Automatic Mapping Lexical Resources: A Lexical Unit as the Keystone
Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC 2014)
Abstract
This paper presents the fully automatic linking of two valency lexicons of Czech verbs: VALLEX and PDT-VALLEX. Despite the same theoretical background adopted by these lexicons and the same linguistic phenomena they focus on, the fully automatic mapping of these resouces is not straightforward. We demonstrate that converting these lexicons into a common format represents a relatively easy part of the task whereas the automatic identification of pairs of corresponding valency frames (representing lexical units of verbs) poses difficulties. The overall achieved precision of 81% can be considered satisfactory. However, the higher number of lexical units a verb has, the lower the precision of their automatic mapping usually is. Moreover, we show that especially (i) supplementing further information on lexical units and (ii) revealing and reconciling regular discrepancies in their annotations can greatly assist in the automatic merging.