Statistics in Language Project

Children learn language from their experiences with linguistic input. Many current theoretical accounts agree that children exploit patterns of co-occurrences to represent the information in their environment – for example, between word sounds to learn transitional probabilities in language, between relevant features to acquire object categories, between names and objects to establish word-to-object mappings, or between words in speech to build structured semantic representations. Some recent work has called into question the scope of encoding patterns of co-occurrences to build structured semantic representations. Specifically, this work argues that early in development, local patterns co-occurrence (i.e., co-occurrences between words experienced close in time) play a central role in the acquisition of structured semantic representations, while global patterns of co-occurrence are much less – if at all – implicated in the acquisition of structured semantic representations. In this project we are interested in testing the idea that this dichotomy of local versus global statistics fails to consider the possibility that local and global statistics are interrelated in the linguistic input and might engage the same learning process at different time scales. To address this question we conduct corpus analysis and modeling of the hypothesized learning patterns, as well as behavioral work with adults and children.