Parsing with Compositional Vector Grammars

摘要: Natural language parsing has typically been done with small sets of discrete categories such as NP and VP, but this representation does not capture the full syntactic nor semantic richness linguistic phrases, attempts to improve on by lexicalizing phrases or splitting only partly address problem at cost huge feature spaces sparseness. Instead, we introduce a Compositional Vector Grammar (CVG), which combines PCFGs syntactically untied recursive neural network that learns syntactico-semantic, compositional vector representations. The CVG improves PCFG Stanford Parser 3.8% obtain an F1 score 90.4%. It is fast train implemented approximately efficient reranker it about 20% faster than current factored parser. soft notion head words performance types ambiguities require information PP attachments.