3 Paraphrasing, Part One: Words

As characterized in chapter 2, the main objection to the argument from the data of chapter 1 to the conclusion that species and other types exist is that we need not conclude that types exist, because each claim that seems to refer to or quantify over types is merely an “avoidable manner of speaking,” a façon de parler for some other claim, one that does not appear to refer to or quantify over species and other types. The present chapter and the next will be concerned with what type talk is supposed to be a façon de parler for, that is, what the paraphrase might be.¹Close The epistemological motivation for this objection was addressed and, I hope, seriously undermined in chapter 2. With the epistemological motivation gone or drastically diminished, perhaps the pressing sense that there simply has to be some way to “paraphrase away” all talk of types will be diminished too, and the reader can take a more objective look at the prospects for an adequate paraphrase.

Here is a sketch of the main argument in what follows. Nominalists usually try to paraphrase sentences that appear to refer to universals or abstract objects by making do with words, terms, or other linguistic items (for example, by saying that what all mockingbirds have in common is that the word ‘mockingbird’ applies to them). But this makes use of words, which are also abstract objects in need of being “analyzed away.” One extremely popular nominalist suggestion (found, e.g., in Goodman 1977b and in Sellars 1963 and mentioned everywhere I have given a talk on this subject) is that ‘Type T has property P’ just amounts to ‘Every token t is P’. In this chapter, I show that in the case of words, there is little hope of obtaining any (nominalistic) property P that is had by every token in view of the lack of similarity to be found among the tokens of a word—no similarity beyond being tokens of that word.

In the next chapter (chapter 4) I show that it is also not the case that ‘Type T has property P’ amounts to ‘Either every (token) t is P, every normal t is P, most ts are P or average ts are P’. Then, borrowing from research in linguistics, I suggest that the best paraphrase would be ‘ts are P’, where this is a generic, or “characterizing” sentence; but I go on to argue that even this does not work. I then spell out the difficulties of paraphrasing even so simple and understandable a sentence as ‘Old Glory had twenty-eight stars in 1846 but now has fifty’. But the coup de grâce against the likelihood of adequate paraphrasing is the virtual impossibility of doing away with apparent quantifications over types, as in, for example, Mayr’s quote in chapter 1: “There are believed to be about 28,500 subspecies of birds in a total of 8,600 species, an average of 3.3 subspecies per species.”

But first, we must take note of the fact that it is not enough merely to claim that there is a paraphrase. One must give it. As Quine (1961b, p. 105, my italics) said:

Consider the man who professes to repudiate universals but still uses without scruple any and all of the discursive apparatus which the most unrestrained of platonists might allow himself. He may, if we train our criterion of ontological commitment upon him, protest that the unwelcome commitments which we impute to him depend on unintended interpretations of his statements. Legalistically his position is unassailable as long as he is content to deprive us of a translation without which we cannot hope to understand what he is driving at. It is scarcely cause for wonder that we should be at a loss to say what objects a given discourse presupposes that there are, failing all notion of how to translate that discourse into the sort of language to which ‘there is’ belongs.

Nor is it enough to provide a piecemeal translation, by providing the odd paraphrase here and there. Given the ubiquity of apparent references to types, as we saw in chapter 1, we need to be assured that they can always be eliminated. There is no assurance of this unless there is a systematic way to eliminate them. But if their eliminativity is to be more than a nominalist article of faith, we need to be given the elimination rules. Tarski would not be so rightly famous had he been content to give a few examples of truth conditions.

Section 1 will focus on the typical pure nominalist strategy for dealing with problems posed by abstract objects/universals (e.g., redness), which is to renounce the abstract objects/universals in question and fall back instead on terms for them (e.g., ‘red’). Often this is seen by the nominalist as unproblematic, as though word types themselves aren’t abstract objects apparent references to which need to be “analyzed away” if the nominalist program is to be successful. The popular Goodman 1977b and Sellars 1963 suggestion for paraphrasing is to replace a reference to a type (‘T is P’) by a quantification over all its tokens (‘every t is P’), the idea being that what is true of the type is true of all its tokens. So for example, ‘the grizzly bear is ferocious’ would just amount to ‘all grizzlies are ferocious’. Goodman suggests doing exactly that for words. Section 2 will consider whether this suggestion is feasible. We’ll examine at some length the nature of words to see whether there is anything linguistically interesting common to all and only tokens of a word type. I will argue that there is not, and hence that the suggested paraphrase fails.

