Language as a Tool for Thought
A typical tool must meet two requirements: it must be able to be used by its intended user, and it must perform its designated function. Often the tool must be firmly gripped and manipulated, so it must fit into the human hand, while its other end performs the task in question. If we think of spoken language as a tool for communication, we should expect the same duality: it must be both usable and effective. What this comes down to is that the words of the language must be pronounceable and audible. The speaker must be able to utter the words and the hearer must be able to hear them. These are nontrivial requirements, familiar as they are: words could be pronounceable yet inaudible (mere whispers) or they could be audible yet not pronounceable (not conform to the human vocal system). Spoken human languages, as we find them, combine these two characteristics, so they are both usable and effective. True, words can be hard to pronounce and fatigue can set in, and ambient sounds can drown utterances out; but generally words meet the two necessary conditions of a successful communicative tool. We can imagine situations in which they don’t, where nothing pronounceable is audible and vice versa (a noisy planet and inept voices). Then there would be no such thing as a spoken language–or a sign language, if visual conditions were not conducive (perpetual thick fog).
But what about language as a tool of thought—what design features would we expect to find built into language that performed this function? It would need to be mentally manipulated and it would have to serve the purposes of thought. Thus it would need to consist of storable, segmented, manageable words that combine to produce complex novel constructions. It’s no use if the complex constructions consist of parts that the mind can’t handle (the analogue of words that can’t be pronounced), or if the parts can be easily handled but can’t form complex constructions like sentences. But natural languages meet the two requirements for being a cognitive tool quite neatly: the human mind can store, process, and manipulate words; and words can combine into phrases and sentences that can express an infinitude of thoughts. The basic architecture of natural languages thus suits them to be tools of thought, since they can be both used by the thinker and also serve the purpose of thinking, viz. encoding complex novel thoughts. The lexicon plus the grammar can be mentally manipulated, and the products of this manipulation serve to encode thoughts, which can enter into reasoning, practical and theoretical, which can lead to action. Without this finely wrought tool thought would not be as effective as it is (it might not exist at all in anything like its present form), and luckily an effective tool exists that can also be handled by the thinking mind. If words could not exist in internalized form, then language could not perform the cognitive function it now performs—at best we would have to speak out loud whenever we entertained a thought. Words and syntax can exist both internally and externally, and because of this language can serve as both a tool of communication and a tool of thought.
I am stating truisms. I have simply set the basic property of language in the context of the notion of a tool—language as a finite stock of discrete digital elements that can be combined by syntactic rules into an infinite array of structured hierarchically organized complex expressions. I have said that this formal object satisfies the two conditions on being a tool of thought: it is both mentally accessible and functionally effective. If it were not finitely based, it might still serve the purposes of thought as a potentially infinite system, but it would not be accessible to finite minds such as ours–while if it could not express a potential infinity of thoughts, its mental accessibility would not help in performing the task at hand. Many animals have access to symbolic systems of some kind (bees, whales), but these systems do not permit the kind of recursive unbounded scope we find in human languages. And the gods might speak languages matching or exceeding ours in expressive power, but are beyond our capacity to master (perhaps because of memory limitations). The design features of a tool are suspended between the two requirements, reflecting those requirements, and are sometimes compromises between them. They are best understood as the joint result of limited human capacities and a desired objective—possibly being a trade-off between the two. There is no point in designing a perfect tool that can’t be used, and usable tools can be effective enough without being perfect.
