Preface This book summarizes and synthesizes the results of a long-term project on the semantics and pragmatics of Lao lexicon and grammar. The project has its roots in my years at the Australian National University and the University of Melbourne, with most of the work carried out in the Language and Cognition Department at the Max Planck Institute for Psycholinguistics, Nijmegen, within the collaborative research environment of two major research project groups: ‘Event Representation’ and “Categories across Language and Cognition’. The Lao and Kri data and analyses were collected and carried out during field expeditions in Laos (Vientiane Municipality and Nakai District of Khammouane Province) between 1996 and 2013. Lao is a Southwestern Tai language spoken by some 25 million people in Laos, Thailand, and Cambodia. Lao speakers are the dominant ethnic group in Laos. Enfield (2007) is a reference description of the grammar, with detailed background information on the language. Kri is a Vietic language of the Austroasiatic family (see Enfield and Diffloth 2009 for a detailed description of the phonology, with notes on morphosyntax and on the Kri speech community). I would like to thank the following mentors, colleagues, and friends for their help and input and ideas: Sasha Aikhenvald, Felix Ameka, Jiirgen Bohnemeyer, Melissa Bowerman}, Penny Brown, Niclas Burenhult, Herb Clark, Tony Diller, Mark Dingemanse, Bob Dixon, Paul Drew, Grant Evans, Nick Evans, Cliff Goddard, Marianne Gullberg, Bill Hanks, John Heritage, Sotaro Kita, Paul Kockelman, Steve Levinson, Asifa Majid, Bhuvana Narasimhan, Gunter Senft, Jack Sidnell, Tanya Stivers, Anna Wierzbicka, David Wilkins, and Chip Zuckerman. | also thank Annelies van Wijngaarden and Maarten van den Heuvel for their expert assistance in preparing the manuscript. I gratefully acknowledge the Max Planck Society, the Max Planck Institute for Psycholinguistics (through Steve Levinson’s Language and Cognition Department), and the European Research Council (Starting Grant project 240853, ‘Human Sociality and Systems of Language Use’) for sup- porting this work. I thank my many friends and consultants in Laos, especially the inhabitants of Doune Ian Village, Vientiane Prefecture, and Mrka Village, Nakai District, Khammouane Province. And once again—this time thanks to Yves Goudineau and Christine Hawinxbrock—the Vientiane office of PEcole francaise d’Extréme-Orient kindly provided the amenities for me to write in peace. At OUP, John Davey generously and patiently supported the idea of publishing this book, and his guidance was, as ever, indispensable. I thank him and Julia Steer for their attentive overseeing of the book’s long gestation.Preface xv I am grateful for the good fortune to have been able to draw directly on the teachings—in personal communication and in print—of four generous and formid- able mentors, each fantastically different from the others, and each equally indis- pensible: Anna Wierzbicka, Nick Evans, Steve Levinson, and Melissa Bowerman‘. Finally, I thank Na for support and Nyssa for perspective, and I dedicate the book to our second daughter Nonnika. By adding one to our unit she doubles our number of relationships!The utility of meaning Here, in two points, is the thesis this book will defend: 1, Word meanings reflect, and create, a deeply subjective view of the world. 2. This is a necessary consequence of the fact that word theanings are the historical product of their utility as means to people’s communicative ends. With case studies and arguments, the book aims to support this thesis, and to grapple with some of its theoretical and methodological implications. A good portion of the work addresses a prerequisite to any question about the causes and effects of word meanings,’ namely: What are word meanings like in the first place, and how do we show it? = Here’s what happens when we consider this question in causal terms. We are forced to deal with the implications of a gap between private cognitive representa- tions of word meanings, on the one hand; and the public careers of words—both in the flow of social interaction and in the population-level emergence of convention— .on-the other hand. Then, a paradox comes to light. We see that the cherished principle of semantic invariance is at once a methodological necessity and a theor- etical near-impossibility, Yet the system works. It works because of the utility of meaning. [ mean this not just in the sense that words are means to ends,” but also in the sense that meanings must.be good enough to serve their functions, but need not be better than that. ‘Meaning is often thought of as representation, but at the core of language use is decision-making, There are constant questions. What did she mean by that? Why did: she say it that way? How shall I respond? What will she think I mean? We need answers on the fly. As in any decision-making process, the heuristics we use for * Tsay ‘word’ meanings, but I also mean to include other symbolic structures of language at scales both smaller than, and'larger than, the word (see eg. sections 5-53, which focus on the meaning and productivity of grammatical constructions). * The idea that words and other parts of language are tools is well worn, from carly psycholinguists Zipf (4935; 1949) and Vygotsky (1934) to philosophers Austin (1962) and Grice 1975), linguists Chafe (1980) and Everett (2012), and anthropologists of language Hanks (1990; 2005), Sidnell (2005; 2010) and Kockelman (2010; 2013), among many others. »2 The utility of meaning coming to solutions will be fast and frugal (Gigerenzer et al. 2011). Pre-i950s psycho- linguistics had this idea (Zipf 1935; 1949), and new research on language is exploring it, for example in the population-level diachronic aggregation of conventions (Barr 2004), the microgenetic comprehension of utterances (Ferreira and Patson 2007; Enfield 2009a), and the enchronic flow of action ‘and response in conversation (Sidnell and Enfieid 2012, 2014; Enfield 2013a).’ Each of these temporal-causal frames isa wheel within wheels, making the whole of language look something like a Mayan calendar. Knowing how just one wheel turns will not suffice. Definitive answers to the questions posed in this book—questions of what words mean and why—demand simultaneous attention to processes in all of the relevant frames (Enfield 2014) ‘Within the scope of this book, we shall focus on the problem of word meaning by thinking not just in synchronic terms but also with due consideration of the dynamic frames of language that account for the creation and maintenance of meaning, and of the rich-cultural-historical contexts in which language is used. Chapters 2-6 argue, in turn, that word meanings are layered, multiple, anthropocentric, cultural, and distrib. uted, thus touching from all angles on the general thesis stated in points 1 and 2 above. The conclusion (Chapter 7) is that meanings are, above all, and necessarily, useful. If word meanings were not useful for people they would not exist. The broad goals of the book are thus (1) to put forward a view of what word meanings are like, (2) to see what this view implies for the ways we study word meanings, and (3) to explore the relevance of this view for our understanding of their causal underpinnings. Point 1 of the thesis will already be familiar to readers. Many authors have championed the idea that the concepts conveyed by language are fundamentally subjective in character, and that language is not a means for reflecting how things are, but rather a means for portraying it in certain ways, depending on a speaker's communicative goal." Few would deny that language is inherently subjective in nature. But this does not detract from the point’s importance, nor from the need to make it, While the subjectivity of language is widely acknowledged, its implications— both theoretical and methodological—are not always appreciated. Point 2 of the thesis concerns questions that usually fall outside the scope of synchronic semantics. But it is essential to deal with these questions. Any version of semantics must be compatible with a natural, causal theory of language (cf. Millikan 2005; Enfield 2014). A complete account must explicate not only the meanings of interest, but also the causes, conditions, and consequences of those meanings. {For the terms ‘microgenetic’ and ‘enchronic’, see below; also Enfield (201335 2014). * Sources for this idea range from Boas (1911) to: Sapir (1921; 1949), Malinowski (1925), Mead (1934), Zipf {(4s4a), Whorf (1956), Brown (19588), Jakobson (1960), Austin (1962), Searle (1966). Slverstein (1976: 1979), Clark and Clark (975), Levinson (2983; 2000), Langacker (1087), Lakoff (i987), Hanks (1990), Dixon (agen), H, Clark (1992; 19962; 1996b), Wilkins (1996), E. Clark (1999), Croft and Cruse (2004), Evans (2010), Everett Gora), and exp. Wierabicka (i985; 1988; 989; 1996), whose indelible influence should be clear throughout the chapters of this book. What this book aims to show 3 Many semanticists aim only to describe meanings within a synchronic perspective, specifying the contents of coded meanings, and relations between those meanings in a system. Others complement this with a diachronic perspective, investigating principles of meaning change, sometimes stating them descriptively, and sometimes positing specific causal accounts—thus invoking not just diachrony but other causal- temporal scales such as enchrony and microgeny—for example drawing on the role of pragmatic inference in conversation (Traugott 1989; Sweetser 1990; Wilkins 1996; ‘Traugott and Dasher 2002; Evans 2003). These lines of work need to be better linked to disciplines that also have a stake in semantics, and thus have a stake in getting semantics right. I am thinking here, for example, of Millikan’s (2005) insistence that meanings have their causal basis in natural population-level processes of cultural evolution (see also Schelling 1978), and Brandom’s (1979; 1994; 2000; 2008; 2014) insistence that logical relations, are grounded in public commitments and social normativity (see also Sacks 19925 Heritage 1984). 1.1 What this book aims to show In this book we ask what the meanings of words are like, using the Lao language as a main source for case studies. Our investigations will support the following conclusions: 11.1 Word meanings have an effectively invariant core Whenever.a words used, that word will dependably invoke a definable core idea in the minds of people who hear the word being used (Wierzbicka 1996: 24 and passim; Goddard 2011). A speaker will be accountable for having intended to convey this core meaning, which is to say that the speaker will be unable to plausibly deny that they had wanted to invoke this understanding in the other. And a listener will be accountable for having understood the word with that meaning, as long as they claim, explicitly or implicitly, to have understood the utterance in which the word was used.