what fossas are, at minimum you need to know something about how fossas are similar to, or different from, other kinds of things. 14 How much you need to know is notoriously context sensitive – in some contexts quite a lot, in others very little. Your understanding of ‘fossa’ is impoverished, not because you don’t know what semantic theory says about it, but because you don’t know what fossas are. 13 Stanley (1999) writes, “[A speaker] understands ‘elm’ if and only if: [The speaker] knows what elms are, and knows that ‘elm’ refers to that.” I take it the view here is that understanding ‘elm’ requires a) knowing that ‘elm’ stands for elms, and b) knowing what elms are. Insofar as Stanley uses this point in defense of semantic descriptivism, he seems to think that semantic theory should say something about what elms are, so this may be a point where we disagree. 14 There is clearly a significant difference between know that constructions such as (6a) and know what constructions such as (6b), but it would take us too far astray here to determine its semantic (or pragmatic) source. For a discussion of know how constructions that has some parallels to my brief sketch of what it is to know what a word means, see Stanley and Williamson (2001). 13 Returning to speaker S, she might know what horses are (having some experience with them), but her competence with ‘horse’ is similarly impoverished because she fails to connect what she knows about horses with ‘horse’. But how do we characterize her impoverished state? I will have more to say about this in chapter six, but here is a sketch. In normal conversations about horses, more information about horses is exploited than is captured by the semantic rule for ‘horse’. In order to hold one’s own in such conversations, one has to know that horses have four legs, a tail, are herbivores, run faster than humans, are usually domesticated and ridden by humans, etc. Perhaps not all of this is required, but knowledge of this kind is presupposed, or “taken as given” in such conversations. Because S doesn’t associate this information with ‘horse’ as normal speakers do, she won’t know this information is being presupposed when she engages others using ‘horse’. There’s an important connection between the notion of a “normal” conversation and being “normally” or “fully” competent with an expression. Competence with ‘horse’ involves knowing what horses are and associating this with ‘horse’. Since knowing what horses are is context sensitive, there’s a sense in which being competent with ‘horse’ is too. But the notion of a “normal” or “everyday” conversation involves a range of regularly recurring types of conversational context that any mature speaker is expected to hold their own in. Thus, there’s also a notion of a speaker being “normally” or “fully” competent with ‘horse’ since she knows what horses are relative to this range of contexts. 15 15 Stanley (1999) makes the point that understanding a word is context sensitive, but doesn’t make the connection between normal understanding and normal conversational contexts. 14 If I walk up to a stranger and strike up a conversation in English, I assume this person is a normal speaker of English, and this assumption cashes out as an assumption that this person will be able to hold their own conversationally on a variety of everyday topics. I would assume this person knows what horses look like and associates this information with ‘horse’, so it would be natural to walk up and ask, “Did you see a horse pass this way?” It would be oddly uncharitable to first ask, “Do you know what horses look like?” By contrast, it would be inappropriate to strike up an unprompted technical conversation about horse physiology. So there’s a limited range of worldly knowledge about horses that contributes to competence with ‘horse’. Being normally or fully competent with ‘horse’ is partly a matter of knowing all (or most) of this information. Knowing what horses look like contributes to competence with ‘horse’, but knowing obscure facts about horse physiology does not. The boundary between knowledge that contributes to competence and knowledge that doesn’t is no doubt vague, but nonetheless there is such a boundary, and it’s grounded in the fact that there are regularly recurring kinds of conversations any speaker of the language is expected to hold their own in. The relation between this worldly knowledge and the word involves a norm, but the norm doesn’t stem directly from commitment to following a semantic rule. Given the kinds of everyday talk about horses we have, there is a norm to the effect that a speaker needs to know what horses look like, and associate this with ‘horse’. What makes this a norm, rather than a widely shared belief about something everyone is thought to know, is that it should be known since it’s relevant to, and widely presupposed in, everyday conversational exchanges that any speaker is likely to have encountered before and likely 15 to encounter again. Not only should a speaker know this information, but he should know that it’s widely presupposed, and therefore know that in virtually any conversational context, others will know and expect that he knows too. This norm depends, in part, on the fact that humans have a common range of interests that dominate much of discourse. Norms exist to provide coordinated means of reaching goals that arise from common interests. Our common interests determine shared perspectives of our world – i.e. the various objects, properties and their relationships we care about. In the case of language, certain norms serve to facilitate discussion of important and recurring topics by providing standards for framing discourse. So competence with a language is in part relative to goals, practices, and interests of its speakers which infiltrate from outside the language. 16 Imagine a scenario in which humans don’t interact much with horses and don’t care much about them (considering them a relatively obscure wild species). Humans still use the word ‘horse’ to refer to horses, and thus speak the same dialect of English we do, 17 but the standard of understanding for ‘horse’ is lower (it might be similar to ‘elm’ in this regard). When humans engaged in normal everyday conversations about horses, not much knowledge of horses would be presupposed. Most speakers might have only the vaguest idea of what horses look like, but this wouldn’t be a significant impairment to their understanding of ‘horse’ (as is arguably the case for us with ‘antelope’). It’s a contingent fact that we have 16 Thanks to David Braun for helping me see this point about my view. 17 This isn’t entirely correct since we have a host of horse-related terminology (i.e. ‘bridle’), so consider the sub-dialect of actual English including ‘horse’, but none of the related terminology, and note that this would be the same dialect of English spoken in the imagined scenario. 16 fairly rich knowledge of horses and exploit this knowledge for conversational purposes. We need to have more knowledge than our imagined compatriots to be fully competent with ‘horse’, but this knowledge isn’t exactly knowledge of English. In the imagined scenario most speakers don’t have this additional knowledge, and aren’t required to, yet speak the same language we do. The difference between us and them is that they bring a different set of shared interests, and hence a different set of presuppositions, to the conversational table. We thus need to distinguish between two kinds of norms that govern language use. There are norms stemming directly from our shared commitment to conventional rules of use, and there are norms stemming from extra-linguistic facts about our shared interests. Our “knowledge of meaning” rides on these two beasts, so a complete account of meaning and how it figures in language use requires two distinct theories. A theory of understanding must admit degrees since competence admits degrees, involving a range of knowledge about a given content that one can have more or less of. The natural way to think about this additional knowledge of content is that it plays the role of background presuppositions in conversational contexts. Hence, the theory of understanding fits on the pragmatic side of the picture. It is of course nothing new to be told that semantics must be supplemented by pragmatics to give an overall theory of communication. The view I’m defending tells a story about the semantics-pragmatics interface where much more work is being done on the pragmatic side, involving a much richer set of presuppositions, than previously 17 supposed (i.e. by the standard “Gricean” view). But this point is recognized by others. 18 What’s new or surprising about the view I’m defending is that pragmatics makes an important contribution to linguistic competence and knowledge of meaning. These notions have traditionally been thought of as only involving specifically linguistic knowledge. But on the view I’m defending, they partly involve worldly knowledge of objects and properties (and the propositions composed out of them). Understanding and Competent Use In the previous section, we saw how two speakers might differ in their level of competence with a give expression even though both know the semantic rule associating expression and content. And I argued that the way to explain the difference between them is by appeal to a difference in what they know about the content, or the different perspectives they have the content. This notion of perspectives on content gives us the requisite graded notion by which we can begin to account for the graded nature of linguistic understanding. However, the theory of understanding must be grounded in a conception of competent use – i.e. in a conception of the abilities requisite for being competent with different expressions in a language. Hence, the link we are forging between semantics and understanding isn’t complete until we relate this notion of having perspectives on content to competent use. In this section, I briefly describe my proposal about how to understand competent use, and motivate the idea that the relevant abilities are manifestations of perspectives on content. 18 See Soames (2009a; 2009b; 2010), Bach (1987; 1994), Recanati (2004). 18 The turn toward abilities is an important one. The theory of syntax has made remarkable progress in the last 50 years, and this is largely due to Chomsky’s methodological insight that a theory of our knowledge of grammar should answer to facts about what speakers can do – i.e. the ability to produce sentences and make felicity judgments about strings of words. 19 A theory of our knowledge of meaning should be similarly grounded. This idea is not new. Semantic theory is grounded in our shared interpretive dispositions towards sentences across contexts, and I’m not proposing a revision on this front. 20 I am, however, arguing that other abilities are relevant to our knowledge of meaning. Our shared interpretive dispositions towards sentences ground claims about what the semantic content of an individual expression is, and this is all that is required for semantic theory. However, for a theory of understanding, we additionally need ground for claims about how speakers grasp content, or what perspectives they have on content, and this is where the abilities considered here come in. Having argued that knowing what ‘horse’ means involves knowing what horses are, we now need an account of how knowledge of horses manifests in distinctively linguistic abilities (similarly for other words). Each of the abilities I will describe is independently recognized in the literature, so what is new here is the claim that all three are aspects of competence. Our knowledge of meaning involves distinct modes of apprehension, a result that isn’t so apparent without considering the abilities. 19 See Chomsky (1965). 20 A caveat: I insist that our interpretive dispositions include the interpretation of discourses, and I’m defending the revision that careful attention to our interpretation of discourses reveals a pragmatic component to our knowledge of meaning. 19 Competence with an expression consists in at least referential competence, recognitional competence, and inferential competence. Referential competence is the ability to use a word with its meaning – using it in various sentences to make assertions, commands or promises, to ask questions and form linguistically mediated thoughts (I will henceforth only mention the case of assertion). 21 This is the kind of competence exhibited in deferential use discussed above, but of course a speaker with richer understanding will also have it. A speaker is recognitionally competent with a word to the extent she can recognize whether or not it applies in a given circumstance – i.e. to the extent she can correctly determine whether or not ‘red’ applies to various objects. 22 A speaker is inferentially competent with a word to the extent she can see how its content relates to the contents of other words in the language. 23 Being disposed to infer ‘Feynman is male’ from ‘Feynman is a bachelor’ shows a degree of inferential competence with ‘bachelor’ (and ‘male’). It’s convenient to describe this ability in the epistemic idiom by saying it’s a matter of knowing that ‘Feynman is a bachelor’ entails ‘Feynman is male’. 24 Sentential entailment, as I understand it here, is a metaphysical modality between sentences (in the 21 Referential competence isn’t widely recognized in the literature, but it has been recognized (explicitly in some cases, more tacitly in others) by Kripke (1980), Putnam (1996), Soames (1989), and Higginbotham (1989; 1998a; 1998b). Putnam (1996) (originally published in 1974) is the first recognition I know of. When explaining the sense in which he is competent with ‘elm’, Putnam writes (p. 13), “...I ‘mean’ elm when I say ‘elm’.” 22 Some form of recognitional competence is discussed by Putnam (1996), Cresswell (1978), Marconi (1997), and King (1998). 23 Some form of inferential competence is recognized by Katz and Nagel (1974), Putnam (1996), Marconi (1997), and King (1998) (among others). 24 To be clear, I think of all three kinds of competence as abilities, and don’t assume that that knowledge always underlies or explains the abilities. Is a speaker disposed to infer ‘Feynman is male’ from ‘Feynman is a bachelor’ because she knows that ‘Feynman is a bachelor’ entails ‘Feynman is male’, or vice-versa? I am agnostic about this issue here. 20 sense that if one sentence is true, the other must be true). But inferential competence is a broad notion including other relations. Here are a few of the relations I have in mind (each is an example of something that normally competent speakers know due to their knowledge of the meaning of the words in question): synonymy (‘horse’ means the same as ‘steed’); redundancy (‘naked nudes’ is redundant); semantic anomaly (‘The rock is happy’ is anomalous); meaning containment (the meaning of ‘apartment’ contains the meaning of ‘dwelling’). 25 The perspective on content involved with mere referential competence is comparatively minimal. As suggested in the previous section, referential competence will at minimum give the speaker a metalinguistic perspective on content. A speaker who is referentially competent with ‘Felicity’ and ‘horse’ will have the metalinguistic perspectives on the contents of ‘Felicity’, ‘horse’, and ‘Felicity is a horse’ indicated by (5′) from the previous section. (5′) ∃x∃p(‘Felicity’ refers to x & ‘horse’ expresses p & S knows that (‘Felicity’ refers to x & ‘horse’ expresses p & ‘Felicity is a horse’ expresses the proposition <x, p>)). A metalinguistic perspective on content is a very minimal perspective, to be sure, but it is important to note here that merely having a metalinguistic perspective on an isolated content doesn’t suffice for referential competence. Suppose I know, of the property of being red, that ‘rouge’ stands for this property in French, but I don’t know anything else about French. This knowledge gives me a metalinguistic perspective on the property of 25 See Katz and Nagel (1974, p. 313). Some of these relations are clearly related (i.e. sentential entailment and meaning containment). 21 being red, but this doesn’t grant me any genuine linguistic ability with French. I’m not referentially competent ‘rouge’ since I can’t use ‘rouge’ in any French sentence to say something in French. Similarly, suppose I merely know, of the proposition P expressed by some sentence of French, that the sentence expresses it. This also gives me a metalinguistic perspective on P, but it doesn’t suffice for saying anything in French. I might utter the sentence with the intention of expressing P, but this would not suffice in my view for actually expressing P. Perhaps I commit myself to P in some loose sense, but I don’t succeed at linguistically asserting P in French. At minimum, it is a necessary condition on linguistic assertion that the utterance of a sentence be accompanied by a predicative act that appropriately matches the structure of the proposition expressed by the sentence. This condition is satisfied in (5′) because S in a position to use ‘Felicity’ to single out its content, and use ‘horse’ to single out its content, for the purpose of performing a predicative act in which she herself predicates the latter content of the former content. Further, she know that this predicative act she performs, which amounts to saying, of Felicity, and of the property of being a horse, that the former has the latter, is the “same saying,” or the same proposition, that is expressed by ‘Felicity is a horse’. Alternatively, suppose I know that ‘rouge’ in French means the same as ‘red’ in English. I don’t know anything else about French, but I am a highly competent speaker of English, and understand ‘red’ as well as any typical English speaker. Since I know that the two words mean the same, one might be inclined to the think that ‘rouge’ and ‘red’ give me an equally rich perspective on the property of being red. But surprisingly this isn’t so. The perspective given to me by ‘rouge’ on the property of being red doesn’t 22 allow me to incorporate this property into a system of linguistic representation I can use to predicate this property of something (for this, I would need to know more about French). This point goes hand in hand with the point that I am not referentially competent in French with ‘rouge’ – because I can’t use ‘rouge’ in any French sentence to say or assert anything in French. The essential point here is that referential competence isn’t secured merely by having a metalinguistic perspective on content. In order to be referentially competent with an expression in a language, one must have the systematic ability to use the expression in a range of sentences to say things in the language. Hence, one must be in a position to systematically relate the content of the expression to the contents of other expressions in such a way that one understands how the syntactic structure in the range of sentences relates, at least in a schematic way, to the structure of the propositions expressed by them. Notice that this condition is satisfied in (5′) above. So I think we have to be careful about saying that referential competence is a minimal ability that stems from a minimal perspective on content. If we compare a speaker who is merely referentially competent with an expression to one who also has a significant measure of recognitional and/or inferential competence, then yes, the speaker with mere referential competence has a comparatively minimal ability, and a comparatively minimal perspective on content. But from the broader perspective of being able to systematically use expressions in various sentences to say things, referential competence actually involves a fairly rich perspective on content, and gives rise to a significant linguistic ability. 23 I hope it is intuitive enough that being recognitionally competent with an expression is a manifestation of knowing things about its content, or having a certain perspective on the content. But I want to say a little more about how to understand inferential competence as arising from a grasp of content. In the typical kind of case, I’m supposing that knowing ‘John is a brother’ entails ‘John is male’ stems from knowing that the property of being a brother contains the property of being male (where property P contains property Q just in case every object instantiating P necessarily instantiates Q). In this kind of case, it is clear that the ability stems from knowing something about the relevant contents. I know of the property of being a brother, and of the property of being male, that the former contains the latter. But suppose I’m initially not familiar with ‘canine’ and ‘mammal’, and someone simply informs me that ‘Rufus is a canine’ entails ‘Rufus is a mammal’. Only knowing this much doesn’t necessarily give me much insight into how the property of being a canine and the property of being a mammal relate. I might reasonably guess that the former contains the latter, but I don’t know this for sure, since it could be that the entailment is secured vacuously (say because Rufus is, unknown to me, a lizard). But in my view, I don’t really gain any significant measure of inferential competence merely from knowing that one sentence entails another. Instead, a significant measure of inferential competence comes from having a relatively systematic ability to recognize entailment relations for a variety of sentences. And it is plausible to suppose that in order to do this, I must have some cognitive grip on the contents in question. No doubt more could be said about these matters, but I hope I have said enough to motivate the idea that the abilities I have in mind can be related to the notion from the 24 previous section of knowing things about a content or having a perspective on content. Putting this section and the previous section together, we have now forged a connection from semantics to a theory of understanding and competent use that shows how the two are distinct, but intimately enough related to work together as part of an overall account of linguistic communication. Of course, I have not yet said enough to show exactly how they can work together as part of a theory of communication. I will put the intuitive picture together in the next section, but the view won’t be fully articulated until chapter six. The Basic Picture The positive view I want to defend is starting to emerge, but here I want to fill in a little more detail. In particular, I want to say a little more here to motivate the idea that recognitional competence and inferential competence are genuine kinds of linguistic understanding essential to an overall account of linguistic communication. Since referential competence secures the ability to say or assert something by means of a sentence, I assume it's uncontroversial that this ability is a significant aspect of linguistic practice, and something that must be accounted for theoretically. However, mere referential competence involves such a minimal perspective on meaning that a speaker can be in a position to say things with a word while having a very poor grasp of what it is that he is saying. It will turn out that a speaker’s grasp of what he has said can be so poor that the speaker’s ability to effectively communicate with others by means of the saying (i.e. to incorporate it effectively into a dialogue with others) will be significantly 25 impaired. Essentially it will be possible for a speaker to say and assert things in a context where her understanding of the conversation is significantly defective. What this shows is that a theory of what is said plays a more attenuated role in an overall theory of linguistic communication than is usually supposed. If the mere ability to say things does not suffice for fully successful communication, and if recognitional and inferential competence can provide at least part of whatever else is needed for successful communication, then we can see how these abilities are essential elements in an overall theory of linguistic communication, and part of linguistic understanding in this sense. We need to think about what perspectives on content, besides what is provided by the mere ability to say things, are required for successful conversational exchange. One requirement is that speakers must understand how the various sayings in a conversation are related to each other. For example, one may need to know that one saying follows from another or is incompatible with another. Inferential competence is what secures this type of ability. If one knows how the meanings of words are related to each other, then one will know how the propositions expressed by these words are related to each other. But merely knowing how the sayings in a conversation are related to each other is not enough. In order for understanding of meaning to be adequate for communication, one must also know how what is said is related to the world one is actively engaged with. This ability involves knowing how to determine whether a given saying is true or false in actual situations we find ourselves in. Our understanding of the world influences what we say about it, and what is said about the world influences both our understanding of it and our actions in it. This two-way exchange would not be possible without understanding 26 recognitionally how language and world are related. Recognitional competence is what secures our understanding of the language-world connections. 26 If, for example, a speaker isn’t recognitionally competent with the predicate ‘mimsy’, then he has no basis other than hearsay for asserting that anything is ‘mimsy’ (since he has no way to determine whether anything is ‘mimsy’). Likewise, if he is informed that something is ‘mimsy’, this assertion cannot significantly modify his understanding of the world (beyond knowing that the thing in question is mimsy) or influence his actions in it. If we are going to successfully communicate about the world around us, it is not enough to merely assert propositions which say things about the world. We must also have a sufficiently robust perspective on the propositions asserted to understand how they relate to the world. This is the basic picture, but it isn’t complete until we flesh it out further by considering a range of examples illustrating the view – this is the project I will take up in chapter six. The Variable Standard of Understanding I pointed out above that some words have a higher standard of understanding than others. That is, being merely referentially competent ‘brother’ constitutes a more impoverished state of understanding or competence than does mere referential competence with ‘elm’. I also offered what I think is the primary reason for the variable standard. In sum, my view is that the variable standard is determined by the kinds of typical or everyday 26 Notice that I am only saying that recognitional competence secures our understanding of the language- world connections, not that it secures the connections in the first place – the connections are secured in my view by different mechanisms, involving the broader activities of the entire linguistic community rather than the activity of a single speaker. 27 conversations that speakers of a language have on a regular basis. What counts as a typical or everyday conversation, in turn, depends on the culturally determined range of interests speakers of a language share. The standards for ‘elm’ and ‘beech’ are relatively low because speakers generally aren’t sufficiently interested in trees to bother themselves with the precise difference between the two, and thus are content with merely knowing that each is a different kind of deciduous tree. And this is reflected in normal conversations because generally one can get along just fine in such conversations without knowing anything more. By contrast, we tend to care a lot about family relationships, and hence there is set of words such as ‘brother’ that describe these relationships. These words come up a lot in ordinary conversations in such a way that, in order to track the conversation, one must have a relatively precise understanding of their meaning (i.e. knowing that all brothers are male and that all brothers are siblings). This part of the story seems right, but there are two other issues related to the variable standard that I want to address in this section. First, I think there is a three-fold distinction we can make about the variable standard. There are words that carry a higher standard, but where it nonetheless is possible to be merely referentially competent with them (such as ‘brother’, ‘red’, ‘horse’, ‘water’, etc.). But there are other words that seem to carry an even higher standard so that mere referential competence them – if it is possible at all – leads to a more serious breakdown in the utterances containing them. I think logical terms fall into this category. For example, if a speaker only knows that ‘most’ is a kind of quantifier, but doesn’t know how it differs, if at all, from other quantifiers, her attempts to use it in a sentence would 28 be quite defective – more so than her uses of a word like ‘brother’ with minimal understanding. Accordingly, we can distinguish between three types of words in terms of the standard of understanding they impose: Type-1 Words for which it isn’t possible to be merely referentially competent without a serious breakdown in utterances involving the word (‘all’, ‘most’, ‘or’, ‘only’, etc). Type-2 Words for which it is possible to be merely referentially competent, but being merely referentially competent constitutes a state of significantly impoverished understanding weaker than the sort of breakdown we get with type-1 words (‘brother’, ‘red’, ‘horse’, ‘water’, etc.). Type-3 Words for which it is possible to be merely referentially competent, but being merely referentially competent constitutes a state of impoverished linguistic understanding that is less severe than it is for type-2 words (‘Chris’, ‘elm’, ‘carburetor’, ‘quark’, etc). I will discuss type-1 words, and offer a characterization of the kind of breakdown involved, in chapter five – for the moment I just want to register the point that the consequences of having the minimal sort of knowledge involved in referential competence is more severe for some words than it is for a word like ‘brother’. The second issue I want to discuss is the notion of the “division of linguistic labor” famously defended by Putnam (1996). Consider Putnam’s ‘beech’/‘elm’ example. Putnam tells us that a speaker can be competent in some sense with both ‘beech’ and ‘elm’ even though the speaker cannot distinguish the meanings of the two words. The speaker can tell us that ‘beech’ and ‘elm’ each refer to a different kind of tree, but cannot tell us anything further about the difference. And when placed in a forest of beeches and elms, the speaker cannot tell which trees ‘beech’ applies to or which trees ‘elm’ applies 29 to. Not only is it possible to have such a speaker, but many normal speakers are in a similar state of impoverished understanding with ‘beech’ and ‘elm’. In spite of this impoverished understanding, however, normal speakers are nonetheless competent with ‘beech’ and ‘elm’ in some sense. In what sense are they competent? Putnam offers a rough characterization of this competence: “…the extension of ‘elm’ in my idiolect is the same as the extension of ‘elm’ in anyone else’s, viz., the set of all elm trees…I “mean” elm when I say “elm” ” (2003, p. 74). So in spite of having the impoverished understanding described above, Putnam (a normal speaker) is competent with ‘elm’ because he can use the word with its meaning in the sense that he means elm when he says “elm.” Even though Putnam cannot discern the difference in meaning between ‘beech’ and ‘elm’, he nonetheless shares a level of competency with the words had by those who can discern the difference – all mean elm when saying “elm,” and all mean beech when saying “beech.” Putnam clearly has the notion of referential competence in mind here. This is possible, Putnam tells us, because those with an impoverished understanding use the words with deference towards the use of those who have more robust understanding. There is a division of linguistic labor. There are certain members of the linguistic community who are “experts” on the meaning of ‘beech’ and ‘elm’. These experts can tell us how ‘beech’ and ‘elm’ differ in meaning, and they furthermore can apply ‘beech’ and ‘elm’ with a high degree of accuracy under normal observational conditions. It is the robust understanding of these experts which fixes the content of 30 ‘beech’ and ‘elm’ for the community. Normal speakers manage to mean elm whey they say “elm” because they defer to the experts. While Putnam is pointing to an important linguistic phenomenon here I don’t entirely agree with his description of the phenomenon. First, the term “expert” suggests an overly dramatic contrast between those with more robust understanding and those with more impoverished understanding. While there may be some whose understanding is extremely impoverished, and some who have the extremely rich understanding Putnam associates with the expert, most speakers are in the gradations in between. In my usage, the term ‘expert’ will simply designate those with more robust understanding, but not necessarily an extreme or very specialized understanding that might be suggested by the term. Second, Putnam misidentifies the role played by the so-called “experts.” Putnam seems to think speakers with expertise are required in order to fix the content of a word, but this isn’t the case. Suppose the experts on ‘beech’ and ‘elm’ died out so that nobody could discriminate between the meanings of the words (other than to say that they refer to different kinds of tree). Still, normal speakers could carry on meaning elm when they say “elm.” This shows that there need not be any living experts in order to fix reference, but we can go further and say that there need not be any experts at all for reference fixation. The words could have been introduced by ordinary speakers, one fixing reference by appeal to a stand