Eating Your Own Cooking: Automatically Linking Wordnet Synsets of Two Languages

Linked wordnets are invaluable linked lexical resources. Wordnet linking involves matching a particular synset (concept) in one wordnet to a synset in another wordnet. We have developed an automatic wordnet linking system that is divided into a number of stages. Starting with a synset in the first language (also referred to as the source language), our algorithm generates a list of candidate synsets in the second language (also referred to as the target language). In consecutive stages, a heuristic is used to prune and rank this list. The winner synset is then chosen as the linkage for the source synset. The candidate synsets are generated using a bilingual dictionary (BiDict). Further, the earlier heuristics which we developed used BiDict to rank these candidate synsets. However, development of a BiDict is cumbersome and requires human labor. Furthermore, in several cases sparsity of the BiDict handicaps the ranking algorithm to a great extent. We have thus devised heuristics to eliminate the requirement of BiDict during the ranking process by using the already linked synsets. Once sufficient number of linked synsets are available, these heuristics outperform our heuristics which use a BiDict. These heuristics are based on observations made from linking techniques applied by lexicographers. Our wordnet linking system can be used for any pair of languages, given either a BiDict or sufficient number of already linked synsets. The interface of the system is easy to comprehend and use. In this paper, we present this interface along with the developed heuristics.