® 5 This may sound odd, but here's how it works. Imagine that John says to Mary There's a dead cat in the rivesay, and she replies Real? as she heads ou to take 2 lok. fn fact the dead thing in the driveway turns out to be a dog, John will be accountable for having lied, mis-spoken, or been mistaken. He won't be able to defend his earlier assertion, a fact that is explained with reference to the core meaning of the English word cat, Something similag applies to Mary. Ifthe dead thing turns oUt to be a Mans, she can later asert that she hadn't expected this John didnt say it was a Manz). But she can’t later assert that she hadn't expected e.g, that the dead thing would be an animal (even though John didn’t say it was an animal either). More accurately: If she were to assert that she hadn't expected it to be a cat, or an animal, she would be accountable forthe intonsistency between this and her explicit signal of understanding in the response she gave to John (Really?).In other words, ifshe acts now like she knows what he means, she can't say later that she didn’t know.4 The utility of meaning By saying that these meanings are effectively invariant, I am explicitly allowing thet word meanings can vary in their actual representation from person to person, but crucially they do this in such a way that any differences are undetectable to users of the language. This is a recurrent theme in this book. The invariance is methodo- logically necessary, but theoretically near-impossible. A few points of clarification are needed, to avoid misunderstandings of what this means. 1, Effective semantic invariance’ does not mean that multiple meanings are not possible. In fact, it is rare for a word not to be polysemous. But for each distinct ‘meaning that a word has, that meaning is effectively invariant. Instead of saying that a word has an effectively invariant meaning, it wil often be more accurate to say that a word.has effectively invariant meanings, 1b. “Effective semantic invariance’ does not mean that words cannot be taken to convey different things in different contexts. In fact, it is rare for a word not to be given a context-particular understanding—via various forms of inference—that is more specific than the word’s invariant core meaning. When a word is used in a context, many more things than just the word’s meaning are available for the hearer to use in constructing an interpretation of what the speaker wanted to say, While a word's invariant core meaning (or at least, ole of them; see (1a)) is always conveyed, many further things may be conveyed in addition, depending on the context. 1c. “Effective semantic invariance’ does hot méan that word meanings cannot change; in fact they change in numerous ways, and in numerous causal-temporal scales, all the time. But it is important to note that the different kinds of extension of meaning implied by such change are based in logically distinct causal mechanisms, and these different kinds of extension should not be confused with one another, The points just made are treated in Chapters 2 and 3: ftom the problems of defining the invariant meanings of words, to teasing these apart from the contextual _ effects of those words when used in context (semantics vs. Pragmatics; see’section 21) to establishing and distinguishing between the multiple meanings of a single word (polysemy, sections 2.2 and 3.2, and monosemy, section 3.4), to different kinds of thing we mean by extension (sections 2.3, 31, and 3.2), to a more specific sense of extension across syntactic categories (heterosémy, section 3,3). 11.2 Word meanings are deeply subjective—both anthropocentric - Gnd culture-centric—in content The contents of word meanings, as wel as the context-situated enriched meanings of words, are highly subjective in that they reflect especially human concetns (Wierzbicka 1985; 1989; Simpson 2002; Evans 2003). This may seem like an uncon- troversial or commonsense claim, but in fact, with increasingly widespread adoption ‘of onomasiological approaches to studying word meaning (good examples being =] What this book aims to show 5 Geeraerts 1997; 2010; Levinson and Meira 2003; Majid et al. 2007; Majid et al. 2008; Hellwig 2006a; 2006b; Majid et al. 2011; 201b; Levinson and Wilkins 2006; Evans etal. 2011), it is important to acknowledge a potential danger of prioritizing exten- sional facts in semantic analysis. [f your application of referent-first methods leads you to conclude that the objective properties of a word's referents are equal to the ‘meanings of that word, then you're doing it wrong. Words can denote things in the world, but words’ core meanings are not grounded in objective properties of the things they denote. Instead, word meanings embody a human perspective. This is true in more than one sense. One is that humans have" special propensity for attributing psychological motives behind actions and behav- iour- We see this, for example, in the meanings of words for simple controlled behaviours such as cutting and breaking (sections 4.1 and 4.2). Another is that even in referential domains that do not denote human agents, actions, or their products— such as the domain of landscape features—a deeply anthropocentric perspective is still encoded, especially on account of the intrinsically subjective nature of affor- dances (sections 4.3-4.5). Yet another is that cultural expectations can guide the interpretation and selection of words in grammatical contexts, showing us how cultural perspectives are manifest both on the syntagmatic axis (sections 51-5.3) and the paradigmatic axis (5.4 and 5.5) 1.13 Word meanings are primarily conceptual, not primarily perceptual If purely perceptual features of referents play any role in determining or shaping word meaning, it