Paper: Learning from errors: Using vector-based compositional semantics for parse reranking

ACL ID W13-3202
Title Learning from errors: Using vector-based compositional semantics for parse reranking
Venue Continuous Vector Space Models and their Compositionality
Session
Year 2013
Authors

In this paper, we address the problem of how to use semantics to improve syntac- tic parsing, by using a hybrid reranking method: a k-best list generated by a sym- bolic parser is reranked based on parse- correctness scores given by a composi- tional, connectionist classifier. This classi- fier uses a recursive neural network to con- struct vector representations for phrases in a candidate parse tree in order to classify it as syntactically correct or not. Tested on the WSJ23, our method achieved a statisti- cally significant improvement of 0.20% on F-score (2% error reduction) and 0.95% on exact match, compared with the state-of- the-art Berkeley parser. This result shows that vector-based compositional semantics can be usefully applied in syntactic pars- ing, and demonstrates the benefit...