Integrating Syntactic and Semantic Tools in Sfy

Despite the large lexicon of even broad coverage Natural Language Processing systems, there are many missing lexical items. We describe two methodologically distinct approaches to augment the lexicon: the look-up approach and the shotgun approach. Both approaches are framed within Sfy, a new research program to pull together several broad coverage systems. The classic look-up approach requires compatible electronic sources (e.g. WordNet). Through system hooks into the source, we can pull as much relational and semantic information as Sfy requires. Since WordNet does not overtly provide Sfy with sufficient syntactic role and semantic information, we use the synonyms of an unknown target word that WordNet provides us with to create templates for a new entry in our lexicon. In a fully realistic model, we cannot always rely on the lookup approach to solve our lexical issues. We need another back-off method. Much like a person intuiting the part-ofspeech of a new term, an unknown word presents a syntactic hole to our parser. Only certain parts of speech will fill that hole. We can try to validate all the possible fillers by naively testing every part of speech. The subset of all these sentences that can be parsed informs us of exactly which parts of speech our unknown word can be. Interestingly, the shotgun approach also provides a means to solve the related problem of a familiar orthogrpahic word functioning in an unfamiliar part of speech.