The 'original resources' available to humans has puzzled the founder of modern neuroscience, Santiago Ramon y Cajal. In his autobiography he writes (Cajal, 1917, pp. 345-350):

At that time, the generally accepted idea that the differences between the brain of [nonhuman] mammals (cat, dog, monkey, etc.) and that of man are only quantitative, seemed to me unlikely and even a little offensive to human dignity ... but do not articulate language, the capability of abstraction, the ability to create concepts, and, finally, the art of inventing ingenious instruments . seem to indicate (even admitting fundamental structural correspondences with the animals) the existence of original resources, of something qualitatively new which justifies the psychological nobility of Homo sapiens? (Cited by De Felipe et al., 2002, p. 299)

Collegium Budapest (Institute for Advanced Study), 2 Szentharomsag utca, H-1014 Budapest, Hungary, Research Group for Theoretical Biology and Ecology, Institute of Biology, Eötvös University, Budapest, Parmenides Center for the Study of Thinking, München, Germany, e-mail: [email protected]

M. Barbieri (ed.), The Codes of Life: The Rules of Macroevolution. © Springer 2008

Natural language is a unique communication and cultural inheritance system. In its practically unlimited hereditary potential it is similar to the genetic and the immune systems. The underlying principle is also similar in that all these systems are generative: they achieve unlimited capacity by the combination of limited primitives. The origin of natural language is the last of the major evolutionary transitions [Maynard Smith and Szathmary, 1995]. Although later in society important transitions did happen in the way of storing, transmitting, and using inherited information, they were not made possible or accompanied by relevant genetic changes in the biology of our species. In contrast, language has a genetic background, but it is an open question how a set of genes affects our language faculty. It is fair to say that with respect to their capacity to deal with the complexity of language, even 'linguistically trained' animals are very far from us.

Understanding language origins and change is difficult because it involves three interwoven timescales and processes: individual learning, cultural transmission, and biological evolution. These cannot be neatly separated from one another (Christiansen and Kirby, 2003). The fact that a population uses some useful language that is culturally transmitted changes the fitness landscape of the population genetic processes.

Language has certain design features, such as symbolic reference, composition-ality, recursion, and cultural transmission (Hockett, 1960). Theories of language and language evolution can be divided into two sets of hypotheses: a nativist versus empiricist account and a non-adaptationist versus adaptationist account, respectively (Smith, 2003).

The nativist paradigm argues that language capacity is a collection of domain-specific cognitive skills that is unique to humans and is somehow encoded into our genome. Perhaps the most famous proponent of this approach is Noam Chomsky, who coined the term 'language organ' and argued in favour of the uniqueness and the innateness of human linguistic skills (Chomsky, 1986). Different scholars agree with Chomsky on this issue (Pinker and Bloom, 1990; Jackendoff, 1992; Pinker, 1994; Maynard Smith and Szathmary, 1995; Pinker and Jackendoff, 2004). The empiricist paradigm, however, argues that linguistic performance by humans can be explained with domain-general learning techniques (Sampson, 1997).

Fisher and Marcus (2006, p. 13) are right in stating that 'In short, language is a rich computational system that simultaneously coordinates syntactic, semantic, phonological and pragmatic representations with each other, motor and sensory systems, and both the speaker's and listener's knowledge of the world. As such, tracing the genetic origins of language will require an understanding of a great number of sensory, motor and cognitive systems, of how they have changed individually, and of how the interactions between them have evolved'. The study of language origins is however hampered by the fact that there is a critical lack of detailed understanding at all levels, including the linguistic one. There is no general agreement among linguists how language should be described: widely different approaches do exist and their proponents can have very tense scientific and other relationships. As a biologist I would maintain that symbolic reference combined with complicated syntax (including the capacity of recursion) is a least common denominator in this debate. Within this broad characterisation I just call attention to two approaches that have, perhaps surprisingly, a strongly chemical flavour. One is the minimalist programme of Chomksy (1995) where the crucial operator is merge, the action of which triggers certain rearrangements of the representation of a sentence. There is a broad similarity between this proposal and chemical reactions (Maynard Smith and Szathmary, 1999). An even closer analogy between chemistry and linguistics can be detected in Steel's Fluid Construction Grammar (Steels, 2004; Steels and De Beule, 2006), in which semantic and syntactical 'valences' have to be filled for correct sentence construction and parsing. We should note (see also chapters on the genetic code) that the roots of genetic inheritance are of course in chemistry, and that even at the phenomenological level Mendelian genetics was a stoichiometric paradigm, influenced by contemporary chemical understanding (elementary units that can be combined in certain fixed proportions give rise to new qualities). Chemical reactions can be characterised by rewrite rules. It will take in-depth study to consider how deep this analogy goes. The deeper it goes, the more benefit one can hope from taking the analogy seriously.

