Applying the Right Statistics: Linking Distributional Learning to Language Production of Relative Clauses

Kerz, E. 1 , Wiechmann, D. 2 & Ströbel, M. 1

1 RWTH Aachen University
2 University of Amsterdam

Statistical learning research has shown that language users are sensitive to the distributional statistics inherent in the linguistic input and rely on knowledge of such statistics to facilitate their language processing and boost their acquisition. A related line of research on language adaptation has shown that they are also capable of continuously adjusting their language in response to changes in the distributional statistics in the language they encounter. Both lines of research have primarily targeted language comprehension and short-term adaptation within the time period of an experiment and to the stimulus material used in laboratory settings. In this study, we present a novel approach to investigating long-term linguistic adaptation in language production that combines natural language processing techniques and large corpora of authentic language use capturing distributional statistics from different communicative contexts (registers). In this approach, we derived accurate and reliable estimates of the distributional frequencies of relative clause structures across the five registers of the Corpus of Contemporary American English (560 million words) and linked them to the patterns of language production in native and non-native speakers. Our findings indicate that both L1 and L2 speakers are capable of successfully adapting to register-specific distributional statistics to a similar degree.