is not a direct result of the fact that some feature of the referent is perceptually salient. Rather, it is because this perceptual salience is a useful tool for people to use in solving a referential coordination game in communication. Percep- tual salience of a referent is a means to an end, that end being successful coordination of reference (sections 4.4, 4.5, 6.1, 6.2). If we discover that a word’s meaning denotes a perceptually salient distinction, this does not explain why the word exists. For that we must ask: What is the social-communicative goal of the coordination of reference that the word enables? z 1.4 Word meanings are distributed, in dialogues and in populations . While a cognitive approach to word meaning must situate word meanings in people's heads (i.e. in their mental representations, including their meta-representations), meanings are publicly distributed, in two ways. First, a word’s meaning cannot be calculated without access to a person’s response that reveals this meaning (Peirce 1955; Kockelman 2005; 2013a; see Enfield 20132: ch. 4 for explication of this point within a broader semiotic framework). This establishes a dialogic basis for meaning—which is to say that meaning is distributed across the contributions of two people in a communicative interaction (sectioris 6.1 and 6.2).6 __ The utility of meaning Zipf (1949: 20) puts it more radically: ‘we shall define a meaning of a word a8 a kind of response that is invoked by the word.’ Normally, in synchronic semantics, we bracket Out the essential role played by thé interpreter. A second sense in which word meanings are distributed is that they are multiplied across a population of heads or minds, and across an enormous (though finite) and interconnected set of past and present contexts of usage (6.3 and 6.4). While meanings can effectively be nailed down, the fact that they are radically distributed in these two ways means that we cannot actually nail them down. In other words, the assumption of semantic invariance is methodologically well-motivated, not theoretically well-motivated. , r We will focus mostly on the above two senses in which meaning can be dynam- ically distributed, but note that these belong to a larger set of dynamic frames, A usefill taxonomy of six such frames spells MOPEDS (Enfield 2014): Microgenetic (invoking cognitive and motoric processes involved in producing and comprehend- ing language), Ontogenetic (invoking lifespan processes by which people, especially as children, acquire linguistic knowledge and skills), Phylogenetic (invoking ways in which the requisite cognitive capacities have evolved in our species), Enchronic (invoking the: sequential interlocking of social actions in linguistic clothing), Dia- chronic (invoking historical change, conducted socially in human populations), and « Synchronic (any approach that does not explicitly invoke notions of process or causation at all, such as language description). The distributed nature of meaning—in the multiple senses just discussed, and in the infteraction between them—creates footholds for change and maintenance by means of the diffusion of innovations in human populations (Rogers 2003; Enfield 2014). This involves splitting a word’s usual overall import into a semantic part (the core invariant, stored in the head) and a pragmatic part (the enriched sense in context, not necessarily stored), and distributing these parts across representations in the mental lexicon (in a synchronic frame) and processes of inference (in micro- genetic and enchronic frames). At the same time, there is a distinction between representations in individual minds (multiplied by people in a community) and the necessity for individuals to assume that there is a functional average for any word meaning in their community—ie. what Frege meant by ‘sense’ (see 6.1 and 6.2 for explication) ‘The assumption that word meanings are truly shared is a piece of meta-reality. It is an assumption about how things really are. But itis effectively vindicated by reality too. If seems to us that things really are that way. “ For convenience, I sometimes use the term ‘online’ to denote the enchronic and microgenetic rames taken together. a What this book aims to show 7 Words exist because they have been used; therefore to know why we people use a word is to know why the word exists Whenever awordis used, ithas ben selected fora function that fis what weare trying do with our words. That function is enabled by the meaning of the word, and it is yp what has led to the selection by people of that word for that function (See sections fa a chapter; also 4.4, 4.5, 6.1, 6.2). A functional approach to meaning should not 1 | jase dlachuonic-and miaogtneti effec of frequency, eytematization, and od atta structure (Croft and Cruse 2004: 291ff), but also the enchronic effects of ranguage use for Social action, Let us consider this point in more detail. . Why do we have different words for referring to what seems to be the same thing’ One answer is so that our utterances can. convey the different vonstruals that we make of a situation we want to describe, This can be viewed as a purely reflective function of language. On this view, whatever construal I happen to make of the situation, I will simply, perhaps automatically, select the words that best reflect this. Note that this is about the relation between a speaker and the state of affairs they are describing, It does not take into account the relation between speaker and hearer. Consider the example of choosing between using passive or active voice in describing