With these preliminary points in mind, then, I wish to make my main positive proposal: human languages make effective tools of thought (partly) because they permit the formation of conceptual distinctions. Words enable us to formulate distinctions that would not be possible without them—not possible given our contingent psychological nature. They may not be ideal tools for this job—no tool is ideal for any job—but they perform it well enough, and it is hard to see what might perform it better. One central goal of thought is the making of fine distinctions (in addition to producing new thoughts) and language is inherently suited to facilitating this process. Lest I be accused of further truism, let me state the thesis more forthrightly: the distinctions that exist between words as such are used as a tool to construct distinctions of meaning, i.e. conceptual distinctions. The mind invokes words qua formal objects to aid it in the process of conceptual distinction making. This is the analogue of the thesis that the mind uses the combinatorial power of language to aid it in combining concepts into thoughts—it recruits syntax as a device of thought assembly. Similarly, the mind uses lexical distinctness as a tool for generating conceptual distinctness. Note that this is not the thesis that conceptual distinctness is lexical distinctness but rather the thesis that lexical distinctness is used as a tool for constructing conceptual distinctness. A hammer is a tool for knocking nails in not a nail that has been knocked in; a language is (inter alia) a tool for making conceptual distinctions not a conceptual distinction itself. It is not just that human language permits an infinite number of thoughts; it also permits an infinite number of finely individuated thoughts. A language is a thought-slicer, a concept-differentiator. It is designed to generate ever finer conceptual distinctions, and it does so by containing fine distinctions itself. 
There are many kinds of conceptual distinction; I shall mention three. First, we have distinct but co-extensive concepts, expressed by co-denoting names and co-extensive predicates. We have separate words for these concepts (“Hesperus” and “Phosphorus”, “creature with a heart”, “creature with a kidney”). There might possibly be a language that failed to provide for such conceptual distinctions—that was completely extensional in meaning: but no human language is like that. It would surely be impoverished relative to ours, if more to the liking of certain logicians. Second, we have near-synonymy—words that are close in meaning but not really synonyms: “house” and “home”, “friend” and “ally”, “stone” and “rock”. These are important because they illustrate the pervasiveness of distinction making: the concepts may be very similar but they are not exactly the same; there is a conceptual distinction to be made. Third, we have straight synonymy—identity of concept expressed: “bachelor” and “unmarried male”—that type of thing. Even here different words are suitable in different contexts, so there are distinctions to be made. Human language is rich in synonyms, as well as near-synonyms and co-extensive non-synonyms: that is, it is rich in conceptual (semantic) distinctions. It seems to make a point of registering such distinctions, as if it matters. It insists on marking subtle differences of meaning. This is not throwaway redundancy, but attentiveness to fine distinctions of human thought. I venture to suggest that it is a universal feature of all human languages (though not all possible languages): they all contain the machinery for recording fine-spun distinctions, and also the machinery for making new fine-spun distinctions. In thought we are constantly recognizing and creating conceptual distinctions: the thesis is that human languages aid in that process, crucially so. Words are the knives of thought: they chop things up finely.
Chomsky talks about the Merge operation that combines words into collections of words, construed as a function from sets to sets.  Clearly language requires such an operation, because it is combinatorial; and the mind must be able to compute Merge if it is to exploit the combinatorial power of language. Let me likewise introduce an operation I call Dissect that operates on concepts to generate more fine-grained concepts. Suppose we start with an undifferentiated concept people and then apply Dissect to this concept to derive the concepts men and women: we have gone from a concept treated as unitary to two concepts that distinguish among the members of the class consisting of people. If we applied Dissect to the set of people (not the concept people), it would partition that set into two subsets, viz. the set of men and the set of women—so it would be the opposite of Merge. If we apply it to concepts (or word-like elements in the language of thought), it generates two new concepts (or inner words)—thereby introducing a conceptual (semantic) distinction. I do not claim that I know how Dissect works (it seems rather mysterious), but evidently it does, since we make new distinctions all the time. We suddenly (or slowly) “see the difference between X and Y” and feel a sense of accomplishment or enlightenment. We implicitly grasped the distinction (whatever quite that means) and now we grasp it fully and consciously. Language enables us to make the leap: our language faculty feeds into our cognitive faculty and awakens us to distinctions that had been blurred or elusive before. Thus we employ Dissect in conjunction with Merge in our mental operations: we merge different elements into one, but we also dissect a single element into two (and those in turn may be dissected). Both operations enlarge our cognitive scope: by permitting the formation of new thoughts by means of combination, and by making finer distinctions that produce increasingly refined thoughts. Both are “generative” or “creative”, but along different dimensions of semantic space (inter-concept and intra-concept, respectively).
