Probing the Robustness of Coreference Resolution Systems
Abstract
Coreference resolution is the process of resolving the coreference between referring expressions, which is critical to discourse analysis and high-level NLP applications. With the advent of deep learning, it is shown that one can process coreference with minimal linguistic resources. In this talk, I will walk through the history of coreference resolution. Then, I will present my recent work challenging on the robustness of off-the-shelf coreference resolution systems. Our finding shows that the state-of-the-art neural system decreases dramatically over rule-based systems, which indicates that linguistic features are prone to harness the robustness onto coreference resolution.