a scene. Tomlin (1997) showed that when a person’s visual attention is on the patient in a transitive scene (a cartoon of a fish being eaten by another fish), then the person will reliably use a passive rather than active utterance when describ- ing the scene. This attributes little agency to a speaker: linguistic formulation is an automatic reflection of attentional state (cf. Chafe 1994). But there is also a creative function of selecting from among linguistic construals. Suppose you see a film of a collision between two cars, while someone describes the scene with words. Loftus and Palmer (1974) carried out an experiment with two conditions. In one condition, the scene was incidentally described by the experi- menter with the verb crash, in the other with bump. People viewed the same scene, but under different descriptions. When they were later asked to simply estimate the speed at which they thought the cars were driving based on what they saw—a task that ostensibly makes no reference to what was said to them by the experimenters— those people who had heard the word crash estimated the speed as higher than those who had heard bump, even though everyone haé witnessed the same scene. People's different perceptions of the scene were creations of the experimenter’s choice of words. ; In this way, a speaker may freely select one of another word as a way to intention- ally influence how their hearer will think about the scene they are describing: Why ‘would we do this? Because when we speak, we are not just saying things, we are doing things (Austin 1962; Sacks 1992; Clark 1996, Levinson 2013). We select different linguistic formulations to fit our utterances to the actions we mean to make.8 __ The utility of meaning A simple and familiar illustration is when choice of words can porcray an action as either intentional (culpable) or not (an accident). Sidnell and Barnes (2013: 331) give * the example of a dispute between two children who have just been reprimanded by an adult for getting too rowdy. One child says She poked it, the other counters with I tapped it. Here, the coded semantics of the English words poke vs. tap—clearly different in meaning, but both equally plausible for reference to certain situations— are used strategically for potentially consequential ends.” There is no one-to-one or otherwise direct relation between words and the things they refer to (Brown ig58a). When I want to talk about a thing, place, person or time, {cannot just ‘use the word for it’ I cannot just label the thing, because there is no one way to label it. I choose words that I trust will be an adequate guide for you to know what thing I mean, and how I want you to take what I am saying about it, to'a degree that is at best perfect, and at least adequate, for communicative purposes. - The points just discussed will be explored in this book with reference to case studies mostly from Lao, in Support of the thesis stated at the beginning of this chapter. ‘Now, we preview three recurring ideas that form thematic strands running through the chapters of the book; first, the idea that the utility of words is what justifies their existence; second, that, psychologically, a word meaning is never more than a hypothesis; and third, that there is a gap between minds and the community—a gap too small to matter for communication, but just big enough for the wedge of semantic change to fit. 1.2 Word utility as raison d’étre for semantic coding Linguistic labels are public conventions that are kept in circulation by public behav- iour that creates joint attention directed at those labels. Just because a concept is useful does not mean it has to have a label. Labels are necessarily public, social entities, while concepts need not be more than private, When a concept has a label, this tells us about the utility of the label. It does not necessarily tell us anything about the utility of the concept. This point is important, because even the best work on meaning tends to focus exclusively on conceptual content—the contents of the signified pole of the sign—without also looking into the career of the public entity that is the sign as a whole, 7 See Schegloff (1972), Sacks and Schegloff (2007), Stivers (2007). Enfield (2013b) presents an extended ‘example showing how references to times can be tailored to the action being crafted with talk, comparing ‘eg. this time tomorrow's. about midday vs: about lunch time, and over Christmas vs. since December 24 Vs, dluring the last week. Word utility and semantic coding 9 This issue is central to a controversy in anthropological research on ethnobiolo- gical classification. Berlin (1992: 80) defined the following puzzle. We need to explain why ‘only a small subset of the species diversity in any one local habitat is ever recognized linguistically by local human populations’. This puzzle—with implica tions for all semantic domains—is the one we are interested in here. Berlin opposes two possible solutions. One'view is ‘intellectualist’. Human beings are inherently interested in acknowledging—by naming—the perceptually most salient discontinuities supplied by nature. The attested labels directly reflect this perceptual salience. [ln the categorization of plants and animals by peoples living in traditional societies, there exists @ specifiable and partially predictable set of plant and animal taxa that represent the smallest fundamental biological discontinuities easily recognized in any particular habitat. This large but finite set of taxa is special in’each system in that its members stand out as