The aim of the conference

The project for the conference came from the fact that both of us were engaged in an ESF Eurocores project on the evolution of language and had — possibly different but strong in any case — skepticisms about popular theories. For instance, quite a few theories have hypothesized that language had emerged from social pressures and that its emergence was made possible by previous mind reading or theory of mind skills. However, such hypotheses never made clear what sort of theory of mind was a necessary condition for language emergence (though Tomasello and colleagues have made a serious effort at spelling out the intention reading prerequisites). Similar claims _were made for the precedence of language acquisition. However, the chronological evidence in both theory of mind development and language acquisition did not seem to agree with such hypotheses in their rough and ready form and it seemed clear that some precision would be needed. A possibility, which some at least of the contributors have followed is that language and theory of mind have coevolved, which of course does not preclude that one or both of them should be necessary for the normal acquisition of the other. Thus, the conference.

Individual contributions

Introducing the conference, Reboul tried to shift through the evidence on both theory of mind and language, using both acquisition and comparative data. Her main hypotheses are that language and ToM are specific to human animals, that ToM is not a monolith but rather a collection of possibly modular abilities, that the acquisition of the lexicon, though it may be helped by relatively elementary mind reading mechanisms, begins without them and, finally, that success at the false belief task may well depend on abilities acquired through language rather than otherwise. In an effort to draw a parallel between the underlying mechanisms of language and ToM, Dominey outlined a mechanism for ToM constructions, _as mappings between behavioral scenario structure and the corresponding social/intentional interpretation or outcome (analogous to the form-meaning mappings in grammatical constructions).  The ensuing debate identified the point that aspects of ToM (notably the ability to metarepresent) are semantic abilities, likely distinct from the structure mapping capability. Newmeyer goes on to puzzle out the different contributions of cognition-aiding factors and vocal interaction-aiding factors in the origins and evolution of grammar.  He concludes that cognitive factors were the first to shape grammars. But with the passage of time, the exigencies of communication came to play an ever-more important role in grammar, and thus human language today therefore reflects _the influence of both types of factors. While Newmeyer concentrates on the evolution of language, Baron-Cohen concentrates on empathy, which he sees as having both a cognitive element (ToM, illustrated by the false belief test) and a more emotional aspect. _Empathy seems to be stronger in females than in males and the ability to identify speaker's meaning is an strong advantage in language acquisition, explaining why females are more precocious in that regard than males. Origgi and Sperber acknowledge that ToM comes in different and probably successive modular abilities (they are skeptical of the false belief test as a test of the kind of mindreading abilities crucial for language acquisition or evolution), and see it as a necessary condition for the evolution of the language acquisition device, given that it would enable communicators to go beyond the restrictions of an eventual protolanguage to more or less propositional forms, corresponding to the speaker's meaning. As they point out, this is something impossible on a code model of linguistic communication and their view (following Relevance Theory) is that linguistic communication rests on both code and inference, inference being done through a special purpose pragmatic module. Fitch adopts a comparative perspective, pointing to three capacities, which are both fundamental for language and specific to humans as opposed to chimpanzees: speech and especially vocal imitation, recursive syntax and intentional semantic communication. He argues that vocal imitation amounts to non-semantic sensorimotor quoting from which semantic representations can develop and advocates a "coevolutionary relationship between quoting and theory of mind", additionally and strongly conditioned by syntactic recursion. Jill de Villiers _outlines an hypothesis, backed by experimental data, according to which ToM (specifically the false belief test), far from conditioning language acquisition or evolution, is on the contrary dependent on language and, more precisely, on mastery of the recursive structure of complementation evidenced by discourse and mental verbs (e.g., saying, thinking). According to her, such structures were first developed to express the reliability of signals and of the signalers who produce them. Ruffman argues that children's initial understanding of mind is likely implicit and manifest in their behavior rather than insights that they can verbalize. This understanding could be innate, but learning likely plays a large role. Over time, and as their language develops, children develop a consciously mediated and verbally based theory on the basis of these implicit intuitions. Diesendruck concentrates on the acquisition of words and, without denying that there may several roads to word learning, claims that only one of them, which goes through mindreading abilities, is going to provide the learner with the kind of words which will enable her to take part in normal linguistic communication. In other words, though very young children, trained animals and, possibly, autistic speakers learn words through a training process of a probably associative nature, only the acquisition of words through (possibly fairly elementary) mindreading abilities is going to provide an individual with the opportunity for normal communication. Nazir and Boulenger consider the problem from a perspective of evolutionary pressure to minimize processing requirements.  In this context, infants are strongly predisposed to seek for optimal solutions to minimize processing requirements, and words provide a perfect clue allowing them to rapidly form categories, including those related to theory of mind.  Arbib outlines a motor-oriented account of language-readiness, based on successive capabilities to recognize and imitate action, leading to protosign, protospeech and language. He identifies the need to investigate whether an account can be given of a shared evolution of  "mirror systems" suiting both ToM and complex imitation, hypothesizing that the ancestral mirror system for manual praxis was distinct from the putative mirror system for facial expression of emotion. The former would support pantomime and thence on to multi-modal symbols; and then the availability of symbols could enrich the latter to yield rudiments of ToM.

