by David J. Lobina
(This is Part 2 of a brand new series of post, this time about the relationship between language and thought; Part 1 is here)
A provocative title, perhaps, and perhaps also counterintuitive. One thinks in the language one speaks, everybody knows that. Why would anyone ask bilingual speakers which language they think in (or dream in) otherwise?
I suspect that what people usually have in mind when they ask such questions is related to the phenomenon of inner speech, the experience of internally speaking to ourselves, which may well be ubiquitous in adults (but probably not in children), though not entirely universal. I certainly think that inner speech plays a role in thinking, but not as central a role as most people seem to think (I will come back to this on a later post, probably in Part 4 of this series, where I will also discuss how writers of fiction use the narrative technique of “interior monologue” to outline some of the mental processes of a given character (thinking, feeling, etc.) – but mostly to argue that authors generally go about it the wrong way!).
The point I want to make in this post is that no-one thinks in any natural language; not in English, or Italian, or whatever, but in a language of thought, an abstract, unconscious and moreover inaccessible, conceptual representational system of the mind. Or at least I intend to provide some of the evidence, anecdotal and otherwise, that suggests that this is indeed the state of affairs.
The idea of a language of thought is in fact a rather old one. It effectively refers to the old doctrine that we think in a mental language that is not a spoken language. Traceable back to Aristotle, Boethius and William of Ockham (among others), the doctrine is to a large extent premised on the general observation that speakers of different languages can refer to the very same “things”, though they may employ different words to talk about them. As the French philosopher Claude Panaccio has aptly put it in a recent historical overview of the mental language, the French can talk about un homme whereas the English would say a man and the ancient Romans homo, but they all would have had the same “idea” in mind – the same concept, as cognitive scientists call such things, and as I myself mentioned last time around. Crucially, the same logic applies to the sentences in which the mentioned words can appear: homo currit, un homme court and a man is running simply describe the same event – the same thought – in different languages.
This, at the very least, suggests a general intertranslatability among different languages, what the philosopher Jerry Katz once called the “effability principle” – namely, the intertranslatibility of whatever thought one might be able entertain in one language into another language (in rough outline, of course, not in precise, linguistic detail, and certainly not in terms of a one-to-one correspondence between words or phrases).
But how would the postulation of a language of thought explain why this is the case, and how? The idea is that behind the words of a language lie concepts and behind the sentences of a language lie combinations of such concepts. To have a belief or a thought is to have a particular combination of concepts in mind. To believe that a man is running, then, is to have the relevant mental concepts, e.g., MAN and RUNNING (concepts are usually written in capital letters in cognitive science), and to have the capacity to put them together (i.e., MAN RUNNING). In this sense, the language of thought is the common code in which concepts are couched, thus explaining how speakers of different languages can at all entertain the same sort of thoughts. We all think in roughly the same mental language, a system composed of concepts that allows us to represent and make sense of the world.
Somewhat relatedly, the philosopher John Searle used to talk about the “principle of expressibility”, according to which whatever can be meant or thought can be said in a language, though the correspondence between language and thought would never been one-to-one. This brings an interesting contrast between these two domains.
Consider the sort of words that are allowed in natural language, as there seem to be some constraints on the type of concepts that can be lexicalised in language —i.e., which concepts can be turned into individual words. A language certainly confers a great deal of flexibility when it comes to inventing new words, especially in the case of nouns, but some verb forms appear to not be possible. The linguist Guglielmo Cinque offers the following sample of non-existent verbs in English, shown in italics below, and with the intended meaning within parentheses:
- He has climbend the tree (‘he has worryingly (for the speaker) climbed the tree’).
- He fightaf/runaf (‘he is afraid of fighting/running’).
- He didish it (‘he did it shamelessly’).
- I sayam you are wrong (‘I am sympathetic in saying you are wrong’).
There has been some discussion in the literature as to why this is the case, the proposed reasons ranging from the metaphysical to constraints internal to language, but the consensus seems to be that such verb forms are in fact impossible – they can’t exist in English. The meanings these verbs convey are not hard to grasp, though, for the corresponding concepts are perfectly entertainable (they can be mentally represented). It is just that these concepts cannot be lexicalised.
These curious data point to the sort of primitives thought allows for, that is, the sort of things we can entertain and think about without the need to employ a linguistic vehicle for such a purpose (at least in terms of individuals words, as the relevant thoughts can be expressed with paraphrases and the like). As Cinque puts it,
of all the concepts and distinctions that populate our system of thought only a fragment receives a grammatical encoding in the languages of the world
Or in other words, there seems to be more in the realm of things we can think and entertain than in the realm of things we can talk about.
Interestingly, the issue can be empirically investigated. A psycholinguistic study a while back reported the results of an experiment on the learnability of unlexicalised words – words that do not exist but which could have, if only circumstances had allowed it. One such concept involves the meaning that would underlie the unattested word fost, the opposite of the existing determiner “most” in English, thereby standing for “less than half” – if only it existed!