of beeches, and the other fixing reference by appeal to a stand of elms. Neither introducer need be familiar with trees of the other kind. Both can use the words and pass them on to others. Some users can be in possession of both, correctly believing them to refer to different kinds of trees, though not themselves knowing how to 31 distinguish them further. There may in fact be no one competent with both who knows how to distinguish them further. Even the original introducers may have lost track of the original stands of beeches and elms they appealed to in introducing the words, and thus may not have the ability to distinguish the words if they were to confer with each other. So ‘beech’ and ‘elm’ could have been introduced and used with their respective meanings even though no experts appeared at any stage. Since in this scenario there are no experts, deferential word use can’t be a matter of deferring to experts. What happens in this scenario is that the introducers introduce objective standards of use when they appeal to the stands, and everyone in the community manages to use the words with their meaning because they are appealing deferentially to these standards of use. Deferential use of a word is not deference to an expert user, but deference to an established standard of use. Even though experts are not required for reference fixation, they do have an important role to play. Notice that without expert users of a word, the word will eventually fall into disuse and may even disappear from the language altogether. Suppose literally every speaker in the community is using ‘beech’ and ‘elm’ deferentially and nobody can discriminate between the meanings, as in the two scenarios described above. If it becomes widely known that nobody can discriminate the meanings, then it will become widely known that nobody really knows what they mean when they say “beech” versus when they say “elm.” As a result, both words will fall into disuse, and likely disappear from English entirely. But the words will not languish because they lack reference or meaning. Speakers can use the words with their meanings, so the problem lies not with the lack of meaning, but rather with the lack of understanding about what 32 exactly the meaning is. The fact that the words will fall into disuse and probably disappear without expertise strongly suggests that there is some important linguistic function, independent of mere reference fixation, which is secured by this more robust understanding of word meaning had by the experts. Even though Putnam overly dramatizes the division of linguistic labor, there is nonetheless a division of labor here, since the more robust understanding of the experts is securing vital work for the words – work without which the words would fall into disuse. So, the important upshot of this discussion I want to emphasize is that the division of linguistic labor Putnam talks about is a division of linguistic understanding. While many or most speakers might be perfectly happy to carry on using words like ‘elm’ or ‘beech’ (and other type-3 words) with relatively minimal understanding – so long as this understanding allows them to follow along alright in normal conversations – they do rely on there being at least some other speakers of the language that have a more robust level of understanding. Putnam remarks that the “division of linguistic labor rests upon and presupposes the division of nonlinguistic labor” (2003, p. 75). The kernel of truth in this remark is that the so-called “experts” aren’t experts about the semantic rules of English. Everyone who is referentially competent with ‘elm’ knows its semantic rule equally well. Rather, the experts know more about the content of the expression. And as I’ve been arguing, this sort of knowledge contributes to linguistic understanding. Putnam also points out that there are some words which don’t exhibit a division of linguistic labor (2003, p. 76), but here too I think this is best understood as a point about the division of understanding rather than reference fixation. I don’t think we should 33 construe the idea that for some words everyone is (or should be) an “expert” in some sense to mean that everyone is (or should be) fixing the content of these words for themselves. Consider ‘bachelor’ for example. It is plausible to suppose that anyone, in order to be fully competent with ‘bachelor’, must know that all bachelors are unmarried, and that all bachelors are male. It is doubtful that anything more than this is required, but something more is required in order to fix the content of the word. A two year old child is probably not a bachelor even though he is an unmarried male. Similarly, it is doubtful that the Pope is a bachelor even though he is an unmarried male. So I don’t think we can consider what most speakers know about ‘bachelor’ to be any kind of “definition” that fixes the content of the word for their uses. It is also doubtful that there are certain individuals in the community who have established for everyone that neither the Pope nor two year old boys are in the extension of ‘bachelor’. In this case, at least, what I think fixes the content of ‘bachelor’ is, at least in part, the dispositions of the community to apply the term. If, as I suspect, a suitable majority of English speakers are reluctant to apply ‘bachelor’ to the Pope or to two year old boys, then this is what makes it the case that ‘bachelor’ doesn’t apply to them – not the “expertise” of any individual. So, again, the essential point here is that this whole idea of the division of linguistic labor is best understood as a division of labor for linguistic understanding. In cases where there isn’t a division of labor (as is plausibly the case for type-1 and type-2 words), this is best understood as the idea that that any speaker should have a relatively high level of understanding for the word in question in order to be fully competent with it. 34 An Alternative Semantic Framework I have been assuming for a semantic framework an externalist theory of reference, and defending it on the grounds that, while it can’t serve as a theory of understanding, it can be conceived as a friendly companion to a plausible theory of understanding. For comparison, let’s consider how the old internalistic, descriptivist framework fares on these points. As we shall see, this framework can’t serve as a theory of understanding either, and furthermore it isn’t so clear that it is as elegantly compatible with a theory of understanding as the externalist framework. While this certainly doesn’t defend the externalist framework from all comers, some of the contemporary alternatives are variants of the old internalist framework. So I think the argument I offer here can at least go some way toward defending externalism against alternatives on the grounds that it is nicely amenable to a supplementary theory of understanding. The internalist picture promises a unified account of meaning, reference fixation, and linguistic understanding. The idea is that meaning is sense, sense fixes reference, and that the transparent grasp of sense explains linguistic competence. It is by now a familiar point that internalism fails to provide a satisfactory account of meaning and reference fixation. 27 It is less widely recognized, however, that the internalist picture of understanding is also untenable. In fact, I think that many externalists still labor unhappily to some extent under the old internalist picture of competence. In terms of the present discussion, the transparency thesis adopted by the internalist amounts to the thesis that it is a necessary condition on merely knowing the 27 At least, the descriptivist interpretation of internalism has been discredited on these points, and this will be the version of internalism that I will consider here – see Kripke (1980). 35 semtantic rule for an expression (i.e. grasping the expression-content relation) that one knows a fair amount about the meaning of the word. Since content is transparent, merely associating a content with a word goes hand in hand with knowing a good deal about the relevant content. The internalist wants to include whatever information about meaning is requisite for being fully competent with a word among the information which is transparently accessible to any speaker who knows its semantic rule. At the very least, in order to use a word with its reference, one must have a transparent view of descriptive information which uniquely picks out the referent. 28 In some cases, any descriptive information may do (so long as it is shared by all speakers in a context), but for some words one must associate a privileged description which displays how the meaning of the word is related to the meaning of other words. The internalist wants to include relations such as synonymy and meaning containment among the information transparently accessible, so the internalist presumably thinks that such information is required for competence with some words. Let’s consider some examples. In the case of names, one must associate the name with a description that picks out the reference of the name, but there is no particular description required. 29 One might associate the description the teacher of Alexander with ‘Aristotle’, but one need not know that Aristotle was a teacher in order to be competent with ‘Aristotle’, and hence this particular description is not necessary for competence 28 On this view, using a word with its reference requires using it with its meaning, since meaning determines reference, but of course reference and meaning are distinct. 29 Frege thought that, relative to a particular context of communication, all participants had to associate the same sense with a name. But beyond this, he did not think it necessary that any particular sense be associated with a name – any sense which picks out the right referent will do. 36 with the name. For a predicate such as ‘brother’, however, one must associate the description male sibling in order to be competent with the term. 30 Any speaker who is competent with ‘brother’, ‘male’, and ‘sibling’ will, due to transparency, have a clear view of their contents, and thus be in a position to know that the meaning of the first contains the meanings of the other two. Similarly, the speaker will also be in a position to know that ‘brother’ is synonymous with ‘male sibling’. So the internalist assumes that this knowledge is necessary for competence with ‘brother’. This assumption is partially correct since having this robust understanding is required to be fully competent with the word. But what is wrong with the classic internalist picture is that linguistic competence comes out as an all-or-nothing affair. Either one doesn’t know the semantic rule for the word and is completely incompetent with it, or one knows the rule and also knows everything about the word’s meaning which is required for competence, and is thus fully competent with the word. The classic internalist picture thus cannot account for the phenomenon of deferential word use since this requires a graded notion of competence. The difference with the externalist comes in with the denial of transparency. Since the externalist denies transparency, this allows the externalist, as I am proposing, to account for the gradation in competence in terms of a gradation in how much a speaker knows about meaning. Transparency doesn’t allow for this gradation, so there isn’t any room for gradation on the classic internalist picture. 30 The description need not be something linguistic. A speaker need not acquire ‘male’ and ‘sibling’ before acquiring ‘brother’. The speaker would, however, need some means of representing the description (say by means of the cognitive concepts male and sibling). 37 It is worthwhile to consider briefly how classic internalism might be modified to give a graded picture of competence. The internalist will have to invoke a graded notion of sense. Speakers who are merely using a word deferentially will associate the word with some minimally descriptive sense (and in light of the discussion above, it is likely that this will have to be a metalinguistic description), while more expert users associate the word with a more robust sense. On the classic picture, the sense of a word is what is expressed – and thus is a component part of the assertion made by a speaker when she uses the word in a normal assertive utterance. 31 But once the internalist invokes a distinction between a minimal sense and a more robust sense, we have the question as to which sense is part of assertion. For a speaker who is using a word deferentially and is merely in possession of the minimal sense, is the minimal sense or the robust sense part of the assertions she makes using the word? The internalist will surely want to say that words express the same sense for both deferential and expert users. 32 Even though the deferential user only associates the word with a minimal sense, he somehow manages to assert the robust sense. The internalist will owe us story here about how speakers can assert the more robust thing even though all they are aware of is the minimal thing. But granting that the internalist can provide such a story, there is a further difficulty. Transparency of content, which is arguably a 31 For example, if a speaker associates the description the teacher of Alexander with ‘Aristotle’, this description will be part of what is asserted when she uses ‘Aristotle’ in a normal assertive utterance. An assertive utterance of ‘Aristotle is a philosopher’ will in her mouth assert the proposition that the teacher of Alexander is a philosopher. 32 The alternatives are unpalatable. Either the internalist will have to invoke an implausible multiplication of ambiguity (all words will be ambiguous between the minimal and robust sense), or he will have to argue that the deferential user isn’t really competent at all since she can’t even use the word to express its conventional meaning. 38 cornerstone of internalism, will be lost. The deferential users will be asserting a content which they do not have a transparent view of. For the deferential users, content will be just as non-transparent as it is under externalism. But transparency is one of the main attractions of internalism, so once it is lost the internalist really owes us an explanation of exactly what the attraction of internalism over externalism is supposed to be. Also, notice that this particular solution essentially invokes an idea similar to what I have proposed for externalism. The view will presumably have to be that the minimally competent user of a word merely knows its semantic rule, without knowing much about its content. The difference between minimal competence and more robust competence will then have to be accounted for by a difference in what is known about meaning. So I think it is going to be challenging for the internalist to account for the fact that linguistic understanding is graded. But regardless of whether or not the internalist can offer a satisfactory account, I hope it is clear at this stage that the problem we are considering is not merely a problem for the externalist. The fact that linguistic understanding is graded arises out of fairly basic observations about linguistic usage and is thus a phenomenon which any theory of linguistic understanding must account for. What seems to be creating the difficulty here, at least in part, is not the externalist conception of meaning, but rather the social nature of language use in which deference plays an important role. So it may be that any semantic framework, in order to provide an account of the social nature of language, will have to face the difficulties raised here. 39 Conclusion We are suffering from an internalist hangover. Insofar as we want (i) to maintain (PC), (ii) to suppose that merely knowing an expression’s semantic rule usually comes packaged with lots of information about its meaning, and (iii) to cling to a discrete, non- graded, conception of linguistic competence, we are merely clinging to false vestiges of the old internalist vision. Transparency is dead. And this is a good thing since the notion of transparency is at odds with the graded reality of competent use. Knowing an expressions’ semantic rule must be conceptually distinguished from fully understanding its meaning, and linguistic competence itself must be understood as a graded notion involving perspectives on content that are essential to linguistic communication. 40 Chapter Two Why Semantic Theory Can’t Be a Theory of Understanding I: Names Introduction In this chapter, I present part of my case against (PC′) A semantic theory can be adequate, explaining everything we want a semantic theory to explain, while also serving as an adequate theory of linguistic understanding, and thus also explaining everything we want a theory of linguistic understanding to explain. Here I present the argument only for proper names. In chapter four, I extend the argument to simple predicates. Preliminary Considerations Consider a semantic theory for English (call it T1) incorporating the notion that names are directly referential. Suppose T1 has the following two axioms: (1.1) ‘Hesperus’ refers to Venus. (1.2) ‘Phosphorus’ refers to Venus. Here I use italics to indicate the metalanguage, since I am not assuming that the metalanguage is English. Now consider another semantic theory (call it T2) which is identical to T1, except that, in place of (1.1) and (1.2), T2 has strictly “translational” axioms: (2.1) ‘Hesperus’ refers to Hesperus. 41 (2.2) ‘Phosphorus’ refers to Phosphorus. As semantic theories, T1 and T2 are equally acceptable. On the assumption that the semantic content of a name is its referent, a semantic theory need only state which names have which objects as referents. And clearly T1 and T2 fair equally well in this regard. Since we are interpreting ‘Venus’, ‘Hesperus’, and ‘Phosphorus’ in the metalanguage as directly referential terms, any of these names is equally capable of indicating the referent of ‘Hesperus’ and ‘Phosphorus’. If anything, we might think that T1 is a better theory than T2 because T1 makes the fact that ‘Hesperus’ and ‘Phosphorus’ have the same meaning (referent) explicit. We are not assuming that it is a known fact about the metalanguage that ‘Hesperus’ and ‘Phosphorus’ are co-referential, and hence (2.1) and (2.2) do not make it explicit that ‘Hesperus’ and ‘Phosphorus’ have the same referent. Thus, T1 contains a little more information about the meaning of English than T2 does, so if either of the semantic theories is superior to the other, surely T1 is superior to T2. This demonstrates an important point about semantic theory. It shows that there is a sense in which an adequate semantic theory can state all the facts about meaning that it needs to state without giving all the information there is to know about meaning. Both T1 and T2 do a perfectly adequate job of stating the facts about meaning which are required by semantics yet T1 actually has more information about meaning than T2. So even though there is a fact about meaning not stated by T2, T2 is perfectly adequate. 42 But now let’s ask which semantic theory does a better job of providing an account of linguistic understanding, on the assumption that understanding is merely a matter of “knowing the theory.” I pointed out in chapter one that axioms such as (1) – (2) express singular propositions, and claimed that it won’t do to suppose that we can explain a speaker’s competence with the relevant names merely by supposing that the speaker knows the propositions expressed by the axioms. I also claimed that the reason this won’t do is that it leaves out any characterization of the perspective the speaker has on these propositions – something which is required for an adequate theory of understanding. The axioms in (1) and the axioms in (2) express the same two propositions, so T1 and T2 will be equally inadequate as theories of understanding according to the claims that I made in chapter one. What I will do here is provide a more full-blown argument for the claims that I made in chapter one (but only for the case of names). My strategy is the following. First, I will grant that there is a way for semantic theory to be interpreted as implicitly indicating the perspectives a speaker has on the propositions expressed by (1) – (2). This, in turn, will provide a way to implicitly indicate the senses the speaker associates with ‘Hesperus’ and ‘Phosphorus’, by which she grasps the referent of each name, and, in part, understands each name. Then I will argue that this means of tacitly bringing in perspective still fails as a theory of understanding, and, a fortiori, this will also show that we cannot account for the speaker’s understanding of the names merely in terms of 43 knowing the singular propositions expressed by the axioms in (1) or (2) (where perspective is left out entirely). Notice that we are now recognizing the fact that any semantic theory will be stated linguistically (i.e. in terms of a metalangage). This presents the possibility that the theory can be taken as implicitly incorporating information about perspective or sense, and thus successfully serve as a theory of understanding. The idea is that we can suppose that the metalinguistic names used in (1) – (2) are associated with a sense, and that the very same sense is associated with the corresponding names in the object language. On this view, we can take axioms (1.1) and (1.2) of T1 as assigning content, but also as indicating that both ‘Hesperus’ and ‘Phosphorus’ are grasped via the same sense – i.e. the sense associated with ‘Venus’ in the metalanguage. 1 Similarly, axioms (2.1) and (2.2) of T2, in addition to making the very same assignment of content as T1, indicate that ‘Hesperus’ and ‘Phosphorus’ are grasped via different senses. ‘Hesperus’ is grasped via the sense associated with ‘Hesperus’ in the metalanguage, and ‘Phosphorus’ is grasped via the same sense which is associated with ‘Phosphorus’ in the metalanguage. On this view, T1 and T2 are equally adequate as semantic theories since they make the very same assignment of content, but potentially differ as theories of understanding since we are supposing that each can be taken as telling a different story about the sense associated 1 More precisely, the referent of ‘Hesperus’ is grasped via the same sense as the referent of ‘Phosphorus’, which is the very same sense by which the referent of ‘Venus’ in the metalanguage is grasped. Since this sense is the means by which the referent is grasped for each name, the sense partly explains the speaker’s understanding of each name. For ease of exposition, however, I will just talk about the names being grasped via a sense. 44 with each name (and the perspective had on the two propositions expressed by the axioms). There are two things to notice, however, about the trick we are now considering. First, it doesn’t tell us anything substantial about the senses speakers associate with names. We are taking the different names in the metalanguage as mere markers of different senses without telling us anything about what the senses are actually like. Second, the trick really involves an interpretation which comes from outside semantics. Semantics itself doesn’t have anything to say about sense. Rather, we are choosing to interpret the axioms of the theory in such a way as to sneak in a vague account of sense. But in order for this trick to work, we really need an explicitly linguistic model of what understanding theories such as T1 and T2 amounts to. According to the model I shall adopt here, we imagine that a rational agent is taught the semantic theory in question and then compare what he would thereby know (or be able to do) to what ordinary competent speakers of English know (or are able to do). We imagine that the rational agent already knows the metalanguage in which the theory is stated, and that teaching the agent the theory is a matter of teaching the agent the sentences which comprise the axioms of the theory. As a result of the instruction, the rational agent will know what is stated by the axioms of the theory and what is stated by any theorems which can be derived (according to a syntactic notion of provability that is known to the agent, and anything else that the agent might reasonably be expected to know) from the 45 axioms. 2 We can think of the metalanguage as a “language of thought” without necessarily committing ourselves to the claim that there really is a language of thought. Rather than committing ourselves to the thesis that natural language is really cognized via a language of thought, we are merely adopting the notion of a language of thought as a kind of “enabling assumption” in order to get a tractable model of “knowing a semantic theory” off the ground. The hypothetical postulation of a language of thought is in a certain sense necessary here, because we want a linguistic model, but in order to explain how a speaker can learn her first natural language, we can’t suppose that the speaker already knows some other natural language. This is a reasonable model with which to analyze the notion of “knowing a semantic theory.” Semantic theory is stated in propositional terms, so it is reasonable to think of a speaker who knows the theory as grasping propositions. The linguistic aspect of the model is also reasonable. First, it has the advantage of clarity, because the clearest handle we have on how propositions are grasped comes via our acquaintance with linguistically mediated thought. Second, the notion of a “theorem” of semantic theory requires the thoroughly linguistic (syntactic) notion of provability. If we expanded the notion of a theorem to include everything propositionally entailed by the propositions expressed by the axioms of the theory, and assumed that these theorems would be known to every competent speaker, the theory would be excessively productive, producing 2 Of course, there will be some compositional rules stated by the theory which aren’t captured merely by the notion of provability. I am assuming here that these compositional rules will be among the axioms of the theory. 46 theorems which shouldn’t be part of the theory because they clearly aren’t known to every competent speaker. Specifically, due to singular propositions, there would be no way to block the proposition that ‘Hesperus is Phosphorous’ is true from counting as a theorem, and thus as something known to any competent speaker. 3 Third, it is an assumption essential to semantic theory that the generative nature of natural language 3 If we have a restricted notion of propositional entailment for structured propositions, we will not, due to singular propositions, be able to block undesirable theorems. Axioms (2.1) and (2.2) of T2 express (2.1*) and (2.2*) (where ‘o’ stands for Venus, and ‘Ref’ stands for the reference relation). (2.1*) <Ref, <‘Hesperus’, o>> (2.2*) <Ref, <‘Phosphorus’, o>> Together with other propositions expressed by other axioms, (2.1*) and (2.2*) will entail (2.3*) (where ‘Exp’ is the expressing relation, and ‘=’ is the identity relation). (2.3*) <Exp, < ‘Hesperus is Phosphorus’, <=, <o, o>>>> We will need to incorporate a theory of truth for propositions into the system (and tie this to a theory of truth for sentences). Because competent speakers will know (2.4*), we will want (2.4*) to be a theorem (where ‘True’ is the property of being true). (2.4*) <True, ‘Hesperus is Hesperus’> Now (2.4*) will be derived, in part, on the basis of (2.5*). (2.5*) <Exp, < ‘Hesperus is Hesperus’, <=, <o, o>>>> But there will be no way to block (2.6*) from being derived on the basis of (2.3*) without also blocking (2.4*) from being derived on the basis of (2.5*). (2.6*) <True, ‘Hesperus is Phosphorus’> So even the most restrictive notion of propositional entailment will yield undesirable theorems such as (2.6*) (or fail to allow desirable theorems such as (2.4*)). Hence, we need a thoroughly linguistic notion of provability in order to block certain theorems while allowing other theorems – and, accordingly, a linguistic notion of what it is to understand or “know” the theory. We can, however, have a linguistic model that incorporates linguistic descriptions of structured propositions. This model will incorporate a linguistic notion of provability. In the linguistic model, (2.1*) – (2.6*) will be replaced by (2.1**) <Ref, <‘Hesperus’, Hesperus>> (2.2**) <Ref, <‘Phosphorus’, Phosphorus>> (2.3**) <Exp, <‘Hesperus is Phosphorus’, <=, <Hesperus, Phosphorus>>>> (2.4**) <True, ‘Hesperus is Hesperus’> (2.5**) <Exp, <‘Hesperus is Hesperus’, <=, <Hesperus, Hesperus>>>> (2.6**) <True, ‘Hesperus is Phosphorus’> In this explicitly linguistic semantic theory, ‘Ref’, ‘=’, ‘Exp’, and ‘True’ are predicates of the metalanguage that stand respectively for the referring relation, the identity relation, the expressing relation, and the property of truth. ‘<=, <Hesperus, Phosphorus>>’ is a linguistic description of the structured proposition <=, <o, o>> (as is ‘<=, <Hesperus, Hesperus>>’). So long as we suppose that it isn’t a known fact about the metalanguage that ‘Hesperus’ and ‘Phosphorus’ are co-referential, we can have a linguistic notion of provability in this semantic theory such that (2.4**) is derivable, but (2.6**) isn’t derivable – as desired. 47 understanding can be adequately modeled by the linguistic notion of provability. However this generative understanding in fact works in human cognition, when we engage in semantic theory, we are at least assuming this generative understanding can be adequately modeled by the linguistic notion of provability. The Central Argument To summarize the discussion so far, we need a linguistic model in order to get a clear handle on what is involved in “knowing” a semantic theory. This potentially brings hope to advocates of (PC′) because the linguistic model provides, as I have suggested, a way to tacitly sneak in an account of sense or perspective and thereby preserve the notion that semantic theory can be a theory of understanding. Even though this model is reasonable, it is important to note that it really involves a view that is outside of semantics. Semantic theory itself doesn’t tell us how to understand the notion of “knowing” the theory, and it doesn’t by itself have anything to say about senses or perspectives. We are supplying a model of what is involved in knowing semantic theory that comes from outside semantics, and it is this model that is allowing us to sneak in an account of sense. Nonetheless, the proposed model, if successful, might provide a suitably tight connection between semantics and understanding that it could reasonably be accepted as a defense of (PC′). What I will now show, however, is that even with the proposed linguistic model of understanding, semantic theory still fails as a theory of understanding. 48 A rational agent who learns T1 will clearly know a bit more than an agent who learns T2. The agent will know (3) since it is a theorem of T1. (3) ‘Hesperus is Phosphorus’ is true iff Venus is Venus. 4 And, on the basis of (3), the agent will know (4), which is also a theorem of T1. (4) ‘Hesperus is Phosphorus’ is true. The agent who knows T2 will know the theorem (5) ‘Hesperus is Phosphorus’ is true iff Hesperus is Phosphorus. (4) is not a theorem of T2, but we can’t conclude on this basis alone that the agent doesn’t know (4). If the agent happened to know that ‘Hesperus is Phosphorous’ is a true sentence of the metalanguage, then the agent would know (4) on the basis of (5) even though (4) cannot be derived from the axioms and the notion of provability. But the agent does not (or need not) know (4) on the basis of (5), since we are not assuming that the agent knows that ‘Hesperus is Phosphorous’ is a true sentence of the metalanguage. So T2 can block the undesirable consequence that every speaker who is competent with ‘Hesperus’ and ‘Phosphorus’ would know (4). However, it is important to emphasize that (4) can only be blocked by means of the claim that it need not be known to the rational agent that the metalanguage sentence ‘Hesperus is Phosphorous’ is true. 4 This will be so regardless of whether the semantic theory is couched within a truth conditional framework or a propositional framework. (3) is straightforwardly a theorem in a truth conditional framework. In a propositional framework, we will need a theory of truth for propositions, which is tied to a theory of truth for sentences, and this will allow us to derive (3) as a theorem. 49 If we consider T1 as a theory of understanding, the theory is clearly problematic. As a theory of what competent speakers know, T1 implies that competent speakers would know (4) simply in virtue of being competent with ‘Hesperus’ and ‘Phosphorus’. But clearly it is possible for speakers to be competent with ‘Hesperus’ and ‘Phosphorus’ without knowing (4). T2 avoids this undesirable consequence, and is thus superior to T1 as an account of understanding. So T1 is, if anything, superior to T2 as a semantic theory while T1 is clearly inferior to T2 as an account of linguistic understanding. There are two important results to register at this stage. First, it is now evident why (PC2) (from chapter one) is false. (PC2) All of the facts about meaning stated by any adequate semantic theory for language L are among the facts known to all competent speakers about the meaning of L. As we have seen, it is possible to have a perfectly adequate semantic theory (i.e. T1) which has more information about meaning than some competent speakers know. 5 The second result is that the desiderata for evaluating a semantic theory are different from the desiderata for evaluating a theory of understanding. All that a semantic theory for a language needs to do is correctly state which words in the language have which meanings (and show how these meanings compose to give us an overall account of “what is said” by various sentences). If the semantic theory gives more information about meaning than 5 This does not rule out the possibility that, for certain carefully stated semantic theories (perhaps T2 is an example), any competent speaker will know every fact about meaning stated by the theory. But if such a theory can be stated, it will not, merely from the standpoint of semantics, be superior to some other perfectly adequate alternative theories having more information about meaning. 