Learning Hierarchies from Ambiguous Natural Language Data

Abstract

This paper presents a method for acquiring a semantic hierarchy and updating an incomplete hierarchy. The creation of a comprehensive hierarchy is one important step in constructing a system for translating Japanese texts into English. The hierarchy is used to bias the learning of rules that indicate the English translation of a Japanese verb. The task is particularly challenging because training examples are ambiguous in the sense that each of the attributes forming an example may have a set of potential values. Each value corresponds to a different word meaning. When acquiring a hierarchy from scratch, translation rules are learned by an inductive learning algorithm in the first step. A new hierarchy is then generated by applying a clustering method to internal disjunctions of the learned rules and new rules are learned biased by this hierarchy. When updating an existing manually-constructed hierarchy, we take advantage of its node structure. We report experimental results showing that the semantic hierarchies generated by our method yield learned translation rules with higher average accuracy.