A Galois Lattice based Approach to Lexical Inheritance Hierarchy Learning

Lexical inheritance hierarchies are used widely as a means of representing lexical information efficiently but there have been few attempts to construct them automatically. This paper presents a two-step construction algorithm in which a Galois lattice is built and then pruned into an inheritance hierarchy. The pruning step utilises a maximum entropy model. This is compared to a pruning method which prunes nodes in the hierarchy randomly. To asses the performance of the two methods an automatic evaluation method has been implemented which matches an automatically derived hierarchy to a manually built hierarchy for the same lexicon and defines precision and recall as a distance function between the two hierarchies.