Jungyeul Park

Alignment-based projection offers a surprisingly effective way to fix broken Chinese word boundaries in noisy text, outperforming direct segmentation and stabilizing annotation pipelines.

Despite superior head prediction accuracy, learned heads fail to consistently improve constituency parsing, especially when evaluated on punctuation, challenging the assumption that better headedness directly translates to better parsing.

Arc-standard dependency parsing isn't just about graphs; it's secretly building ordered trees, unlocking a new perspective on projectivity and parsing.

Bock's 1971 minimum spanning tree algorithm, a cornerstone for non-projective dependency parsing, gets a modern makeover, making it finally understandable and implementable.

Learning constituent headedness as a supervised task achieves near-ceiling accuracy and significantly boosts constituency-to-dependency conversion compared to traditional rule-based approaches.