Identifying Nonlocal Dependencies in Incremental Parsing

Abstract

SUMMARYThis paper describes a method of identifying nonlocal dependencies in incremental parsing. Our incremental parser inserts empty elements at arbitrary positions to generate partial parse trees including empty elements. To identify the correspondence between empty elements and their fillers, our method adapts a hybrid approach: slash feature annotation and heuristic rules. This decreases local ambiguity in incremental parsing and improves the accuracy of our parser.

Transition-Based Left-Corner Parsing for Identifying PTB-Style Nonlocal Dependencies

“…Figure 10 shows the resolution rules. These heuristic resolution rules are similar to the previous work (Campbell, 2004;Kato and Matsubara, 2015), which also utilizes c-command relation. An important difference is that we design heuristic rules not for fully-connected parse tree but for stack of parse trees derived by left-corner parsing.…”

“…Several methods of identifying nonlocal dependencies have been proposed so far. These methods can be divided into three approaches: pre-processing approach (Dienes and Dubey, 2003b), in-processing approach (Dienes and Dubey, 2003a;Schmid, 2006;Cai et al, 2011;Kato and Matsubara, 2015) and post-processing approach (Johnson, 2002;Levy and Manning, 2004;Campbell, 2004;Xue and Yang, 2013;Xiang et al, 2013;Takeno et al, 2015). 1 In preprocessing approach, a tagger called "trace tagger" detects empty elements.…”

Left-Corner Parsing for Identifying PTB-Style Nonlocal Dependencies

“…These heuristic resolution rules are similar to those described in the previous work (Campbell 2004;Kato and Matsubara 2015), which also utilize the c-command relation. However, whereas previous rules were designed for fully-connected parse trees, our heuristic rules apply to stacks of parse trees derived by left-corner parsing.…”

“…Previous pre-and in-processing approaches yield relatively low constituent parsing accuracies, probably because parsing models based on these approaches are probabilistic context-free grammars, which cannot easily capture the global features. Finally, we compared our method with that proposed by Kato and Matsubara (2015), which is similar to ours in many respects. Both methods augment the Penn Treebank with nonlocal dependency information, and both employ a transition-based parser.…”