As is evident from chapter 1, there are all types of types we talk about: biological, chemical, physical, linguistic, aesthetic, and so on. I will confine my attention here mainly to linguistic types, especially words—words, that is, of a natural language like English (although I do not mean to imply that the very same word can’t be a word in more than one language; ‘cannelloni’, ‘lo-mein’, and ‘souvlakia’ prove otherwise). There are a number of reasons for starting with linguistic types, but the most important for my purposes is that it is quite hard to do without them (as will be seen in this and the next two chapters, chapters 4 and 5). Philosophers anxious to deny the existence of one sort of abstract object or another—for example, universals or mathematical entities—typically retreat to linguistic entities to do the job. This is practically the definition of ‘nominalism’; The Oxford Dictionary of Philosophy characterizes nominalism as “the view that things denominated by the same term share nothing except that fact: what all chairs have in common is that they are called ‘chairs’” (p. 264). But usually, the linguistic entities themselves can only be construed as types, and hence as abstract objects. Locke, for example (although not quite a nominalist in the preceding sense, leaning as he does on concepts to do the job that realists assign to universals), relies heavily on an apparently realist semantics for word types in his discussion in Locke 1975 (book III, chap. 3) of “general terms.” Even Quine relies on word types to express his preference for desert landscapes in Quine 1961a. There he renounces redness, but retains the word ‘red’ when he says:

the word ‘red’ or ‘red object’ is true of each of sundry and individual entities which are red houses, red roses, red sunsets; but there is not, in addition, any entity whatever, individual or otherwise, which is named by the word ‘redness’ … (p. 10)

The nominalistic trend of jettisoning universals/abstract objects for linguistic objects reaches its zenith in the philosophy of mathematics called “formalism,” where untold infinities of mathematical objects are consigned to Hilbert’s Hell in favor of “signs.” David Hilbert, Hartry Field, and Harold Hodes are our representative formalists. Formalism is, roughly, the philosophy of mathematics that holds that math is not “about” anything, least of all numbers, sets, and spaces; it is the mere manipulation of symbols, as for example occurs when we are doing “proofs.” “In mathematics,” says Hilbert (1967, p. 465), “what we consider is the concrete signs themselves,” which elsewhere (Hilbert 1983, p. 143) he says “in number theory [are] …. the numerical symbols”

1, 11, 111, 1111 ….

The problem is that these “concrete signs” cannot be construed as physical tokens if Hilbert is to derive the mathematics he wants. (I will not argue for it here, having argued for it at length elsewhere—Wetzel 1984, chapter 3.) Field (1980, p. 1), after claiming that “nominalism is the doctrine that there are no abstract entities” claims that “in defending nominalism therefore, I am denying that numbers, functions, sets or any similar entities exist.” He then goes on to help himself to an unlimited number of expressions; his theory “contains, besides the usual quantifiers ‘∀’ and ‘∃’, also quantifiers like ‘∃₈₇’ (meaning ‘there are exactly 87’)” (p. 21). In all probability, there are only a few of these numerical quantifiers having actual tokens, so to get the job done Field must be relying on expressions other than those of the actual physical variety. Hodes, too, although his view in Hodes 1984 is quite different from Field’s, rejects numbers but requires an unlimited quantity of expressions. Hodes (1984, p. 143) says, for example, “[i]n making what appears to be a statement about numbers one is really making a statement primarily about cardinality object-quantifiers.” Thus the most plausible candidates for the sort of expressions formalists require are expression types.²Close But expression types are abstract objects.

Broadly speaking, formalists want to reduce mathematics to proof theory, which studies the relation of deducibility between sentences where this is understood purely syntactically and without reference to semantic notions such as truth. But what of proof theory itself?

Quine sometimes wears a formalist hat, as for example in Quine 1961a (p. 18) when he says “In speaking of [mathematics] as a myth, I echo that philosophy of mathematics to which I alluded earlier under the name of formalism.” And so does Nelson Goodman. But unlike the other formalists mentioned, Quine and Goodman try to shoulder their ontological responsibilities by coming up with a purely nominalistic syntax for proof theory.

In Goodman and Quine 1947, they start with just the following six symbols (types or tokens? I’ll let you decide):

∨ ‘ () | ∊

(How successful their project is will be discussed in chapter 5.) Goodman (1977a, p. 262) argues that

it is [linguistic] types that we can do without. Actual discourse, after all, is made up of tokens that differ from and resemble each other in various important ways. Some are “now”’s and others “very”’s just as some articles of furniture are desks and others chairs; but the application of a common predicate to several tokens—or to several articles of furniture—does not imply that there is a universal designated by that predicate. And we shall find no case where a word or statement needs to be construed as a type rather than as a token.