50 is strictly required for an adequate theory of what is said, this does not necessarily impugn the theory. It can, if anything, make the theory better, since the theory has more information about meaning, and after all, the semantic theory is supposed to be a theory of meaning. But when evaluated as a theory of understanding, we do not want the theory to have more information about meaning than is required in order to be competent with the language. Knowing that ‘Hesperus’ and ‘Phosphorus’ are co-referential is clearly not a necessary condition for being competent with both names. Even though the desiderata for a semantic theory and a theory of understanding are different, it does not follow from this that we can’t have a theory which satisfies both sets of desiderata. But can we have a theory which satisfies both sets of desiderata? I shall argue that we cannot. Given what has been said so far, it looks like T2 might be able to satisfy both sets of desiderata. Certainly T2 is adequate as a semantic theory. But there is reason to think that T2 isn’t adequate as a theory of understanding. As an account of understanding, T2 needs to explain how it can be that speakers can be competent with the names ‘Hesperus’ and ‘Phosphorus’ without knowing that they have the same reference. We are modeling the linguistic knowledge of competent speakers on what a rational agent would know after learning the linguistically stated theory. Hence, in order to explain how it is that competent speakers of English can fail to know that ‘Hesperus’ and ‘Phosphorus’ have the same reference, we imagine a rational agent who is competent with the metalanguage of the theory and who is similarly confused or ignorant about two 51 of the names in the metalanguage (i.e. ‘Hesperus’ and ‘Phosphorus’). But this explanation is unsatisfactory because all it does is pass the explanatory buck up to the metalanguage. We are faced with a mysterious phenomenon about linguistic understanding: how can a speaker be competent with two words which have the same meaning yet not know that they have the same meaning? And the explanation of this mysterious phenomenon provided by (2.1) and (2.2) is, in effect, to say, “Let’s suppose we have the same mysterious phenomenon for the metalanguage.” But if we are going to explain a mystery about understanding in the object language, it is reasonable to demand that the explanation not invoke the same mystery in the metalanguage. For comparison, consider the following “explanation” for why electrons are negatively charged: “Electrons are negatively charged because they work just like another particle, called the schmelectron, and schmelectrons are negatively charged.” Even if there were schmelectrons that worked in a relevantly similar way to electrons, nobody would think that this “account” really explains anything, at least not without a full-blooded account of why schmelectons are negatively charged. The explanation offered by T2 isn’t much better – both “accounts” simply pass the explanatory buck. Of course, there is nothing wrong with passing the explanatory buck so long as it is cashed out clearly wherever it is passed to, but T2 doesn’t do this. Although I think the explanation offered by T2 is bad, the objection I want to maintain here is not that the explanation fails entirely, only that it is uninformative. We 52 are after all allowing the idea that different names in the metalanguage are associated with different senses, and that these senses carry over to the corresponding names in the object language. So we are allowing that T2 can be taken as telling us that ‘Hesperus’ and ‘Phosphorus’ are (or can be) associated with different senses, and this is why competent speakers can be ignorant of (4). But beyond this vague explanation, T2 can’t offer anything more informative. What is missing from the explanation is a positive account of sense. 6 Let me describe some axioms which would properly explain the mystery. Consider a semantic theory (call it T3) which has the following axioms: (3.1) ‘Hesperus’ refers to the first celestial object visible in the evening. (3.2) ‘Phosphorus’ refers to the last celestial object visible in the morning. Without further details about how T3 works, it isn’t clear how T3 is going to capture the fact that ‘Hesperus’ and ‘Phosphorous’ are directly referential, but let’s set this worry aside. Also, (3.1) incorrectly suggests that in order to be competent with ‘Hesperus’ one must know that ‘Hesperus’ refers to the first celestial object visible in the evening (a similar point holds for (3.2)). It is highly problematic to assume, as T3 does, that a speaker must associate a particular definite description (or any definite description at all) 6 Since we are allowing that the metalanguage is some kind of “language of thought,” one might think that a sophisticated theory of mind might supply the needed account of sense. Perhaps this is so, but notice that theories of mind are really in the business of giving an account of mental representation, and not in the business of giving an account of sense. A theory of sense needs to be tied to linguistic competence, because the theory needs to tell us what competent speakers know about reference, and how this figures in an overall theory of linguistic communication. So a theory of sense is really part of the theory of language, not part of the theory of mind. 53 with a name in order to be competent with the name, but set this problem aside as well. The important point here is that T3 does not account for the mystery it is trying to explain by invoking the same mystery at the metalanguistic level. It is not a mystery to us how an agent could be competent with ‘the first celestial object visible in the evening’ and ‘the last celestial object visible in the morning’ without realizing that these two expressions designate the same object. In this regard, T3 offers a better account of competence than T2 because it explains a puzzling phenomenon about understanding names (how a speaker can be competent with two co-referential names without realizing they are co- referential) in terms of a phenomenon that isn’t puzzling (how a speaker can be competent with two definite descriptions which designate the same object without realizing they designate the same object). Furthermore, the theory gives us a positive account of the sense associated with each name (but it is an implausible account). Returning to the problems with T2 as a theory of understanding, there are other aspects of understanding that a semantic theory such as T2 will not be able to explain, even by this uninformative move of passing the explanatory buck to the metalanguage. One interesting facet of our understanding of names is that learning a new name can grant us the ability to have singular thoughts about the referent of the name even though we did not have this ability prior to learning the name. Consider a speaker who first learns a name for Venus (‘Hesperus’ say). Learning the name ‘Hesperus’ grants the speaker the ability to have singular thoughts about Venus. But the speaker need not have had this 54 ability prior to learning the name. Prior to learning the name ‘Hesperus’, the speaker need not have learned another name for Venus, and need not be directly acquainted with Venus (i.e. by visual perception) or learn anything about it that would suggest the speaker is capable of having singular thoughts about Venus. We have to allow for the possibility, with names and other words, that a speaker can acquire a word and use it to represent its content without first associating it with another representational item with the same content. But this aspect of learning names cannot be explained by T2. Notice that according to our model of understanding, the rational agent must already have acquired the metalanguage before we can teach him the theory. In order to learn axiom (2.1), our rational agent must already have acquired a name for Venus in the metalanguage (i.e. ‘Hesperus’). But if the agent has already acquired ‘Hesperus’ prior to learning ‘Hesperus’, then the agent already has the ability to have singular thoughts about Venus prior to learning ‘Hesperus’. The model incorrectly suggests that a speaker can only acquire a name and use it to represent its content by linking it with another representational item with the same content. So as a theory of competence, T2 gives an implausible picture of the representational resources needed to acquire a name (T1 has the same problem of course). So far we have seen that there are perfectly adequate semantic theories such as T1 and T2 which, as theories of understanding for names, cannot solve certain difficulties. In 55 order to solve these difficulties, I think we will be forced to weaken the theory to the point where it is no longer an adequate semantic theory. Here is a suggestion. Suppose that competence with names such as ‘Hesperus’ and ‘Phosphorus’ involves the following sort of schematic knowledge: 7 (5.1) ∃x(x = the referent of ‘Hesperus’ & for any predicate F, to assertively utter ⎡Hesperus is F⎤ is to say of x that it is F). (5.2) ∃x(x = the referent of ‘Phosphorus’ & for any predicate F, to assertively utter ⎡Phosphorus is F⎤ is to say of x that it is F). 8 The idea here is that only the name and its referent will be common to every speaker of the language who has minimally acquired the name and is able to use the name to refer to its referent – i.e. every speaker who is at least referentially competent with the name. Similarly, the only perspective on the referent that will be shared by every speaker who is referentially competent with the name will be the minimal metalinguistic perspective indicated by the metalinguistic descriptions in (5.1) and (5.2). This minimal perspective on the referent will form part of the perspective that the speaker has on propositions expressed by sentences in which the name occurs (as is also suggested by (5.1) and (5.2)). 7 Soames (2002, p. 65) may have something like this in mind when he writes, “In order to be a competent user of a name n of an object o…one must realize that to assertively utter n is F is to say of the referent, o, of n that it “is F.” ” (Here Soames uses bold italics to indicate Quine corners.) 8 I won’t worry here about the distinction between object language and metalanguge. We can just assume that both languages shares a set of simple predicates, and that the variable ‘F’ is ranging over these predicates. I will make a similar assumption for similar clauses below. Also, (5.1) and (5.2) obviously only cover simple subject-predicate constructions where a name forms the subject and ⎡is F⎤ forms the predicate. We will need different clauses for different types of construction, but I won’t go into this here. 56 Notice (5.1) and (5.2) do not specifically state what the referents of the names are. Thus (5.1) and (5.2) fail to satisfy a minimum constraint on semantic theory – that the theory state what the meaning of each word in the language is. (5.1) and (5.2) are consistent with an indefinite number of (directly referential) interpretations of the names and thus also with an indefinite number of interpretations of the language. But as a theory of competence for names, (5.1) and (5.2) can solve all of the difficulties that we have seen. First, it is plausible that (5.1) and (5.2) are known to all speakers who are at least minimally competent with the two names. Second, (5.1) and (5.2) can explain how a speaker can understand both names without understanding that they are co-referential, and do so without passing the explanatory buck. In (5.1) and (5.2), the failure to understand that the names are co-referential is attributed to the failure to understand that the descriptions (6.1) and (6.2) designate the same object. (6.1) The referent of ‘Hesperus’. (6.2) The referent of ‘Phosphorus’. But as we have already noted, it is no mystery as to how a speaker can understand two definite descriptions without realizing that they designate the same object, so (5.1) and (5.2) do not problematically pass a puzzle about how names are understood up to the metalanguage. Finally, (5.1) and (5.2) do not implausibly suggest that the only way to acquire a name and use it to represent its content is by linking the name with another representational item with the same content. (5.1) and (5.2) do not require that the 57 metalanguage have any names for Venus. There are probably alternative approaches for solving the difficulties that I have raised here, but I suspect that all of the approaches will similarly involve merely “schematic” information about meaning such as we have in (5.1) and (5.2), and thus would be similarly inadequate from the standpoint of semantic theory. Let me be clear at this stage that I am not denying that a speaker who is merely referentially competent with ‘Hesperus’ and ‘Phosphorus’ knows (what is expressed by) (2.1) and (2.2). What I am denying is that the speaker’s knowledge of (2.1) and (2.2) explains her referential competence with the names. Consider the following two claims. (i) S knows the propositions expressed by (2.1) and (2.2). (ii) S is referentially competent with ‘Hesperus’ and ‘Phosphorus’. In my view, these two claims express coeval facts about S. (ii) entails (i), and (i) can also be plausibly take to entail (ii), 9 but this not because (i) explains (ii). Instead, both (i) and (ii) are explained, in part, by (iii). (iii) S knows the propositions expressed by (5.1) and (5.2). It is not my view that (iii) is a sufficient condition for (i) or (ii). Suppose S is given a list of 100 names she has never encountered. She never encounters anyone actually using 9 I think it is reasonable to interpret the notion of knowing an axiom of semantic theory such as (2.1) and (2.2) in such a way that referential competence is coeval with such knowledge. However, one might be able to concoct clever examples where a speaker in some sense knows (2.1) and (2.2) but nonetheless fails to be referentially competent with ‘Hesperus’ and ‘Phosphorus’. Such examples would, if anything, place further pressure on the thesis that linguistic understanding can be explained in terms of knowing a semantic theory. For the purposes of discussion here, however, I will assume that such cases would not be genuine cases of knowing (2.1) or (2.2). 58 them, and doesn’t even know the language they are part of. All she learns are versions of (5) corresponding to them. Does S thereby know the propositions expressed by the versions of (1) or (2) corresponding to each name, and is she referentially competent with each name? Maybe, but this is by no means obvious, and certainly isn’t a claim I want or need to commit to. Perhaps no propositional knowledge over and above the versions of (5) is required, but probably something else is – forming the appropriate referential intentions with the names, having extensive knowledge of the language the names are part of, experience with its users, being part of a causal chain of communication for each name extending back to its referent, etc. Likewise, our speaker who is merely referentially competent with ‘Hesperus’ and ‘Phosphorous’ would need to meet these further conditions in addition to knowing (5.1) and (5.2). The important point here is that (iii) together with the further conditions (whatever exactly they may be) is what explains both (i) and (ii). 10 We can now see why (PC4) (from chapter one) is false. (PC4) All of the facts stated by an adequate semantic theory are facts which serve to explain linguistic competence. (2.1) and (2.2) are two facts stated by the adequate semantic theory T2. Even though these two facts are known by any speakers who are referentially competent with 10 Again, if we granted examples where a speaker in some sense knows (2.1) and (2.2) but can fail to be referentially competent, then the view could not be that (i) and (ii) are coeval. Even if we granted such examples, however, it would still turn out that at least some of the extra conditions beyond knowing (5) required for (ii) would also be required for (i). So (i) would still be explained in part by some, though not all, of the extra conditions that explain (ii). 59 ‘Hesperus’ and ‘Phosphorus’, this is not knowledge which explains their competence with the names. Let me fill out some more detail on how the explanation of (i) and (ii) will go. Suppose a speaker has picked up ‘Hesperus’ and ‘Phosphorus’ conversationally so that S at least knows (5.1) and (5.2). The following attributions are true. (7.1) S knows that ∃x(x = the referent of ‘Hesperus’ & for any predicate F, to assertively utter ⎡Hesperus is F⎤ is to say of x that it is F). (7.2) S knows that ∃x(x = the referent of ‘Phosphorus’ & for any predicate F, to assertively utter ⎡Phosphorus is F⎤ is to say of x that it is F). Since the speaker has picked up the names conversationally, she will be part of a causal chain of communication extending back to the referent of each name. Further, suppose that the speaker meets the other conditions I mentioned above (she has formed an appropriate referential intention for each name, has extensive knowledge of the language the names are part of, experience with its users, etc.). All this justifies exportation of the quantifiers in (7), so that (8.1) and (8.2) are also true. (8.1) ∃x (S knows that (x = the referent of ‘Hesperus’ & for any predicate F, to assertively utter ⎡Hesperus is F⎤ is to say of x that it is F)). (8.2) ∃x (S knows that (x = the referent of ‘Phosphorus’ & for any predicate F, to assertively utter ⎡Phosphorus is F⎤ is to say of x that it is F)). Also, (8.1) and (8.2) entail (8.3) – (8.6). (8.3) ∃x (S knows that x = the referent of ‘Hesperus’). (8.4) ∃x (S knows that x = the referent of ‘Phosphorus’). 60 (8.5) ∃x (S knows that for any predicate F, to assertively utter ⎡Hesperus is F⎤ is to say of x that it is F). (8.6) ∃x (S knows that for any predicate F, to assertively utter ⎡Phosphorus is F⎤ is to say of x that it is F). (8.3) and (8.4) entail (i). Furthermore, notice that (8.5) and (8.6) justify the claim (made more generally in chapter one) that a speaker who is merely referentially competent with a name knows, of the proposition P expressed by a sentence in which the name occurs, that the sentence expresses P. This will also allow her to form the appropriate intention to assert P by means of the sentence. This of course further justifies the idea that (8.1) and (8.2) entail (ii). Similarly, the speaker will generally be able to entertain singular thoughts about the referent by means of the name. 11 But linguistic communication demands, at least in some holistic way, a more robust understanding of names than is given by mere referential competence. We can see this point by noting that if the entire linguistic community were merely referentially competent with ‘Hesperus’ and ‘Phosphorus’, then the two names would fall into disuse and eventually disappear from the language. The names would fall into disuse and 11 Of course, (8.5) and (8.6) only cover the case of assertion, and we would need different clauses to cover other cases. But the basic argument would be the same. For instance, one way to form a belief is by sincerely accepting a sentence. For a speaker who picks up ‘Hesperus’ conversationally, (8.7) will be true. (8.7) S knows that ∃x(x = the referent of ‘Hesperus’ & for any predicate F, to sincerely accept ⎡Hesperus is F⎤ is to believe of x that it is F). From (8.7), we can argue in a manner exactly parallel to the argument above and arrive at (8.8). (8.8) ∃x (S knows that for any predicate F, to sincerely accept ⎡Hesperus is F⎤ is to believe of x that it is F). Thus, we can see how referential competence with a name grants a speaker the ability to entertain singular thoughts about the referent of the name by sincerely accepting sentences containing the name. 61 disappear form the language because speakers of the language would have no motivation to use them, since they would have no understanding of how the referents of the names related to other objects or properties they wished to talk about. So there is a more robust understanding of names which does vital work in a language. Clearly we cannot capture this richer sense of understanding with clauses such as (5.1) and (5.2), or any of the further conditions on referential competence, so some stronger sense of understanding will be required. Competence with a name ranges from very minimal (mere referential competence) to something more robust. Accounting for this graded understanding doxastically in terms of propositional knowledge is problematic. The minimal level of competence will require a very weak knowledge attribution such as (5.1) and (5.2), but as I have suggested the account will have to supplement this weak schematic knowledge with a further story, at least some elements of which won’t be propositional in nature. The more robust level of understanding will, if it is to be accounted for propositionally, require a knowledge attribution stronger than (5.1) and (5.2). But if we are to preserve (PC′), what could this stronger knowledge attribution be other than the claim that the speaker knows (the propositions expressed by) (2.1) and (2.2)? Saying that the more robustly competent speaker knows (2.1) and (2.2) won’t explain the speaker’s more robust level of competence because, as we have already seen, even the minimally competent speaker knows (2.1) and (2.2). Since for a name semantic theory only offers an axiom stating the referent of the name, this is all we have to work with in trying to 62 make semantic theory account for linguistic competence. But it is impossible to see how the notion of “knowing” such an axiom can be understood in a graded way that adequately captures the gradations in understanding. Saying that a speaker knows (2.1) and (2.2) is explanatorily impotent regardless of whether we are trying to account for the minimal level of competence or the robust. The difficulty with names, however, is that it is hard to say exactly what the richer level of understanding beyond mere referential competence amounts to. As I pointed out in chapter one, a name is a word for which being merely referentially competent generally does not constitute a state of significantly impoverished linguistic understanding (i.e. names are what I called “type-3” in chapter one). One view which is tempting here is to be a descriptivist, not about the semantics of names, but about understanding for names. Since I am divorcing the theory of understanding from semantics, this possibility is open to us. Suppose that, in addition to knowing (5.1), many, but not all, English speakers associate ‘Hesperus’ with the description the first celestial object visible in the evening, and that this association is what gives these speakers a more robust understanding of ‘Hesperus’ compared to those who are merely referentially competent (and thus only know (5.1)). If so, this would suffice for the name to remain viable in the community. I think this sort of view is appealing, but I won’t investigate it any further here (however, I will return to this idea in chapter six as part of my positive account of understanding). 63 Conclusion In order to advance the negative argument against (PC′), we really need to consider words for which being merely referentially competent constitutes a state of significantly impoverished linguistic understanding (i.e. what I called “type-2” words in chapter one). This is the advantage of discussing predicates, since many predicates are type-2. Also, this discussion may contribute, in an indirect way, to the account of competence with names. Mere referential competence with names (and all type-3 words) is only possible in the context of a richer understanding of the language in general, and this richer understanding is secured, in large part, by a speaker’s richer understanding of type-2 words. We will turn to the consideration of predicates and type-2 words in chapter four, after a more thorough investigation of the three linguistic abilities (referential, recognitional, and inferential competence) in chapter three. 64 Chapter Three Referential, Recognitional, and Inferential Competence Introduction In my view competence with an expression consists in at least referential competence, recognitional competence, and inferential competence. The three kinds of competence are abilities – literally “things we can do with words” – that reflect what we know about meaning. The abilities point the way toward a general theory of sense, understood as a theory of what speakers know about reference, or how they cognize reference. The three abilities can thus be thought of as involving three distinct modes of presentation of reference. These modes of presentation are similar to what Stanley and Williamson have called practical modes of presentation. 1 The idea is that understanding a referent via a practical mode of presentation entails having certain dispositions towards the referent. I don’t think the abilities I have in mind are mere dispositions, so for my purposes, a practical mode of presentation is a way of grasping a referent (or a proposition) which entails certain practical abilities. Referential competence certainly isn’t a mere disposition, however the other two abilities can perhaps be thought of in dispositional terms. In this chapter I discuss these three abilities in greater detail. 1 See Stanley and Williamson (2001, p. 429). 65 Referential Competence We use words to “say” or “assert” things. 2 This ability typically manifests itself at the level of a sentence since typically we assertively utter a sentence in order to say something. But even though the ability manifests itself at the level of a sentence, the ability is traceable to an ability had at the level of words. This kind of competence is what I’m calling referential competence. To be referentially competent with a word W is to be able to exploit its conventional meaning. The meaning (or “semantic content”) of a word is a social convention governing its use. When a speaker uses a word in a sentence to say something, she exploits the convention governing its use. For example, if a speaker is able to use the sentence ‘Bill Clinton is happy’ to assert that Bill Clinton is happy, this is due in part to the speaker’s ability to exploit the conventions according to which ‘Bill Clinton’ refers to Bill Clinton, and ‘happy’ expresses the property of being happy. 3 A more precise characterization of referential competence is going to be sensitive to the details of one’s views about the relation between meaning and assertion, but for the moment we can remain fairly neutral on the matter by simply saying the following: In the context of a sentence, the conventional meaning of a word serves (in conjunction with the other words in the sentence) to limit what a competent speaker can reasonably be understood to assert when he assertively utters the sentence. Without conventional meaning, human communication would be more difficult – it would at best be like a 2 Again, the focus on assertion is a simplification since there are (presumably related) abilities such as the ability to “ask” things, or believe things (i.e. by accepting a declarative sentence). 3 For an extensional semantics, the convention for ‘happy’ would be a rule to the effect that ‘happy’ applies to (or is true of) an object x if and only if x is happy. Nothing in this alternative conception of the rule impacts the intuitive point about a speaker’s ability to exploit the convention. 66 game of Charades where players try (and fail in all sorts of amusing ways) to communicate via pantomime. But since conventional meaning places rather strict limits on what a speaker can be understood to be asserting by means of a sentence, linguistic communication proceeds at a relatively easy and brisk pace. To get a clearer sense for what referential competence amounts to, it is helpful to considering some examples where a speaker isn’t referentially competent. The conventional meaning of a word is set up by a linguistic community which speaks the language of which the word is a part, but it doesn’t follow from this that any given member of the community is in a position to exploit the convention. Suppose for example that Jeff has never met or heard of anyone who goes by the name ‘Cornelius’. Jeff is aware that ‘Cornelius’ is a name, and being a generally competent speaker, Jeff knows that a name is used to refer to its bearer. But Jeff hasn’t acquired this name as the name of any person or object. Now suppose that Jeff suffers a hard knock on the head, and as a result spontaneously forms the false belief that he has met or heard of an individual who goes by the name ‘Cornelius’. Believing that the person he met or heard of is a jerk, Jeff assertively utters the sentence (1) Cornelius is a jerk. The central point here is that the name ‘Cornelius’ cannot, in Jeff’s mouth, serve to limit what Jeff might be understood to be saying in the way it would for someone who is competent with the name. Since Jeff has the word ‘jerk’ as part of his linguistic repertoire, and since he knows in general how names are used, we can perhaps narrow the interpretive possibilities down to those where Jeff is saying of some individual that he 67 is a jerk. We might further limit the interpretive possibilities to those where Jeff is saying of some individual who goes by the name ‘Cornelius’ that he is a jerk. However, if we were to line up all the individuals who go by the name ‘Cornelius’ and ask which of these individuals Jeff predicated jerkhood of, there would be no determinate answer to this question. Jeff hasn’t acquired the name ‘Cornelius’ as the name of any actual individual, hence this name cannot serve in Jeff’s utterance to single out which individual Jeff is predicating jerkhood of. And since the occurrence of ‘Cornelius’ in Jeff’s utterance cannot provide any means of singling out which individual Jeff is speaking of, Jeff has not managed to assert anything specific. 4 But let’s modify the example to make it more plausible that Jeff has asserted something specific. Suppose that Jeff receives his bump on the head from Fred, and Jeff (as he sees Fred fleeing the scene) spontaneously forms the false belief that that very man fleeing the scene is the individual named ‘Cornelius’. So Jeff points to Fred as he is fleeing the scene and assertively utters (1). Perhaps Jeff has asserted (what is expressed by) (2) Fred is a jerk. But even if this is granted, the crucial point is that Jeff’s use of ‘Cornelius’ in asserting (1) does not serve to limit the possible interpretations of Jeff’s utterance. If Jeff has 4 Perhaps Jeff manages to assert that someone is a jerk. But even if this is so, Jeff still hasn’t managed to exploit any specific convention governing the use of ‘Cornelius’ (i.e. according to which the name refers to some specific individual). There is no such specific convention for the name that can serve to constrain our interpretation of Jeff’s utterance. The only thing that can serve to constrain our interpretation is Jeff’s knowledge of the general convention according to which names are used as devices of reference to individuals. 68 managed to assert (2), he has done so by means of his pointing at Fred, not by means of exploiting the linguistic convention governing ‘Cornelius’. By contrast, a speaker who is referentially competent with a word can exploit its meaning, and when she uses the word in an assertive utterance of a sentence, the word serves (together with the others words in the sentence) to constrain what she could reasonably be taken to have asserted by means of the utterance. In order for a speaker to assert anything by means of an utterance, there must be sufficient constraints on the interpretation of the utterance to narrow the candidate assertions down to some reasonable number (but of course the speaker could assert many things by means of a single utterance). And typically such constraints are provided, at least in part, by the semantic content of words. There may be some strange cases, such as the “modified” case involving Jeff above where a speaker manages to assert something in the course of uttering a sentence even though he is not referentially competent with one of the words in the sentence. But such cases do not harm the point that the word in question is not serving to constrain the interpretation of the speaker’s utterance. I have so far remained neutral about the relation between semantic content and assertion, but I will now discuss the notion of referential competence in the context of more specific views of this relation. According to the “classic” picture of the relation between semantic content and assertion, when a speaker assertively utters a sentence S in a context (where she is speaking literally and without the presence of any pragmatic effects that might cancel the normal force of the utterance), she thereby asserts the 69 semantic content of S relative to the context. 5 Of course, the speaker might assert other things besides the semantic content of S. Anything that is pragmatically entailed by the content of S might be counted among the things the speaker has asserted. 