Models of monolingual and bilingual language acquisition

Children learn language despite the very impoverished nature of the input. Since the 1960s, the symbolic-deductive paradigm has explained this with reference to an innate mental language system. For about two decades, an alternative to symbolic accounts of language and language acquisition has been offered by connectionism, which can be viewed as one of the main subsymbolic-inductive paradigms. The recent models in this paradigm test detailed models against large databases of utterances. A general conclusion from this research is that, despite being very noisy and inconsistent, the nature of language input is nevertheless sufficient to support inductive mechanisms by which seemingly rulelike behavior emerges from a data-driven learning process. Constraints on the learning process are imposed by the architectures of the models. Several models within the symbolic-deductive paradigm have now also been worked out in much more detail, and a lively discussion between proponents of the two paradigms is currently taking place. We review some of the prominent models in both paradigms, with an emphasis on the connectionist models. In particular, we look at models of the acquisition of stress assignment, phonology, past tense formation, pluralization, and certain aspects of semantics.