Automatic detection of semantically primitive words using their reachability in an explanatory dictionary
Grigori Sidorov, Alexander F. Gelbukh · 2002
We suggest the method that permits building a set of candidates to be considered semantic primitives from the standard explanatory dictionary. Our method is based on the frequencies of the words that are reachable in a semantic network constructed from the dictionary. The method implements word sense disambiguation techniques, network construction, and reachability analysis. In part of word sense disambiguation we use an improved Lesk's algorithm. In the part of analysis of reachability we show that the words to which our algorithm assigns high weight, are plausible candidates to be semantic primitives. It is also shown that better candidates to semantic primitives should be included in short vicious cycles, which is detected by our algorithm. We applied the method to a rather large Spanish explanatory dictionary.