beacons on the landscape of biological reality, figuratively crying out to be named. (Berlin 1992: 53) Another view is ‘utilitarian’, The naming of a given plant or animal in a language will be culture-specific, guided by the ‘practical consequences of knowing or not knowing {that} plant or animal’ (Hunn 1982: 834). Berlin (1992: 80-90) mounts an empirical «attack on utilitarianism, showing, for example, that it is common, indeed normal, for languages to have a great many words for biological taxa whose referents are of no obvious utility to speakers, whether this utility be concerned with the need to pursue, avoid, or otherwise be practically interested in the life form. . But there is a logical argument for the necessity of utilitarianism in a different sense, which suggests an empirical research trajectory not yet seriously pursued, despite shelves of literature, It begins from the realization that the question of what the referent is useful for is only indirectly relevant to why the label exists. Rather, we must ask what the label (ie. the word in the language) is useful for. To rephrase Hunn: We will have a name for something when there are practical consequences of knowing or not knowing the word for it, Itis clearly reasonable to assume that if a referent is useful, it-will probably have a label. This is because people will need’a label for solving recurrent problems of social coordination that relate to the said useful referent. So, if'a variety of Cyperacea reed (labeled phittiz in Laoy'see section 4.5) is useful to us because we use it to make floor mats, then having a word for the plant is strongly motivated. But note that the utility of the thing is logically distinct from the utility of a word for the thing. The word is needed in order to coordinate social behaviours that are in some way oriented to that thing (e.g. saying things like ‘Do you want to go to the swamp tomorrow and collect some Cyperacea reeds"). So, ifa referent has.what we might call direct utility—the referent itself is useful to people—then it will probably be labelled, for the reasons just mentioned. In Berlin’s10 The utility of meaning critique of the utilitarian hypothesis, he seems to imply that ifa referent does not have direct utility, then a utilitarian account should predict that it will not thérefore be labelled. But this does not follow.’ He bases his argument in part on the fact that there are words for plants that appear to be of no direct utility to people who know / the word—i.e. where there are no consequences of knowing or not knowing those plants. But there is what can be called indirect utility, where the utility of knowing the referent is less obvious. Here is an example of the indirect utility of knowing a particular tree species and being able to distinguish it from others, This utility is completely independent from whether the species itself is in any way useful, eg, for fruit, building material, hosting edible insects. It is a utility for solving a problem of coordination in communication (Schelling 1960; Clark 1996a). The example concerns the word Ikéém, a label in the Kri language for the tree Prerocarya tonkinensis or Tonkin Wingnut. Why do Kri speakers have a word for this tree? The only way to tell is to look at what Kri speakers are doing when they use the word in real life. In the-example, from a video-recorded conversation, two men are discussing locations where they are planning to establish their swidden fields in the following year: (1) ° ciep nuu nit lee mee —r6dqg_sbaaiel-Ikéém nag chop.up toc here and make fence ‘side tree-sp. _dem.there ‘(IM)) chop (the vegetation) up here and make a fence on the side of that (other) Ikéém tree there,” In using reference to this species of tree, the speaker is indeed exploiting the fact that the tree is perceptually salient to his addressee, But nothing here suggests that the tree itself is useful to the men, nor that this reference is motivated by intellectual interest, a motivation that Berlin proposes. This example stipports Berlin’s view that percep- tual salience of the referent is important, but it shows that the role of this salience is primarily to provide a basis for communicative coordination, Interest in the referent itself is irrelevant here. The only thing that matters is that a specific tree is mutually identifiable for the purpose of referring to a location. The reference to it as a laridmark is opportunistic. The example suggests a mechanism. for how perceptual salience can motivate naming. Thé motivation is not mere intellectual interest. If it were just curiosity, this would imply something that a person can engage in privately. It would not provide a mechanism by which the perceptual distinction ‘that @ person finds ® For one thing, it is an instance of the logical fallacy post hoc ergo propter hoc: ‘Ifa referent i useful, then the referent is labelled’ does not entail that ‘Ifa referent i labelled, then the referent is useful. There «an be other reasons itis labelled, One such reason, as argued here, is thatthe word for the ferent is useful, regardless of whether the referent itself i Meanings are hypotheses noticeable or interesting would become conventionalized in the public communica- tive system of language. The same applies to any account of the motivations for linguistic meanings that appeal to individual-based psychological principles, for example claims in cognitive linguistic literature relating to principles of embodiment, metaphor, metonymy, or directionality in grammaticalization. We can claim that a certain kind of meaning is conceptually natural—e.g. a body-part metaphor like leg of a table or a metonymic place name like Palm Creek—but this will have no éxplana- tory value if no causal linkage between the private domain of cognition and the public domain of language convention is posited. Conceptually natural meanings enter language not just becausé they are natural, but because they have provided the best ways for people to get other people to know what they wanted to say. If conceptual naturalness plays a role, it is indirect Conceptual naturalness leads to the patterns we see because it provides dyads with ‘optimal solutions to problems of coordination of reference in communication. 