Children learn a robust representation of lexical categories at a young age. We propose an incremental model of this process which efficiently groups words into lexical categories based on their local context using an information-theoretic criterion. We train our model on a corpus of childdirected speech from CHILDES and show that the model learns a fine-grained set of intuitive word categories. Furthermore, we propose a novel evaluation approach by comparing the efficiency of our induced categories against other category sets (including traditional part of speech tags) in a variety of language tasks. We show the categories induced by our model typically

Significant work is being done to develop NLP systems that can detect writing errors produced by non-native English speakers. A major issue, however, is the lack of available error-annotated training data needed to build statistical models that drive these major systems. As a result, many systems are trained on well-formed text with no modeling of typical errors that non-native speakers produce. To address this issue, we propose a novel method of using geographic region-specific web counts to detect typical errors in the writing of non-native speakers. In this paper we describe the approach, and present an analysis of the issues involved when using web counts. 1