Improving String Processing for Temporal Relations

摘要

This paper presents a refinement of the superposition operation on strings which are used to represent temporal relation information such as is found in documents annotated with TimeML. Superposition is made demonstrably more efficient by interleaving generation with testing, rather than generating and then testing. The strings offer compact visual appeal while remaining an attractive option for computation and reasoning. Motivated by Freksa's semi-interval relations, a suggestion is also made for a potential method of representing partial information in these strings so as to allow for analysis at different granularities, and for more flexibility when dealing with cases of ambiguity.