Narrow Results

Natural languages do not differ arbitrarily, but are constrained so that certain properties recur across the languages of the world. These constraints presumably arise, at least in part, from the nature of the human brain, but the nature of the mapping from brain to language structure is unclear. One theory argues that strong (or even absolute) constraints are built into the language faculty and imposed by individual learners (Chomsky, 1965). An alternative suggestion (e.g. Kirby, Dowman & Griffiths, 2007) is that the same typological distributions could arise given only weak biases in individual learners, as a consequence of cultural transmission within populations. This debate has profound implications for theories of the origins and evolution of language, because the culturally-mediated mapping between learner biases and language structure complicates the biological evolution of the language faculty (see e.g. Smith & Kirby, 2008).

A test-case for the relationship between cognitive biases of individuals and structural properties of language is linguistic variation. Variation in language tends to be predictable: in general, no two linguistic forms will occur in precisely the same environments and perform precisely the same functions. Instead, usage of alternate forms is conditioned in accordance with phonological, semantic, pragmatic or sociolinguistic criteria. Experimental studies (e.g. Hudson Kam & Newport, 2005) show that, given a language in which two forms are in free variation, adult learners tend to probability match (i.e. produce each variant according to its frequency in the input), whereas children are more likely to regularize, suggesting that unpredictable variation is absent from natural languages simply because it cannot be acquired by children.

We show here that iterated learning can produce linguistically-conditioned stable variability (see also Reali & Griffiths. 2009). 25 adult participants were trained on an artificial language exhibiting unpredictable variation (plurality could be marked using two forms, which alternated freely). These participants showed no evidence of having eliminated this variability, i.e. they appeared to probability match. Ten of these participants were then used as the first generation for ten independent iterated learning chains, with the language produced by the first generation on test being used to train the second generation, and so on. Variability was preserved across five generations in seven of the ten chains. However, the predictability of that variability gradually increased, until nine often chains exhibited entirely predictable plural marking: the choice of marker became conditioned on the noun being marked. This demonstrates that adult learners have a relatively weak bias against unpredictable variation (not detectable in a sample of 25 individual learners), which nonetheless becomes apparent through iterated learning. The predictability of variation in natural language might therefore be explained as a consequence of either strong learner biases against unpredictability (in children), or the repeated application of far weaker biases (in adults, or children, or both).

Cultural transmission may act to amplify weak biases, and therefore obscure the relationship between learner biases and linguistic consequences of those biases. This implies that we cannot simply read off the biases of learners from population-level behaviour, nor extrapolate with confidence from individual-based experiments to population-level phenomena. Furthermore, evolutionary pressures acting on the language faculty face a similarly opaque mapping between the structure of the languages (over which selection presumably acts) and the cognitive traits that produce those linguistic structures

Note from Publisher: This article contains the abstract and references.

Natural languages do not differ arbitrarily, but are constrained so that certain properties recur across the languages of the world. These language universals presumably arise from the nature of the human brain, but the mapping from brain to language structure is unclear: linguistic universals may reflect hard constraints on language acquisition, or may arise given only weak biases in individual learners, as a consequence of cultural transmission in populations (e.g., Chater & Christiansen, 2010). Understanding this mapping from properties of learners to language structure has profound implications for theories of the origins and evolution of language, because the culturally-mediated mapping between learner biases and language structure potentially obscures those learner biases from selection and therefore influences the biological evolution of the language faculty (Smith & Kirby, 2008)…