Home | People | Projects | Publications | Resources
TRIPS Architecture | TRIPS Movies | TRIPS Domains | TRAINS vs. TRIPS
The Parser is a best-first bottom-up chart parser similar to that described in Natural Language Understanding (Allen '95). The grammar and lexicon encode both syntactic and semantic features which are used to disambiguate and produce a logical form representation. Because speaker independent speech recognition can introduce errors and because no grammar is perfect, the Parser component does very robust parsing. Even if a full parse is not achieved, the Parser will create a logical form for all constituents the chart parser was able to find. Many times, this will give the system enough information to make an intelligent reply and to continue the dialogue, even though not all information was conveyed.
From this logical form, the Parser then uses linguistic clues to find a set of plausible speech acts which could describe the input. This set of speech acts is then later interpreted in context by the Interpretation Manager.
Grammar and Lexicon
Here at Rochester, we believe that there is a core set of language that is relevant to any practical dialogue, regardless of the domain. In the TRIPS project, we are trying to come up with a core dialogue "engine" for practical dialogue, which can then be supplemented domain specific information where needed for a particular domain.
In this vein, the grammar and lexicon are comprised of domain dependent and a domain independent pieces. The domain independent piece is a set of core English which seems, so far, to be independent of any domain. We then merge with it a domain dependent piece which has more of the specific predicates, etc. of the specific domain. Using this process, we have been able to "port" TRIPS to several different domains. (See here for details)