As can be seen from the previous quick summing up of the various contributions to the conference, the positions taken seem fairly widespread. However, we think that some major points of agreement emerged during the conference, among other things during the discussions, and that differing positions can indicate some interesting areas for experimental and or theoretical debate. It is to them that we now turn.

Putting all the strands together

There are a few questions that naturally come to mind on the topic of the coevolution of language and theory of mind. Here are some of them: Are either language or theory of mind as such the product of evolution and, if so, what are their adaptive features? Are either language or theory of mind monolithic abilities (i.e., should one consider either of them a module in a Fodorian sense)? If either of them is not, how should it be divided and should its resulting components be regarded as modular in the Fodorian sense? And, perhaps the central questions in the present conference, have language and theory of mind, supposing them to have evolved, coevolved? If not, is there a relation of dependence between language and theory of mind and if there is, in which direction does it go?

Let us begin with the question regarding whether or not language and theory of mind are monolithic abilities. Clearly, as regards language - especially if this question is linked to the evolution of language -, one should presumably distinguish between phonology and syntax (including the part of semantics that is heavily dependent on syntax) on the one hand and the lexicon on the other. For exemple, it could make sense to say that syntax has evolved and is more or less innate, but it clearly does not make sense to say that the lexicon as such has evolved and is innate.

This is why it makes sense to speak of a language acquisition device being innate, noting that this does not include the lexicon, which is language-specific. This does not exclude the parallel evolution of general mechanisms such as those described by Boulenger and Nazir, where a new word triggers the acquisition of a new concept or category nor the general role of mindreading in lexical acquisition, as described by Diesendruck. Regarding this, some contributors (i.e., Origgi and Sperber, de Villiers) clearly consider that language, in more or less the above sense, has evolved as such, while others do not, either because they do not pronounce on the issue or because they think that language did not evolve as such, i.e., some linguistic abilities have evolved but they did not evolve as linguistic abilities (Dominey, Arbib, Fitch, Newmeyer). Regarding theory of mind, there seems to be a more or less general consensus among the contributors who deal with it, that it presumably is not a monolithic ability, though there may be disagreement over what specific distinct abilities make it (i.e., over how we should cut the cake) as well as over whether these distinct abilities are or not modular in the Fodorian sense. Note as well that some of these abilities might be modular while others might not be. None of the contributors have advocated a new division in ToM abilities, though quite a few have taken up Baron-Cohen's suggestions (1995) regarding the existence of an eye direction detection (EDD) and an intentionality detection (ID), followed by a shared attention mechanism (SAM) and, much later by a Theory of Mind Mechanism (ToMM), which enables children to pass the false belief test at around four years of age.