As a matter of fact, fost could exist, as it exhibits the same semantic properties as “most” and there are no internal or metaphysical reasons barring its lexicalisation (it seems that this determiner does not exist for pragmatic reasons, but we need not be detained by this here). Since fost could in fact exist, it ought to be possible to design an experiment in which participants are required to learn how to use the non-existent word, drawing an interesting distinction between lexicalisation and learnability (such a strategy is in fact a staple in studies of language acquisition). According to the results, both young children and adults can indeed learn and appropriately use the meaning of fost when put to the test. Viewed this way, the experiments provide a way to evaluate whether certain concepts are entertainable or not, and it is reasonable to suppose that the meanings of unattested words would only learnable if the corresponding concepts are in fact available in a language of thought.
In addition, this particular difference between natural languages and the conceptual representations of the language of thought seems to be borne out in a record of brain activity as well. There is some psycholinguistic and neurolinguistic evidence regarding the combinatory rules underlying language and thought – linguistic rules in one case, conceptual in the other. In a number of MEG studies meant to map out some of the areas of the brain involved in syntactic, semantic, and conceptual rules, it was found that the left anterior temporal lobe (LALT) is sensitive to combinatorial effects that are conceptual in nature rather than syntactic or semantic. By employing simple noun-noun combinations varying along a specificity metric (tomato soup vs. vegetable soup, for example, where tomato is more specific than vegetable), some studies have been able to track the times, for an MEG record specifies both space and time, at which different combinatory effects arise in the brain. The results show that form-based, syntactic effects arise at 100-200 milliseconds (ms.), LALT effects at around 200-300 ms., and lexical-semantic effects at around 300-400 ms.
As advanced, the LALT is only activated when the combinations are conceptual in nature, and this was reflected in the data. LALT activity was largest when the first item in noun-noun combinations was specific rather than general, a distinction that is regarded as intrinsically conceptual as it does not elicit any effects in the syntactic or semantic time windows. This is certainly plausible: specificity would not affect syntactic or semantic rules, as syntax and semantics would operate the same way regardless of whether nouns in noun-noun combinations are specific or general.
Let us take stock. To think, then (or rather, to entertain a thought) requires the ability to entertain conceptual representations of various kinds, that is, to combine concepts with each other into thought representations. And this, I have claimed, necessitates a mental language of a special kind, the language of thought I introduced last month and which I have elaborated upon a little bit here.
But none of what I have said so far rules out any role for natural language in our thinking processes. After all, a languageless mind is an impoverished mind – there is no much sophisticated thought in a mind with no language, think of so-called feral children – and we do acquire loads of concepts through language (‘look at the picture of this animal, it’s an x’, and thus you have the concept X). And one way to approach this issue is to consider the sort of structure thought representations must have, and then to probe whether natural language can accommodate it, at any level. And to do that we must trust the philosophers!
Consider Gareth Evans’s Generality Constraint first, according to which thoughts must be structured, not in terms of their internal elements, something Evans in fact rejects, but in terms of “their being a complex of the exercise of several distinct conceptual abilities”. Evans is here drawing attention to the apparent fact that if one can entertain a thought in which a given property, call it F, can be ascribed to one individual, a, this is the result of two abilities: understanding F, understanding a, and applying the predicate F to the argument a. Consequently, if one also understands property G and individual b – that is, one understands the two sentences/thoughts, Fa and Gb – then there are “no conceptual barriers” to entertain the sentences and thoughts Fb and Ga.
Ultimately, this is an ability “to think of an object in a series of indefinitely many thoughts”. The Generality Constraint is related to, but is not quite the same as, what Jerry Fodor has come to call the systematicity of thought, the claim that our ability to entertain some thoughts is intrinsically connected to our ability to entertain similar thoughts. Accordingly, this property is a reflection of constituent structure (the decomposition of complex representations into primitive constituents), given that the stated similarity amongst thoughts is a matter of the form these thoughts have, their internal structure (this is precisely the aspect of systematicity Evans would not have accepted). Thus, if one can entertain the thought that, in the language of formal logic, P®[Q®R] (to be read as “if P, then if Q then R”, where P, Q and R can stand for thoughts), one should ipso facto be able to entertain the thought that [P®Q] ®R (that is, “if P then Q, then R”); an ability of entertain clustered thoughts, as it were.
From Evans’s structured cognitive abilities to Fodor’s structured propositions, but what does all this yield exactly? Nothing much, but it does provide the desiderata that a medium of thought must meet; namely, a language of thought must be able to:
- Appropriately represent the contents of thoughts.
- Accurately distinguish the contents of different thoughts.
- Faithfully represent the propositional attitudes.
- Play a causal role in mental processes.
Critically, these four requirements for a language of thought can provide the yardstick against which natural language can be measured as a potential medium of thought, at whatever level of human cognition. The fun part of this series, in other words, but this will have to wait until next month!