作者: Eugene Charniak , Mark Johnson
关键词:
摘要: Discriminative reranking is one method for constructing high-performance statistical parsers (Collins, 2000). A discriminative reranker requires a source of candidate parses each sentence. This paper describes simple yet novel sets 50-best based on coarse-to-fine generative parser (Charniak, generates lists that are substantially higher quality than previously obtainable. We used these as the input to MaxEnt (Johnson et al., 1999; Riezler 2002) selects best parse from set sentence, obtaining an f-score 91.0% sentences length 100 or less.