Thought without language and the hypothesis of a language of thought

But it's still possible to argue that language supports thought, that it's an essential tool. What then is its role ? Andy Clark, following Frege, defends the idea that the expression of thought in linguistic form enables its explicit manipulation: " Linguistic formulation makes complex thoughts accessible to the processes of mental attention, enabling them to be subjected to a variety of other operations. This enables us, for example, to isolate different elements of a complex thought in order to examine them in turn; to stabilize highly abstract ideas in working memory ; and to inspect and criticize our own reasoning ". Benjamin Whorf goes even further. For him, the formation of abstract ideas depends on language : " the linguistic system (...) of each language is not only an instrument of reproduction serving to vocalize ideas, but is itself a formator of ideas (...). " Language is also responsible for the formation of mental categories : " we cut up nature according to the lines that our language has drawn ".

Experimental results from ethology, and also from child development psychology, are completely at odds with these theses. Due to lack of time, the lecture has been limited to citing just a few examples. The prefrontal cortex of the macaque monkey categorizes a continuum of images, without words of course, simply on the basis of the rewards it receives (Freedman et al., 2001). Moreover, the principles underlying the learning of mental categories appear identical in humans and baboons (Chemla et al., 2019). Finally, contra Andy Clark, macaque monkeys, although lacking any language, are capable of paying attention to their mental representations and making second-order similarity judgments, or an introspective judgment about their level of ignorance (cf. the 2010-2011 lecture on metacognition and introspection).

And yet, isn't it the learning of symbols that distinguishes our species and enables the mental representations of the human species to surpass those of other animals ? An examination of animal representations shows that, although many species can learn dozens of arbitrary signs, their mental representations are not identical to ours. Their signs differ from our symbols in at least two ways. Firstly, signifier-signified links are not spontaneously reversible : when a monkey learns that the symbol A predicts the object α, it does not generalize this knowledge in the opposite direction, i.e. from the object α to the symbol A. Thus, animals seem unable to move flexibly back and forth between signifier and signified. Secondly, and most fundamentally, animals fail to combine multiple symbols in a genuine syntax. This seems to prevent them from conceptualizing mathematical or scientific regularities as deeply as we do. The lecture dwelt on the example of geometric shapes : given a rigorously identical test, even a simple square is the subject of an eminently different representation in humans and baboons : the human species represents them using a Combinatorics of symbolic features (parallelism, right angles, etc.), while the baboon confines itself to an approximate perceptual representation, similar to that of artificial convolutional neural networks (Sablé-Meyer et al, 2021).

But is this a linguistic effect? No, because the geometric regularity effect is observed in all members of the species, including pre-school children who have not yet learned the vocabulary of geometry, or Himba adults from Namibia, who have no access to formal education in this field. So this is not a Whorfian effect, but a universal characteristic of the human species. We hypothesize that only the human species possesses :

• a system of symbolic, arbitrary, bidirectional links between signifiers and signifieds ;
• an internal Combinatorics capability, a " language of thought " that enables us to combine these symbols.

According to this hypothesis, we possess the samecore knowledge as other primate species: objects, shapes, people, numbers, positions and so on. But we recombine them into sentences in a " language of thought ", enabling us to form an infinite pyramid of embedded concepts. This internal language is universal: all humans possess this ability and can therefore access the same concepts. However, the conceptual space opened up by this recursive capacity is so vast that no human being can explore it in its entirety, and different cultures don't make the same choices or make the same discoveries. The role of linguistic communication, through a specific language, could be to direct attention towards certain combinations that a given culture, with the experience of generations, deems more important than others. There would thus be both " universality " (all members of the human species are capable of thinking the same thoughts) and " cultural specificity " (the availability of certain words, particularly during learning, biases our attention and affects our perceptual and cognitive judgments).