Non-adaptationist accounts of language evolution rely heavily on 'spandrels' (Gould and Lewontin, 1979). The idea is that language or linguistic skills evolved not because it gave fitness advantage to its users; rather it evolved as a side effect of other skills as spandrels are side effect of architectural constraints. Chomsky again has a prominent role in this debate as the protagonist of the non-adaptationist approach. In the latest reworking of the theory (Hauser et al., 2002), Chomsky and colleagues distinguish between the 'Faculty of Language in the Broad Sense' (FLB) and 'Faculty of Language in the Narrow Sense' (FLN). They argue that FLB consists of skills that evolved in other animals as well as in humans, whereas FLN consists of only one skill (merge), which evolved in a different (unspecified) context and was then co-opted for linguistic use. However, the finding that European starlings appear able to recognise context-free grammatical structures (i.e. hierarchical syntax (Gentner et al., 2006) is somewhat contrary to Chomsky's position given that it shows that precursor of the skill they have assigned to FLN (i.e. merge) may have independently evolved in other animals too; although a strict proof of appropriate parsing of these structures by starlings is lacking (M. Corballis, personal communication, 2007, email].

The first adaptationist account of human language was by Darwin (1971), later defended by Pinker and Bloom (1990) in their influential paper about the Darwinian account of language. More specifically, these authors argued that language, as any complex adaptations, could only be explained by means of natural selection. This paper catalysed many linguists and biologists to study language and language evolution from the perspective of evolutionary biology and was followed by many influential publications (Jackendoff, 1992; Maynard Smith and Szathmary, 1995; Knight et al., 2000; Christiansen and Kirby, 2003). Recently, Pinker and Jackendoff (2005) made a forceful defence of the adaptationist paradigm in response to Chomsky and colleagues (Hauser et al., 2002).

Language needs certain prerequisites. There are some obvious prerequisites of language that are not especially relevant to our approach. For example, apes do not have a descended larynx or cortical control of their vocalisations. Undoubtedly, these traits must have evolved in the human lineage, but we do not think that they are indispensable for language as such. One could have a functional language with a smaller number of phonemes, and sign language (Senghas et al., 2004) does not need either vocalisation or auditory analysis. Thus, we are mostly concerned with the neuronal implementation of linguistic operations, irrespective of the modality. It seems difficult to imagine the origin of language without capacities for teaching (which differs from learning), imitation, and some theory of mind (Premack, 2004). Apes are limited in all these capacities. It is fair to assume that these traits have undergone significant evolution because they were evolving together with language in the hominine lineage. To this one should add, not as a prerequisite, but as a significant human adaptation the ability to cooperate in large non-kin groups (Maynard Smith and Szathmäry, 1995). These traits together form an adaptive suite, specific to humans. I suggest that in any selective scenario, capacities for teaching, imitation, some theory of mind, and complex cooperation must be rewarded, because an innate capacity for these renders language emergence more likely.

On the neurobiological side I must call attention to the fact that some textbooks, [e.g. Kandel et al., 2000] still give a distorted image of the neurobiological basis of language. It would be very simple to have the Wernicke and Broca areas of the left hemisphere for semantics and syntax, respectively. But the localisation of language components in the brain is extremely plastic, both between and within individuals (Neville and Bavelier, 1998; Müller et al., 1999). Surprisingly, if a removal of the left hemisphere happens early enough, the patient can nearly completely retain his/her capacity to acquire language. This is of course in sharp contrast to the idea of anatomical modularity. It also puts severe limitation on the idea that it is only the afferent channels that changed in the evolution of the human brain: modality independence, and the enormous brain plasticity in the localisation of language favour the idea that whatever has changed in the brain that has rendered it capable of linguistic processing must be a very widespread property of the neuronal networks (Szathmäry, 2001). Components of language get localised somewhere in any particular brain in the most functionally 'convenient' parts available. Language is just a certain activity pattern of the brain that finds its habitat like an amoeba in a medium. The metaphor 'language amoeba' expresses the plasticity of language but it also calls attention to the fact that a large part of the human brain is apparently a potential habitat for it, but no such habitat seems to exist in non-human ape brains (Szathmäry, 2001).

A dogma concerning the histological uniformity of homologous brain areas in different primate species has also been around for some time. Recent investigations do not support such a claim (DeFelipe et al., 2002). In fact the primary visual cortex shows marked cytoarchitectonic variation (Preuss, 2000), even between chimps and man. It is therefore not at all excluded that some of the species-specific differences in brain networks are genetically determined, and that some of these are crucial for our language capacity. But, as discussed above, these language-critical features must be a rather widespread network property. Genes affect language through the development of the brain. One could thus say that the origin of language is to a large extent an exercise in the linguistically relevant developmental genetics of the human brain (Szathmäry, 2001).

