Paper: Fast and Accurate Unlexicalized Parsing via Structural Annotations

ACL ID E14-4032
Title Fast and Accurate Unlexicalized Parsing via Structural Annotations
Venue Annual Meeting of The European Chapter of The Association of Computational Linguistics
Session Main Conference
Year 2014
Authors

We suggest a new annotation scheme for unlexicalized PCFGs that is inspired by formal language theory and only depends on the structure of the parse trees. We evaluate this scheme on the T?uBa-D/Z treebank w.r.t. several metrics and show that it improves both parsing accuracy and parsing speed considerably. We also show that our strategy can be fruitfully com- bined with known ones like parent annota- tion to achieve accuracies of over 90% la- beled F 1 and leaf-ancestor score. Despite increasing the size of the grammar, our annotation allows for parsing more than twice as fast as the PCFG baseline.