What we hear is how we speak: The role of frequency in language acquisition
03 / 2011 - 02 / 2014
Kinderen leren hun moedertaal zonder enige moeite, aan de hand van spraak in hun omgeving. Maar sommige woorden, zinsdelen, en zinnen horen zij vaker dan andere. Door middel van computersimulaties zal onderzocht worden hoe kinderen deze informatie gebruiken.
Children learn their mother tongue from the ambient speech in their environment. Some words, phrases and sentences they hear more often than others. This project asks how frequency of occurrence influences developmental patterns in language acquisition. Frequency effects, where more exposure leads to faster acquisition, have been demonstrated in many areas of language. Counting the relative frequencies of structures, however, is too coarse a predictor of the order of acquisition. Consistent meaning-to-form mappings, lexical diversity, and similarities with other structures can facilitate learning despite low frequency. Novel measures of frequency will be developed which take into account the availability, variability, and analogy of distributional information in child-directed speech. The mechanisms of statistical learning which cause frequency effects in child-language are largely unknown. There is a blatant gap between psycholinguistic theories and developmental data in our understanding of the role of frequency. A cognitively adequate computational model of human language acquisition will be developed to simulate frequency-driven differential learning in a mechanistic framework. Crucially, this model adds semantic representations to the study of statistical learning in a novel way. Thus, it allows us to address the question how children acquire meaning-to-form mappings from plausible input distributions and how the learnability of structures is affected by manipulations of this input. The study of frequency effects in child-language acquisition using techniques from computational linguistics is a novelty in developmental psychology. The project will not only provide fine-grained frequency measures, but also models and explanations of concrete linguistic phenomena---frequency effects in the acquisition of prepositional phrases, relative-clauses and non-adjacent dependencies. Furthermore, quantitative model predictions will be derived and tested in behavioral experiments. Finally, the software tools developed during this project will be made available to the community of linguists.