The use of language is central to human social dynamics. Current techniques for modeling human language are widely based on simplifying assumptions that underestimate human language capacities by ignoring statistical patterns. The present project starts from a strongly contrasting approach which can lead to a breakthrough in the understanding of linguistic dynamics by linking changes in language across the different time scales of speech, the human life cycle, and the history of communities. This approach models grammar as a highly plastic cognitive system sensitively tuned to the probabilities of the environment. The project combines detailed statistical analyses of patterns found in recorded speech with laboratory experiments on language production in adults and language learning in children. The phenomena considered include high-level grammatical choices (such as whether to say "give someone a job" or "give a job to someone"), low-level pronunciation choices (such as where "to" is reduced to "tuh"), and overregularizations (such as when children say "goed" instead of "went"). Studies will involve work with data from multiple dialects of English, as well as German and Chinese. The project involves cross-disciplinary collaboration among researchers trained in Linguistics, Psychology, and Computer Science. This research will lay the groundwork for improved language technologies, and may find other applications, for example in the treatment of language disorders.