It might be wondered why thought dissects: what is the point of making concepts ever more acutely distinct? The answer is that finely individuated thoughts are useful thoughts: the distinctions are real and they affect the ability of thought to function effectively in the world. There is a selective advantage to having thoughts with this degree of differentiation: semantic differentiation is biologically adaptive. Thought is relatively crude without it. We may surmise that many animals have thoughts of some kind, but it is also likely that they don’t do well on the score of conceptual differentiation. Some animals may even have the ability to entertain a potential infinity of thoughts, but their thoughts are not chopped as finely as ours. Productivity and differentiation are distinct properties of a cognitive system, so they can in principle be dissociated. And the reason our thoughts differ from animal thoughts in this way is that we have a language rich in lexical distinctions and they don’t. It is estimated that human speakers have mastery of 30,000 to 50,000 lexical items, as well as the ability to combine these to produce indefinitely many distinct expressions for what is intuitively the same thing. Our language consists of clusters of distinct but semantically related words, the result of Dissect. Dissect evolved at some point and it opened up the possibility of creating new conceptual distinctions from old concepts: it took us from an initial relatively crude lexicon to a more sophisticated lexicon rich in semantic distinctions. Thus we became advanced thinkers compared to other animals, and we made adaptive use of this cognitive advance. Dissect is what made science possible, as well as religion, philosophy, linguistics, mathematics, politics, and so on. A creature stuck at the level of purely extensional concepts (or worse) would not have our advanced cognitive powers. Part of this is that more fine-grained concepts are needed for the purposes of explanation—an area in which other animals are conspicuously lacking. In explanation we need to distinguish between different properties of things, not just different things: for example, we need to distinguish weight and mass in physics, species and genus in biology, and belief and knowledge in psychology. One wonders whether there are any such fine distinctions to be found in bee or whale language (and thought).
The standard emphasis on the combinatorial power of language stresses the quantitative results of this power—the sheer number of sentences that can be generated. Thus its utility as a tool of thought rests on its ability to increase the quantity of thoughts a creature can entertain. But the emphasis on the differentiating power of language stresses the qualitative results of that power—the acuity and finesse of the thoughts it can facilitate. It can generate superior thoughts (a normative notion): as Descartes would say, clear and distinct ideas in contrast to unclear and blurred ideas. We obviously value our distinction making capacities, for reasons both practical and non-practical (e.g. philosophy). Not to be able to make sharp conceptual distinctions is a definite intellectual handicap. Somehow the thinking mind taps into the structure of language and uses that structure in acts of Dissectto produce ever more refined conceptual distinctions. It uses distinct words as concept markers, thereby eliciting concept division. That, at any rate, is the hypothesis—though we have virtually no understanding of how the process works, computationally or cerebrally. We know that language serves as a tool in the making of distinctions (or we have good reason to suspect that it does), but we don’t know how it does this—except that it appears to have something to do with the distinctness of words themselves (whatever they are exactly). The lexicon itself is quite mysterious, intrinsically and evolutionarily, and its mental operations inherit that mystery—thus Dissect is shrouded in obscurity. Concepts give rise to other concepts by a process of differentiation, but how precisely this comes about is unknown. Language acts like a tool in the process, but the underlying mechanics are not apparent. 
 Mental images were once viewed as the elements of thought, or at least vital tools of thought. That idea has come in for a lot of criticism and is no longer widely accepted. Although images have some combinatorial powers, it is unclear that they can match those of concepts, which are essentially unlimited. But it is also true that images don’t have the kind of fine-grained individuation that meanings have: several meanings can correspond to the same image (say, an image of a man fishing), and meanings can be abstract in a way that images cannot (as with the abstract concept of a triangle). Images don’t have the right properties to generate fine-grained thoughts, but words do. Language as a formal object consisting of a lexicon and a set of syntactic rules is exactly the right kind of structure to serve as a tool for facilitating thought, because it maps so neatly onto the structure of thought.
 I am not saying Dissect is a permanent mystery, only that our current understanding of it is quite limited. Mergeis also not free of mystery at the psychological and neurological levels, despite its clarity as an abstract set-theoretic operation.