6 But the semantic content of S is the “primary” proposition asserted by the speaker. The semantic content is primary in the sense that it will normally be the thing that the speaker primarily intends to assert and convey to her audience. It is also primary in the sense that any other assertions that might be attributed to the speaker are pragmatically entailed by it. So on this view the semantic content determines the primary assertion and the primary assertion places further constraints on what else can count as being asserted in the context. In this framework, it is relatively easy to define the notion of referential competence. A speaker who is referentially competent with a word will be in a position such that when she assertively utters a sentence S containing the word (and is speaking literally and in the absence of pragmatic canceling effects), then the word will contribute its conventional content and the semantic content S has relative to the context of the utterance will be the primary assertion made by the speaker. So on this view, a word W semantically contributes its conventional meaning to sentences in which it occurs, thus playing its role in determining the semantic content of the sentence. A speaker who is referentially competent with W is in a position to exploit 5 Those who eschew the notion of meaning in favor of a theory of truth and reference, but nonetheless take themselves to be giving an analysis of the notion of what is “said” or “asserted” might say the following. If the canonical T-theorem for sentence S is ‘S’ is true iff P, then when a speaker assertively utters a sentence S (where she is speaking literally and without the presence of any pragmatic effects that might cancel the normal force of the utterance), she thereby asserts P. 6 A proposition P is pragmatically entailed by the semantic content of a sentence S if and only if P is an obvious and relevant consequence of the conjunction of the content of S and any background information that is salient in the context. 70 its conventional content. So when the speaker uses a sentence in which W occurs, W will, together with other words in the sentence, determine the semantic content of the sentence as the primary assertion made by the speaker. And since the primary assertion places indirect constraints on what else might count as being asserted by the speaker, W also serves in the speaker’s mouth to constrain what she might reasonably be interpreted to be saying. On the other hand, if the speaker is not referentially competent with W, she will not be in a position to use W in an assertive utterance of a sentence such that W serves to determine the primary assertion (at least W won’t help determine the primary assertion in the way it normally or conventionally does). In such a case the speaker will usually not succeed in saying anything at all. Since W doesn’t serve to help determine the primary assertion, there won’t be any primary assertion. And since other assertions are triggered via the primary assertion, nothing will be asserted. Let’s turn to a view that departs from the classic picture sketched above. In recent work, Scott Soames has proposed a view about the relation between semantic content and assertion which differs from the classic view in several major respects. 7 My aim in discussing Soames’ view is neither to defend it, nor to explain the motivations for it, but merely to show how the notion of referential competence can be characterized in an alternative framework. Before discussing Soames’ view, however, let me first make a distinction which will prove useful. On the classic view, there are two senses in which the semantic content of a sentence is primary. It is primary in the sense that it is the content that a speaker primarily asserts and intends to convey to her audience, and it is primary in 7 See for example Soames (2009a) and Soames (2009b). I am also indebted to Soames for the discussion of the “classic” picture. 71 the sense that it is necessary for determining anything else that the speaker might be understood to assert by means of the utterance. On the classic view, if the semantic content isn’t asserted, then typically nothing else will be since other assertions trigger (via pragmatic implication) from the semantic content. To avoid confusion, let’s reserve the phrase “primary assertion” for the assertion that the speaker primarily intends to assert and convey to her audience. And let’s use the phrase “initial content” for the content that is initially made available by the utterance and serves to constrain what counts as being asserted by the sentence (here it isn’t required that the initial content be asserted). With these distinctions in place, we have a convenient way to contrast Soames’ view to the classic picture. Soames’ view agrees with the classic view in that the initial content of an utterance is the semantic content of the sentence uttered. For both views, semantic content provides the initial content from which other contents or assertions are generated by pragmatic mechanisms. This is perhaps not surprising, because it is difficult to see how we could account for the fact that the role of semantic content is to limit the domain of possible interpretations of a speaker’s utterance without semantic content being identical to initial content, since presumably the only way to impose the necessary constraint would be to limit (by means of convention) the content made available by an utterance from which pragmatic mechanisms trigger. But Soames’ view differs from the classic view in that the primary assertion made by an utterance can differ from the semantic content of the sentence uttered. On Soames’ view, the semantic content of a sentence need not be asserted – it will only count as being 72 asserted if it is an obvious and relevant consequence of the primary assertion together with background information salient in the context of utterance. Semantic content still imposes a limit on what the primary assertion can be, however, because the primary assertion is always something which is an acceptable pragmatic augmentation or enhancement of the semantic content of the sentence uttered. So on Soames’ view, there is still an initial content made available by an utterance (identical to the semantic content of the sentence uttered) that limits what can be asserted by means of the utterance. The primary assertion is either identical to the initial content or is an acceptable pragmatic augmentation of the initial content, and anything else which might be asserted by means of the utterance is pragmatically implied by the primary assertion We are now in a position to define the notion of referential competence within Soames’ framework. A speaker who is referentially competent with a word W will be in a position such that, when she assertively utters a sentence S containing W, W (together with the other words in the sentence) will determine the initial content of the utterance. This initial content will in turn place a constraint on what the primary assertion made by the utterance can be. On the other hand, if the speaker is not referentially competent with W, she will not be in a position to use W to determine the initial content of the utterance. In such a case the speaker may not succeed in saying anything at all. 8 Since W cannot serve to help determine the initial content, there won’t be any initial content. And since the primary assertion will either be identical to the initial content or an acceptable 8 Again, the speaker may actually succeed in saying something very minimal. For example, if the name ‘n’ fails to refer, a speaker’s use of ‘n is F’ might still assert that something is F. 73 pragmatic enrichment of the initial content, there won’t be any primary assertion, or any assertion at all (at least in typical cases). To my knowledge, referential competence has not been widely recognized in the literature, with three notable exceptions. First, I think Putnam essentially has referential competence in mind when, in discussing the ‘beach’/‘elm’ example, he writes “…the extension of ‘elm’ in my idiolect is the same as the extension of ‘elm’ in anyone else’s, viz., the set of all elm trees…I “mean” elm when I say “elm.” 9 Second, James Higginbotham writes of “merely possessing a word, or having it in one’s repertoire, and so being able to use it with its meaning,” where I take him to be sketching the notion I have in mind. 10 Finally, Scott Soames writes, “[T]he explanation of my belief that Pluto is a distant planet involves the fact that, (i) I accept the sentence ‘Pluto is a distant planet’; (ii) the sentence expresses the proposition that Pluto is a distant planet; and (iii) I am a competent speaker and thereby understand the sentence.” 11 Here I take it that the relevant sense of competence or understanding is referential competence. While I have defined referential competence in terms of assertion for the sake of simplicity, being able to use a word with its meaning extends beyond assertion. Being referentially competent with all the words in the sentence ‘Pluto is a distant planet’ grants one the ability to form the belief that Pluto is a distant planet by means of accepting the sentence, to assert that Pluto 9 Putnam (2003, p. 74). 10 See Higginbotham (1998a) and Higginbotham (1989). 11 Soames (1989b, p.588). 74 is a distant planet by means of assertively uttering the sentence, or even to ask whether Pluto is a distant planet by putting forth the query, ‘Is Pluto a distant planet?’. There are two general points I want to make about referential competence. For the purposes of this discussion, I will adopt the “classic” picture of the relation between semantic content and assertion described above, but this simplification does not impact either of the points. First, as I argued in chapter one, the conditions sufficient to be referentially competent with many words are minimal. Certainly this is the case for names. Suppose Smith picks up the name ‘Richard Feynman’ in conversation from another speaker. When he picks up ‘Richard Feynman’ from the other speaker, he forms the deferential intention to use this name to refer to the same person as the speaker from whom he learned the name. 12 It seems to me that the only linguistic ability with ‘Richard Feynman’ that Smith acquires as a result of picking up the name is that he is now referentially competent with ‘Richard Feynman’. Suppose Smith assertively utters a sentence of the form ⎡Richard Feynman is F⎤. What has Smith said? Smith’s intention to use ‘Richard Feynman’ to refer to the same person as the speaker from whom he picked up the name determine Feynman as the referent of ‘Richard Feynman’. And because of this, there is a determinate answer to the question – Smith asserts the proposition which attributes the property of being F to Richard Feynman. But notice that Smith can acquire referential competence with ‘Richard Feynman’ without knowing much at all. Smith just picks up the name in conversation and forms a 12 Actually, the exact nature of his intention is unclear, but I won’t delve into this issue here. 75 deferential intention with regard to the use of the name. Smith need not be able to recognize Feynman or know anything substantive about him. Nonetheless, Smith is able to use ‘Richard Feynman’ in assertive utterances to express content with definite truth conditions, even though Smith himself has, in a certain sense, a rather dim grasp of these truth conditions. 13 Can the same be said for other kinds of words? Consider again the ‘elm’/’beech’ example from Hillary Putnam. 14 In effect, Putnam argues that he is merely referentially competent with the terms ‘elm’ and ‘beech’. This seems correct. If, for example, Putnam were to assertively utter the sentence ‘There is an elm in my yard’, he would be saying something with determinate truth conditions. Thus he is minimally competent even though he has very little relevant knowledge – he cannot recognize the difference between elms and beeches, nor can he say how they differ. All Putnam knows is that ‘elm’ and ‘beech’ refer to different kinds of trees. So I think the conditions sufficient for referential competence with “natural kind” terms are also minimal. What about other kinds of words, such as ‘brother’? If no distinctions are made about different kinds of linguistic competence, one might be inclined to say that a speaker is linguistically competent with ‘brother’ just in case she knows such things as (12), (13) and (14) (and claims to this effect are often made in the literature). 13 Recall Kripke’s discussion in Naming and Necessity of a speaker’s ability to use a name even when she does not take herself to have any descriptive information that uniquely determines the referent of the name See Kripke (1980). 14 Putnam (1996). 76 (12) ∀x(x is a brother → x is male). (13) ∀x(x is a brother → x is a sibling). (14) ∀x(x is a brother → x shares a mother with someone else). If this were correct, then the conditions for competence with ‘bachelor’ would not be minimal as they are for ‘elm’. But I think this is mistaken. A speaker can be referentially competent with ‘brother’ without knowing (12), (13), or (14). To see this, suppose that Jessica, a young child, constantly hears her parents using the term ‘brother’. Her parents say things to her like, “Where is your brother?” or, “You and your brother should go play outside.” Jessica understands that ‘brother’ applies to Ben (her brother), and understands that it is a predicate. She may understand that the term indicates some sort of familial relationship, but does not know what the relationship is exactly. In fact, let’s suppose that Jessica sometimes misapplies ‘brother’ without any inkling that she has done something incorrect, as for example when she says that her cousin Emily is her brother, or that the family dog is her brother. 15 Jessica does not know (12), (13), or (14). Jessica does not know that ‘brother’ applies to an object only if it is male, or that ‘brother’ applies to an object only if it is a sibling, or that ‘brother’ applies to an object only if it shares a mother with someone else. In spite of her ignorance, Jessica uses ‘brother’, and when she does so she has the deferential intention to mean by this term what her parents mean by it. Now suppose that 15 Jessica does not yet have a clear enough conception of “blood ties” to form any clear conception of how the dog’s membership in the family might differ significantly from other family members. Perhaps she does have some hazy idea that the dog’s familial status is somehow different from the other family members, but she doesn’t think this prevents the dog from being a brother. 77 Jessica assertively utters the sentence ‘Ben is my brother’. It seems that in this case Jessica has managed to assert the proposition which attributes the property of being Jessica’s brother to Ben. Ben is in fact her brother, so Jessica has managed to say something true, even though she herself is not in a particularly good position to recognize that what she said is true (aside from the testimony of her parents). Furthermore, I don’t think this ability is limited to assertions about Ben. For example, I think Jessica also has referential competence with expressions such as ⎡n’s brother⎤ for names n of her friends and neighbors. Earlier I said I had two general points to make about referential competence. The second point is that referential competence is a relatively discrete rather than a graded notion. A speaker is either referentially competent with a word or not – there is no matter of a speaker being “more” or “less” referentially competent with a word than another speaker. Consider again Putnam’s referential competence with ‘beech’ and ‘elm’. Putnam initially knows very little about the meaning of these two terms, other than that they indicate two different kinds of tree. Now let’s suppose that Putnam becomes something of an expert on beeches and elms. As a result of his newfound expertise, he can readily distinguish the two kinds of tree when he is out walking in the woods, and he can explain in very precise terms how the two trees differ biologically. These newfound abilities do not make Putnam more referentially competent with ‘beech’ and ‘elm’. Before becoming an expert, Putnam was already able to use ‘elm’ in the sentence ‘There is an elm in my backyard’ to assert the proposition expressed by this sentence, and there is no way for him to improve on this ability. After becoming an expert, Putnam is able to use ‘beech’ 78 and ‘elm’ to say or assert things no better than he was before, even though there is a significant sense in which he is now in a better position to understand what he says or asserts when he uses these terms. Recognitional Competence A speaker who is recognitionally competent with a word has appropriately linked the word with a recognitional capacity which enables her to reliably recognize the conditions under which it is correctly applied. For example, Smith is recognitionally competent with ‘red’ to the extent that he is correctly disposed to apply ‘red’ to red things and abstain from applying it to non-red things. Putnam is recognitionally competent with ‘elm’ to the extent that he is disposed to apply it to elms and to refrain from applying it to non-elms. The notion of recognitional competence presumably can, in principle at least, be applied to whole sentences as well as sub-sentential expressions, though it is perhaps more difficult to say what the notion amounts to for the case of sentences. Max Cresswell seems to have some notion recognitional competence in mind when he writes, “What I want to put forward as the semantic competence of a native speaker is nothing more nor less than his ability, when presented with a sentence and a situation, to tell whether the sentence, in that situation, is true or false.” 16 Cresswell doesn’t offer much in the way of qualification or explanation here, so it is difficult to say what exactly he has in mind. 17 Presumably a “situation” is something like a possible circumstance, state of affairs, or 16 Cresswell (1978, p. 10). 17 He clearly seems to think that this conception of competence is a natural companion to his possible world semantics, but he isn’t very clear about how the connection should be understood. 79 world. This is still vague, and surely something needs to be said about how the situation is presented to a speaker. Is the speaker’s ability to be analyzed in terms of counterfactual claims about how he would be disposed to evaluate the sentence if he was in the situation? Or imagining the situation? Or considering a verbal description of the situation? Cresswell doesn’t say, and each of these alternatives might yield a significantly different view. The most natural way to extend the idea of recognitional competence to a sentence would be to suppose the speaker is placed in the circumstance, and observing with his senses, and for the purposes of discussion, I will assume this is what Cresswell has in mind. So, for example, the extent to which one is recognitionally competent with the sentence ‘Felix is meowing’ is the extent to which one is disposed to correctly determine whether this sentence is true or false in various circumstances considered observationally. Cresswell seems to think that this sense of competence is all there is to being competent with a sentence, which I deny, since in my view there are a variety of ways in which a speaker can be linguistically competent. But Cresswell is right here in that recognitional competence is one way in which speakers are competent with sentences. However, Cresswell is mistaken to imply, as he seems to, that speakers who are recognitionally competent with a sentence have some kind of infallible ability to determine whether or not the sentence is true in any given circumstance. 18 I think it is better to say that speakers who are recognitionally competent with a sentence have a 18 Presumably Cresswell would agree, if it were pointed out to him, that the ability can’t be infallible. But in his paper he doesn’t offer any qualification to this effect. 80 generally reliable capacity for determining whether or not the sentence is true given a wide (but suitably restricted) range of circumstances. There is obviously some relation between a speaker’s recognitional competence with a sentence (or phrase) and his recognitional competence with the words that make up the sentence (or phrase). My recognitional competence with the sentence ‘Felix is meowing’ no doubt has something to do with my recognitional competence with ‘Felix’ and ‘meowing’. However, it is rather difficult to say exactly what this relation is. One difficulty is that a speaker’s recognitional competence with a sentence doesn’t always compose in any obvious fashion out of the speaker’s recognitional competence with the expressions that make up the sentence. For example, suppose that I have a high degree of recognitional competence with ‘Felix’, ‘dark’, and ‘alley’; it does not follow that I thereby have a high degree of recognitional competence with ‘Felix is in a dark alley’. I may generally be very poor, when observing situations where Felix is in a dark alley, at recognizing that ‘Felix is in a dark alley’ is true. Here are a few more examples (based on Fodor’s critique of recognitional concepts). 19 One might be recognitionally competent with ‘pet’ and ‘fish’, but not with ‘pet fish’. Or one might be recognitionally competent with ‘pet fish’ but not more generally with ‘pet’ or ‘fish’. Similarly, one might be recognitionally competent with ‘red’ and ‘hair’, but not with ‘red hair’. Another important point about recognitional competence is that it is a highly graded notion. There is a very clear sense in which a speaker can be “more” or “less” 19 Fodor (1998). 81 recognitionally competent with an expression compared to another speaker. This point is rather obvious for natural kind terms such as ‘elm’. Clearly one speaker can be better than another at recognizing whether or not ‘elm’ applies to a given object. Similarly, one speaker can be better than another at recognizing whether or not ‘red’ applies to a given object. I want to emphasize (as I did in chapter one) that recognitional competence is part of linguistic understanding. Beyond the mere ability to say things, one must also know how what is said is related to the world one is actively engaged with. This ability requires knowing how to determine whether a given saying is true or false in actual situations we find ourselves in. 20 It also requires, say for a simple subject-predicate sentence, the ability to determine what object in the world the subject designates, and what objects in the world instantiate (or fail to instantiate) the property expressed by the predicate. Our understanding of the world influences what we say about it, and what is said about the world influences both our understanding of it and our actions in it. This two-way exchange would not be possible without understanding how language and world are related. Recognitional competence is what secures our understanding of the language- world connections. Recall the point from chapter one that, if a speaker isn’t recognitionally competent with the predicate ‘mimsy’, then he has no basis other than hearsay for asserting that anything is ‘mimsy’ (since he has no way to determine for 20 Evaluating propositions or sentences relative to descriptions of counterfactual circumstances is a more complicated case. Is this a species of recognitional competence, or inferential competence? A mix of both? I will set this sort of case aside here. 82 himself whether anything is ‘mimsy’). 21 Likewise, if he is informed that something is ‘mimsy’, this assertion cannot significantly modify his understanding of the world or influence his actions in it. If we are going to successfully communicate about the world around us, it is not enough to merely assert propositions which say things about the world. We must also have a sufficiently robust perspective on the propositions asserted to understand how they relate to the world. Recognitional competence will count as part of linguistic understanding more for some words than others, and this is determined by what is presupposed in normal conversational exchanges. For example one can be more or less recognitionally competent with a name, but this in general won’t be part of linguistic understanding for names. Speaker’s who are referentially competent with ‘Cary Grant’ will vary in terms of their ability to recognize when Cary Grant is about, and thus in their ability to correctly apply the name. In spite of this variation in recognitional competence, all the speakers will, for the most part, be uniformly competent with ‘Cary Grant’ because in normal conversational exchanges it will not generally be presupposed that conversational participants know what Cary Grant looks like (i.e. it would not be presupposed that any normal speaker competent with ‘Cary Grant’ knows what the man looks like). Of course, there will be specific conversational contexts where such knowledge is presupposed. In a conversation amongst a group of friends who have just watched a Cary Grant film together and are discussing the film, it will be presupposed that every party to the conversation knows what Cary Grant looks like. This presupposition may even play a key 21 i.e. no basis from his own observations of the world around him. 83 role in conveying certain bits of information in the conversation. However, the point here is that knowledge of what Cary Grant looks like wouldn’t be presupposed in virtually any conversational context (and the same goes for any other name). In the case of ‘red’, however, it will be widely presupposed in normal conversational exchanges that participants can recognize which things ‘red’ applies to and which it doesn’t. To give a very simple example of this, consider a command where one speaker says to another, “Bring me the red book lying on the table.” Here it is presupposed in the conversation that all parties to the conversation know how to determine which things are “red” and which aren’t. Inferential Competence The third kind of lexical competence is what I call inferential competence. Inferential competence has to do with a speaker’s knowledge of various kinds of “semantic relationships” which hold between words. The relationships are “semantic” in the sense that the relations hold between the words in virtue of some relation holding between the semantic contents of the words in question. Here is a list of some of the semantic relations I have in mind along with an example to illustrate (each example is an example of something that most competent speakers supposedly know due to their knowledge of the meaning of the words in question): synonymy (‘brother’ means the same as ‘male sibling’); redundancy (‘naked nudes’ is redundant); analyticity (nightmares are dreams); entailment (if ‘George ate an omelette’ is true, then so is ‘George ate some food’); semantic anomaly (‘The rock is happy’ is anomalous); meaning containment 84 (‘dwelling’/‘apartment’). 22 Being inferentially competent is a matter of recognizing relations such as these. Additionally, we might include abilities such as the ability to paraphrase a sentence, or give the definition of a word 23 (though arguably these abilities are simply founded on a speaker’s ability to recognize the synonymy relation). It is beyond the scope of the present discussion to discuss all of these semantic relations in detail, but let’s take a closer look at entailment. What I have in mind here is a speaker’s ability to make correct judgments of the following schematic form: If the set of sentences S is true, then the sentence P must also be true. Hence, I take the entailment relation to be a modal relation between sentences. The set of sentences S entails the sentence P just in case P must be true if all the sentences in S are true. 24 For example, if a speaker knows that sentence (4) must be true if (3) is, this is an instance of inferential competence with both ‘brother’ and ‘male’. (3) Chris is a brother. (4) Chris is male. A speaker who knows that (3) entails (4) will generally know this by way of his understanding of the meaning ‘brother’ and ‘male’. The speaker has a grasp of the meaning of ‘brother’ and the meaning of ‘male’ which is non-transparent yet sufficiently 22 See Katz and Nagel (1974, p. 313). 23 Marconi (1996). 24 There is no circumstance of evaluation relative to which all the sentences in S are true and P is false. 85 robust to secure his knowledge of the entailment. 25 Thus, knowledge of the entailment relation is a manifestation of knowledge of meaning. To give another example, knowing that sentence (6) must be true if sentence (5) is true is an instance of inferential competence with ‘dog’ and ‘animal’. (5) Rufus is a dog. (6) Rufus is an animal. Now it should not be assumed that, because I am describing the ability to judge correctly that (6) must be true if (5) is as a kind of competence with ‘dog’ and ‘animal’, I am suggesting that this judgment is analytic (i.e. that the judgment would be available to any speaker who understand (5) and (6)). Far from it, in my view a speaker could be merely referentially competent with ‘dog’ and ‘animal’ and thus in some sense understand both (5) and (6), but would not be in a position to judge that (6) must be true if (5) is (more on this below). I also should not be understood as suggesting that it is knowable a priori that if Rufus is a dog then Rufus is an animal, though I do of course think that the proposition expressed by (6) is a necessary consequence of the proposition expressed by (5). Inferential competence is distinct from both referential competence and recognitional competence. A speaker could be referentially competent with both ‘dog’ and ‘animal’, yet not recognize the entailment between (5) and (6). Likewise, a speaker 25 A speaker might come to know that (3) entails (4) merely on the basis of metalinguistic knowledge. Another speaker might inform him that anything to which ‘brother’ applies is also something to which ‘male’ applies. While I think that this merely metalinguistic understanding would constitute a minimal advance in the speaker’s understanding of the words in question, I really have in mind something more robust. What I really have in mind is a scenario where the speaker has some language-independent (or at least non-metalinguistic) way of grasping both the property of being a brother and the property of being male, and is recognizing the entailment relation on this basis (and of course on the basis of his understanding how the words are associated with the properties). 86 could be recognitionally competent with ‘dog’ and ‘animal’ yet not know that (5) entails (6). If recognitional competence were infallible, then one might be able to argue that even though a speaker might not assent to the sentence ‘If ‘Rufus is a dog’ is true, then ‘Rufus is an animal’ is true’, she might still in some tacit sense know that what is expressed by this sentence is true because she is disposed to apply ‘dog’ only to those things which she is also disposed to apply ‘animal’. However, since recognitional competence isn’t infallible, it is difficult to see how recognitional competence with ‘dog’ and ‘animal’ could really amount to the knowledge that (5) entails (6). A speaker could have a fairly high degree of recognitional competence with both ‘dog’ and ‘animal’ yet refrain from applying ‘animal’ to an object she is disposed to apply ‘dog’ to. Inferential competence also seems to be a graded notion. A speaker is inferentially competent with a word to the extent that she in some sense recognizes various relations that hold between the content of the word and the content of other words, and clearly this can come in degrees. One speaker can be more or less inferentially competent than another speaker. One speaker might be able to give a richer and more accurate definition of a word than another speaker. Or one speaker might be able to list more synonyms for a given word than another speaker. One speaker might recognize that both ‘The rock is happy’ and ‘The rock is truculent’ are anomalous while another speaker recognizes the anomaly in the first sentence but not in the second (because he is unclear about the meaning of ‘truculent’). In each instance, it seems to me that the first speaker is more inferentially competent with the word in question than the second. Notice that the last example also shows that inferential competence is holistic. A 87 speaker can be less inferentially competent with ‘rock’ due to an impoverished understanding of some other word. Again, I want to emphasize the point from chapter one that inferential competence is a vital part of linguistic understanding. Inferential competence secures the ability to understand how the various sayings in a conversational exchange are related to each other. This ability is vital to understand and competently participate in any coordinated conversational exchange, and it is thus an integral part of linguistic communication. Conclusion The purpose of this chapter has been to describe three kinds of linguistic abilities that contribute to competent use. There are many tasks yet unfinished. There is the task of showing that semantic theory alone can’t account for these three abilities (chapter four). And there is the task of determining what the limits of mere referential competence may be (chapter five). But most importantly, there is the task of explaining how recognitional and inferential competence are vital to linguistic understanding. I have so far sketched a few reasons why I think they are part of linguistic understanding, but more needs to be said (in chapter six) to fully defend this view. 88 Chapter Four Why Semantic Theory Can’t Be a Theory of Understanding II: Predicates Introduction In this chapter, I complete my case against (PC′) A semantic theory can be adequate, explaining everything we want a semantic theory to explain, while also serving as an adequate theory of linguistic understanding, and thus also explaining everything we want a theory of linguistic understanding to explain. In chapter two, I presented the argument for proper names. Here I extend the argument to simple predicates. At this stage I will also appeal to the three kinds of lexical competence discussed in chapter three. Linguistic understanding is a graded and multifaceted phenomenon, and an adequate account of it must reflect this. Recall that semantic theory is supposed to account for linguistic understanding doxastically in terms of propositional knowledge. If we had an adequate semantic theory T of some natural language (i.e. which assigned the correct contents to each word of the language, and told the correct story for how these contents compose to give the contents of complex expressions), then the idea is that we can account for the linguistic understanding of an individual speaker S by saying that (1) S knows T. My claim, however, is that (1) can’t provide an adequate account of the graded and multifaceted nature of linguistic understanding. Specifically, my claim is that we can’t account for the fact that S can have different kinds and levels of competence with 89 individual expressions by appeal to (1) – and the project for this chapter is to demonstrate why this is so. Preliminary Considerations Let’s review some of the results from chapter two. Consider a speaker who is merely referentially competent with ‘Hesperus’ and ‘Phosphorus’. One of the challenges facing a theory of linguistic understanding is to explain how such a speaker could fail to know that ‘Hesperus’ and ‘Phosphorus’ are synonyms. Knowledge of this synonymy relation is a species of inferential competence, but this kind of inferential competence isn’t part of competence with names. Competence with a name, regardless of whether it is minimal or more robust, doesn’t require that one know which other names in one’s repertoire are co- referential with it. Being at least minimally competent with the names, the speaker does in some sense understand both ‘Hesperus’ and ‘Phosphorus’, yet this understanding falls short of knowing that the names have the same meaning (referent). Essentially, the challenge is to explain how such a speaker can fail to know (2) ‘Hesperus is Phosphorus’ is true. One of the key arguments from chapter two showed that a satisfactory account of how such a speaker can fail to know (2) drives us toward a knowledge attribution that is considerably more schematic than would be had from knowing a semantic theory. What was argued in chapter two is that supposing our speaker knows (3.1) and (3.2) doesn’t adequately explain how she fails to know (2). (3.1) ‘Hesperus’ refers to Hesperus. 