Thursday, December 9, 2021 - 3:30pm to 5:00pm
Location:Virtual Presentation - ET Remote Access - Zoom
Speaker:YANG YANG, Masters Student /YANG%20YANG
Natural-Language Input for the Scone Knowledge-Base System
Recent advances in AI have brought various natural language processing tasks to an extremely high level, for example in text classification, translation, semantics analysis, etc. However, these techniques do not achieve any real natural language understanding, by which we mean the creation of a language-independent, concept-level representation of the meaning of the utterance. This thesis describes a prototype implementation of an NLU system, comprising a Construction Grammar (CxG) engine that processes incoming text statements and represents their meaning as new knowledge structures in the Scone knowledge base system (KBS).
Scone is an expressive, efficient, and extensible knowledge base system for storing symbolic representations of knowledge. Scone acts as a “smart memory” that supports simple inference, flexible search, and reasoning over knowledge in multiple contexts – distinct but overlapping world-models in the knowledge base. Construction grammar studies the pairing of linguistic forms and underlying meanings. This combination offers many advantages for NLU, which are illustrated by this proof-of-concept implementation. The current implementation only covers a limited subset of English forms, but it is able to understand simple sentences and, consequently, provides a practical way for users to put new knowledge into Scone. It also is designed to provide a foundation for future development in this area.
Scott E. Fahlman
Zoom Participation. See announcement.