To emphasize the fact that words and statements are utterances or inscriptions— i.e., events of shorter or longer duration—I shall sometimes use such terms as “word-events,” “noun-events,” “‘here’-events,” “‘Paris’-events” and so on, even though the suffix is really redundant in all these cases …. A word-event surrounded by quote-events is a predicate applicable to utterances and inscriptions; and any

“‘Paris’ consists of five letters”

is short for any

“Every ‘Paris’-inscription consists of five letter-inscriptions.”

Goodman’s specific suggestion concerning ‘Paris’ will be considered at length in chapter 5. In the meantime consider his suggestion as a general one for paraphrasing—one found also in Sellars 1963. It is that to say a type T has property P is just to say that all tokens of T have P. So, for example, to say ‘The atom consists of a heavy nucleus whose radius is less than one ten-thousandth the radius of the atom’ is to say ‘every atom consists of a heavy nucleus whose radius is less than one ten-thousandth the radius of the atom’. In general, the suggestion is that

(1) To say ‘The T is P’ (or ‘T is P’) is to say ‘Every t is P’.

In the case of the word ‘Paris’, to say that it consists of five letters is just to say that every token (or, rather every inscription token) of it does.

This may work for some types and some properties, in that there may be some types T and properties P such that T has P if and only every token of T does. Perhaps every atom consists of a heavy nucleus whose radius is less than one ten-thousandth the radius of the atom (although it is doubtful whether hydrogen can be said to have a heavy nucleus). But it fails for many types and properties—the grizzly bear is brown, but not all grizzlies are brown; some are blond, some are black. In particular, it fails for words. To see why, we need to embark on what might appear to be a rather lengthy digression on words—what they are, and what, if anything, their tokens have in common. The elimination rule we are considering entails that whatever is true of words must be true of all their tokens. But a discussion of words is essential, not only because words are a paradigm case of types (recall how Peirce drew the type–token distinction with the word ‘the’), but because it will show how utterly hopeless (1) and several other paraphrasing suggestions are.

What do I mean by a word? Ziff’s (1972) amusing and instructive article “What Is Said” illustrates that there is a host of different identity conditions for ‘what is said’, among them phonetic, phonemic, morphological, syntactic, and semantic conditions. Some of the same distinctions apply to words as to what is said; The Oxford Companion to the English Language (McArthur 1992), for example, lists eight kinds of word.³Close Yet there is an important and very common use of the word ‘word’ that stands out. A rough characterization of this kind is the sort of thing that merits a dictionary entry. (‘Rough’, because some entries in the dictionary, e.g., ‘il-’, ‘-ile’, and ‘metric system’, are not words, and some words, e.g., place names and other proper names, do not get a dictionary entry.) It is claimed in The Story of English (McCrum, Cran, and MacNeil 1986, p. 102), for example, that “Shakespeare had one of the largest vocabularies of any English writer, some 30,000 words, [that] estimates of an educated person’s vocabulary today vary, but it is probably about half this, 15,000.” If I were a certain sort of philosopher, in lieu of the characterization given, I would just give some examples of words. I could probably give about 15,000 examples but let me just mention a few: consider the noun ‘color’. The OED (Murray et al. 1971, vol. 2, pp. 636–638) entry for it lists: a pronunciation [kɒ′ lər]; two “modern current or most usual spellings” (colour, color); eighteen earlier spellings (collor, collour, coloure, colowr, colowre, colur, colure, cooler, couler, coullor, coullour, coolore, coulor, coulore, coulour, culler, cullor, cullour); and eighteen different senses—divided into four branches— with numerous subsenses. There is a separate heading for a verb with the same pronunciation and spellings (pp. 638–639). We may also add that in the sense under consideration, English ‘red’ and French ‘rouge’ are different words. Now if we are to take these dictionary entries seriously—and I think we should—then we can see that

Still, if we were to continue to try to generalize from these two dictionary entries we might be tempted to think that at least a word has only one pronunciation, since only one is listed for ‘color’. But if that means that everyone pronounces ‘color’ identically, then of course it is incorrect. Can we instead conclude at least that every word has only one correct pronunciation? No; for consider ‘schedule’. My Webster’s (Mish et al. 1993, p. 1044) lists four current pronunciations: [′ske-(,)jü(ə)l](Am), [′ske-jəl] (Am), [′she-jəl] (Can) and [′she-(,)dyü(ə)l] (Brit). Thus to (i) through (vi) we may add:

The very idea of there being only one correct pronunciation of each word—such that whole dialects are simply wrong—has, I think, pretty much died out, at least among linguists. True, there is something called “Received Pronunciation” (RP, or “Public School Pronunciation,” “the Queen’s English,” or “the King’s English”), which many people take to be standard English. Amazingly, research in Britain has shown, according to McCrum, Cran, and MacNeil (1986, p. 29), that

speakers of RP—identifiable only by voice—tend to be credited with qualities such as honesty, intelligence, ambition, even good looks. After RP, there is a league table of acceptable accents. Dublin Irish and Edinburgh Scottish are high on the list, which then descends through Geordie …, Yorkshire and West Country, until we reach the four least valued accents in Britain: Cockney, Liverpool Scouse, Birmingham and Glaswegian.

