Multi-Sense Embeddings for Semantic Role Labeling
Yong Zuo (1651134)
We present an approach to applying multi-sense embeddings to semantic role labeling. Semantic Role Labeling(SRL) is a fundamental natural language processing task to discover the predict-argument structure of a predicate in a given sentence. Although word embeddings are applicable widely to various natural language processing tasks, polysemy in natural language is always a difficult for obtaining fine-grained embeddings. “Multi-sense” method, which train one vector, or embedding for each sense of a word instead of each single words, could be a possible resolution to this problem. Therefore, we propose a predicate-based method for multi-sense embeddings by setting different hyperparameters for predicates and other words and employ it to semantic role labeling task.