Realistic Evaluation Principles for Cross-document Coreference Resolution
This addresses evaluation issues for researchers in natural language processing, but it is incremental as it focuses on methodology rather than new models.
The paper tackles the problem of inflated results in cross-document coreference resolution evaluation by proposing two principles: using predicted mentions instead of gold mentions and avoiding exploitation of synthetic topic structures, which reduces a competitive model's F1 score by 33 points.
We point out that common evaluation practices for cross-document coreference resolution have been unrealistically permissive in their assumed settings, yielding inflated results. We propose addressing this issue via two evaluation methodology principles. First, as in other tasks, models should be evaluated on predicted mentions rather than on gold mentions. Doing this raises a subtle issue regarding singleton coreference clusters, which we address by decoupling the evaluation of mention detection from that of coreference linking. Second, we argue that models should not exploit the synthetic topic structure of the standard ECB+ dataset, forcing models to confront the lexical ambiguity challenge, as intended by the dataset creators. We demonstrate empirically the drastic impact of our more realistic evaluation principles on a competitive model, yielding a score which is 33 F1 lower compared to evaluating by prior lenient practices.