1.3 Meanings are hypotheses ‘We now preview a simple but consequential element of the vision of word meaning outlined in this book. The idea is this: Word meanings are, by necessity, hypotheses. ‘An individual's mental representation of what a word means can never be more than a hypothesis of what that word means.” This follows from the assumption that there is no telepathy. The idea that meanings are hypotheses is simple, and not new, but-its implications for semantics, and for categorization more generally, are underestimated, Humans are separate and mobile individuals embedded in dense and complex social environments. Because we are not telepathic, our interactions with others depend on joint attention to environmental stimuli. We depend on external signals to coordinate with others. Language, the uniquely human semiotic system for social coordination, relies heavily on one type of semiotic relation, the symbolic éelation. Sound strings corresponding to words like animal, cat, and Manx stand in symbolic relations with the conceptual categories to which they refer. It is commonly stated that the defining feature of a symbolic relation such as that between the sound of the word cat and the idea ‘cat's its arbitrariness, often said to mean that there is nothing in the form of the word, or in the context(s) in which it occurs, that causes it to have the meaning it has. But this only says what does not cause a symbolic relation. It is 1 do not mean this herein the obvious sense that the semanticis’s proposed statement of a word's ‘meaning sa hypothesis; Iam talking about ll uses of a language. Al speakers’ mental representations of| linguistic meanings are, necessarily, hypotheses at best.12 __ The utility of meaning important to note what does cause a symbol to have the meaning it has: namely, Social convention. It is a symbolic relation if we link just this sound to just this meaning because others act as if they linked just this sound with just this meaning So, symbols mean what they mean because people effectively agree that they do. The sound cat denotes the idea ‘cat’ because speakers of English tacitly agree to hold anyone using the word cat to be accountable for intending to convey the idea ‘cat’ That cat means ‘cat’ is not a natural fact. It is a fact constructed by social agreement (Searle 1995; 2010). But if words like cat and Manx mean what they mean because ‘we agree’ for them to do so, how do we as individuals know just what the content of their meanings are? Assuming (a) that the concepts we refer to as word meanings or signifieds cannot exist anywhere but in people’s minds, and (b) that humaas are not telepathic, the answer must be this: We cannot know what others mean by the words they use. We can only hypothesize. But we can hone these hypotheses well, because we test them at every chance in the flow of communication . Hypothesizing as to the meanings of words is how word learning begins in infancy. The form-meaning pairings we call words are a special type, of complex category which is not given to us, but which we each have to construct (Brown 1958a3 1958b; Tomasello 2003). We first need to recognize categorical distinctions in a phonological system in order to recognize distinct signifiers (cf. bear vs. pear). Once we recognize distinct forms, we can attribute meanings to them based on the behaviour of people using them. Notice that this way of putting it has the learner beginning with the label, and then looking for the concept. Rather than taking concepts and labelling them, word learning is more like taking labels and constructing concepts that fit them. Different claims have been made as to the nature of the concepts which comprise the signifieds of words—eg, as strings of primitives, or as innate gestalt concepts, or what have you—but none can dispute the fact that for any given language, each word requires the creation of a complex category consisting of a phonological category and a semantic, category, linked together as a unit (Saussure 1986{1916]). Even if you believe that word meanings are innate and cannot be the product of construction by hypothesis (Fodor 1998), the non-telepathic person still needs to hypothesize which of these meaning categories are linkéd to which of the sound categories in the lexicon. Let us now summarize. To know a word is to have learned a word. To learn a word is to do three things: (2) construct a phonological concept based exclusively on exposure to exemplars of the category, in contrast to exemplars of similar categories,!? (2) assume thet people's utterance of this sound is intended to communicate some- thing, a meaning, and (3) assign to that form a semantic concept corresponding to that meaning. Depending on one’s theory of meaning, the semantic concept may be "© ‘The learner's construction of phonological representations is a crucial part of the story, but it is beyond our scope in