90 (3.2) ‘Phosphorus’ refers to Phosphorus. A better explanation of mere referential competence with names drives us in the direction of supposing that our speaker has more schematic knowledge, along the lines of knowing (4.1) and (4.2) (4.1) ∃x(x = the referent of ‘Hesperus’ & for every predicate F, to assertively utter ⎡Hesperus is F⎤ is to say of x that it is F). (4.2) ∃x(x = the referent of ‘Phosphorus’ & for every predicate F, to assertively utter ⎡Phosphorus is F⎤ is to say of x that it is F). Again, (4.1) and (4.2) are consistent with any (directly referential) interpretation of ‘Hesperus’ and ‘Phosphorus’ and thus aren’t adequate for semantic theory because they don’t specifically state what the contents of ‘Hesperus’ and ‘Phosphorus’ are. So in attempting to give a doxastic account of the competence of a speaker who is merely referentially competent with two co-referential names, we are driven to suppose that her competence stems from knowing something weaker and more schematic than is required for semantic theory. Since (4.1) and (4.2) fail to identify the referent as the same in both cases it is obvious that knowing these weakened theorems will not result in knowing (2) – as desired. Before turning to a discussion of predicates, let’s review the working model introduced in chapter two for the notion of “knowing a semantic theory.” According to this model, we imagine that a rational agent is taught the semantic theory in question and then compare what he would thereby know to what ordinary competent speakers of English know. We imagine that the rational agent already knows the metalanguage in 91 which the theory is stated, and that teaching the agent the theory is a matter of teaching the agent the sentences which comprise the axioms of the theory. As a result of the instruction, the rational agent will know what is stated by the axioms of the theory and what is stated by any theorems which can be derived (according to a syntactic notion of “provability” that is known to the agent) from the axioms. I think this is a reasonable model in which to analyze the notion of “knowing a semantic theory,” however I won’t defend this claim here (see chapter two for a defense of the model). One thing I do want to discuss more extensively here is the idea that the metalanguage is a so-called “language of thought.” It isn’t a natural language, but it is a language-like medium of cognitive representation, and is acquired in a fundamentally different way than natural language is acquired. We really need an assumption like this in order to account for the fact that speakers learn a natural language without first knowing any other natural language. So on this model, learning and understanding a semantic theory for a natural language is a matter of learning and understanding the axioms and theorems stated in a language of thought. I will also assume for my purposes that a speaker’s understanding of expressions in the language of thought doesn’t derive parasitically from her understanding of expressions in any natural language she knows. Rather, I will assume that the speaker’s grasp of expressions in a natural language she knows always derive parasitically from her grasp of expressions in the language of thought. In order to explain first language acquisition in the model under consideration, it must be assumed that, in some cases at least, the speaker’s understanding of an expression in natural language derives from her 92 antecedent understanding of the language of thought, but I will assume that all expressions of natural language are parasitic on expressions of the language of thought in this way. It is entirely plausible, for example, to suppose that a baby can think, “I’m hungry” before he can express this thought in words, and the baby could even have this thought if he never learned a natural language. According to our model, we suppose that the baby is entertaining this thought via the language of thought. And it is plausible to suppose that, when the baby later comes to understand the sentence ‘I’m hungry’, his understanding of this sentence derives, at least in part, from his understanding of its counterpart in the language of thought. This assumption is merely a simplification adequate for our purposes here, but it may be an oversimplification. When an individual is thinking privately, it is plausible that this is, at least in some cases, partly a matter of “talking to oneself” in a natural language. If one wants it to come out that whenever an individual is thinking privately, the thinking is mediated by the language of thought, then it is plausible that some expressions in the language of thought might be parasitic on expressions of natural language. I am making the simplification to avoid confusion, and nothing turns on it. The only consequence is that, in my terminology, there can be cases where an individual’s private thoughts are mediated, at least in part, by natural language rather than the language of thought. I should emphasize that all this is a model for understanding only. I am not assuming that there “really is” a language of thought distinct from natural language, or that in order to know a semantic theory, a speaker would literally need to have the theory 93 inscribed in their language of thought. All I claim is that for my purposes here the notion of knowing a semantic theory can be analyzed in terms of this model. Now for some terminology. According to the ordinary philosophical usage of the term “concept,” one is generally taken to have a concept in any case where one has the capacity to represent, express, grasp, or entertain an appropriate content. For example, if one is able to represent the property of being a dog, one is taken to have the concept “dog.” One problem with this usage is that it obscures, in terms of our model, the very important distinction between representational capacities that are secured by natural language, and representational capacities which are secured by the language of thought alone. We need to allow for the possibility that one can, through using the language of thought, come to understand and use some natural language words which have a content that cannot be expressed in the language of thought. Likewise, there may be contents expressible in the language of thought which cannot be expressed in natural language. A speaker might be able to think about dogs via her language of thought, and thus be able to represent the property of being a dog in her language of thought even though she doesn’t know any natural language word (or phrase) that expresses the property of being a dog. But a speaker might also be in a position where she is merely referentially competent with ‘dog’, and thereby able to say things about dogs, and even entertain thoughts about dogs, yet not have any means by which she can represent the property of being a dog in her language of thought (set aside for now the question of how the speaker might be in this position, the important point for the moment is that I want to allow for this possibility). It of course could also turn out that the speaker is referentially competent 94 with ‘dog’ and in addition has a word in her language of thought which represents the property of being a dog, but does so in a way that isn’t in any sense parasitic on the speaker’s use of ‘dog’. Similarly, the speaker’s use of ‘dog’ may not be parasitic in any sense on the item in her language of thought which represents the property of being a dog. In this case, the speaker has two entirely distinct means by which she can represent the property of being a dog. She can entertain the property by means of her language of thought, or she can entertain it by means of ‘dog’. We thus have the distinction between representational capacities which are given to an individual in virtue of knowing a natural language, and representational capacities which are given to the individual in virtue of her language of thought. To keep this important distinction straight in terms of concept-talk, I will reserve the phrase “linguistic concept” to indicate words of natural language. “Having a linguistic concept” will simply be a matter of being referentially competent with a word of natural language, since being referentially competent with the word will grant one the capacity to talk and think about (i.e. represent) the content of the word. Since on my usage, linguistic concepts are words, I will refer to linguistic concepts by the same quotational device used to mention words. I will speak, for example, of an individual who is referentially competent with ‘dog’ as having the linguistic concept ‘dog’. I will reserve the phrase “cognitive concept” for words in the language of thought (which is the metalanguage in our model). I will use italics to mark cognitive concepts (i.e. to distinguish them from words of English). For example, ‘dog’ is a cognitive concept whose content is the same as the word ‘dog’. The same conventions for use and 95 mention apply for cognitive concepts. I will speak, for example, of an individual having the cognitive concept ‘dog’. In a use of the metalanguage sentence, ‘Smokey is a dog’, the occurrence of ‘dog’ does not refer to the cognitive concept ‘dog’, but rather this cognitive concept is being used in this language of thought sentence to express the property of being a dog. To rehearse the terminology, imagine we have an English speaker who initially doesn't have the word ‘red’ in her repertoire. She has had experience with many colors besides red, but she hasn’t had any perceptual contact with red objects, so she doesn't have the cognitive concept ‘red’ either. Now suppose she gains experience with red things without learning the word ‘red’. Her experience with red things, together with her experience of different colors, presumably grants her the relevant sort of distinctions so that she can form the cognitive concept ‘red’. But she doesn’t yet have the word ‘red’, so she doesn't have the linguistic concept ‘red’. Alternatively, suppose the speaker picks up the word ‘red’ in conversation. But in the course of the conversation the speaker doesn't learn much of anything about what ‘red’ means, just that it indicates a type of color. Also suppose that she hasn’t yet had perceptual contact with any red object. The speaker is merely referentially competent with ‘red’. In this case, the speaker has the linguistic concept ‘red’, but doesn't plausibly have the cognitive concept ‘red’. Now suppose that the speaker gains some experience with red objects so that she now has the cognitive concept ‘red’. But suppose that she still doesn’t know that ‘red’ indicates this color she has recently experienced. In this case, the speaker has both the linguistic concept ‘red’ and the cognitive concept ‘red’, but she doesn’t have the linguistic concept in virtue of 96 having the cognitive concept. Likewise, she doesn’t have the cognitive concept in virtue of having the linguistic concept. The facts which explain her possession of the cognitive concept ‘red’ are facts about her experience with different colored objects (including red ones). The facts which explain her possession of the word ‘red’ are facts having to do with her deferential intention to use the word with its meaning, her communicative participation in a community of language users who use the word ‘red’ to express the property of being red, etc. Some of the other ‘red’-users in her community presumably will have the cognitive concept ‘red’ appropriately linked with the linguistic concept ‘red’, and this fact may have to figure in the overall explanation of the speaker’s referential competence with ‘red’. But the speaker herself need not have her linguistic concept ‘red’ and her cognitive concept ‘red’ linked in any significant way. The facts which explain her possession of the linguistic concept may include facts about other speakers possessing the cognitive concept, but it need not include the fact that she possesses the cognitive concept. So in this way the facts which explain her possession of the linguistic concept need not depend on facts about her possessing the cognitive concept. And the facts which explain her possession of the cognitive concept need not depend on facts which explain her possession of the linguistic concept. In this case, the speaker has two distinct means of representing the property of being red. Let me address two difficulties that might arise at this stage. First, the terminology might suggest that cognitive concepts aren’t linguistic and that linguistic concepts aren’t cognitive. I do not intend this suggestion – cognitive concepts are, by 97 hypothesis, language-like in nature, and linguistic concepts are just as much a part of a speaker’s cognitive architecture as are cognitive concepts. Second, for simplicity, the model precludes the possibility that a speaker might learn and understand a new natural language word via one or more linguistic concepts. This can and does happen. For example, a speaker might learn ‘brother’ by picking up a dictionary and learning that ‘brother’ means the same as ‘male sibling’. This is arguably a case where the speaker learns and understands ‘brother’ via the linguistic concepts ‘male’ and ‘sibling’. However, it would complicate the model unnecessarily to allow for this possibility. Nothing I have to say will turn on this possibility being precluded from the model. The Central Argument We are finally ready to discuss predicates. Consider a young girl, Kelly, who is merely referentially competent with ‘sister’. Kelly has an older sister, Sara. Both of Kelly’s parents died shortly after she was born, and since then Kelly has been raised by Sara alone. Sara says things to Kelly such as, “You’re my sister,” “Come here sister,” and “We are sister’s.” When Kelly visits her friends who have sisters, she overhears her friend’s parents saying things to her friends such as “Go tell your sister it’s time for dinner,” or “Where is your sister?” So Kelly has picked up ‘sister’ conversationally and is referentially competent with the word. When Kelly says “sister” she expresses the property of being a sister. When she assertively utters ‘Sara is my sister’, she correctly asserts a proposition which says that Sara is her sister. 98 But Kelly’s grasp of ‘sister’ is nonetheless very dim. Even though she has the linguistic concept ‘sister’, it is doubtful (I shall argue) that she has the cognitive concept ‘sister’. Kelly does not know (5) or (6). (5) ∀x(x is a sister → x is female). (6) ∀x(x is a sister → x is a sibling). Knowing (5) and (6) is not sufficient for being robustly competent with ‘sister’, but it is necessary, and will serve here as a mark of the difference between being merely referentially competent and being robustly competent. Kelly mistakenly thinks that sisters can be male. She mistakenly thinks that one of her male friends, John, and his older brother, George, are sisters (she hasn’t yet voiced this opinion to anyone, so she hasn’t yet been corrected). Kelly rejects the sentence, ‘Every sister is female’, and isn’t in any sense disposed to apply ‘female’ whenever she applies ‘sister’ (being instead in some cases disposed to apply ‘male’ rather than ‘female’). Also, since Kelley doesn’t have the cognitive concept ‘sister’ (I offer more motivation for this claim below), there isn’t any language of thought sentence she could accept which would amount to knowing (5). So Kelly does not know (5). Kelly also doesn’t know (6). Suppose that Kelly simply doesn’t know the word ‘sibling’, and doesn’t in any sense have the cognitive concept ‘sibling’. Kelly doesn’t know such words as ‘mom’, ‘dad’, ‘parent’, or ‘family’, and doesn’t have the cognitive concepts ‘mom’, ‘dad’, ‘parent’, or ‘family’. Sara decided not to tell Kelly about their parents until she is older. Kelly has never heard any of her friends use the words ‘mom’, 99 ‘dad’, ‘parent’, or ‘family’ (Sara asks everyone to carefully avoid such terminology when they are around Kelly). Kelly doesn’t attach any special significance to the older members of her friend’s homes. As far as she is concerned, they are simply older people who live in a house together with her friends, just the way that she lives in a home with her sister. Kelly sometimes visits a house full of unrelated college roommates (one of whom baby sits for Kelly occasionally). Kelly doesn’t in any sense distinguish between how these college roommates are related and how she is related to her sister, or how her friends are related to the other members of their households. Kelly doesn’t have any notion of “family” that is distinct from “people who live together in a dwelling.” Kelly simply doesn’t have the conceptual resources to grasp the cognitive concepts ‘family’ or ‘sibling’. She can’t conceive of relations such as “having the same parent as,” or “being in the same family as.” A fortiori, Kelly doesn’t have the cognitive concept ‘sister’ either because she likewise doesn’t have the conceptual resources to form this concept. Her understanding of the meaning of ‘sister’ is essentially that of “young housemate”. So Kelly doesn’t know (6). Even though Kelly has experience of people who are in fact sisters, this experience doesn’t put her in a position to make the sort of discriminations that she would need to make in order to form the cognitive concept ‘sister’. The only basis for discrimination she has (the only way to single out the property of being a sister) 100 comes for her linguistically via the word ‘sister’. So she only has the linguistic concept ‘sister’. 1 Let’s further suppose that Kelly has picked up the word ‘sis’ conversationally. Sara sometimes says things such as, “Come here sis” or “You’re my sis” to Kelly, and Kelly overhears the older sisters of her friends saying similar things to her friends. Kelly’s competence with ‘sis’ secures for her a linguistic concept of sisterhood (i.e. she has the linguistic concept ‘sis’), but she doesn’t have the cognitive concept ‘sis’ (or ‘sister’) as an independent item in her language of thought. 2 Because Kelly has only overheard older sisters calling younger sisters “sis,” she forms the mistaken impression that in order for x to be the “sis” of y, y must be older than x. Kelley is otherwise just as much in the dark about ‘sis’ and she is about ‘sister’, so her understanding of ‘sis’ certainly doesn’t include knowledge of (5) or (6). Kelley is not aware that ‘sister’ and ‘sis’ have the same meaning. Accordingly, she doesn’t know (what is expressed by) (7) or (8). (7) ‘Every sister is a sis’ is true. (8) ‘Every sis is a sister’ is true. 1 Kelly might be able to form a cognitive concept on the basis of the discrimination made available to her by means of the word. However, such a cognitive concept would be parasitic on the linguistic concept, and I am for simplicity ruling out such a possibility here. But even if it isn’t ruled out, this point won’t help to explain Kelly’s minimal competence with the word. If the possession of the cognitive concept is to help explain possession of the linguistic concept, then the cognitive concept can’t be parasitic on the linguistic concept. 2 Given my explicitly linguistic model of concepts, there is a distinction between the linguistic concept ‘sister’ and the linguistic concept ‘sis’. Likewise, there is a distinction between the cognitive concept ‘sister’ and the cognitive concept ‘sis’. These are all distinct concepts with the same content. 101 In fact, Kelley would explicitly deny (7) since she thinks that ‘sis’ expresses an asymmetric relation while ‘sister’ expresses a symmetric relation. There is no doubt that Kelley’s competence with both ‘sister’ and ‘sis’ is severely impoverished. One mark of her impoverishment is that she doesn’t know (5) or (6) (again, knowing (5) and (6) is only a necessary and not a sufficient condition on being robustly competent with both ‘sister’ and ‘sis’). However, when she uses ‘sister’ she has the deferential intention to use the word with the same meaning as those in her community. 3 Similarly, she uses ‘sis’ with the same kind of deferential intention. This at least secures for her mere referential competence with both words. Let’s turn to the thesis that linguistic understanding can be explained as knowledge of a semantic theory. Kelly’s minimal competence with ‘sister’ and ‘sis’ will have to explained by saying that she knows a semantic theory for English. The more robust competence of a speaker who knows (5) and (6) will likewise be explained by saying that such a speaker knows the same semantic theory for English. The thesis looks immediately implausible since we are offering exactly the same explanation in both cases. Comparing Kelly to a speaker who is fully competent with ‘sister’ and ‘sis’, we have a very striking difference in the degree of understanding. But the thesis gives precisely the same explanation for both cases. The radical difference in competence is somehow supposed to be explained by saying, “Both Kelly and the more robustly competent speaker know the same semantic theory for English.” On the face of it, this 3 Also, she has sufficient grasp of the word’s syntactic category to be able to use ‘sister’ with grammatical felicity in a wide variety of sentences. Similarly for ‘sis’. 102 doesn’t even look like an explanatory strategy that can get off the ground, but let’s dig in and see if the details offer any hope for the view. First, let’s consider the prospects for explaining Kelly’s impoverished competence in terms of her knowing a semantic theory for English. We will need to explain how it is that Kelly fails to know (7) and (8). Let’s suppose that the (correct) semantic theory for English has the following axioms for ‘sister’ and ‘sis’. (9) ‘sister’ expresses the property of being a sister. (10) ‘sis’ expresses the property of being a sis. 4 As usual, the italics are used to indicate the metalanguage (i.e. the language of thought) in which the theory is stated. In the model, explaining how Kelly fails to know (7) and (8) will require the same metalinguistic passing of the explanatory buck that we saw for names in chapter two. We have a bit of a mystery on our hands. In spite of her ignorance, Kelly does in some sense understand ‘sister’ and ‘sis’, as is evident in her referential competence with each word. How is it that she understands the words yet fails to recognize that they have exactly the same meaning? A semantic theory with (9) and (10) as axioms forces us into the following explanation. It is “as if” Kelly knows a metalanguage in which the theory is stated but she is similarly in the dark about two words in the metalanguage – ‘sister’ and ‘sis’. She understands these two words (she knows them as part of her knowledge of the 4 For truth conditional semantics, the relevant clauses would be (9*) ‘sister’ applies to x iff x is a sister. (10*) ‘sis’ applies to x iff x is a sis. Nothing in my argument turns on the choice between a truth conditional semantics and a propositional semantics, though I will couch the argument in the latter framework. I assume for the purposes of discussion that ‘sister’ and ‘sis’ are one-place predicates. Perhaps some would argue that they are two-place predicates, but in any case nothing in my argument turns on this. 103 metalanguage), yet she fails to realize that they have the same meaning. In particular, she fails to realize that (what is expressed by) (7*) and (8*) are true. (7*) ‘Every sister is a sis’ is true. (8*) ‘Every sis is a sister’ is true. Here again we really aren’t explaining anything, at least not in an informative way. All we are doing is brushing the mystery under a metalinguistic rug. We have mystery about Kelley’s impoverished understanding of ‘sister’ and ‘sis’ and the “explanation” offered is, in effect, to say, “Well, suppose we have the same mystery at the metalinguistic level.” Notice also that the “explanation” offered by (9) and (10) for Kelly’s failure to know (7) and (8) is even worse than the corresponding explanation for names we had in chapter two. Since Kelly doesn’t have the cognitive concepts ‘sister’ and ‘sis’, we can’t avail ourselves of the trick considered for names in chapter two. In chapter two, we allowed the trick of taking names in the metalanguage as (mere) markers of sense. The corresponding trick here would be to take ‘sister’ and ‘sis’ as markers of different senses (i.e. as markers of different modes of presentation of the property of being a sister). Doing so allows us to interpret (9) and (10) as telling us that Kelly understands ‘sister’ and ‘sis’ via different senses, and this in turn allows a minimal (though not very informative) explanation of why Kelly fails to know (7) and (8). More specifically, what the trick says is that Kelly has two cognitive concepts for representing the property of being a sister, and each provides a different perspective on the property. The two perspectives then carry over for Kelly in her understanding of the two corresponding words, since she learns the words via the cognitive concepts. Even though this little trick 104 is not particularly informative, we cannot even avail ourselves here of its minimal explanatory power, because Kelly cannot plausibly be taken to have the cognitive concepts ‘sister’ or ‘sis’. This doesn’t mean that we cannot avail ourselves at all of the idea that, in understanding the two words, Kelly entertains the property of being a sister via two different (but very minimal) perspectives. I do ultimately want to exploit this idea. The point is that we can’t, in Kelly’s case, reasonably make (9) or (10) speak about senses or perspectives at all, because (9) and (10) simply don’t apply to Kelly’s situation. Just as we saw with names, there is another difficulty here. Consider Kelly’s cognitive state immediately prior to picking up ‘sister’ in conversation (and suppose she picks up ‘sister’ before ‘sis’). There is no reason to suppose that she can at that stage entertain any proposition that has the property of being a sister as a component part. She need not be aware that her and Sara have any special familial relationship. She surely thinks of Sara as someone who takes care of her, who loves her, who she loves, and who lives with her. But I see no reason to suppose that her experience has given her the cognitive concept ‘sister’, or put her in the position to entertain propositions which have sisterhood as a component part. But notice that according to our model of understanding, Kelly would first need to learn ‘sister’ before she could learn (9), and thus would have the capacity to entertain thoughts about the property of being a sister before she acquired (9). So supposing that Kelly’s competence with ‘sister’ is explained by her knowing (9) really paints her in an overly robust cognitive state prior to acquiring the word. Furthermore, as I have argued, it is implausible to think that Kelly has the cognitive concept ‘sister’ even after she has acquired ‘sister’. After Kelly has acquired 105 ‘sister’ (and before she acquires ‘sis’), Kelly only has the linguistic concept ‘sister’, and has no other means of representing the property of being a sister. But on our model of understanding, (9) effectively assumes that she does have another means of representing the property of being a sister – namely, through the cognitive concept ‘sister’. But this assumption fails for Kelly, and thus (9) cannot possibly explain her competence with ‘sister’. Similarly, (10) cannot possibly explain her competence with ‘sis’. As we saw with names, seeking a more satisfactory explanation for Kelly’s state of understanding really pushes us in a direction where we attribute weaker, more schematic knowledge. Suppose that Kelly’s competence with ‘sister’ and ‘sis’ is due to knowing (11) and (12) instead of (9) and (10). (11) ∃x(‘sister’ expresses x & for any name n, to assertively utter ⎡n is a sister⎤ is to say of n, and of x, that the former has the latter). (12) ∃x(‘sis’ expresses x & for any name n, to assertively utter ⎡n is a sis⎤ is to say of n, and of x, that the former has the latter). 5 As we saw with names, this proposal can explain everything that (9) and (10) have trouble explaining. We can explain how Kelly fails to know (7) and (8) without simply passing the explanatory buck up to the metalanguage. 6 Also, we can see how Kelly might acquire ‘sister’ or ‘sis’ without first having the cognitive capacity to entertain thoughts about the property of sisterhood, and how she can be merely referentially competent with these two words without having the cognitive concepts ‘sister’ or ‘sis’. 5 I won’t worry here about the distinction between object language and metalanguge. We can just assume that both languages shares a set of names, and that the variable ‘n’ is ranging over these names. I will make a similar assumption for similar clauses below. 6 Her ignorance would be traceable to the fact that she fails to believe that ∃x∃y(‘sister’ expresses x & ‘sis’ expresses y & ∀z(z has x iff z has y)). 106 As we saw before, however, (11) and (12) do not specify the content of ‘sister’ or ‘sis’. (11) and (12) are consistent with any interpretation which assigns a property as the content of ‘sister’, and a property as the content of ‘sis’ (the interpretation need not even assign the same property to both words). 7 So (11) and (12) would, if they were the only axioms of a semantic theory governing ‘sister’ and ‘sis’, fail to meet a basic constraint of semantic theory – that for each meaningful expression in the language, the theory must specify what the content of the expression is. As I did with names, let me clarify that I am not denying that Kelly knows (9) and (10). I grant that she does know (9) and (10). What I am denying is that we can explain her mere referential competence with ‘sister’ and ‘sis’ in terms of her knowing (9) and (10). My suggestion instead is that her competence is to be explained, at least in part, in terms of her knowing something more schematic such as (11) and (12). Similarly, Kelly knows (9) and (10) at least in part because she knows (11) and (12). Her knowledge of (11) and (12) thus serves to explain her mere referential competence with ‘sister’ and ‘sis’, and also her knowledge of (9) and (10). The fact that Kelley knows (9) and (10) and the fact that she is referentially competent ‘sister’ and ‘sis’ are coeval facts about Kelly with essentially the same explanatory basis. This explanatory basis includes the fact that she knows (11) and (12), together with other facts, such as the fact that she has reasonably extensive knowledge of English, 8 the fact that she has acquired both ‘sister’ and ‘sis’ in a causal chain of communication with other English speakers (extending back 7 Of course, the properties assigned would have to be of the right type to fit with the syntactic category of the words, but this is a minimal constraint. 8 Her knowledge can still be fairly extensive even though she lacks words to express most familial notions. 107 to things instantiating the property of being a sister), and the fact that she has formed appropriate deferential intentions to use these words with the meaning they have for others in her community. We can fill out the explanation just as we did for names. Kelly knows (11) and (12), so the following attributions are true. (11.1) Kelly knows that ∃x(‘sister’ expresses x & for any name n, to assertively utter ⎡n is a sister⎤ is to say of n, and of x, that the former has the latter). (12.1) Kelly knows that ∃x(‘sis’ expresses x & for any name n, to assertively utter ⎡n is a sis⎤ is to say of n, and of x, that the former has the latter). Since Kelly is, among other things, part of a causal chain of communication extending back to (instances of) the property of being a sister, we can export to get: (11.2) ∃x(Kelly knows that (‘sister’ expresses x & for any name n, to assertively utter ⎡n is a sister⎤ is to say of n, and of x, that the former has the latter)). (12.2) ∃x(Kelly knows that (‘sis’ expresses x & for any name n, to assertively utter ⎡n is a sis⎤ is to say of n, and of x, that the former has the latter)). (11.3) ∃x(Kelly knows that ‘sister’ expresses x). (12.3) ∃x(Kelly knows that ‘sis’ expresses x). (11.4) ∃x(Kelly knows that for any name n, to assertively utter ⎡n is a sister⎤ is to say of n, and of x, that the former has the latter). (12.4) ∃x(Kelly knows that for any name n, to assertively utter ⎡n is a sis⎤ is to say of n, and of x, that the former has the latter). (11.3) and (12.3) show that Kelly knows (9) and (10). (11.4) shows that Kelly knows, of the proposition P expressed by the sentence ‘Sara is a sister’, that the sentence expresses P. Similarly, (12.4) shows that she knows, of P, that ‘Sara is a sis’ expresses P. Further, 108 she would intend to assert P by means of uttering each sentence. Obviously her two different perspectives on the proposition are minimal in each case, and are non- transparent so that she doesn’t know the two sentences express the very same proposition, but her minimal perspectives suffice, together with other facts about her, for mere referential competence, which after all is a minimal ability. 