Regarding the relation of language and theory of mind, some contributors (e.g., Origgi and Sperber) have taken the strong view that some of the modules making up ToM are a prerequisite for the evolution of a language acquisition device (LAD). On this scenario, both ToM and LAD have evolved, but the evolution of ToM, in a possibily fairly elementary form, must precede the evolution of LAD, being a necessary (though not sufficient) condition for it. This view is apparently contradicted by de Villiers who claims that language came first and was indeed the condition for ToM (though she concentrates on false belief test). This seems echoed, up to a point, by Fitch, who also points out the intimate relation between ToM and recursive syntax, though he might favor a more coevolutionary scenario, given his insistence on imitation. A more balanced view is proposed by Diesendruck who insists on the necessity of ToM for lexicon acquisition, but allows for two different learning processes (possibly successive in normal development), only one of which would allow the learner to successfully communicate through language. On Diesendruck's view, some ToM abilities, which are roughly in place by two years of age (note that these might include SAM, on Baron-Cohen's division of ToM abilities), spur language acquisition on. These would be enough for normal language acquisition, though they would still fall short of passing the false belief test. On this count, Diesendruck's and de Villiers' view are not as incompatible as they seem: granted Diesendruck says ToM first, while de Villiers says language first, but they do not mean exactly the same thing by ToM. Diesendruck does not include false belief ascriptions (FBA) in the ToM abilities that are taken to precede language, while de Villiers is only interested in FBA. This might lead to a mixed model under which, though some pretty limited lexicon acquisition would take place before 18 months of age, ToM modules (not including FBA) would then kick in, spurring language acquisition on and allowing normal linguistic communication; language acquisition would proceed, allowing the child to develop more sophisticated and highly language dependent ToM abilities, such as FBA (allowing her to pass the false belief test) and, ultimately, the ability to detect opaque contexts (allowing her to pass, about a year later, _the opaque context tests). By the way, it should be clear that such a scenario might ultimately solve the ToM as theory versus ToM as non-theoretical (either emotional empathy or simulation) debate by going more or less in the direction indicated by Baron-Cohen (2003) through his distinction between the affective and the cognitive components of empathy: one could suppose that the affective (non-theoretical) components would more likely be innate, though susceptible of degrees (hence, the average differences which he describes between males and females), and would condition language acquisition, while the cognitive components (as tested through false belief and opaque contexts tests) would be language dependent and learned (and theoretical).