Most Americans too are dazzled by the Queen’s English, and anyway have their own hierarchy of accents. But although it’s psychologically understandable to think that an accent represents “standard pronunciation,” it is incorrect to do so. Accents constitute merely a social hierarchy; not all accents are on the same footing socially. It’s a social advantage to speak the Queen’s English. It’s also a social advantage to a man to be tall. A survey of graduates of the University of Pittsburgh found that tall men (over 6‘2″) commanded higher starting salaries than those under six feet; and among 6,000 23-year-old Britons, short men made less money than those taller, even after other obvious variables were factored out.⁴Close Clearly it is not incorrect to be short—or poor, or perceived as homely—and it is not incorrect to speak one’s regional dialect. Until Victorian times, there were only regional dialects in Britain; there was no “standard English.” And in this century, according to McCrum, Cran, and MacNeil (1986, p. 28), the “standardness” of a certain accent has largely been perpetuated by the BBC, who consciously adopted it as “the standard” in the 1920s— even though at the time Queen’s English was only spoken by about 3 percent of the British population. Surely, we do not want to say that only a fraction of 1 percent of native speakers of English speak “the right way.” Different accents or dialects can equally well perform the job for which language exists; so all are on the same footing linguistically, even if not socially. Thus although (vii) is stated rather minimally, the gloss on it is that not only may a word have more than one “correct” pronunciation; owing to different accents or dialects it may have quite a few.

Enough has been said to suggest some of the so-called identity conditions that are relevant to words. We may now ask the following key question.

Is there anything all and only tokens of a particular word have in common other than being tokens of that word (i.e., any linguistically nontrivial projectible property)?

If the answer is “no,” then as an elimination rule, (1) is hopeless.

One popular answer to the above question is spelling. That was the jist of Goodman’s claim that any ‘ “Paris” consists of five letters’ is short for any ‘every “Paris”-inscription consists of five letter-inscriptions’. Unfortunately, it is not spelling, for four reasons. First, as we saw above, the word ‘color’ has tokens spelled twenty different ways; even today there are two correct ways to spell it. So being spelled the same way is not necessary for word identity. Second, words can be misspelled—for example, some tokens of ‘cat’ are actually spelled ‘k’-‘a’-‘t’ (although those fond of the spelling theory would probably deny this. Probably they would also hold that young children, who pronounce words strangely, are saying no words at all).⁵CloseThird, different words have the same spelling—for example, the noun ‘color’ and the verb ‘color’—or, for another example, three of the seven OED entries that are spelled ‘d’-‘o’-‘w’-‘n’: the adverb meaning ‘in a descending direction’, which is derived from Old English; the substantive meaning ‘the fine soft covering of fowls’, derived from German; and the substantive meaning ‘open expanse of elevated land’ (as in ‘on the broad downs, Stonehenge was visible’), derived from Celtic (OED, pp. 624–626). Same spelling, different words—so having the same spelling is not sufficient. Fourth, and most important, not all word tokens are inscriptions; some are utterances. (Not that Goodman’s suggestion runs afoul of this, since his is limited to inscriptions.) And utterances are not composed of letter tokens, and hence don’t have a spelling (except, perhaps, in an indirect sense through their types, if they happen to be tokens of a written language). This is evident if we think of a natural language before its writing is established; speakers of such languages utter words every day, but their utterances have no spelling, even in the indirect sense just mentioned.

Notice that even if, contrary to fact, all tokens of a word had the same spelling, we would have analyzed word types in terms of letter types, but we would still need an account of what the latter are (since they seem to be types, too), and of whether their tokens have anything in common (so as to apply elimination rule (1)).

