The Unity of the Word
What is a word? It seems safe to say that a word is something that combines with other words to produces phrases and sentences. We can also describe words as discrete units of meaning, finite in number: they are digital not analogue, never shading into each other or blending together. They are the atoms of discourse. But what are they? We can say what they do and we can characterize their general structure, but we have not yet said what they consist in—what constitutes their essential being. What are words intrinsically? We have a kind of functional-structural description of them, but not a description of their inner nature (compare the physicist’s description of elementary particles).
A first thought might be that words are acoustic signals—the kind of thing that can be recorded with a speech spectrograph. This is certainly the kind of thing we are looking for—it attempts to say what a word is—but it is clearly inadequate. First, the same word can be uttered in many different ways acoustically. Second, the same acoustic signal could be produced by a non-speaker and it wouldn’t count as a word. Third, we must account for words as they are produced inwardly and these are not acoustic signals. Fourth, we are seeking the psychological nature of words not the physical nature of utterances of words—the sound that is made when a word is externalized vocally. Fifth, words can occur in sign language when no sound is made. The sounds (or signs) of speech are not the same thing as words themselves—a word is not what it is in virtue of the sounds speakers make when they utter it. A word is something that underlies external utterance: it is part of linguistic competence (not performance). Our competence consists of a set of stored lexical items (inter alia) and these must be realized somewhere in our mind or brain (usually called “memory”, though this can be misleading). 
A second suggestion brings in reference: words differ if and only if their references differ. This also is clearly wrong: not all words refer to begin with, but even when they do they are not individuated by their reference, since many words can have the same reference (and vice versa). Words are individuated even more finely that senses, so reference will not do the job. Besides, reference is not intrinsic to words as such, so it can hardly be what distinguishes one word from another. Nor can word difference be explicated in terms of associated mental images, any more than meanings can (even less so). Words also have no distinctive phenomenology, so this is not a viable way to individuate them. Is a word some sort of disposition? What kind of disposition? We do have certain dispositions in virtue of mastery of a word, but it is not plausible that a word is a disposition—say, a disposition to utter it. We might not have such a disposition in conceivable cases, and anyway the word is the ground of any such disposition, not the disposition itself. Is the word perhaps a neural state of some sort? But that is far too reductionist, implying that words require specific neural structures (“W-fibers firing”). Again, there is no doubt a neural basis for word storage, but words are not identical to such a basis–words are not synaptic clefts. Words are entities in their own right.
We are having trouble finding a fact for words to be: they are psychological entities of some sort, but we can’t locate a suitable entity for them to be. At this point we might turn skeptical: we might say that there is no fact of word possession, so there are no words. Just as the rule-following skeptic says there are no linguistic rules, so the lexical skeptic says there are no words.  There are sounds, but there is nothing word-like underlying those sounds: there are no such segmented psychological units. Thus an eliminative position will be recommended: the ontology of words is just misguided folk linguistics, not sustainable under analytic scrutiny. Just as some have denied the existence of grammatical rules, so the existence of the lexicon will be denied: there is nothing psychologically real corresponding to the intuitive notion of word—just the flux of sound. Talk of discrete words is nothing but a useful instrument for summing up something much more amorphous—instrumentalism about linguistics. These are all familiar options, which is why I have passed so quickly over them. And a final option is also familiar: we simply don’t know what a word is. Maybe we can’t know, but in any case we don’t know. Words have an unknown inner nature. They have various known extrinsic properties—such as sounds, dispositions, a neural basis—but what they are intrinsically remains mysterious. It is sometimes remarked (e.g. by Chomsky) that we have no idea about the evolutionary origin of the lexicon, though it is clearly innately based; now we see that we have no idea about the constitutive nature of the lexicon either. Indeed, it is hard to see how we could make progress with the question of origin without a better understanding of that which evolved—i.e. without understanding the nature of words. Words are mysterious—welcome to the club.