This scenario makes sense if envisaged relative to Newmeyer's insistence on the importance of cognition for shaping language. On Newmeyer's scenario, the fundamental structure of language is determined by cognitive constraints (though it should be noted that his view is compatible with a concepts first and language as a means of expressing them perspective), though some linguistic variation should be accounted for through communicative constraints. This might lead (introducing ToM, which Newmeyer does not take into account _and giving to language not only cognitive constraints, but a cognitive role) to a view in which language first evolved as a cognitive tool, to organize information (a view very near in some respects to Nowak's and his team (2000, 2001, 2002 - though one would not necessarily accept Nowak's self-organisational outlook), and only then evolved as a communicative medium. In that scenario, the advantage for which language first evolved was cognitive, not communicative, though the size and social complexity of hominoid and early human groups (as well as the long period during which children are dependent on adults, see Baron-Cohen 2003) led to the evolution of mindreading (empathy) devices, facilitating language transition from a cognitive to a communicative tool, _which in turn led to more sophisticated ToM, leading to FBA and ultimately to the detection of opaque contexts and allowing communicative constraints to slightly change the previous structure of language. This would explain both the syntactic difference between language and animal systems of communication, but also its communicative differences with them, well described by Fitch: animals do not seem to have recursive structures in their communicative system neither do they have anything like Gricean cooperation in their communicative use of these systems. It would, as well, explain why linguistic communication cannot be made sense of through a code-only perspective and why a mixed view (both code and inference) is needed. Finally, to go further, if language primarily emerged (either through evolution, self-organization or exaptation from systems evolved for other reasons) as a cognitive tool, and if it emerged with the Sapiens Sapiens lineage, it might explain why that lineage had better evolutive success than did the Neanderthals with whom it co-habited for some seventy thousand years (from the apparition of Sapiens about a hundred thousand years ago to the disparition of Neanderthals about thirty thousand years ago) and may go some way to explain the so-called cultural explosion which occurred some forty to fifty thousand years ago. This is mysterious, because Sapiens Sapiens does not seem to have undergone biological (and especially cerebral) evolution since its first apparition. Thus, the big question is: why did it take Sapiens Sapiens some fifty thousand of years to radically modify its behavior, inventing art and dramatically improving its technology? One frequent answer is through language. This however would be in direct contradiction with the notion that language is (the product of) a biological evolution. But innovations are in some ways very cultural human manifestations - a fact acknowledged by Newton when he said "If I have seen further it is by standing on the shoulders of giants" - and one very obvious means of cultural transmission is through communicative language. A way to reconcile the idea that language may be a biological evolution is to see Sapiens Sapiens as possessing a cognitive language (possibly inherited from a common ancestor with Neanderthals), but going on to use it communicatively, through both better empathy and a radically different vocal tract. On this view, language qua cognitive tool would be a biological evolution, but linguistic communication may have emerged culturally. And one might go the way of Baron-Cohen and beyond here and speculate that the transition of language from cognitive tool to communicative tool might have been very much a female affair…

The conference has raised several points that will be of practical use in the study of ToM and language via simulation and robotic experiments.

One of these concerns the language and ToM capabilities that are employed in code vs inferential communication.  Several authors indicate that linguistic communication is impossible at the code level - Diesendruk, Origgi and Sperber -, though it may be that initially — both for development and evolution — the system begins in a state that is more code level (or associative) than inferential — Reboul, Diesendruck. This suggests a point of departure for simulation to begin with a code level system, and then to introduce a pragmatic function that may either be an innate "module" - Sperber - or something that can be at least in part developped - Ruffman. In this context de Villier's suggests that recognition of one's own false beliefs may provide the bootstrap for a more generalized ability to attribute mental states in a recursive manner. Though de Villiers certainly does not endorse construction based learning mechanisms, it should be noted that her position on ToM (and FBA) is consistent with the construction based learning mechanism proposed by Dominey  and the statistical learning mechanisms cited by Ruffman.  The scenario would then be to consider systems with an initial representation of the own and other's goals and intentions, and a code level communication capability, with the potential to make the transition to communication itself becoming part of the action, goal, intention system.

Conclusion

In conclusion, as co-organisers of the present conference, we would like to thank all of the contributors and discussants for having made it lively, intellectually stimulating and great fun. We can only hope that all the participants have enjoyed it as much as we have and that we have not made any major error in the above summary of contributors’ positions.

Special thanks go to Gloria Origgi, who, as administrator of the web conference, has made this event possible and is, as such, in great part responsible for its success.

References

Baron-Cohen, S. (1995), Mindblindness: An essay on autism and theory of mind, Cambrige, MA, The MIT Press.

Baron-Cohen, S. (2003), The essential difference, London, Penguin Books.

Fitch, T. and Hauser, M. (2004), “Computational constraints on syntactic processing in a nonhuman primate”, Science 303, 377-380.

Nowak, M.A. and Komarova, N.L. (2001), “Towards an evolutionary theory of language”, Trends in Cognitive Science 5/7, 288-295.

Nowak, M.A., Plotkin, J.B. and Jansen, V.A.A. (2000), “The evolution of syntactic communication”, Nature 404, 495-498.

Nowak, M.A., Komarova, N.L. and Nyogi, P. (2002), “Computational and evolutionary aspects of language”, Nature 417, 611-617.