Sensitive to this consideration, those who favor the spelling theory are likely to add that letter types are just shapes (or classes of similarly shaped objects) and hence that words are shapes too.⁶Close Of course, as an addendum to the spelling theory, this view is subject to all the objections that the latter faces. What, for example, is a shape? With good reason Quine (1961c, pp. 73–74) classifies shapes as abstract objects. But it is subject to an additional objection. The spelling theory by itself would at least classify Braille tokens of the word ‘cat’ together with signed tokens of it and Morse Code tokens, since they all have the same spelling; but the shape theory could not classify them together. Worse, the shape theory would not even classify all printed tokens of the letter ‘A’ together. Tokens of radically different fonts are not similar in shape (in anything like the Euclidean sense). Witness the differently shaped tokens of ‘A’, shown in figure 3.1, a short while in the library uncovered.

This illustration should kill enthusiasm for the shape theory, but presenting these ideas publicly has taught me that the shape theory dies hard. Let me take another stab at it. Suppose that someone still wants to insist that all he means by a letter is a very particular shape—as in ‘vee-shaped’, ‘ess-shaped’, ‘ell-shaped’. Then either: (i) he also wants to insist that all the tokens of ‘A’ just illustrated are similarly shaped, in which case he is employing some esoteric notion of ‘similarly shaped’ not based on Euclidean geometry—not even based on topology—and it is incumbent upon him to spell out what it is; or (ii) (employing the Euclidean notion) he denies that all of the ‘A’ tokens are similarly shaped, and points to one of them as his “exemplar.”⁷Close In that case he must deny that the nonsimilarly shaped tokens on the page are ‘A’s. The problem with this is that they are ‘A’s. The shape theory of letters does not accord with how we categorize. The letter ‘A’ has a long and distinguished history; historians say ‘A’ comes in many “forms”—a fact that is at odds with the shape theory. According to Compton’s Encyclopedia:

The letter A probably started as a picture sign of an oxhead, as in Egyptian hieroglyphic writing (1) and in a very early Semitic writing used in about 1500 BC on the Sinai Peninsula (2). In about 1000 bc, in Byblos and other Phoenician and Canaanite centers, the sign was given a linear form (3), the source of all later forms. In the Semitic languages this sign was called aleph, meaning “ox.” The Greeks had no use for the aleph sound, the glottal stop, so they used the sign for the vowel a. They also changed its name to alpha. They used several forms of the sign, including the ancestor of the English capital A (4). The Romans took this sign over into Latin, and it is the source of the English form. The English small a first took shape in Greek handwriting in a form (5) similar to the present English capital letter. In about the 4th century AD this was given a circular shape with a projection (6). This shape was the parent of both the English handwritten character (7) and the printed small a (8). [See figure 3.2]

The shape theory would also incorrectly classify the first letter token of Samuel Adams’ signature (‘S’) on the Declaration of Independence as a token of the same type as the first letter token of James Wilson’s signature (a ‘J’) rather than of Samuel Chase’s, since it is much more similar in shape to Wilson’s than to Chase’s. It would incorrectly classify the shape in figure 3.3 as an eight, since it is more nearly similar in shape to that in figure 3.4 than to a six. (The example is from Ludlow 1982, p. 420.) Charitably construed, (ii) is revisionism, pure and simple (“let’s revise our linguistic habits to accord with the shape theory”). Uncharitably construed, it is Humpty Dumptyism.

Someone partial to shape theory might at this point concede that tokens of the letter ‘A’ do not have the same simple Euclidean shape, but urge that nevertheless there is a disjunctive “shape” they all are: being shaped like the first ‘A’ in the illustration, or the second, or … or the last—and that that is all there is to the letter ‘A’. Of course, we would need quite a large disjunction just to specify all the fonts there are (and we’d have to get exemplars for Morse code, Braille, and sign language). But the real problem is that no disjunction we specify would do, for two reasons. First, the disjunction theory would still classify some tokens incorrectly—for example, that of the first letter token of Samuel Adams’s signature as a token of ‘J’.⁸Close Second, Madison Avenue may invent a new font tomorrow. If we revise the analysis to include any form that may ever be considered “the letter A,” we rule out almost nothing. This is just to give up on the shape theory entirely.

However, the main problem with any spelling theory is that it ignores the spoken word. Linguists put a priority on the spoken word. On, then, to the next theory.

Suppose, then, that we identify a word with a sequence of audible sounds. True, doing so has the disadvantage of ignoring the written word, but perhaps that can be justified on the grounds that the spoken word came first; that written words came only relatively recently and only for some languages; and, most important, that the written word (for most natural languages today) merely represents an attempt to symbolize the sound pattern of the spoken word. The English orthographic system is just a rather crude phonetic one that was hit upon four or five hundred years ago, when ‘one’ was obviously not pronounced as ‘won’.