It might be wondered whether the problem of words is just the problem of concepts in another guise—we don’t have much insight into their nature or evolutionary origins either. But the problem of words is a separate (though related) problem, simply because you can have several words corresponding to the same concept. Words are the vehicle of concepts not concepts themselves. What is surprising is that even words present serious problems of understanding—weren’t they supposed to be the solid ground in our understanding of language? We utter and hear words all the time, yet we have no idea what they are. We know their functional-structural properties well enough, but we can’t say what they consist in—they could be states of a Cartesian substance for all we know. Concepts are elusive, admittedly, but words promised to make themselves transparent. It turns out, however, that the level of syntax is as problematic as the level of semantics. Once we appreciate that words are not to be identified with the sounds made when uttering them, we can quickly see that nothing else suggests itself. We can’t say what words themselves are. We are reduced to thinking of them as primitive constituents of the mind, beyond introspection. 
What is the relationship between words and concepts? Here we must guard against a misconception arising from spoken language: acoustic signals can be said to express concepts, so we may be tempted to model the word-concept relation on the sound-concept relation. That relation is entirely arbitrary and conventional, eminently breakable: but words are not sounds, so they don’t express concepts as sounds do. Words are more abstract and psychological than physical sounds, so they may well be more intimately related to concepts than sounds are. Maybe the relationship is not arbitrary at all, but natural, integral. Words in the language of thought are not external to concepts or imposed on them; they are designed for concepts—expressing concepts is their purpose. Real internal words have concept-specific architecture, not a merely conventional association. We draw a blank when we try to say what words are, so we can’t describe this architecture—but surely the relationship to concepts is anything but conventional or arbitrary or stipulated. We should not assume that it resembles the relationship between sounds and concepts. I see words as the mitochondria of the mind: isolable units performing essential biological work, operating under the radar, surrounded by grosser tissue (concepts, thoughts). Words are somehow encoded in the DNA, though they are of mysterious origin and nature; they provide the sturdy underpinnings of language, used both as a tool of communication and a tool of thought. We can’t say what they are exactly, though we know well enough what they do (though not how they do it). The basic lexicon is no doubt shared by all humans and inherited from a common ancestor. What are vulgarly called words, i.e. vocal or graphic products, are far removed from the underlying psychological reality of lexical competence: it is not sounds or marks that are mentally combined according to grammatical rules to produce sentences (which are not sounds or marks either), but psychological elements of unknown nature. We must not confuse the outward utterance of words with words themselves—the items stored in memory or some other compartment of the mind-brain. Words are constituents of psychological reality, capable of various uses and types of expression, both internal and external.  The unity of a word is the unity of this unknown entity.
 It can be misleading because if the basic lexicon is innate it is not stored in memory but in some other data storage facility. Memory stores what has been learned not what is inherited. What memory contains are the particular sounds of spoken languages as they externalize an innately given internal vocabulary.
 This has a bearing on the nature of quotation. A quotation name of a word is certainly not a name of the physical material appearing literally inside the quotation marks, but a name of the word itself, which is far removed from that physical material. Depending upon what we think a word is, the quotation name is a name of that thing—as it might be, a brain state or a disposition. One can imagine a form of epistemological skepticism arising from this: how do we know what word is named in a quotation name? How do we know, for instance, whether the quotation name ‘“addition”’ refers to the word “addition” or to the word “quaddition”? We think we refer to particular words by our quotation names, but how can we be sure we are referring to the words to which we think we are referring? Maybe we are referring to quite different words. This is a radical form of skepticism because it questions our ability to know what language we are speaking, i.e. our lexicon: we don’t know what symbols our language contains! Intuitively, words are more remote from perception than we are apt to suppose. (Note that this kind of epistemological skepticism is different from the constitutive skepticism mentioned in the text.)
 Do words really “go through the mind” when we “think in words”? Or is it that images of the sounds of words go through the mind? How accessible are words to consciousness? Is the lexicon unconscious?
 There still exists a stubborn behaviorism or materialism about language that insists that words must be some kind of corporeal entity—an aspect of stimulus and response. We must get beyond such peconceptions: words are as internal as the rest of the mind.