Abstract not found

Why do languages share structural properties? The functionalist tradition has argued that languages have evolved to suit the needs of their users. By what means functional pressures may come to shape grammar over time, however, remains unknown. Functional pressures could affect adults ’ production; or they could operate during language learning. To date, these possibilities have remained largely untested. We explore the latter possibility, that functional pressures operate during language acquisition. In an artificial language learning experiment we investigate the trade-off between word order and case. Flexible word order languages are potentially ambiguous if no case-marking (or other cues) are employed to identify the doer of the action. We explore whether language learners are biased against uncertainty in the mapping of form and meaning, showing a tendency to make word order a stronger cue to the intended meaning in no-case languages.

We simulate the evolution of a domain language in small speaker communities. Data from experiments (Garrod et al., 2007; Fay et al., 2008) show that human communicators can evolve graphical languages quickly in a constrained task (Pictionary), and that communities converge towards a common language even in the absence of feedback about the success of each communication. We postulate that simulations of such horizontal evolution have to take into account properties of human memory (cue-based retrieval, learning, decay). We implement a model that can draw abstract concepts through sets of non-abstract, related concepts, and recognize such drawings. The knowledge base is a network with association strengths randomly sampled from a natural distribution found in a text corpus; it is a mixture of knowledge shared between agents and individual knowledge. In three experiments, we show that the agent communities converge, but that initial convergence is stronger when communities are structured so that the same pairs of agents interact throughout. Convergence is weaker in communities when agents do not swap roles (between recognizing and drawing), predicting the necessity of bi-directional communication in domain language evolution. Average and ultimate recognition performance depends on how much of the knowledge agents share initially.

Abstract. Converging findings from English, Mandarin, and other languages suggest that observed “universals ” may be algorithmic. First, computational principles behind recently developed algorithms that acquire productive constructions from raw texts or transcribed child-directed speech impose family resemblance on learnable languages. Second, child-directed speech is particularly rich in statistical (and social) cues that facilitate learning of certain types of structures. Having surveyed a wide range of posited universals and found them wanting, Evans and Levinson (E&L) propose instead that the “common patterns ” observed in the organization of human languages are due to cognitive constraints and cultural factors. We offer empirical evidence in support of both these ideas. One kind of common pattern is readily apparent in the six examples of child-directed speech in Figure 1, in each of which partial matches between successive utterances serve to highlight the structural regularities of the underlying language. Two universal principles that allow such regularities to be learned can be traced to the work of Zellig Harris (1946; 1991). First, the discovery of language structure, from morphemes to phrases, can proceed by cross-utterance alignment and comparison (Harris, 1946; Edelman and Waterfall, 2007). Second, the fundamental task in describing a language is to state the departures from equiprobability in its sound- and word-sequences (Harris, 1991, p.32; cf. Goldsmith, 2007).

Computational experiments have been used extensively to study language emergence by simulating the evolution of language over generations of interacting agents. Much of this work has focused on understanding the mechanisms of how language might have evolved. We propose a complementary approach helpful in understanding why specific properties of language might have emerged as an adaptive response to joint pressures from the environment and constraints on an agent’s cognitive architecture. The approach suggests that linguistic systems can be described as boundedly optimal policies in multi-agent dynamic control problems defined by specific environments, agent computational structures, and task-oriented (vs. communication oriented) rewards. We illustrate the approach with a set of computational experiments.

We simulate the evolution of a domain vocabulary in small communities. Empirical data show that human communicators can evolve graphical languages quickly in a constrained task (Pictionary), and that communities converge towards a common language. We propose that simulations of such cultural evolution incorporate properties of human memory (cue-based retrieval, learning, decay). A cognitive model is described that encodes abstract concepts with small sets of concrete, related concepts (directing), and that also decodes such signs (matching). Learning captures conventionalized signs. Relatedness of concepts is characterized by a mixture of shared and individual knowledge, which we sample from a text corpus. Simulations show vocabulary convergence of agent communities of varied structure, but idiosyncrasy in vocabularies of each dyad of models. Convergence is weakened when agents do not alternate between encoding and decoding, predicting the necessity of bi-directional communication. Convergence is improved by explicit feedback about communicative success. We hypothesize that humans seek out subtle clues to gauge success in order to guide their vocabulary acquisition.

There are certain universal properties of language that are taken to be definitional to the concept of language itself, such as the arbitrary relationship between sounds and meanings of words. Another possibility is that these “design features ” of language may instead be the expressed consequences of general purpose learning constraints within the cognitive system learning the language. To test this, generations of an inverse model learning to map between sounds and meanings of words was tested. In this model, learning to associate phonology to semantics influences the model’s production of phonology from semantics, and phonological productions of one model were used as input to the next generation. Over generations of the model’s learning, the language became easier to acquire, and demonstrated increased arbitrariness of mappings between phonology and semantics. The iterative modelling demonstrated that design features of natural language can spontaneously emerge in a general purpose learning system.

Human languages vary in many ways, but also show striking cross-linguistic universals. Why do these universals exist? Recent theoretical results demonstrate that Bayesian learners transmitting language to each other through iterated learning will converge on a distribution of languages that depends only on their prior biases about language and the quantity of data transmitted at each point; the structure of the world being communicated about plays no role (Griffiths & Kalish, 2005, 2007). We revisit these findings and show that when certain assumptions about the independence of languages and the world are abandoned, learners will converge to languages that depend on the structure of the world as well as their prior biases. These theoretical results are supported with a series of experiments showing that when human learners acquire language through iterated learning, the ultimate structure of those languages is shaped by the structure of the meanings to be communicated.

Extending learned patterns to previously unseen ones is a key hallmark of complex cognition. This paper presents evidence that learners are able to generalize learned patterns to novel stimuli with very different surface properties within and across modalities. Using a statistical learning paradigm, adult learners were exposed to a repetition (reduplication) pattern in which the first element of a three-element sequence repeated (e.g., AB�AAB). The pattern was presented as either spoken repetition (e.g., bago, babago) or a non-linguistic visual analogue (i.e., repetition of non-nameable shapes). Learners showed significant transfer from a non-linguistic repetition pattern to a linguistic reduplication pattern, and vice versa. However, we found a small bias towards linguistic reduplication, as responses to linguistic patterns were numerically higher. This suggests that while learners are able to extend learned patterns to novel patterns in other domains, factors such as familiarity and naturalness may privilege linguistic patterns over non-linguistic analogues.

Speech sounds within a linguistic system are both categorical and combinatorial and there are constraints on how elements can be recombined. To investigate the origins of this combinatorial structure, we conducted an iterated learning experiment with human participants, studying the transmission of an artificial system of sounds. In this study, participants learn and recall a system of sounds that are produced with a slide whistle, an instrument that is both intuitive and non-linguistic. The system they are exposed to is the recall output of the previous participant. Transmission from participant to participant causes the system to change and become cumulatively more learnable and more structured. This shows that combinatorial structure can culturally emerge in an artificial sound system through iterated learning.