this book. The gap between mind and community 13 arrived at by selecting from among an innate inventory.of gestalt meanings (Fodor 1975; 1998), or it may be constructed using innate primitive conceptual building blocks (Wierzbicka 1996; Jackendoff 1983; 1990; Goddard and Wierzbicka 2002; 2013), or delineated by highlighting an area within a semantic space (Croft and Cruse 2004) But in all cases the idea that a particular meaning is signified by-a particular phonological string is a hypothesis based on exposure to a finite number of context- situated examples. | literally mean that it is a hypothesis in the sense of a conjecture based on, and further falsifiable by—but never really verifiable by empirical data, and if necessary, revised in light of the results. ; I do not mean that speakers are necessarily testing hypotheses consciously, discussing competing proposals with their homunculi and anxiously awaiting results of the next experiment. They may do this. But in many cases they will not consciously articulate hypotheses in this way. A key period of first language acquisition is characterized by a type of speech error which may be taken as evidence that the child is beginning to generalize beyond the data they've been exposed to (Bowerman 19824; 1982b; Tomasello 2003). An example from Bowerman’s work is a 3-year-old’s utterance Don’t giggle me. This type of error can be seen as the correct output of a not-yet-correct hypothesis as to which meaning is the one signified by a particular sound string or construction. The reason the hypothesis does not fit yet is because the child has not yet béen able to test it against a-wide enough range of contexts in order that it should conform with all subsequent data (see Tomasello’s 2003 analysis of the class of words that are susceptible to this: giggle, but not laugh and not chortle.) This process, multiplied by the individuals in a population, results in the effective con- vergence of sound-meaning pairings across that population 1.4 The gap between mind and community We observe that different individuals in the community can have different ideas of what words mean—and yet that should not be possible, since, afterall, we want to say that members of a group speak the same language. As we shall see, words are, at best, tolerable friends. We assume that others in our community define words in the same ays as we do, and as long as their usage of words is consistent with our expectations, this assumption is not contravened, and we can cohabit perfectly well. This doesn’t mean that words can mean anything. Meanings cannot be negotiated or contested without limit. We move within strict confines of traditional usage. But even when we {feel that there's no room to move at all, there is room. How does this work? The idea that a word has a meaning implies, first, that there is something conceptual in the word, and second, that this conceptual thing is a convention among members of a population. Now see that there is a gap. Meaning is located in individual minds, because it is conceptual, But meaning is independent of any individual; rather, it is common to the whole community, because it is a convention. °14‘ The'utility of meaning The critical factor in making sense of this tension is a notion that has attracted relatively little attention in the linguistic study of semantics: social accountability (see Enfield 2013a: 109 and passim). The notion of entailment in linguistic meaning is normally taken to refer to purely logical relations between an assertion and other possible assertions that may follow from it. But it really means that someone who makes an assertion is then socially accountable for also making a range of other assertions that would necessarily follow from the first one. This is the sense in which propositions entail interpersonal commitments (Brandom 2000; 2014). Both analysts and users of language tend to assume that because all members of a speech community speak the'same language, and appear to get by without noticing failures of communication, then each member must have the exact same individual representation. It is a good heuristic assumption, but it need not follow. Slight differences in individuals’ representations of words’ meanings may have zero effect on the success of words for their communicative functions. Individual words’ meanings are only as convergent among language users as they have to be, but are free to vary within those limits of tolerance. The possibility of slight, undetected differences in individuals’ representations of word meaning may at the same time provide the starting conditions for significant effects in the historical development of a language. ‘The key to understanding this tension is to see that there are two causal processes of selection of word meanings that do not contradict each other, because they take place in different causal frames. On the one hand, a selectional pressure on individ- uals’ semantic representations of words to have tolerably similar meanings has its effect in the enchronic frame—the frame of conversational moves and their inter- pretive responses—ané is regimented by the link between utterances and their immediate effects. On the other hand, a selectional pressure for conventions to aggregate at the population level has its effect in the diachronic frame, and is regimented by complex mechanisms of the diffusion of innovation. This ends our preview of the central issues to be explored in this book. In Chapter 2 we identify the layers that make up the whole meanings of words and other linguistic items.