🧑‍🔬 Dmitriy Dligach

Dmitriy Dligach works in computational lexical semantics, with a focus on how the meanings of words and the roles they play in sentences can be identified automatically and evaluated fairly. Their papers here centre on word sense disambiguation and semantic role labeling. Both papers are shared-task overviews: they design common benchmarks, assemble annotated English text from resources such as OntoNotes, PropBank, VerbNet and WordNet, and compare competing systems under one protocol.

The most cited of these is SemEval-2007 Task-17, which assessed sense tagging and role labeling together on Wall Street Journal text. It offered a fine-grained all-words track, a coarse-grained lexical sample built on OntoNotes senses, and a role-labeling track scored against both PropBank and VerbNet labels. The VerbNet labels were mapped from PropBank through SemLink. The paper also attached bootstrap confidence intervals to every score and reported results with a frequent, trivially easy predicate removed.

The later work, SemEval-2010 Task 14 on word sense induction, moves from fixed sense inventories to senses learned without supervision from raw web text. Systems induced sense clusters from one corpus and then applied them to separate test instances. This made the task a test of whether a clustering model carries over to unseen data. To expose biases in individual metrics, the evaluation paired V-Measure with paired F-Score and averaged supervised scores over several random splits.

Dligach is currently at Loyola University Chicago. Earlier bylines include the University of Colorado Boulder, the period of these lexical-semantics papers, and later Boston Children's Hospital and Harvard University. Co-authors on the papers here include Sameer Pradhan, Martha Palmer and Suresh Manandhar.

Dmitriy Dligach has 2 papers in our library, in 2007–2010, most of them on Word sense disambiguation, Semantic role labeling; often with Sameer Pradhan, Martha Palmer, Suresh Manandhar.