If a word is a sequence of audible sounds, then perhaps all tokens of the same word sound the same. But of course they don’t. Tokens of the same word sung for thirty seconds at full volume by an operatic bass or whispered quickly by a young child will sound utterly different along almost every acoustic dimension. Still, it might be said that there is one dimension they share: the phonological dimension. That is, both speakers will be uttering the same phonemes. So let us consider the phonological hypothesis:

Every token of a word is composed of tokens of the same phonemes.

It faces problems similar to those of the spelling theory (although matters are more complicated here). For one thing, a phoneme, like a letter, is itself an abstract object, a type with tokens, and so we’d also need an account of what a phoneme is, and what its tokens have in common (if anything). This task promises to be at least as hard as identifying letters. As we saw in chapter 1, phonology (the study of phonemes) is distinct from phonetics (the scientific study of speech production). Phonetics is concerned with the physical properties of sounds produced and is not language relative. Phonemes, on the other hand, are language relative: two phonetically distinct speech tokens may be classified as tokens of the same phoneme relative to one language, and as tokens of different phonemes relative to another language. Phonemes are theoretical entities, and abstract ones at that. They are said (e.g., in Halle and Clement 1983, p. 8) to be sets of features; the English phoneme [p], for example, is {−sonorant, +labial, −voiced, −continuant}. The appeal to both sets and features is not likely to be pleasing to a nominalist.

Another difficulty for the phonological hypothesis is that sameness of phonemes is not sufficient for word identity, as shown by homonyms like ‘red’ and ‘read’, or the earlier example of ‘down’. (It’s not even sufficient for sentence identity. My favorite example is [Ah ‘key ess oon a ‘may sah] which means ‘Here is a table’ in Spanish and ‘A cow eats without a knife’ in Yiddish.) This particular difficulty might be avoided if we modify the phonological hypothesis by requiring in addition that the sequence of phonemes have the same sense. But this is too strong; we saw earlier that the noun ‘color’ has eighteen senses. Besides, this move will not help us with the third difficulty, namely, that sameness of phonemes is not necessary for word identity. We noted earlier that owing to accents/dialects, not even every correct pronunciation of a word will be phonologically identical to every other. Recall [′ske-(,)jü(ə)l] and [′she-jəl]. A Cockney ‘know’ is like RP ‘now’; RP ‘know’ is like Scottish ‘now’; and a Yorkshire ‘know’ is like RP ‘gnaw’ (Fudge 1990, p. 39). Yet we understand one another. Even within a single person’s speech, the same word will receive various pronunciations. For example, the word ‘extraordinary’ is variously pronounced with six, five, four, three, or even two syllables by speakers of British English: it ranges “for most British English speakers from the hyper-careful [′ekstrə′ʔɔ:dɪnərɪ] through the fairly careful [ɪk′strɔ:dņrɪ] to the very colloquial [′strɔ:nrɪ]” (Fudge 1990, p. 40).

This last example demonstrates what we saw in chapter 1: that there may be no phonetic signal in a token for every phoneme that is supposed to compose the word: it is “missing” several syllables. This is also demonstrated by reflection on ordinary speech: [jeet?] for ‘did you eat?’ and [sem] for ‘seven’. There is a humorous handbook on Australian pronunciation entitled “Let Stalk Strine.” No wonder, then, that many phoneticians have given up on the attempt to reduce phonological types to acoustic/articulatory types. (See Bromberger and Halle 1986.) Even the physicalist Björn Lindblom concedes (in Lindblom 1986, p. 495) that “for a given language there seems to be no unique set of acoustic properties that will always be present in the production of a given unit (feature, phoneme, syllable) and that will reliably be found in all conceivable contexts.” Not only is this true for a given language; the example of ‘extraordinary’ illustrates that it is true for a given idiolect. Sameness of phonemes is neither necessary nor sufficient for word identity.

One might, at this point, want to back up. The diversity that tokens of the same word manifest might suggest that the concept of a word we started with was too abstract. Perhaps there is a hierarchy of types of words, and we started “too high” on it. That is, the thought goes, perhaps we should renounce for the time being the question of what lexicographic words have in common, and instead focus on what “lower-level” words on the hierarchy have in common, and then later construct what it is that lexicographic words have in common. In other words, we might first gather together those tokens that are phonetically (and perhaps semantically) identical on the grounds that this is a perfectly good notion of a word. So, for example, [′ske-(,)jü(ə)l], [′ske-jəl], [′she-jəl], and [′she-(,)dyü(ə)l] would qualify as four different “words,” rather than four pronunciations of the same word. A Cockney ‘know’ would be a different “word” from an RP or Yorkshire ‘know’, [sem] would not count as the same word as ‘seven’, and [jeet?] would not count as the same sentence as ‘did you eat?’.