9 Now let’s consider how knowledge of (9) and (10) might explain a more robust competence with ‘sister’ and ‘sis’. Suppose we have a speaker, John, who knows (5) and (6). How might John know this in virtue of knowing (9) and (10)? If (9) and (10) are going to explain John’s knowledge, it can’t be that he just happens to know these other things in addition to (9) and (10). Instead, his knowledge of (5) and (6) must somehow be required in order to know (9) and (10). If knowing the semantic theory is going to have any explanatory significance for a fully competent speaker’s knowledge of (5) and (6), then knowing (5) and (6) will have to be a necessary condition on learning the semantic theory. But this then blocks any possibility of explaining the impoverished competence of a speaker such as Kelly in terms of knowing semantic theory. So we cannot simultaneously account for the minimal competence of Kelly and the robust competence of John in terms of knowing (9) and (10). In order to account for Kelly’s minimal competence, we need to suppose that Kelly can know (9) and (10) without knowing (5) and (6). This supposition is correct, even though, as I have argued, (9) and (10) ultimately can’t explain Kelly’s competence. But in order to explain John’s competence, we need to suppose (incorrectly) that knowing (5) and (6) are a necessary condition on knowing (9) 9 We would of course need other clauses to cover other kinds of sentential construction, but I won’t go into this here. 109 and (10). The requirements for explaining the competence of Kelly and John in terms of semantic theory are mutually exclusive. We thus cannot account for the spectrum of competence that we see with ‘sister’ by supposing that competence is a matter of knowing semantic theory. Notice, however, that there is one possibility left open. We cannot simultaneously account for Kelly and John, and we cannot account for Kelly alone. We might, however, be able to account for the robust competence of John alone. Let’s imagine we have a semantic theory with the following axioms. (9*) ‘sister’ applies to x iff x is a sister. (10*) ‘sis’ applies to x iff x is a sis. (13) ∀x(x is a sister iff x is a female sibling). (14) ∀x(x is a sis iff x is a female sibling). Here I have switched to the truth conditional versions of axioms (9) and (10) to make it clear what can be derived on the basis of them, but nothing I have to say turns on the difference. Let’s further suppose that the syntactic rules of derivation for the metalanguage allow the derivation of (15) – (27). (15) ∀x(x is a sister only if x is a sibling). (16) ∀x(x is a sister only if x is female). (17) ∀x(x is a sis only if x is a sibling). (18) ∀x(x is a sis only if x is female). (19) ∀x(‘sister’ applies to x iff x is a female sibling). 110 (20) ∀x(‘sis’ applies to x iff x is a female sibling). (21) ∀x(‘sister’ applies to x only if x is a sibling). (22) ∀x(‘sister’ applies to x only if x is female). (23) ∀x(‘sis’ applies to x only if x is a sibling). (24) ∀x(‘sis’ applies to x only if x is female). (25) ∀x(‘sister’ applies to x iff ‘sis’ applies to x). (26) ‘Every sister is a sis’ is true. (27) ‘Every sis is a sister’ is true. A rational agent who knows this theory will understand and accept sentences (9*), (10*), (13) – (27), and thereby know the propositions they express. 10 Such an agent would be fully competent with both ‘sister’ and ‘sis’. The agent would know (5) and (6) on the basis of accepting (15) and (16). Also, the agent would know (7) and (8) on the basis of accepting (26) and (27). This is a plausible result, because it is plausible to suppose that a speaker who fully understands the meaning of ‘sister’ and fully understands the meaning of ‘sis’ would realize that they are synonyms, and thus know (7) and (8). We thus might be able to account for the robust competence of John in terms of knowing the semantic theory just described. This move, however, won’t in any sense preserve (PC′) as I understand it. Recall that I assume a semantic theory is nothing more than a theory of “what is said” by sentences (relative to a context). The theory presently under consideration cannot be 10 One might understand the notion of accepting a language of thought sentence according to the familiar metaphor of having it in one’s “belief box.” 111 understood strictly as a theory of what is said, because it has a lot of extra information (albeit information about meaning) that is extraneous to a determination of what is said. 11 There is, however, an alternative conception of semantic theory which fits better with the present proposal. According to this alternative conception, a semantic theory is a theory about what an ordinary competent speaker knows about meaning. The justification for including (13) and (14) (which are essentially meaning postulates) in the theory (together with everything which can be derived from them), is that (13) and (14) express something which is presumably known to ordinary speakers who are competent with ‘sister’ or ‘sis’. One potential advantage of this conception of semantic theory (something which could even be cited as a motivation for it) is that it may capture an ordinary notion of “knowing the meaning of a word.” Notice that there is an ordinary notion of “knowing meaning” according to which most would agree that Kelly doesn’t know the meaning of ‘sister’ and that John does. We can do justice to this ordinary notion under the present proposal by saying that John knows the semantic clauses for ‘sister’ while Kelly does not. My project here is not to defend my favored conception of semantic theory, but to defend a certain conception of linguistic understanding. I therefore won’t attempt here to argue against this alternative conception of semantic theory. I do insist, however, that this alternative conception be properly understood in light of the fact that linguistic understanding is a graded, multifaceted phenomenon. What this brand of semantic theory gives us is a picture of what the typical or ordinary speaker of a language knows about meaning, and nothing more. It cannot give us an account of the full spectrum of 11 Since the theory is couched in truth conditional terms, there is also a gap between the truth conditions and meaning (what is said), but I won’t address this difficulty here. 112 competence. One might think that the view could account for minimal competence in terms of “partial” understanding of the semantic theory, but it cannot. One might think that the way to account for Kelly on this view is to say that she has partial knowledge of the semantic theory. I don’t know what partial knowledge could mean here other than the idea that Kelly only knows (9*) and (10*), and not (13), (14), or any of (15) – (27). But as I have already argued, clauses such as (9*) and (10*) cannot explain Kelly’s mere referential competence with ‘sister’ and ‘sis’. So it won’t do as a way to explain Kelly here to say that she has “partial” knowledge of the semantic theory. I am somewhat skeptical of the value of accounting for our “ordinary” intuitions about “knowing the meaning of a word,” if such intuitions are insensitive to the fact that linguistic understanding is a graded phenomenon. There is an inclination to think about linguistic understanding in discrete terms, and I suspect our “ordinary” intuitions about knowledge of meaning are part of this mistaken tendency. So it is possible to have a brand of semantic theory that gives an account of ordinary competence with a word like ‘sister’. The appeal of such a theory is, however, considerably attenuated in light of the fact that linguistic competence is graded. And when we consider other kinds of terms, this sort of theory starts to look even more problematic. In the case of ‘sister’, the robust level of competence is just a matter of having inferential competence, and the number of “inferences” one needs to know is relatively circumscribed. But for other terms, such as natural kind terms, the more robust level of competence often involves at least some recognitional competence, and what one needs to know (inferentially) in order to be robustly competent is considerably more open 113 ended. As we shall see, it will be impossible for the sort of semantic theory currently under consideration to give an account of the more robustly competent speaker for such terms. Consider the natural kind term ‘water’. Let’s suppose we have a young speaker, Kathy, who is merely referentially competent with ‘water’. Kathy is part of an extremely wealthy family who live together on a large compound in an extremely arid desert. Kathy’s parents fancy themselves aristocrats and above the common folk. In order to mark themselves off from the common folk, Kathy’s parents have adopted the odd practice of shunning all common forms of contact with water. They don’t drink plain water. They drink juice, milk, soda, etc, but no plain water. Also, they do not shower or bathe directly in water. To clean themselves, they have hired help that prepare moist towels anointed with water, alcohol and lavender which they use to clean themselves. Kathy has had contact with water. There is water in her drinks and in the towels she uses to clean herself. But she hasn’t had any experience with water that allows her to identify it as a distinct substance. She thinks (correctly) that her drinks and the liquid her towels are moistened with are a mix of different substances. The area where she lives is so arid that she has never experienced rain, or seen water in a lake or river. So she is never in a position to single out water as “this kind of liquid.” As a result, she simply doesn’t have the cognitive concept ‘water’. But Kathy has picked up the word ‘water’ conversationally. She sometimes overhears her parents using the term. She has gleaned enough from their usage to know that ‘water’ is a mass term for some kind of substance, but nothing beyond this. She has 114 asked her parents, “What is water?”, but they are evasive, saying “We’ll tell you when you are older.” Her parents are worried that, if she learns what water is, she might drink it and bathe in it like the common folk. Kathy is, like Kelly, the victim of an extremely odd upbringing. She is, relative to larger society, culturally stunted, and she is clearly lacking in her competence with ‘water’, but nonetheless she is referentially competent with ‘water’. When she uses the term, she has the deferential intention to use it with the same meaning as her parents. Kathy has correctly surmised that there is water in the milk she drinks (but she doesn’t know if it is a liquid, or some kind of powder that is mixed in, or what). When Kathy assertively utters the sentence, ‘There is water in milk’, she asserts the true proposition expressed by this sentence. Arguably the term ‘agua’ has become part of the English language, but in any case let’s suppose for the purposes of this example that it is a word of English and is synonymous with ‘water’. Let’s further suppose that Kathy is merely referentially competent with ‘agua’ in much the same way that she is with ‘water’, and that she doesn’t know that the two words are synonyms. We can make similar points about Kathy and ‘water’ as we did for Kelly and ‘sister’. We cannot explain Kathy’s mere referential competence with ‘water’ and ‘agua’, and her failure to know (32), in terms of knowing (33) and (34). Kathy does know the propositions expressed by (33) and (34), but this fact doesn’t explain her minimal competence with ‘water’ or ‘agua’, nor does knowing (33) and (34) offer any explanation of her failure to know (32). (32) ‘Water is agua’ is true. 115 (33) ‘water’ expresses the property of being water. (34) ‘aqua’ expresses the property of being agua. Instead, Kathy’s mere referential competence, her failure to know (32) and her knowledge of (33) and (34) are explained, at least in part, by her knowledge of something more schematic, as in (35) and (36). (35) ∃x(‘water’ expresses x & for any sample substance s and any name n which designates s, to assertively utter ⎡n is water⎤ is to say of s, and of x, that the former has the latter). (36) ∃x(‘agua’ expresses x & for any sample substance s and any name n which designates s, to assertively utter ⎡n is agua⎤ is to say of s, and of x, that the former has the latter). So we have the point that we can’t explain Kathy’s minimal level of competence in terms of knowing semantic theory. And as we saw for ‘sister’ above, we also have the point that we won’t be able to simultaneously explain Kathy’s minimal level of competence and the more robust level of competence (whatever exactly this amounts to) in terms of knowing a semantic theory. In the case of ‘sister’, we saw that there is an alternative conception of semantic theory (not strictly a theory of what is said) which can give and account of the richer level of competence with ‘sister’. Can we do the same for ‘water’? I will argue that we cannot. Let’s consider some of the elements that will go into richer competence with ‘water’. First, a speaker who is robustly competent with ‘water’ will have a fairly high degree of recognitional competence with the term. The speaker will be quite adept at 116 determining whether or not ‘water’ applies to a given sample (at least under “normal” observational circumstances). It is highly doubtful that such recognitional competence can be characterized in terms of propositional knowledge, and hence it is highly doubtful that this sort of competence can be accounted for by a semantic theory. Nevertheless, there is arguably some propositional knowledge that speakers will have in virtue of being robustly competent with ‘water’, and arguably some of this propositional knowledge may be a necessary condition on recognitional competence with the term. So perhaps semantic theory, construed as a theory of what typical speakers know about meaning, can at least offer a partial account of what typical speakers know about the meaning of ‘water’. Let’s consider some of the candidate propositions that robustly competent speaker’s might know. (37) ∀x((x is a liquid & x is water & x is pure) → x is clear, colorless, and odorless). (38) ∀x(x is water → x is a type of substance necessary for life on earth). (39) ∀x((x is liquid & x is water) → x is a type of liquid found in the oceans, lakes, and rivers on earth). (40) ∀x((x is liquid & x is water) → x will, when covering a smooth surface, generally make the surface more slippery). (41) ∀x((x is liquid & x is water) → x will, when covering a smooth surface, generally make the surface shiny in appearance). (42) ∀x((x is liquid & x is water) → x will make a good cleaning agent). Obviously knowing (37) – (42) would not be sufficient for being robustly competent with ‘water’. We would at the very least need metalinguistic counterparts to (37) – (42) since a 117 speaker could know all this without even being referentially competent with the term. But even with the metalinguistic counterparts we certainly couldn’t consider (37) – (42) as being sufficient for robust competence with ‘water’. For example, it is part of being recognitionally competent with ‘water’ that one knows that ice is water in a solid state. But this isn’t covered by any of (37) – (42). Are (37) – (42) jointly necessary for robust competence? Even this isn’t plausible. If a speaker didn’t know (42), but knew (37) – (41), and lots of other things about water besides (along with all the metalinguistic counterparts), she would still count as being robustly competent. At best, all we can say is that knowing a significant portion of (37) – (42) might be necessary for robust competence with ‘water’. Certainly a speaker who didn’t know any of (37) – (42) would have fairly impaired competence with the term. The problem here is that, compared to ‘sister’, the knowledge normal speakers have about water (i.e. the meaning of ‘water’) is very open ended. Not only is it open ended, but normal speakers, all of whom would count as being robustly competent with ‘water’ will have different sets of knowledge about water, and it will be impossible to identify any “core” set of knowledge they all share which could plausibly be construed as being responsible for their robust competence. Another difficulty comes when we consider actually adding anything like (37) – (42) to our semantic theory. If we add any of (37) – (42) to the theory, then the theory will predict that what is added is a necessary component of normal or typical linguistic understanding. But this simply isn’t true. None of (37) – (42) are by themselves necessary for robust competence with ‘water’. 118 Ultimately this approach to semantics can’t give an account of competence with ‘water’. It can’t account for mere referential competence with ‘water’ alone, it can’t account for the full spectrum of competence with ‘water’, from minimal to robust, and it can’t account for robust competence alone. The problem with ‘water’ is, first, that robust competence with ‘water’ is at least partly a matter of recognitional competence, and it is implausible that recognitional competence can be explained in term of propositional knowledge. Second, even though a speaker who is robustly competent with ‘water’ has some propositional knowledge about its content, there isn’t any knowledge which is by itself (or jointly as part of a set) necessary for robust competence, and hence a plausible candidate for addition to the theory. Let’s consider another natural kind term – the term ‘red’. Suppose we have a speaker, Jack, who has a very high degree of recognitional competence with ‘red’. Jack’s dispositions to apply ‘red’ are exactly in line with the community as a whole. Jack is disposed to appy ‘red’ to just those objects that the community would generally agree are red, disposed to apply ‘not red’ to just those objects that the community would generally agree are not red, and disposed to withhold judgment for just those objects where there would be controversy in the community as to whether the object is red or not. However exactly one wants to cash this out, the idea is that Jack’s recognitional competence with ‘red’ is impeccable. In the case of ‘red’, one might think there are better candidates for propositional knowledge which might be necessary for robust competence with the term. One might think knowing (45) is necessary for robust competence with ‘red’. 119 (45) ‘If an object is red all over, then it can’t be green all over’ is true. So let’s suppose provisionally that we add axioms to our theory so that (45) can be derived as a theorem. The difficulty with this is that, in spite of any initial plausibility the proposal may have, it really isn’t clear that knowing (45) is in any sense required for full competence with ‘red’. To see this, suppose that Jack denies (45). Jack is a philosopher, and denies (45) due to an esoteric (and mistaken) theory of vagueness that he holds. Due to the theory he holds, Jack believes that there are some color samples which really are both red and green all over. But Jack believes that it is entirely beyond his ken to determine visually which color samples have this conjunctive property, so his denial of (45) makes no impact on his impeccable recognitional competence with ‘red’. Does Jack’s denial of (45) in any way harm his overall competence with ‘red’? It seems to me that it does not. Jack is as competent with ‘red’ as anyone else. In fact, Jack is more competent with ‘red’ than most other speakers who accept (45) but have a lower degree of recognitional competence with the term. So it doesn’t look like (45) is required for robust competence with ‘red’. I want to suggest that, in the case of ‘red’, there may not be any propositional knowledge required for being robustly competent with ‘red’, not even in the loose sense we had for ‘water’. Thus, there wouldn’t be any meaning postulates to add to the theory for ‘red’. All we would have would be the axiom for ‘red’ which specifies its content, and as I have been arguing, such an axiom cannot by itself capture anything about linguistic competence with ‘red’. 120 Conclusion When we consider predicates, the prospects for (PC′) look especially dim, even more so than for names. If semantic theory is taken as an externalist theory of reference and “what is said,” (PC′) is impossible to maintain. Linguistic understanding is a graded phenomenon that such a semantic theory simply can’t capture. It is doubtful that any semantic theory can capture the various shades of understanding, but I haven’t defended such a sweeping claim here. If we abandon the requirement that semantic theory account for the shades of understanding, then (PC′) can perhaps be maintained, in some limited sense, by conceiving of semantic theory as a theory of what “normal” or “typical” speakers know about meaning. Such a move works reasonably well for predicates where normal understanding amounts to a circumscribed degree of inferential competence (i.e. for such words as ‘sister’, ‘brother’, ‘bachelor’, etc.). But for predicates where normal understanding involves at least some recognitional competence which can’t be described in terms of propositional knowledge, or where the inferential competence required is open ended, the approach can at best only offer a partial theory (i.e. for such words as ‘water’). And for predicates where normal understanding is primarily a matter of having recognitional competence, the approach breaks down entirely (i.e. color terms). All this suggests that it is overall a much better approach to divorce semantic theory from the theory of understanding, at least to a certain extent. One minimum constraint on semantic theory is that it needs to be a formal system based on formal rules 121 of composition – we need this much to account for the productive nature of language. I think we can take semantic theory as a theory of understanding, at least in this limited way – it offers an account of how the meanings of expressions contribute to the meaning of larger constructions in which they occur, and accounts for the productive nature of understanding in this sense. But there are other aspects of linguistic understanding, having to do with our understanding of individual words, which cannot be adequately captured within a formal axiomatic system. This is essentially the problem we have been bumping up against over and over again. Once we divorce the theory of understanding from semantics, we are no longer bound by the rather severe constraint that anything we say about understanding must fit within the confines of a formal system. The two parents simply can’t get along, so a divorce is the only reasonable solution. But they can remain close friends, and together raise their children much better than they ever could while joined at the hip with all tension that we have seen. 122 Chapter Five The Limits of Mere Referential Competence Introduction I have been arguing that it is possible to be merely referentially competent with a fairly extensive class of words, including words where mere referential competence constitutes a state of severely impoverished competence with the word in question (i.e. type-2 words such as ‘red’, ‘sister’, ‘bachelor’, etc.). In actual practice, a speaker will typically have minimal understanding for a relatively small set of words in her repertoire, and will typically not use more than one or two of these words in any sentence she utters. So in the normal case, a word that is minimally understood will be standing against a rich background of understanding. It is plausible to expect that there are limits to how many words a speaker can be merely referentially competent with – in a sentence or an entire language – before some kind of severe breakdown ensues. Also, there are words for which the consequences of merely having the minimal understanding typically required for referential competence are particularly severe (i.e. type-1 words); in these cases the lack of recognitional or inferential competence is apt to lead rather swiftly to breakdown, even when the word stands against a rich background of understanding. So the issue at hand for the present chapter concerns the limits of mere referential competence. I’m going to argue that mere referential competence is limited by theses (I) – (III). 123 (I) A speaker can’t be merely referentially competent with all, or even most, of the words in a language. (II) If a speaker A is merely referentially competent with all (or most) of the words in a sentence S, then any assertive utterance of S by A will be a defective speech act. (III) There are words W such that, if a speaker A lacks the sort of understanding of W that goes beyond mere referential competence, then there will be sentences S containing W such that any assertive utterance of S by A will be a defective speech act – even if the speaker has a robust understanding of all the other words in S. I will explain later the sense in which the speech act will be defective in (II) and (III). I will focus mainly on the case of asserting semantic content, but this is a simplification, since referential competence involves more than assertion, and assertion isn’t necessarily limited to semantic content. 1 Thesis I Let’s begin with thesis (I). Suppose we have a simple language that some community actually speaks. The language consists of names (‘a’, ‘b’, ‘c’, etc.), which refer to objects, and predicates (‘A’, ‘B’, ‘C’, etc.), which express properties. Sentences are formed by concatenating these (predicate followed by name) to form the “atomic sentences” of the language (so ‘Ac’, ‘Fa’, and ‘Lg’ are among the atomic sentences of the language). Each atomic sentence expresses a proposition which says, of the object bearing the name, and of the property expressed by the predicate, that the former has the latter. Atomic sentences can be combined to form complex sentences using any of five two-place truth 1 That is, in some cases, what is said or asserted by the utterance of a sentence in a context can go beyond the semantic content of the sentence relative to the context. 124 functional logical connectives (‘α’, ‘β’, ‘χ’, ‘δ’, and ‘ε’). The language employs Polish prefix notion so that the meaning of any complex sentence is unambiguously determined by the order of the symbols. Some examples of complex sentences in this language are ‘βFaGb’ and ‘χβFaGbHc’. Assume for the moment that we were referentially competent with the expressions in the language based only on the minimal information given above (in what follows I will be giving a reductio of this assumption). We know that each name refers to an object, but we don’t know which object, and we know each predicate stands for a property, but we don’t know which property. We also know each logical connective stands for a truth function, but we don’t know which function. We don’t even know how negation is handled in this language. Perhaps ‘α’ is equivalent in meaning to the Sheffer Stroke so that the negation of ‘Fa’ is indicated in this language by ‘αFaFa’ – or perhaps negation is handled in some other way. So we have no way of inferring the meaning of any of the symbols from our understanding of the syntax. We know what type of content each symbol has, we have some purely schematic information about how contents are composed, and we know how to construct well-formed sentences, but nothing more. Nevertheless, suppose we went about in the community uttering sentences of the language with the deferential intention to assert their contents. We might know that we were saying things, and that others in the community were saying things, but our perspectives on the relevant propositions would be so impoverished that there would be no genuine linguistic communication between us and the community. Not only would we 125 fail to understand what we were saying, and what others were saying, but it would be immediately obvious to the speakers in the community that we lacked any real understanding of the language. One thing that would make the speakers wise to our ignorance is the fact that we couldn’t make any conversationally relevant contributions. The assertions that we deferentially offered wouldn’t (setting aside any short-lived luck we might have) be relevant to anything the speakers said to us, or to anything relevant going on in the world around us. We would be offering entirely jack-in-the-box assertions that made no coherent conversational contribution. The second thing that would reveal our ignorance is that we might utter obviously false sentences (including logically false sentences). Or we might utter two or more sentences in succession that were obviously inconsistent. Such blunders, if sufficiently egregious, would reveal that we were not merely dumb, or logically inept, but that we simply failed to understand our utterances. The third thing that would reveal our ignorance is that our utterances would, unless we were being deceptive, lack a normal show of confidence. A normal assertive utterance is a speech act performed with a certain air of confidence. The marks of this confidence are things such as tone of voice, facial expression, body language, etc. This confidence is not born from the bold desire to merely put forth any assertion whatsoever, but rather from understanding the assertion sufficiently well to know that one is making a conversationally relevant contribution to the discourse at hand (and which fits with one’s beliefs, thoughts, intentions, etc.). But we couldn’t make our assertions with the appropriate level of confidence. We would know that our assertions were completely 126 arbitrary, and this knowledge would be revealed to the community by the insecure tone of voice, facial expression, and body language that accompanied our utterances. When the speakers in the community said things to us, they would similarly be looking for evidence that we understood what they were saying. Without being deceptive, however, we couldn’t offer any appropriate marks of comprehension. These marks of comprehension would, to repeat, be things such as facial expression, nods of comprehension, etc. This point about marks of comprehension is not only plausible, but there is linguistic research supporting it. Much of this research is based on a corpus of conversations known as the “London-Lund” corpus. In English, the marks of comprehension include not only facial expressions and nods, but expressions such as “Uh-huh,” and “Oh.” Clark and Schaefer, for example, discuss some of the corpus, and remark: 2 A closer look at actual conversations, however, suggests that they are much more than sequences of utterances produced turn by turn. They are highly coordinated activities in which the current speaker tries to make sure he or she is being attended to, heard, and understood by the other participants, and they in turn try to let the speaker know when he or she has succeeded. Contributing to a discourse, then, appears to require more than just uttering the right words at the right time. It seems to consist of collective acts performed by the participants working together. We might try to fake the marks of comprehension for a while, but our deceptive game would be exposed rather quickly, if for no other reason than that we couldn’t utter the right words at the right time. But if we were sincere, we couldn’t offer marks of comprehension at all. I agree with Clark and Schafer’s suggestion that discourses are 2 Clark and Schaefer (1989, p. 259). 