IIS: Rapid Development of a Frame Semantic Lexicon

International Computer Science Institute, Berkeley CA

Investigators

Abstract

The FrameNet Project at the International Computer Science Institute is building a uniquely detailed lexicon of English, both readable by humans and usable by computers for natural language processing (NLP). It is based on the theory of Frame Semantics, originated by Charles Fillmore, which recognizes that the meanings of many words are best understood in relation to semantic frames defining the set of participants and props in an event or state. For example, in "Alicia boiled the potatoes for 20 minutes", the word "boiled" evokes a frame involving Alicia as the Cook, the potatoes as the Food, and 20 minutes as the Duration---but also some kind of Container (a pot?) and some Heating instrument (a stove?)--these are not mentioned, but are essential parts of the frame. Frames range from very simple, like Placing, to very complex, like Revenge. Since 1997, the FrameNet project has been describing semantic frames (currently > 750, containing roughly 10,000 word senses), and annotating sentences which evoke them (> 135,000), showing which words play which roles in the frame. Annotated sentences are then automatically tabulated to produce a lexical entry exemplifying all possible syntactic patterns for each frame-evoking word. This lexicon is used in hundreds of NLP research projects around the world; and similar projects are underway for other languages. This methodology is, however, very labor-intensive, requiring careful examination of sentences in the corpus to guarantee that all semantic and syntactic combinations are extracted and annotated. This award is for a three-year effort to integrate sophisticated tools for grouping corpus examples, (developed by Adam Kilgarriff), into the FrameNet extraction and annotation work, enabling staff to find syntactic patterns faster and annotate examples semi-automatically, thus rapidly increasing the number of words covered and making the FrameNet lexicon useful in more practical NLP applications. For further information, visit http://framenet.icsi.berkeley.edu.