This will not work. It would classify different words as the same, for example, Cockney ‘know’ with RP ‘now’; RP ‘know’ with Scottish ‘now’; and a Yorkshire ‘know’ with RP ‘gnaw’. Moveover, it has the undesirable consequence that different dialects of the same language would have far fewer “words” in common—if they had any—than one would have supposed, and similarly for different idiolects within the same dialect. Worse, even the very same idiolect would distinguish as different “words” (what one would have thought was) the same word. But worst of all, it would distinguish as different words what are just different representations for the same idiolectal word spoken by the same person. For example, the five pronunciations of ‘extraordinary’ would come out as different words. Not only would a phonologist take this as excessively complicated (see Fudge 1990, p. 43), but the representation types themselves can receive realizations that are acoustically very different (for the small child and the man may speak the same idiolect). According to the phonologist Eric Fudge (1990, p. 31), “it is very rare for two repetitions of an utterance to be exactly identical, even when spoken by the same person.” We would be driven inexorably toward viewing each word token as a different “word.” This is completely unacceptable for a linguistic theory.

However, the last nail in the coffin for the suggestion according to which all tokens of the same word have the “same sound” is that words can be mispronounced—no doubt in many ways. Kaplan (1990, p. 105) made a case for the claim that “differences in sound” between tokens of the same word can be “just about as great as we would like.” He supports this by means of the following thought experiment. There is an experimenter and a subject. The experimenter says a word; the subject is supposed to wait five seconds and then repeat the word. ‘Alonzo’; ‘Alonzo’. ‘Rudolf’; ‘Rudolf’. The subject performs well; he is highly motivated, sincere, reflective, not reticent, and so on. Then the experimenter starts tampering with the subject’s speech mechanism—and not just by making him inhale a little helium. The experimenter puts weird filters of all kinds into the poor subject. Kaplan claims that we would say “yes, he is repeating that word; he is saying it in the best way that he can,” however dissimilar the imitation (p. 104).

Kaplan’s extremely clever thought experiment draws our attention to an often ignored fact: that intention is very important to the identity of a word token. As a vector toward determining the identity of a word token, it is much weightier than is usually appreciated (think back to the rigid thinking of the spelling enthusiasts). Of course, it is doubtful that something that sounds like ‘supercalifragalisticexpialidocious’ can be a token of the word ‘Alonzo’, but certainly, as any parent can tell you, words can receive pretty strange pronunciations and still retain their identity. Not that Kaplan is committed to this, but let us consider, then, whether intention is the thing every token of a word has in common. The intention hypothesis is that:

Every token of a word is caused by an intention to produce it.

First, is it necessary for a particular utterance to be a token of ‘cat’ that the speaker intended to utter ‘cat’? No, because a speaker might intend to remain silent, but against her will utter ‘cat’ because a neurosurgeon is stimulating a nerve in her brain during an operation. Second, is it sufficient? No, because a speaker might have the intention of uttering ‘cat’ but die before she gets the word out. Suppose a speaker intends to utter ‘cat’ and succeeds in making a noise. Is this sufficient for her to have uttered ‘cat’? No, because she might have aphasia and utter ‘cow’ instead of ‘cat’.

What if a person is in control of her faculties (including those of speech production), is awake, and is in “normal” circumstances? Is the intention to utter ‘cat’ sufficient for uttering ‘cat’? No; she might indeed utter something, and yet might fail to utter ‘cat’. She might simply utter ‘cow’ by mistake. Now we could drag in talk of “unconscious intentions,” but then the account will be only as plausible as the thesis that every slip of the tongue is a Freudian slip. (I’m reminded here of the joke about the two psychoanalysts who are discussing their Easter visits home. One says: “I made a terrible Freudian slip at dinner; I intended to say to my mother ‘Please pass the hot-cross buns,’ but instead I said ‘You’ve ruined my life, you witch.’”)

I don’t wish to nitpick about necessary and sufficient conditions. It is probably true that in most cases in which ‘cat’ is said by someone who is awake, in possession of her faculties, and in normal circumstances, and so on, she intended to utter ‘cat’, and in most cases in which such a person intends to utter ‘cat’, she utters ‘cat’. There is a high degree of correlation. Important and interesting as this might be, it doesn’t help us with our current project. There is also a high degree of correlation between surfboards and intentions to produce surfboards. But it would be putting the cart before the horse to analyze surfboards, or words, in terms of intentions-to-produce-them. Any account of what an intention-to-utter-‘cat’ is will probably presuppose some account of what the word ‘cat’ is. (This is not to say that intentions should not be considered when doubt arises as to the identity of an utterance or a bit of styrofoam, but this fact bears more on the question of what makes us think that something is a token of a certain word than on what the word is.)

