Structured representations for coreference resolution
Sebastian Martschat (Heidelberg)

Coreference resolution is the task of determining which mentions in a text are used to refer to the same entity. Inherently, coreference resolution is a structured task, as the output consists of sets of coreferring expressions. This complex structure poses challenges for model development and error analysis.

In this talk, we present machine learning and error analysis frameworks for coreference resolution that account for the structure. Our machine learning framework yields a unified representation of approaches, ranging from simple mention pair models to sophisticated entity-centric approaches. We employ the error analysis framework to perform an in-depth analysis and comparison of a wide range of approaches on a benchmark data set.