The close genetic similarity between humans and chimps strongly suggests that the majority of changes relevant to the human condition are likely to have resulted from changes in gene regulation rather than from widespread changes of downstream structural genes. Recent genetic and genomic evidence corroborates this view. In contrast to other organs, genes expressed in the human brain seem almost always upregulated relative to the homologous genes in chimp brains (Caceres et al., 2003). The functional consequences of this consistent pattern await further analysis.

We know something about genetic changes more directly relevant to language. The FOXP2 gene was discovered to have mutated in an English-speaking family (Gopnik, 1990; Gopnik, 1999). It has a pleiotropic effect: it causes orofacial dyspraxia, but it also affects the morphology of language: affected patients must learn or form the past tense of verbs or the plurals of nouns case by case, and even after practice they do so differently from unaffected humans (see Marcus and Fisher, (2003) for review). The gene has been under positive selection (Enard et al., 2002) in the past, which shows that there are genetically influenced important traits of language other than recursion (Pinker and Jackendoff, 2005), contrary to some opinions (Hauser et al., 2002). There is a known human language, apparently with no recursion (Everett, 2005). It would be good to know how these particular people (speaking the Piraha language in the Amazon) manage recursion in other domains, such as object manipulation. Apes are very bad at recursion both in the theory of mind or 'action grammar' (Greenfield, 1991).

It does seem that the capacity to handle recursion is different from species to species. Although the relevant experiment must be conducted with chimps as well, it has been demonstrated that tamarin monkeys are insensitive to auditory patterns defined by more general phrase structure grammar, whereas they discover violations of input conforming to finite state grammar (Fitch and Hauser, 2004). Human adults are sensitive to both violations. Needless to say it would be very interesting to know the relevant sensitivities in apes and human children (preferably before they can talk fluently). It will be interesting to see what kind of experiment can produce consistent patterns in such a capacity in evolving neuronal networks, and then reverse engineer proficient networks to discover evolved mechanisms for this capacity.

I share the view that language is a complex, genetically influenced system for communication that has been under positive selection in the human lineage (Pinker and Jackendoff, 2005). The task of the modeller is then to try to model intermediate stages of a hypothetical scenario and, ultimately, to re-enact critical steps of the transition from protolanguage (Bickerton, 1990) to language. It cannot be denied that language is also a means for representation. This is probably most obvious for abstract concepts, for which the generative properties of language may lead to the emergence of a clear concept itself. This is well demonstrated for arithmetics: for instance, an Amazonian indigenous group lacks words for numbers greater than 5; hence they are unable to perform exact calculations in the range of larger numbers, but they have approximate arithmetics (Pica et al., 2004).

I mentioned before that the fact that language changes while the genetic background also changes (which must have been true especially for the initial phases of language evolution) the processes and timescales are interwoven. This opens up the possibility for genetic assimilation (the Baldwin effect). Some changes that each individual must learn at first can become hard-wired in the brain later. Some have endorsed (Pinker and Bloom, 1990), while others have doubted (Deacon, 1997) the importance of this mechanism in language evolution. Deacon's argument against it was that linguistic structures change so fast that there is no chance for the genetic system to assimilate any grammatical rule. This is likely to be true but not very important. There are linguistic operations, performed by neuronal computations, related to compositionality and recursion that must have appeared sometime in evolution. Whatever the explicit grammatical rules are, such operations must be executed.

Hence a much more likely scenario for the importance of genetic assimilation proposes that many operations must have first been learned, and those individuals whose brain was genetically preconditioned to a better (faster, more accurate) performance of these operations had a selective advantage (Szathmary, 2001). Learning was important in rendering the fitness landscape more climbable (Hinton and Nolan, 1987). This view is consonant with Rapoport's (1990) view of brain evolution. This thesis is also open for experimental test.

The origin of language is an unsolved problem; some have even called it the 'hardest problem of science' (Christiansen and Kirby, 2003). It is very hard because physiological and genetic experimentation on humans and even apes is very limited. The uniqueness of language prohibits, strictly speaking, application of the comparative method, so infinitely useful in other branches of biology. Fortunately, some elements of language lend itself to a comparative approach, as we shall see in relation to bird song. Nevertheless, limitation of the approaches calls for other types of investigation. I believe that simulations of various kinds are indispensable elements of a successful research programme. Yet, a vast range of computational approaches has brought less than spectacular success (Elman et al., 1996). This is attributable, I think, to the utterly artificial nature of many of the systems involved, such as connectionist networks using back-propagation (e.g. see Marcus (1998) for a detailed criticism). In Section 6, I present an alternative, potentially rewarding, modelling approach.

Was this article helpful?

0 0

Post a comment