Abstract
We introduce a method for learning to extract vocabulary and encyclopedic information to assist second language (L2) learners acquiring deep knowledge of target vocabulary. In our approach, grammar patterns, collocations, representative examples are extracted, aimed at providing rich lexical information for any target words. The method involves word sense disambiguation on target words, automatically parsing the sentences in a large-scale corpus, automatically generating grammar patterns, collocations, examples, and quizzes for every target word, and automatically linking named entities to corresponding Wikipedia information. We present a prototype vocabulary learning system, Linggle Booster, that applies the method to corpora and web pages. Evaluation on a set of target words shows that the method has reasonably good performance in terms of generating useful and correct information for vocabulary learning.