Similarly, it may be generally true that t is a linguistic token of type T if and only if members of the relevant linguistic community would agree that it is.⁹Close (The “relevant linguistic community” for tokens of English could not include everyone who speaks English, since it is often hard to understand dialects too dissimilar from one’s own.) The problem with this otherwise excellent suggestion is that it, too, puts the cart before the horse. It may be generally true, but it does not offer a linguistically interesting property. To see this, consider that it may be generally true that something is a surfboard if and only if members of the relevant community (presumably surfers) would agree that it is. Yet this tells us nothing about the nature of surfboards, which may well have something functional in common. It may be generally true that something is a musical token of a Mozart sonata if and only if members of the relevant musical community would agree that it is. But again, this is not a musically interesting property. Mentions of both Mozart and sonata form are essential.

The two previous suggestions are in line with Ned Block’s suggestion¹⁰Close that in view of the multiple realizability of words as tokens, perhaps a functional definition is in order. The most promising to my mind are the two just considered. But both seem to presuppose that we know what words are already, before we can identify “intentions to produce them” or “what the community accepts.” This is not like defining a mousetrap as “anything that can trap a mouse.” It is like defining ‘the’ as “anything the community accepts it as a ‘the’.” We haven’t gotten anywhere.

It might be thought that Sellars (1963) solved this problem by appealing to the notion of a linguistic role, which Loux (1998, p. 79) defines two word tokens as having when they “function in the same way as responses to perceptual input; they enter into the same inference patterns; and they play the same role in guiding behavior.” It is dubious whether this notion can be unpacked without referring to abstract objects (same inference patterns?), but in any event it cannot be used to pick out all tokens of a word, as we have been using the word ‘word’. The reason is that ‘red’ and French ‘rouge’ are different words in our sense, but their tokens play the same linguistic role for Sellars.

Thus my answer to the question posed earlier, “Is there anything all and only tokens of a particular word have in common other than being tokens of that word (i.e., any linguistically nontrivial, ‘natural,’ projectible property)?” is no.¹¹Close (I say “in general” because there may be exceptions. Maybe all tokens of ‘eleemosynary’ are traceable back to something Shakespeare wrote.) But this is not very different from grizzly bears. Not all adult grizzlies are big, not all are brown, not all have humps, and so forth. Almost any generalization about all grizzlies will be false if there is one midget albino grizzly who happens to terrify easily. Yet it is still true that the grizzly is a big, humped, brown bear native to North America. It is also true that the word ‘cat’ is correctly spelled ‘c’-‘a’-‘t’ nowadays, and correctly pronounced [′kœt].

Two disclaimers should be mentioned. The first is that I am not saying that it is just a brute fact that word token t is a token of word T. Plenty of tokens of ‘cat’ are spelled ‘c’-‘a’-‘t’ or pronounced [′kœt], just as plenty of grizzlies are big and humped. Spelling and pronunciation are factors that help determine, for each word token t, what word type T it is a token of and why. Other factors include: the linguistic context (phrase, sentence, paragraph, … the linguistic community) in which t occurs, and, as Kaplan (1990) rightly emphasizes, the intentions of the producer of t and perhaps of the producer’s audience, if there is one. The second, related, disclaimer is that nothing that has been said is meant to rule out the possibility that what makes a token a token of type T supervenes on purely physical properties and relations.¹²Close

Since there is no linguistically nontrivial, “natural,” projectible property that all tokens of a word have in common—other than being tokens of that word—the paraphrase provided by (1) of ‘The T is P’ as ‘Every t is P’ is woefully inadequate. No property that one can correctly predicate of a word is likely to be had by all the tokens. In particular, to use Goodman’s example, ‘Paris’ consists of five letters, but it is not the case that every ‘Paris’-inscription consists of five letter-inscriptions. For example, ‘Parrys’ and ‘Pareiss’, listed under ‘Paris’ in the OED, do not. (More discussion of Goodman’s example is to be found in chapter 5.) Similarly, the CDK4 protein inhibits the p16 protein—but not always. And as is clear from the preceding discussion, it won’t work for most claims about a species, since its members vary so much among themselves.

The upshot is that the most popular way to paraphrase type talk does not work. The next chapter will consider more sophisticated suggestions, but in the end these too will be seen not to work.