 |
|
The goal of the first phase of the FrameNet project was to create a lexicon that recorded the semantic and syntactic combinatory possibilities of English words, supported by evidence taken from a large electronic corpus and based on a descriptive framework known as Frame Semantics. At the heart of the FrameNet database is a collection of semantically annotated sentences, each of them exemplifying a way of realizing the semantic/syntactic Valences of the lexical item in question. The "entries" are derived automatically by an algorithm which assembles information from the annotations into displays of all recorded valence possibilities. The FrameSQL interface makes it possible to summarize for individual words (or for semantically related groups of words) the manner in which particular semantic relations are syntactically related, the semantic functions expressed by particular syntactic patterns, and so on. Most of the words examined in FrameNet I were verbs and adjectives and deverbal nominalizations.
The goal of the second phase of the project - FrameNet II - is to continue building the lexical database, greatly expanding both the number of frames and also the lexical units, by moving from the "governing" words to "dependent" words (the words that participate in phrases that fill the slots created by the governing words). Also, in the second phase, we plan to work on evaluating our results by testing their potential to improve performance in a variety of Natural Language Processing applications, including word sense disambiguation, automatic translation, and information extraction.
Back to main page