127 collective acts, and due to our extremely impoverished understanding of the language, we wouldn’t be able to participate in the collective conversational activities of the community. Since it would be clear to the speakers from the outset that we don’t understand the language, we wouldn’t be accepted into the community as generally competent speakers of the language. Without being accepted as speakers of the language, we wouldn’t be speakers of the language, and our utterances wouldn’t count as linguistic assertions at all. Linguistic assertion involves a kind of public commitment. When a speaker asserts something by means of an utterance, he isn’t merely entertaining the content for private purposes, but is publicly committing himself to the content. His audience will take him to believe the assertion, and he will be evaluated in light of that belief. If the belief is found wanting, either because it is deemed false, or not to cohere with other beliefs the speaker has divulged, then the speaker will be faulted for this. But this public commitment requires that the speaker has a standing in the community as a generally competent user of the language, since, without this standing, he will not be deemed competent to testify to his beliefs. So without this standing, his utterances will not be accepted by the community as saying or asserting anything, and hence he can’t say or assert anything without this acceptance. In our example, the other speakers would recognize that we were in a bad way, so much so that they would deny us standing as generally competent speakers. For this reason, they would be reluctant to accept our utterances as genuine assertions. Thus, the supposition that we were merely referentially 128 competent with the hypothetical language is highly doubtful. For these reasons, I think it isn’t possible to be merely referentially competent with any language. In order to become a speaker of a language (by being accepted into the community as such), one must have a richer understanding of the language than is afforded by mere referential competence. Before being accepted into a linguistic community, a speaker must demonstrate his understanding of the language in a public way. The public marks of comprehension are (i) that he can hold his own in a variety of everyday conversations by offering consistently relevant contributions to the discourse at hand; (ii) that his assertions are not obviously false or inconsistent on a regular basis, (iii) that his assertive utterances are accompanied by appropriate marks of confidence; and (iv) that he shows appropriate signs of comprehension in response to the utterances of others. Of course, the picture may be considerably more complex than I have described here. There may be other marks of comprehension needed, and I have for simplicity focused on the case of assertion. But at least (i) – (iv) are required, and a speaker must have a robust level of understanding for a majority of words in a language in order to satisfy these conditions. Most crucially, the speaker will need to have a robust level of understanding for most of the words that come up regularly in normal everyday conversations – because it will really be his ability to hold his own in these kinds of conversations that will form the basis for his being accepted as a speaker of the language. It is important to note, however, that these marks of comprehension are holistic in nature. A speaker’s good standing in the community will not be threatened by the occasional gaff. Furthermore, the speaker can get by with minimal understanding (i.e. mere 129 referential competence) for a significant number of words, even using one or two of them in a sentence in many cases without his lack of understanding coming to light. Theses II and III Turning to thesis (II), let’s begin by reviewing the minimal understanding involved in mere referential competence. If an individual is referentially competent with a word, then she will be a speaker of the language of which the word is a part, have robust understanding for a majority of expressions in the language, have extensive experience communicating with other speakers of the language, and will be part of a causal chain of communication extending back to the object or property which is the content of the word. 3 Also, the speaker will know enough about the syntactic category of the expression to give her minimal insight into the category of content the expression has. Let’s focus on a name ‘n’ having an object as its content and a simple predicate ‘F’ having a property as its content. A speaker who is merely referentially competent with ‘n’ and ‘F’ will know (1) and (2). (1) ∃x(‘n’ refers to x & for any predicate G expressing property p, by assertively uttering ⎡n is G⎤ I am saying that n has p). (2) ∃x(‘F’ expresses x & for any name a, by assertively uttering ⎡a is F⎤ I am saying that a has x). The speaker can assert the relevant semantic content when she uses ‘n’ or ‘F’ in a range of sentences where she has a more robust understanding of the other words in the sentence (at least this much is required if she is to be referentially competent with the 3 In the case of properties, the casual chain will extend back to instances of the property in question. 130 terms). But what happens when the speaker assertively utters ‘n is F’? It’s tempting to think that the speaker doesn’t know what she is saying by uttering the sentence, and hence can’t assert the semantic content of the sentence. If so, the defect in (II) would be that the speech act fails entirely because no assertion is made. But it isn’t clear that such a strong view is warranted. The speaker will know (3). (3) ∃x∃y(‘n’ refers to x & ‘F’ expresses y & by assertively uttering ‘n is F’ I am saying that x has y). And (3) will also form the basis for her deferential intention to use ‘n is F’ with its meaning when she assertively utters the sentence. So we have (where A is our speaker) (4) A knows that: ∃x∃y(‘n’ refers to x & ‘F’ expresses y & by assertively uttering ‘n is F’ I am saying that x has y). And, when the speaker assertively utters ‘n is F’, (5) A intends that: ∃x∃y(‘n’ refers to x & ‘F’ expresses y & by assertively uttering ‘n is F’ I am saying that x has y). As I have argued elsewhere, exportation is justified here by the fact that A is, among other things, part of a causal chain of communication extending back to the content of ‘n’ and ‘F’. Hence, we have (6) and (7). (6) ∃x∃y(‘n’ refers to x & ‘F’ expresses y & A knows that: by assertively uttering ‘n is F’ I am saying that x has y). (7) ∃x∃y(‘n’ refers to x & ‘F’ expresses y & A intends that: by assertively uttering ‘n is F’ I am saying that x has y). (6) and (7) show that A both knows, of the proposition ϕ expressed by the sentence, that the sentence expresses ϕ, and that she intends to assert ϕ. What is amiss with A is not that 131 she doesn’t know that she is asserting ϕ, or that she isn't intending to do so, but rather that she has an extremely impoverished perspective on ϕ. For a sentence S with content P, (8a) and (8b) are necessary conditions on A being able to assert P by uttering S. (8a) A knows that: by assertively uttering S I am saying that P. (8b) A intends that: by assertively uttering S I am saying that P. (6) and (7) show that A does know the relevant versions of (8) for the sentence ‘n is F’. Of course (8) is not by itself sufficient for assertion, but A is in a better position than is characterized by (8). In addition to satisfying (8), A has some insight into the structure of the proposition expressed by ‘n is F’, and how this structure relates to the structure of the sentence. A has extensive experience asserting other sentences, which she understands more robustly, with the same syntactic structure and ‘n is F’. Additionally, A is a speaker of the language, part of a causal chain of communication extending back to the contents of n and F, etc. Even though (8) isn’t sufficient for assertion, it may be that (8), together with the further conditions that A satisfies in virtue of being referentially competent with ‘n’ and ‘F’, is sufficient for A to be able to assert the proposition expressed by ‘n is F’. So I don’t think it is clear that the impoverished understanding of ‘n is F’ A has really defeats her ability to assert ϕ by means of it. There are a few other considerations that also make it unclear that a speaker can’t assert in the case where she is merely referentially competent with all (or most) of the words in a sentence. Suppose that A is having a conversation with another group of 132 speakers, and in the course of the conversation, one of the speakers, Jones, assertively utters (9) (and suppose A has good reason to believe that Jones is being sincere). (9) Cornelius is a teetotaler. This is the first time that A has encountered the word ‘teetotaler’ or heard anyone referred to by the name ‘Cornelius’. But A understands that ‘Cornelius’ is a name and that ‘teetotaler’ is a count noun. A is merely referentially competent with both words. It’s clear in this case that A can accurately report on what Jones said and what Jones believes by assertively uttering (10) and (11). (10) Jones said that Cornelius is a teetotaler. (11) Jones believes that Cornelius is a teetotaler. A can use ‘Cornelius’ and ‘teetotaler’ with their respective meanings in (10) and (11). It is then puzzling if A can’t use ‘Cornelius’ and ‘teetotaler’ with their respective meanings in (9), and be able to use (9) on her own to assert its content. One might try to argue that there is a variable standard for using a word with its meaning, and that the standard is somehow lower when a word occurs in the scope of an attitudinal verb. But if so, one would need a view about why this is so, and it isn’t clear what the view would be. 4 A similar point holds for questions. A can ask (12) of one of the other conversational participants. (12) Is it true that Cornelius is a teetotaler? 4 Notice that, if one could defend this view, it would mean that referential competence is itself a graded phenomenon, and not fairly discrete as I have been supposing. If so, however, this would only show that linguistic understanding is even more graded than I have been supposing. 133 Given the reasonable assumption that names and predicates contribute the same content when occurring in interrogative sentences (with interrogative force) as they do when they occur in declarative sentences (with assertive force), A is managing to use ‘Cornelius’ and ‘teetotaler’ with their respective meanings when she puts forth query (12). Again, it would be odd if A could use these words with their respective meanings in (12) but not in (9). There is the possibility here too that the standard for using a word with its meaning is lower when used in the context of interrogative force as opposed to assertive force. But one would like to have a view about why this would be so, and again it isn’t clear what the view would be. As tempting as it might be to say that a speaker can’t assert the semantic content of a sentence if she is merely referentially competent with all the words in it, it simply isn’t clear that this is so. If A assertively utters (9), there is clearly something defective about the utterance, but the defect may not be that the speech act fails entirely (with no assertion made). Similar points hold for thesis (III). If there are any words for which it is problematic to be merely referentially competent, logical terms are surely a paradigmatic example. Suppose for the sake of discussion that ‘stroke’ is a logical connective in English. And suppose our hapless speaker A picks up ‘stroke’ conversationally, only learning that it is a kind of logical connective, in the same category as ‘and’, ‘or’, etc. Further, suppose that A overhears Jones assertively utter (13) (and has good reason to believe Jones is sincere). (13) Snow is white stroke grass is red. 134 Here too it seems that A could accurately report on what Jones said and what he believes by uttering (14) and (15). (14) Jones said that Snow is white stroke grass is red. (15) Jones believes that Snow is white stroke grass is red. If A can use ‘stroke’ with its meaning in (14) and (15), then it is difficult to see how she can fail to also use it with its meaning in (13); there is no compelling reason to accept this. Similarly, if A doubts that Jones is speaking truly, she can voice her concern to one of the other conversational participants by uttering (16). (16) Is it true that Snow is white stroke grass is red? If A is using ‘stroke’ with its meaning in (16), then it isn’t clear how we can deny that she can also use the expression with its meaning in (13). For these reasons, there is no clear justification for concluding that the relevant defect in theses (II) and (III) is that the speech act fails entirely. There is, however, a different sort of defect which does a perfectly adequate job of explaining what is wrong with being merely referentially competent with all (or most) of the words in a sentence, or being merely referentially competent with a logical term. It is part of the view I am defending that a complete account of linguistic communication must account not just for assertion, but for dialogue and conversational exchange. The proper focus of investigation for semantics is the relation between sentences and content, which is central to accounts of assertion and thought. However, the focus of linguistic understanding is broader. The proper focus of investigation for linguistic understanding is at the level of 135 discourse, and it is this expanded focus that creates a gap between semantics and the theory of understanding. Clearly a conversational exchange requires more than the mere ability to assert things. A conversational exchange is a set of assertions which are planned or coordinated, and the coordination must be something which is understood and recognized by a conversational participant if the participant is to count as a fully competent party to the conversation. Linguistic competence is, at least in part, a matter of being conversationally competent. A speaker must, in addition to asserting, understand how the assertion fits into the overall conversation of which it is a part. The conversation may be further embedded in some non-linguistic goal or activity, and in such a case, it is part of understanding the conversation to understand how the conversation relates to the activity. So in order to fully understand the conversational exchange, speakers must understand how individual assertions are related to each other, and how each are related to the world of action with which the conversational participants are engaged. Understanding how assertions relate to each other is a matter of having inferential competence with the words in the sentences used to make the assertions. And an understanding of how the assertions relate to the world is secured by recognitional competence. Conversational competence can come in degrees, and this is one of the primary reasons why linguistic understanding is a graded phenomenon. The extent to which an individual speaker understands the conversation she is participating in can be a matter of degree. If a speaker is merely referentially competent with one or two words used in the 136 conversation, this can certainly impair her ability to understand the conversation, but so long as her understanding of the other words used is robust, she will still largely be able to follow the conversation and participate in it. She may even be able to fill in the few gaps in her understanding by means of contextual clues stemming from her richer background of understanding. If she is merely referentially competent with all or most of the words used in the conversation, however, she really won’t be able to participate coherently in the conversation at all. Linguistic communication is a rational activity, and as such (17) is a condition on any speech act. (17) A knows that her speech act forms a relevant contribution to the conversation of which it is a part, and thus can rationally intend it as such. It may be tempting to say that (17) is a necessary condition on the speech act being successful at all, but it isn’t clear how to justify this claim, and I’m only going to argue here for the weaker claim that the speech act is defective in a weaker sense if (17) fails. If (17) fails, the speech act will be a species of irrational action. A comparison to non-linguistic action is helpful. Suppose I perform a series of intentional actions in a room over the course of an hour. First I rub my stomach, then I scream, then I hop on one foot, then I stand on my head, then I cycle the lights on and off 10 times, and so on. Evaluated in isolation, each action might be perfectly rational in the sense that I intend to perform the action, and have a sound plan for performing the action which I know is a reliable means for executing the action in question. 5 But we can still 5 For example, I know that lifting my arm, grasping the light switch, and moving my hand up and down in a certain way is a reliable means of cycling the lights on and off. 137 find fault with each action when evaluated in a broader context because none of the actions fit into any larger plan or project that I have underway. I am acting randomly and irrationally during the hour. Each act was “successful” in the sense that I acted in accord with my intentions, yet each action can still be faulted for irrationality because it fails to fit into any broader plan or goal. I would be even more blameworthy if, during the hour, I was trying to participate in some group activity with others. In such a case, I would understand the need for my actions to make some coordinated contribution to the group project underway. But in spite of this understood need, I just perform the random series of actions with no understanding of how these actions might form a contribution to the social project. Even if, by accident, some of my actions do further the group effort, I am still highly blameworthy for the irrationality of my actions. But in spite of the highly defective nature of my actions, each action is an intentional action that I perform. Similarly, if (17) fails for some speech act, then the speech act may be “successful” in the sense that the speaker manages to assert something, or put forth a query, or whatever. But the speech act will nonetheless be an irrational action, because the speaker doesn’t understand how the speech act forms a contribution to the conversation, and can’t rationally intend for it to do so. We are now in a position to see the case for (II) and (III). If a speaker is merely referentially competent with all (or most) of the words in a sentence, she won’t be able to see how assertively uttering the sentence could constitute a relevant contribution to the conversation, and hence (17) will fail if she does assertively utter the sentence. Similarly, in the case of (13), it is clear that a speaker who is merely referentially competent with 138 ‘stroke’, wouldn’t generally be able to assertively utter (13) without violating (17). Even if the speaker has a robust understanding of ‘snow’, ‘white’, ‘grass’, and ‘red’, her ignorance of ‘stroke’ will prevent her from knowing how (13) could make a relevant contribution to virtually any conversation. The view I am outlining can also explain why there is a lower standard for questions or attitude reports. Even though A will violate (17) if she assertively utters (9) or (13), she might assertively utter any of (10), (11), (14), or (15) without violating (17). If for example, the conversation turns to the topic of what Jones believes, A can offer (10) and (15), and know that she is making a relevant contribution to the conversation, even though she herself is in a certain sense in the dark about which propositions she is claiming Jones to believe. In the case of a question, it is plausible to think that it is easier to formulate a question which fits coherently into a conversation than it is to formulate an assertion which fits. This partly explains why, for A, assertively uttering (9) or (13) seems very defective, but asking questions such as (12) or (16) seems much less defective (or perhaps not defective at all). So I contend that the primary defect in (II) and (III) is the failure of (17). This view offers a plausible explanation of what is amiss with a speaker who is merely referentially competent with all the words in a sentence, or with a speaker who is merely referentially competent with certain words such as logical terms (i.e. type-1 words). The failure of (17) constitutes a significant breakdown in the speech act (though it may not fail entirely). It is inappropriate and deceptive for a speaker to offer an utterance when she can’t rationally intend the utterance to contribute to the conversation of which it is a 139 part. In actual practice, speakers will refrain from performing the speech act in this sort of circumstance, which effectively takes away their capacity to make the relevant assertion (even if the speaker could, strictly speaking, make the assertion by means of the utterance). While I think that the failure of (17) is the primary defect in (II) and (III), the speech act is defective in at least one other sense as well. This is a defect noted earlier – namely, that the speaker wouldn’t (if sincere) be able to offer the assertion with an appropriate level of confidence. If a speaker is merely referentially competent with all the words in a sentence, the speaker won’t be able to assertively utter the sentence with an appropriate level of confidence. Similarly, if the speaker is merely referentially competent with a logical term, there will be sentences containing the term for which the speaker isn’t able to assertively utter the sentence with confidence, even if the speaker has a robust understanding of all the other words in the sentence. But this confidence is primarily born of (17), so the defect in confidence is derivative from the defect resulting from a failure of (17). As I have suggested, there may be a stronger view in the philosophical space here. The stronger view would take the defect in (II) and (III) as a complete failure of the speech act, where the semantic content of the sentence isn’t asserted at all. But the stronger view is harder to defend and doesn’t offer any ready explanation of the lower standard we observed for questions and attitude reports. The weaker view I am defending here is obviously true, and offers a plausible explanation of the variable standard we observed. In light of this, there is simply no need to defend the stronger view. 140 Parallel remarks can be made for the case of private thought, and by way of closing, let me sketch the parallels here. Just as a conversation is a coordinated set of utterances (assertions or otherwise), so thinking is a coordinated series of individual thoughts. This means that an individual’s ability to think will also be a graded phenomenon, depending on her ability to see how individual thoughts are connected to each other. This ability will be tied in turn to the speaker’s grasp of individual concepts. Also, one’s thought is surely bound up with one’s actions in the world, so one additionally needs to see how one’s thoughts relate to the world. But thinking is, at least in part, linguistically mediated, so a speaker’s capacity for thinking will depend in part on her linguistic competence. To employ the terminology developed in chapter four, an episode of thinking may deploy both linguistic concepts and cognitive concepts, and the success of the episode will depend crucially on the individual’s grasp of both kinds of concepts. The individual will need to be able to see how both kinds of concepts relate to each other, and to the world. Focusing on linguistically mediated thought, we can make comments parallel to what was said above for the case of conversation. The parallel to (6) (where “entertaining” a sentence takes the place of utterance) would be (6*) ∃x∃y(‘n’ refers to x & ‘F’ expresses y & A knows that: by entertaining ‘n is F’ I am thinking that x has y). Intention seems to fit more with social communication than with private thinking, so I won’t speculate here about the need (or lack thereof) for any version of (7). (6*) shows that A knows, of the proposition ϕ expressed by ‘n is F’, that by entertaining the sentence, 141 she is thinking ϕ. Again, what is amiss with A is not that she isn't thinking ϕ, but rather that she has an extremely impoverished perspective on ϕ. And just as for the case of assertion, it isn’t clear here that the speaker’s perspective on the proposition is so minimal that she can’t form the relevant thought by means of entertaining the sentence. Her minimal perspective may not by itself be adequate for thought, but again there are other facts to cite. The speaker has formed appropriate referential intentions with the words in the sentence, is part of a causal chain of communication involving them, has a rich understanding of the language more generally, etc. Also, it is relevant to her ability to think by means of the sentence that she has entertained similar sentences, more richly understood, with the same type of structure, and these other sentences have figured in successful chains of thought. Note that we have the same variable standard here as we had for the case of utterance. It is clearly more problematic to suppose that A can form a thought means of entertaining (9) or (13) than it is to suppose that A can form a thought by means of entertaining any of (10) – (12), or (14) – (16). 6 Furthermore, there is arguably a similar necessary condition here on rational, linguistically mediated thought. (17*) A knows that her act of entertaining a sentence forms a relevant contribution to the chain of thinking of which it is a part, and thus can rationally intend it as such. Just as we saw before, this condition is perfectly plausible, and can explain the variable standard. Therefore, I think we should adopt the same modest view about thought and 6 I assume provisionally here that asking a question privately in thought is analogous to entertaining an interrogative sentence with interrogative force. 142 belief as I argued above that we should adopt for the case of communication. An individual thought may come very cheaply, and the higher standard comes with understanding the thought well enough to coordinate it with a chain of other thoughts, and with seeing how such thoughts relate to the world, both of which are things private thought needs to do for us. Conclusion It is important to keep in mind here that we are wrestling with the fact that content isn’t transparent to the speaker. We thus need an account, both for the case of uttering sentences in communication, and entertaining sentences in thought, which doesn’t require that the speaker have completely transparent access to content. Transparency is part of the internalist paradigm that we are struggling here to make a clean break from. A speaker’s understanding of representational content lies in the smooth gradations between complete transparency and complete lack of transparency. Given how smooth the gradations are, there is bound to be a slippery slope problem in trying to raise the bar for assertion or thought above the minimal level of mere referential competence. So, the most vigorously defensible position is going to be the one I have defended here – according to which uttering a sentence or entertaining a sentence can be cheaply, but minimally, successful. 143 Chapter Six How Recognitional and Inferential Competence Contribute to Understanding Introduction While I have been arguing that recognitional and inferential competence are vital aspects of linguistic understanding, we still stand in need of concrete examples showing how this is so. In the following sections, this will primarily be done by showing how recognitional and inferential competence figure in linguistic communication (I assume it’s uncontroversial that referential competence plays a role, since it secures central functions such as assertion). Understanding Names In this section I show how the view I’m developing can shed new light on a traditional problem involving descriptive uses of names (i.e. Frege’s Puzzle). This will be done by showing how the view can extend and further justify a recent pragmatic solution to the puzzle defended by Scott Soames. Suppose Felicity goes by two names – ‘Felicity’ and ‘Hillary’. ‘Felicity’ is her “track name,” and ‘Hillary’ is the name she goes by around Calumet Farm (CF) where she trains. Jones lives near CF and knows Felicity as ‘Hillary’ from his visits to the farm. Felicity is a black mare and Jones associates the description a black mare at CF with ‘Hillary’. Jones is also an avid fan of horse racing. Felicity won the 2006 Kentucky Derby (KD) and Jones saw her do this on television. Jones associates the description a 144 winner of the KD with ‘Felicity’. Further, suppose that Jones doesn’t know that ‘Felicity’ and ‘Hillary’ (and their associated descriptions) apply to the same horse. Jones is referentially competent with ‘Felicity’ and ‘Hillary’ and knows what semantic theory says about these names. He knows, of Felicity, that ‘Felicity’ refers to her, and of Hillary (i.e. Felicity) that ‘Hillary’ refers to her. Hence, (1) is true. (1) ∃x(Jones knows that (‘Felictity’ refers to x & ‘Hillary’ refers to x & x is a winner of the KD & x is a black mare at CF)). But since (1) attributes singular knowledge to Jones, it’s consistent with (2) and (3), which are also true. (2) ¬ Jones believes that ∃x(‘Felicity’ refers to x & ‘Hillary’ refers to x). (3) ¬ Jones believes that ∃x(x is a winner of the KD & x is a black mare at CF). Jones has different non-transparent perspectives on Felicity, mediated by the descriptive information he associates with each name (I assume here that the “metalinguistic” descriptions in (2) are reasonably salient for Jones, giving him two minimal metalinguistic perspectives on Felicity). Jones can express his lack of belief by assertively uttering (4). (4) I don’t believe that Felicity is Hillary. (4) presents a powerful prima facie objection to the direct reference theory of names. (4) seems true, but according to the theory its semantic content is a false proposition (i.e. that Jones doesn’t believe the proposition attributing the relation being identical to felicity and herself). Intuitively, what Jones is trying to convey by uttering (4) is (2) or (3) – or at 145 least something similar to this. This is an instance of an objection that has been raised against direct reference. The objection observes that a name can be used to convey descriptive information associated with the name. Furthermore, this usage seems so natural that it’s hard to avoid the intuition that names have descriptive meaning. In recent work Scott Soames has shown how the direct reference theorist can give a pragmatic reply to this objection. The objection assumes that what a speaker primarily asserts and intends to convey to his audience by the assertive utterance of a sentence (in a context) is always identical to the semantic content of the sentence (in that context). Soames argues that this assumption is mistaken, offering an alternative picture of the relation between assertion and semantic content. 1 On Soames’ alternative picture, the proposition that a speaker primarily asserts and intends to convey to his audience can be an acceptable “pragmatic enhancement” of semantic content. For our purposes, the important point about pragmatic enhancement is that it can consist in adding descriptive information associated with the relevant names in the context. The propositions expressed by (4a) and (4b) both count as acceptable pragmatic enhancements of the semantic content of (4). (4a) ¬ Jones believes that Felicity, the referent of ‘Felicity’, is Hillary, the referent of ‘Hillary’. 1 Soames’ view involves an alternative conception of the notion of linguistic meaning. The meaning of an expression is properly understood as the contribution made by that expression to what is said by various utterances across all competent speakers and contexts of use. The best theory of meaning for names is one which identifies meaning with reference, because reference is what is common to the use of names across speakers and contexts. On this conception, meaning only serves as a constraint on – without entirely determining – what may be asserted by an utterance in a context. Soames and I thus agree that the meaning of an expression constitutes a minimal rule governing its use. See Soames (2009a; 2009b; 2010). 146 (4b) ¬ Jones believes that Felicity, a winner of the KD, is Hillary, a black mare at CF. 2 Jones can count as asserting (what is expressed by) both (4a) and (4b) by uttering (4). 3 Intuitively, this is the right result, because it would be artificial in this case to decide that Jones is asserting one and not the other. In this, Soames’ view has an advantage over forms of descriptivism that attempt to account for descriptive uses of a name semantically by assigning a single description to the name. Such a view is forced to make a choice about whether Jones is asserting the descriptive content in (4a) or the descriptive content in (4b), but either way the choice is artificial. 4 One problem with thinking that descriptive content is part of semantics is that a speaker can associate different descriptions with the same name in the same context, and this makes it unclear which description is being asserted by sentences such as (4). Soames’ view handles this difficulty nicely by saying that all the descriptive information can count as being asserted. But the view I’m defending further illuminates Soames’ solution to the puzzle. First, by connecting the use of names to the three abilities described in chapter three, my view clearly identifies the essential role played by descriptive information associated with names. Not only can names be associated with descriptive information, but they usually must be in order for any significant thought or communication to take place. A 2 The proposition expressed by (4b) is, according to Soames, the proposition expressed by ‘¬ Jones believes that [the x: x is a winner of the KD & x = z][the y: y is a black mare at CF & y = w][x = y]’ relative to the assignment of Felicity to ‘z’ and ‘w’. Similarly for (4a). 3 In conversation, Soames has expressed some doubt about how salient metalinguistic descriptions are for most speakers. I assume they are for the purposes of discussion, but nothing I say turns on this. 4 To my knowledge, Soames doesn’t make this specific point against descriptivism. 147 speaker who is merely referentially competent with a name can refer to the object bearing the name, but such references will be cognitively vacuous for her, since she has no understanding of how the object relates to other objects or properties she wishes to talk about. Descriptive information secures relational understanding for names, and manifests as recognitional and inferential competence with the name. The description a black mare at CF grants Jones a degree of recognitional and inferential competence with ‘Hillary’. Jones knows that ‘Hillary’ applies to x only if x is a black mare. Jones also knows that the sentence ‘That’s Felicity’ entails ‘That’s a mare’. 5 It was traditionally thought that descriptions primarily serve to secure the reference of names. But Kripke (1980) famously showed that descriptions don’t generally serve this purpose. My view highlights the correct function of descriptions that individuals associate with names – their primary purpose is not to secure reference, but to secure relational understanding. 6 Second, my view can account, in a more striking way, for the intuition noted earlier that the meaning of names involves descriptive content. On my view, pragmatics makes an important contribution to knowledge of meaning. In the case of names, there is a pragmatic dimension to our knowledge of their meaning that has descriptive import. Fully understanding a name involves more than merely knowing its semantic rule – one must also know something about who or what the name refers to, and this is typically 5 It is very often the case that having some degree of recognitional competence with a word will go hand-in- hand with having some degree of inferential competence. However, recognitional competence and inferential competence can dissociate to a large extent. 