Robust Induction of Parts-of-Speech in Child-Directed Language by Co-Clustering of Words and Contexts

摘要

We introduce Conflict-Driven Co-Clustering, a novel algorithm for data co-clustering, and apply it to the problem of inducing parts-of-speech in a corpus of child-directed spoken English. Co-clustering is preferable to unidimensional clustering as it takes into account both item and context ambiguity. We show that the categorization performance of the algorithm is comparable with the co-clustering algorithm of Leibbrandt and Powers (2008), but out-performs that algorithm in robustly pruning less-useful clusters and merging them into categories strongly corresponding to the three main open classes of English.