Competence and Performance Multiplied
In psycholinguistics it is customary to distinguish competence from performance.  Competence is an internal cognitive structure (“knowledge of grammar”) while performance is the outward expression of this knowledge in actual speech. Performance is conceived as the externalization of competence. But this dualism is too simple: we need to recognize several competence-performance distinctions—at least three that I can think of. This is because there are several kinds of linguistic competence: competence in the language of thought, competence in the language of inner speech, and competence in the language of outer speech (call these T-language, I-language, and O-language, respectively). T-language is the innate language that is coded in the genes in all normal humans (what Chomsky calls Universal Grammar and Fodor calls the Language of Thought): it is the formal computational basis for thought. Here we need to distinguish the internal representation of the language (competence) from its deployment in actual thinking (performance)—the cognitive structure from its expression in thought. We need not suppose that this structure takes the form of knowledge—we are not obliged to maintain that each thinker knowshis or her T-language (its constituents and rules of combination). All we have to accept is that the brain-mind codes the T-language in some way, presumably unconsciously. What is important is that the process or activity of thinking is not identical to this internal structure: it is one manifestation of the structure and is susceptible to various kinds of breakdown that don’t undermine the integrity of the underlying competence (fatigue, drugs, disease, etc). Note that here the performance aspect is purely internal, not a type of external behavior (though we can think of it as inner behavior). A being could in principle exhibit a competence-performance distinction with respect to its T-language and not behave at all (externally).
Now consider inner speech: this is not the same as thought, though there is doubtless overlap. It is the passing of words through consciousness, and is best understood as a hybrid of one’s native language and one’s T-language. For simplicity, suppose that English speakers use English as their I-language: then again we have a distinction between competence and performance in this language. On the one hand, there is the mastery of the language in question conceived as a cognitive structure (even a straightforward case of knowledge); on the other hand, there is the deployment of this structure in acts of inner speech. Again, this performance is subject to factors that leave the basic competence untouched: the knowledge can remain the same though the actual course of inner speech can vary according to circumstances. One person might have a lot of inner speech and another person relatively little, even though they both are equally competent in the underlying language. It is to be assumed that this competence embeds competence in the T-language: our competence in the language of thought is part of our competence in the language of inner speech. But we can’t identify the two kinds of competence or performance: they involve different layers of psycholinguistic reality. Note too that this second kind of performance is also purely an inner reality: episodes of inner speech are precisely that, not overt behavioral events. It might also be more basic than outer speech (like T-language), preceding it in evolutionary history.
Finally, we have outer speech: the external expression of the previous two levels of linguistic reality. A speaker has mastery of his or her native language and also has the ability to externalize that mastery in acts of communication. These are separate things that can in principle be dissociated: you could have competence without performance (paralysis) and performance without competence (a well-trained parrot). Various factors can influence performance and leave competence unaffected. Presumably the previous two levels are embedded in this third level: we use our competence in T-language and I-language in our acts of external speech. The performance aspect includes the motor system, so it is different from the previous two levels of performance, which don’t engage any muscles. Each competence-performance division coexists with the others in normal humans, but they are distinct psychological realities: thinking, speaking inwardly, and speaking outwardly. All are exercises of language, but they involve different types of competence and performance, subtly interrelated.
I stated that the third level involves outer speech, making noises with the mouth in most cases, but actually that is strictly wrong. Here again the performance is essentially mental in nature: this is because a speaker could in principle engage in linguistic articulation and not be engaging in overt behavior. The brain in a vat proves this most dramatically: the motor centers of your brain issue orders to your mouth but your mouth isn’t there. You are performing acts of would-be external speech but nothing is coming out—still you are engaging in linguistic performance. Competence in your native language is connecting with motor instructions to body parts that happen not to exist: it would seem to you just as if you were speaking (we can’t refute skepticism by claiming certainty that we are speaking out loud). There is still a competence-performance distinction, but it does not involve any audible speech. Strictly speaking, bodily performance is not essential to “outer” speech. We can’t be behaviorists about any kind of performance, not to mention competence, since thought, inner speech, and “outer” speech are not bodily in nature. True, a brain in a vat is not performing acts of audible speech, but psychologically such a brain is just like an ordinary embodied speaker. Strictly, then, all performance, like all competence, takes place within the subject, behavior being a dispensable effect.
There are three uses of language—in thinking, inner speech, and outer speech—and each admits of a competence-performance distinction.  The situation is therefore more complex than the simple competence-performance distinction has made it seem. There are several types of linguistic competence and several types of linguistic performance, not a single dualism.
 This distinction goes back to Chomsky’s earliest work on language and is central to his rejection of behaviorism. It is analogous to the distinction between the categorical basis of a disposition and its manifestations—say, molecular structure and chemical reactions. In psychology it is a special case of the distinction between knowledge (or mental representation in general) and acting on knowledge—cognition and volition basically. The distinction is hard to contest.
 I leave aside the question of whether there might be other uses of language, say in memory, imagination, or dreaming. Is it possible that dreaming involves linguistic performances different in kind from the three I have identified? Dream talking is certainly not much like conscious inner speech. Wittgenstein spoke of there being different language-games; maybe it would be better to speak of different language-worlds—the world of thought (conscious and unconscious), the world of silent inner monologue, the world of outer perceptible utterance, and the world of somnolent dream chat. (The fantastic world of dreams is apparently more tolerant of verbal nonsense—think of someone remarking in a dream, “Colorless green ideas sleep furiously”). The underlying formal structure of language lends itself to a variety of possible modes of performance—not just within external communicative speech but also across other types of linguistic reality. Sounds (and signs) are not the essence of language by any means.