6 It may be, however, that metalinguistic descriptions such as the referent of ‘Felicity’ serve, at least in part, to secure the reference of names for individual speakers. It may be that forming an appropriate referential intention for ‘Felicity’ involves the description the referent of ‘Felicity’. So my claim here is about more robust descriptive information. 148 secured by descriptive information. Since I’m driving a wedge between semantics and understanding, I can say that fully understanding a name involves descriptive information, and thus grant the intuition that names typically have descriptive import for speakers, without thereby being committed to an implausible semantic theory that assigns descriptive content to names as a linguistic rule. Competence with names, however, is highly context specific. A speaker must associate descriptive information with a name in order to understand it in a rich way, but this information can vary across speakers, and result in varied competence across contexts. If a speaker associates the description a black mare at CF with ‘Felicity’, she may not be fully competent with the name relative to a context where all the other speakers fail to associate this description with ‘Felicity’, but rather associate the description a winner of the KD. Competence with names differs significantly from competence with predicates in this regard. 7 For predicates, the same information may be presupposed by uses of the predicate in virtually any normal communicative context, so a speaker must know this information in order to be generally competent with the predicate across normal contexts. Let’s turn now to predicates. Understanding, Discourse, and Predicates In this section, I consider the case of predicates, which involves a shift from considering individual assertions to the broader context of conversational exchange. A conversation is 7 However, it may be, for names of famous individuals who are frequently the subject of broad public conversation, that speakers must associate a common set of descriptive information about these individuals across a wide range of contexts. 149 a set of assertions coordinated with one another, and with the non-linguistic activities of its participants. So in order for a speaker to fully understand a conversation, she must not only be able to make assertions, but also know the community standards for how the contents of expressions used in a conversation are typically understood to relate to each other (secured, at least in part, by inferential competence), and to things agents are able to recognize in the world (secured by recognitional competence). The argumentative strategy here is as follows. I will consider examples of defective conversational contexts, and show how the defect can be attributed to a speaker’s lack of inferential or recognitional competence. This will show how these abilities contribute to linguistic competence and understanding. I will also show how the abilities are tied to presuppositions that facilitate discourse. Finally, I discuss some of the interesting features of these presuppositions. One thing a speaker must do in order to understand a conversation is recognize when the same content is presented again. This need arises most obviously in the case of synonymy relations between words. Suppose speakers A and B have the following conversation. (5a) A: How many animals do you own? (5b) B: I own a horse. My faithful steed has been with me for 7 years. B is primarily asserting the two propositions semantically expressed by (5b) in the context. Each of these propositions attributes ownership of an individual instantiating the property of being a horse to B. Let’s put this point by saying that each of the propositions is “about” horse-hood. In uttering (5b), B is trying to convey to A that he only has one 150 animal, even though this isn’t part of what is semantically expressed by (5b). B is relying on pragmatic mechanisms to get across the idea that he only has one animal. B assumes A knows that B owns no other animals besides those he mentions, and that he has only mentioned one animal. In order to get this across, B assumes A knows that ‘horse’ and ‘steed’ are synonyms. 8 A needs to know this in order to understand that B is only telling him about one animal instead of two. Suppose B had instead replied, (5b′) B: I own a horse. My faithful dog has been with me for 7 years. Though it may be a little odd, here B is intending to convey to A that he has two animals. Suppose A is merely referentially competent with ‘steed’, only knowing that it refers to a kind of animal. 9 Since A knows what semantics says about ‘steed’, we need to be careful about what A does and doesn’t know; (5c) is true, but (5d) is false. (5c) ∃p(A knows of each of the propositions expressed by B’s use of (5b) that it is about p). (5d) A knows that ∃p(each of the propositions expressed by B’s use of (5b) is about p). A knows that the proposition expressed by ‘I own a horse’ is about horse-hood, and that the proposition expressed by ‘My faithful steed has been with me for 7 years’ is about horse-hood (i.e. the property of being a steed). Hence (5c) is true. But (5d) is false, so 8 I assume for the purposes of discussion that ‘horse’ and ‘steed’ are synonyms. It may be that they are only “near” synonyms, being subtly different in meaning, and B only needs to assume this much, but this complication doesn’t significantly impact what I say about the example. 9 A will have a modicum of recognitional and inferential competence, since he knows that ‘steed’ applies to x only if x is an animal, and that the sentence ‘That’s a steed’ entails ‘That’s an animal’. But for my purposes, A is, in effect, merely referentially competent. 151 there is a sense in which, for all A knows, B has told him about two animals. 10 A’s ignorance is the result of his failure to understand that ‘horse’ and ‘steed’ are synonyms. B might have replied, (5b′′) B: I own a horse. My faithful horse has been with me for 7 years. (5b′′) expresses the same propositions as (11b) – that B owns a horse and that B’s horse has been with him for 7 years. If B had replied with (5b′′), then both (5e) and (5f) would be true. (5e) ∃p(A knows of each of the propositions expressed by B’s use of (5b′′) that it is about p). (5f) A knows that ∃p(each of the propositions expressed by B’s use of (5b′′) is about p). A would have known that B was only telling him about one animal because he would have recognized that each occurrence of ‘horse’ in (5b′′) expressed the same content. The way to account for these observations is to say that A has two different perspectives on the propositions that B owns a horse and that B’s horse has been with him for 7 years. One is mediated by his understanding of (5b) (as B uses it). This perspective is impoverished because it doesn’t give A the insight that both propositions are about the same property (in the sense that (5d) is false). The other perspective is mediated by his understanding of (5b′′) (as B might have used it – I assume here that A 10 I assume for the purpose of discussion that A is being a bit daft, since it arguably would be very natural for A to infer from B’s response that ‘steed’ is a synonym for ‘horse’ (or at least that ‘steed’ stands for a kind of horse). This is an interesting point, because it shows that speakers can learn something about how the meanings of words relate from conversational context. But since this point isn’t central to my discussion, I won’t delve into it here. 152 has an interpretive disposition towards (5b′′), even though this isn’t B’s actual response). This perspective is richer because it does give A the insight that both propositions are about the same property (in the sense that (5f) would be true). In assuming A knows that ‘horse’ and ‘steed’ are synonyms, B is, in effect, assuming A’s understanding of (5b) grants A the richer perspective on the propositions that B owns a horse and that B’s horse has been with him for 7 years that allows him to see that these propositions are about the same property (i.e. B effectively assumes (5d) is true). This thought experiment yields some important results. First, it’s clear that a speaker’s grasp of, or perspective on, a proposition can be mediated linguistically (but not that such perspectives are only mediated linguistically). Second, such perspectives can include (or fail to include) information about how two or more propositions are related, and this information (or lack thereof) can be crucial to a speaker’s understanding of a conversation. Third, a speaker’s understanding of how two propositions are related can be tied to a speaker’s understanding of how the contents of two words are related, so a speaker’s understanding of individual words can impact his understanding of conversations. 11 11 Also, there is a holistic element to competence operating here that isn’t captured by semantic theory. Semantic theory has nothing to say about the perspectives by which speaker’s grasp content. But for common synonyms, competence often demands that speakers have a unity of perspective, because this will be assumed and exploited in conversational contexts. A’s understanding of ‘steed’ is greatly impoverished, to be sure, but the only facet of impoverishment relevant here is his failure to know that it means the same as ‘horse’. This impoverishment is symmetric and applies equally well to A’s understanding of ‘horse’. The example above could just as well motivate a story about A’s lack of competence with ‘horse’. Hence, there is a holistic element because synonyms can place extra demands on competence with an individual word. Not only does a speaker need to know what a word means, but in some cases must know that other words mean the same. If there were no synonym for ‘horse’ A would be perfectly competent with it. But since there is, A’s competence with ‘horse’ is diminished since he doesn’t know ‘steed’ means the same, even though he might know what horses are and associate this information with ‘horse’ just as well as any normal speaker. What B has, and A has not, is a unity of perspective on the property of being a horse for both words. 153 The example above illustrates how inferential competence contributes to the ability to recognize the same content as presented again. This ability is also relevant to recognitional competence, since we often have to recognize that a property expressed by a word is also instantiated in the local environment. This ability is crucial to understanding how a conversation relates to practical pursuits. Suppose A is the valet for a restaurant. The parking lot for the valet service is across the street, so when patrons come out of the restaurant, they can see their cars across the street. B has used the valet service, is finished with his meal, and needs to have his car brought around. B exits the restaurant and has the following conversation with the valet. (6a) A: Which car is yours? (6b) B: It is burgundy. B has looked across the street at the valet lot, noticed that his is the only burgundy car in the lot, so he reasonably assumes that A will be able to determine which car belongs to B by being told that it is burgundy. For the purpose of discussion, let’s assume that B’s use of (6b) asserts the singular proposition (6c), where o is B’s car, and B* is the property of being burgundy. 12 (6c) <o, B*>. In uttering (6b), B is assuming A knows that ‘burgundy’ applies to o. 12 This is merely a simplifying assumption. It may be that B’s use of ‘it’ in (6b) is anaphoric on the contextually supplied description B’s car. And arguably B expects A to infer that o is B’s car. Even if this is so, however, it will still be true, as I argue, that A needs to know which car ‘burgundy’ applies to in order to appropriately understand B’s contribution to the conversation, and thus do what he needs to do in response to it – namely retrieve the burgundy car for B. 154 But suppose A is, in effect, merely referentially competent with ‘burgundy’, having at best very little recognitional competence with the term. A knows ‘burgundy’ is a color term, but nothing more. So A doesn’t know that ‘burgundy’ applies to o. This is the crucial point, but again we have to be careful about what A does and doesn’t know. Let’s grant that A knows B’s utterance of (6b) expresses (6c). 13 Further, let’s assume A has a normal visual system, has experience with burgundy objects, and that this ensures he knows that o instantiates B*. But A doesn’t know that there is a single property that is both instantiated by o and predicated of o by B’s utterance. A has to understand this in order to understand B’s response to his query – given that they both know the conversation is bound up with the practical goal for A to retrieve B’s car. So (6d) is true, but (6e) is false. (6d) ∃p(A knows that (o instantiates p & B’s utterance of (6b) asserts <o, p>)). (6e) A knows that ∃p(o instantiates p & B’s utterance of (6b) asserts <o, p>). B is, in effect, reasonably (but incorrectly) assuming that (6e) is true. As with the previous example, this shows that A has two non-transparent perspectives on (6c). The first perspective is quite rich, and tied to A’s visual capacities. But the second is an impoverished perspective tied to his impoverished understanding of ‘burgundy’. When A looks at B’s car, he entertains (6c) in the first way, and when he considers B’s utterance, he grasps (6c) in the second way. Due to the impoverished guise 13 I assume that A knows B’s use of ‘it’ in (6b) refers to o. But this doesn’t mean that A can visually locate the car B has referred to. 155 tied to A’s impoverished understanding of ‘burgundy’, A doesn’t know that o instantiates the same property that is attributed to o by B’s utterance (in the sense that (6e) is false) – i.e. A doesn’t know that the two perspectives are of the same proposition (or better, the same property). In order for A to do what he needs to do in response to his conversation with B, his visually mediated perspective on (6c) must line up sufficiently well with his linguistically mediated perspective on (6c) so he can see that these are simply two different views of the same proposition (as in (6e)). Normal speakers do this by means of their recognitional competence with ‘burgundy’. A speaker with a normal level of recognitional competence with ‘burgundy’ will be able to tell which car in the lot this word applies to, and this is the crucial link uniting the two perspectives. So we can trace A’s failure to understand the conversation in a way that is adequate to the practical demands of the context to his lack of recognitional competence with ‘burgundy’. 14 That a speaker can fail to know two expressions are synonyms even though she is in some sense competent with both has traditionally been taken as the mark of non- transparency. The example above shows that non-transparency is part of a more general phenomenon including cases where a speaker doesn’t know that the property expressed by a word is presented again in her local environment. We can go further and show that 14 Similarly to the first example, there is a unity of perspective assumed here by B, but there is an important difference. In discourse (5a) – (5b), A could be pretty ignorant about horses and still understand B alright as long as he knows the relevant synonymy. But in the case of (6a) – (6b), the required unity of perspective must be achieved “on the fly” in the context by means of recognitional competence. So in (6a) – (6b), A’s failure to have the unity of perspective assumed by B is more directly due to his failure to connect what he knows about the color burgundy (i.e. knowing what burgundy objects look like) with the word. The general point is that recognitional competence can be tied to action and perception in ways inferential competence isn’t. 156 non-transparency is revealed by a speaker’s failure to see other kinds of relationships between contents (besides identity). Let’s consider the meaning containment relation. The meaning of both ‘stallion’ and ‘mare’ contain the meaning of ‘horse’. Suppose we have the following conversation. (7a) A: How many horses do you own? (7b) B: I own three stallions and two mares. Clearly B has appropriately answered A’s question. Even though the semantic content of (7b) (relative to the context) isn’t the proposition that B owns five horses, B counts as having asserted this proposition. The semantic content of (7b) is the proposition B primarily asserts and intends to convey. But any propositions that are pragmatically implied by the proposition a speaker primarily asserts and intends to convey in a context will also count as asserted by the speaker in the context. 15 This explains the sense in which B has appropriately answered A’s query, since an appropriate answer to this query must involve B saying how many horses he owns – in this case, five. In order for B to count as saying that he owns five horses, we must assume that (what is expressed by) (7c) and (7d) are among the salient presuppositions of the context. (7c) ∀x(stallion(x) → horse(x)). (7d) ∀x(mare(x) → horse(x)). 15 This is a plausible and widely accepted view. Proposition 1 is pragmatically implied by proposition 2 in a context if and only if proposition 1 is an obvious and relevant consequence of proposition 2 in conjunction with other presuppositions which are salient in the context. In order to cash the notion of an “obvious” consequence, we need to invoke some notion of “provability” for propositions. Thus, one proposition being an obvious consequence of others will not simply be a matter of the proposition being propositionally entailed by others. This point is crucial for understanding why (7c) and (7d) must be thought of as salient background presuppositions of the context. 157 But suppose A is merely referentially competent with ‘stallion’ and ‘mare’. A doesn’t know that both ‘stallion’ and ‘mare’ contain the meaning of ‘horse’ and hence doesn’t know that (7c) – (7d) are relevant to B’s response (A might know both, but not know that (7c) is relevant to uses of ‘stallion’ or that (7d) is relevant to uses of ‘mare’). For all A knows, B has responded to his query inappropriately, along the lines of (7b′). 16 (7b′) B: I own three bulls and two hens. Since A knows the semantic rules for ‘stallion’ and ‘horse’, however, he knows, of the proposition that B has three stallions and two mares, that B has asserted it. Although A’s ignorance makes the context defective, because there is communicative failure, the defect is entirely on A’s head. It’s reasonable to describe the conversation by saying B asserted that he owns five horses, but A didn’t realize this. To see this point more clearly, suppose there are 100 others in the conversational audience, all of whom are normally competent with ‘stallion’ and ‘mare’ and thus know (7c) – (7d) are relevant. In this case we surely wouldn’t want to say B didn’t assert that he owns five horses due to the ignorance of one. A’s ignorance doesn’t harm the fact that B has responded appropriately to A’s query, and it follows from this that (7c) – (7d) are salient presuppositions of the context even though A doesn’t know they are salient. In my view, (7c) is a pragmatic presupposition normatively associated with ‘stallion’ in the sense that speakers need to know it’s generally relevant to uses of the 16 As with discourse (5a) – (5b), I assume here that A is being a bit daft, since it arguably would be natural for A to infer from B’s response that ‘stallion’ and ‘mare’ contain the meaning of ‘horse’ (by noticing that this is required for B’s response to be appropriate). Again, this is interesting since it shows how speakers can infer from conversational context something about how the meanings of words are related, but since this isn’t central to my discussion, I won’t go into it here. 158 word in order to hold their own in everyday conversations (similarly for ‘mare’ and (7d)). Presuppositions of this kind are not salient because every party to the conversation is disposed to accept them as salient (e.g. as in Stalnaker’s view), but rather because speakers need to know that they are salient in order to do what they are expected to do – competently converse in normal contexts. 17 This explains why the defect is on A’s head. Due to his impoverished competence – his failure to meet established pragmatic norms for ‘stallion’ and ‘mare’ – A fails to realize that (7c) – (7d) are salient presuppositions of the context. The context isn’t dragged down to A’s impoverished level, making B not count as asserting that he owns five horses, but rather A is failing to meet operative standards of the context. This doesn’t mean, however, that these presuppositions are indefeasible. (7c) – (7d) are sustained as presuppositions under the reasonable assumption made by B (and others in the audience) that everyone (including A), is a normal speaker, and thus knows what the relevant words mean (which requires knowing (7c) – (7d)). If A goes on to say or do something suggesting he doesn’t know what the words mean, then (7c) – (7d) are temporarily canceled, and the conversation digresses until A is brought up to speed. The examples we have considered illustrate a type of pragmatic presupposition with several interesting features. 18 First, as already mentioned, presuppositions of this kind are normatively associated with an individual word, as a standard of what needs to 17 See Stalnaker (1973), and the discussion of his views in Soames (1989). My language of “disposed to accept” alludes to the strategy of “accommodation” Soames discusses in relation to Stalnaker’s view. 18 I think this type of pragmatic presupposition hasn’t been recognized, but it would take us too far astray here to catalogue the various theories of presupposition for comparison, though I have briefly indicated how it differs from Stalnaker’s (1973) view. 159 be known about its content to count as fully competent with the word. An utterance containing a token of the word will, in general, automatically “trigger” one or more presuppositions associated with it (so long as it can be reasonably assumed that all parties to the conversation are fully competent with the word). Second, these presuppositions facilitate the representation of discourses. A conversation is an exchange coordinated by its participants, so in order for a speaker to track the exchange, she must be able to represent for herself the salient ways in which various assertions are related to each other and to things in the world. 19 Third, these presuppositions partly govern the rationality of utterances. A degree of rationality is one target of any discourse, and this places a normative constraint on individual contributions to the effect that the utterance can’t express any jack-in-the-box assertion entirely disjointed from the established flow. 20 It’s part of the rationality of a speech act that a speaker forms a reasonably accurate intention about how her assertion makes a coherent contribution. In doing so, she relies on a shared representation of the discourse. The intuitive idea is that a speaker generates a fairly accurate “picture” of the discourse as it unfolds. Then, when the speaker makes a new contribution to the discourse, the rationality of her utterance is partly governed by her being sensitive to how the new contribution coherently adds to the existing picture. This sensitivity may include new presuppositions that get triggered by the new contribution, and facilitate her 19 One might say that the presuppositions serve to frame the representation of discourses in a speaker’s “language of thought.” I’m not committed to their literally being a language of thought, but I do think that a speaker’s representation of a discourse is “cognized” in some fashion. 20 At least, if it is, it must be marked as such – i.e. “I’m sorry to change the subject, but…” or “I’m sorry to interrupt, but…” 160 understanding of how her utterance contributes. Depending on how well she understands the words in the new contribution, her sensitivity to how it contributes to the discourse can be better or worse, and as a result, the utterance can be more or less rational. In the case of discourse (7a) – (7b), the view is that (7c) and (7d) are salient presuppositions triggered by B’s utterance of (7b), and they are a required part of B’s rational intention that his utterance forms a coherent contribution to the conversation so far (i.e. an appropriate response to A’s query). Suppose alternatively that we have discourse (8). (8a) A: How many horses do you own? (8b) B: # I own three bulls and two hens. In this case, B clearly has answered A’s query inappropriately (as marked by ‘#’ in (8b)). Suppose that all parties to the conversation are normally competent with the relevant words, so everyone knows that B has answered inappropriately. The infelicity of B’s answer lies, at least in part, in the fact that it’s irrational. B has an accurate representation of the discourse at the moment of his utterance, knows what ‘bull’ and ‘hen’ mean (i.e. knows that neither bulls nor hens are horses), and hence can’t rationally intend his utterance as a coherent contribution to the discourse (i.e. an appropriate response to A’s inquiry). But the infelicity of B’s response clearly does not prevent him from successfully (but uncooperatively) asserting that he own three bulls and two hens. So the rationality of an utterance is not entirely contained by the isolated act of assertion, but also by the extent to which a speaker can rationally intend her assertive utterance as a coherent contribution to the discourse at hand. 161 Conclusion Language is first and foremost a tool for discourse. Because a speaker’s ability to converse is graded, linguistic competence and knowledge of meaning are too. The gradations are found in almost every aspect of understanding and usage. A speaker can understand a sentence better or worse, and her speech acts using it can as a result be more or less rational. The gradation filters all the way down to the understanding of an individual word such that a speaker can have more or less knowledge of its meaning – and thus be more or less competent with it – even though she knows everything semantic theory has to say about it. Semantic theory cannot account for the gradations in understanding, and thus must be supplemented by a graded theory of understanding. I have shown here how culturally regimented perspectives on the referents of words play a role in facilitating discourse. Such perspectives provide shared frames for understanding how assertions in a discourse relate to each other, and to things in the world its participants commonly engage. I have argued that a direct reference theory of content is ideally suited to accommodate this insight via a separate theory of understanding. The resulting picture rings true because it places the shared perspectives where they belong – on the pragmatic side. The shared perspectives on content ultimately stem from culturally shared interests speakers bring to the conversational table. Knowing which perspectives are assumed for a community of language users is largely a matter of having experience communicating with them – being acclimated not only to their social culture, but also to their communicative culture. 162 References Asher, N. (1988). Semantic competence, linguistic understanding, and a theory of concepts. Philosopical Studies, 53, 1-36. Antony, L. M. (2003). Rabbit-pots and supernovas: On the relevance of psychological data to linguistic theory. In A. Barber (Ed), Epistemology of Language (48-67). Oxford: Oxford University Press. Bach, K. (1987). Thought and reference. Oxford: Clarendon Press. Bach, K. (1994). Conversational impliciture. Mind and Language, 9, 124-162. Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge: MIT Press. Clark, H. H., & Schaefer, E. F. (1989). Contributing to discourse. Cognitive Science, 13, 259-294. Creswell, M. J. (1978). Semantic competence. In F. Guenthner and M. Guenthner-Reutter (Eds.), Meaning and translation: Philosophical and linguistic approaches (9-27). London: Duckworth. Davidson, D. (1967). Truth and meaning. Synthese, 17, 304-323. Dummett, M. (1996a). What is a theory of meaning? (I). In The seas of language (1-33). Oxford: Oxford University Press. Dummett, M. (1996b). What is a theory of meaning? (II). In The seas of language (34- 93). Oxford: Oxford University Press. Fodor, J. (1998). There are no recognitional concepts; Not even red. In In critical condition: Polemical essays on cognitive science and the philosophy of mind (35- 48). Massachusetts: MIT Press. Foster, J. A. (1976). Meaning and truth theory. In G. Evans and J. McDowell (Eds.), Truth and Meaning (3-32). Oxford: Oxford University Press. Higginbotham, J. (1983). Is grammar psychological? In L. S. Cauman (Ed.), How many questions? Essays in honor of Sidney Morgenbesser (170-179). Indianapolis: Hackett. 163 Higginbotham, J. (1989). Knowledge of reference. In A. George (Ed.), Reflections on Chomsky (153-174). Oxford: Basil Blackwell. Higginbotham, J. (1992). Truth and understanding. Philosophical Studies, 65, 3-16. Higginbotham, J. (1998a). Conceptual competence. Philosophical Issues, 9, 149-162. Higginbotham, J. (1998b). On knowing one’s own language. In C. Wright, B. C. Smith and C. Macdonald (Eds.), Knowing our own minds: Essays on self-knowledge (429-441). Oxford: Oxford University Press. Katz, J. J. (1990). The metaphysics of meaning. Cambridge: MIT Press. Katz, J. J. (1996). The unfinished Chomskyan revolution. In Mind and Language II, 270- 294. Katz, J. J., & Nagel, R. I. (1974). Meaning postulates and semantic theory. In Foundations of language, 11, 311-340. King, J. (1998). What is a philosophical analysis? Philosophical Studies, 90, 155-190. Kripke, S. (1980). Naming and necessity. Cambridge: Harvard University Press. Marconi, D. (1997). Lexical competence. Cambridge: MIT Press. Putnam, H. (1996). The meaning of “meaning.” In The Twin Earth chronicles (3-52). New York: M. E. Sharpe. Putnam, H. (2003). Meaning and reference. In M. Richard (Ed.), Meaning (70-81). Oxford: Blackwell Publishing. Recanati, F. (1993). Direct reference: From language to thought. Oxford: Blackwell Publishing. Recanati, F. (2004). Literal meaning. Cambridge: Cambridge University Press. Salmon, N. (1989). Illogical belief. Philosophical Perspectives, 3, 243-285. Soames, S. (1984). Linguistics and psychology. Linguistics and Philosophy, 7, 155-180. Soames, S. (1989a). Presupposition. In Handbook of philosophical logic (Vol. IV, pp. 553-616). D. Reidel Publishing Company. 164 Soames, S. (1989b). Semantics and semantic competence. Philosophical Perspectives, 3, 575-596. Soames, S. (2002). Beyond rigidity. Oxford: Oxford University Press. Soames, S. (2006). Reply to critics. Philosophical Studies, 128, 711-738. Soames, S. (2009a). Naming and asserting. In Natural language: What it means and how we use it (philosophical essays, Vol. 1, pp. 251-277). Princeton: Princeton University Press. Soames, S. (2009b). The gap between meaning and assertion. In Natural language: What it means and how we use it (philosophical essays, Vol. 1, pp. 277-297). Princeton: Princeton University Press. Soames, S. (2010). Philosophy of Language. Princeton: Princeton University Press. Stalnaker, R. (1973). Presuppositions. Journal of Philosophical Logic, 2, 447-457. Stanley, J. (1999). Understanding, context-relativity, and the description theory. Analysis, 59 (1), 14-18. Stanley, J., & Williamson, T. (2001). Knowing how. The Journal of Philosophy, 98 (8), 411-444. 165
Abstract (if available)
Abstract
This book defends a theory of linguistic understanding, or what is alternatively called linguistic competence or semantic competence
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Meaningfulness, rules, and use-conditional semantics
PDF
Constraining assertion: an account of context-sensitivity
PDF
Iffy confidence
PDF
Understanding semantic relationships between data objects
PDF
Telling each other what to do: on imperative language
PDF
Reasoning with uncertainty and epistemic modals
PDF
Existential constructions: a syntactic predication approach
PDF
Superlative ambiguities: a comparative perspective
PDF
Instantial terms, donkey anaphora, and individual concepts
PDF
Comparative iIlusions at the syntax-semantics interface
PDF
Reduplication and distributivity in Kannada
PDF
Conceptually permissive attitudes
PDF
Narrowing the focus: experimental studies on exhaustivity and contrast
PDF
Investigating coordination of lexical and structural information cross-linguistically
PDF
Flexibility in language production
PDF
Reasons, obligations, and the structure of good reasoning
PDF
Reference time in the dynamics of temporal dependency in Korean
PDF
Building a knowledgebase for deep lexical semantics
PDF
Representation, truth, and the metaphysics of propositions
PDF
Vagueness, legal content, and legal interpretation
Asset Metadata
Creator
Bowman, Brian
(author)
Core Title
Linguistic understanding and semantic theory
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Philosophy
Publication Date
07/25/2012
Defense Date
04/30/2012
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
direct reference,knowledge of meaning,knowledge of reference,linguistic competence,linguistic understanding,OAI-PMH Harvest,reference,semantic competence,semantic externalism,semantic theory
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Soames, Scott (
committee chair
), Schein, Barry (
committee member
), Wilson, George (
committee member
)
Creator Email
bbowma@gmail.com,bcbowman@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-63752
Unique identifier
UC11290240
Identifier
usctheses-c3-63752 (legacy record id)
Legacy Identifier
etd-BowmanBria-984.pdf
Dmrecord
63752
Document Type
Dissertation
Rights
Bowman, Brian
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
direct reference
knowledge of meaning
knowledge of reference
linguistic competence
linguistic understanding
reference
semantic competence
semantic externalism
semantic theory