Quoref: A Reading Comprehension Dataset with Questions Requiring Coreferential Reasoning
This addresses the problem of evaluating coreference resolution in machine reading comprehension for NLP researchers, though it is incremental as it builds on existing dataset creation methods.
The authors tackled the lack of complex coreferential reasoning in reading comprehension benchmarks by creating Quoref, a dataset with over 24K span-selection questions from Wikipedia paragraphs, and showed that state-of-the-art models perform at 70.5 F1, significantly below human performance of 93.4 F1.
Machine comprehension of texts longer than a single sentence often requires coreference resolution. However, most current reading comprehension benchmarks do not contain complex coreferential phenomena and hence fail to evaluate the ability of models to resolve coreference. We present a new crowdsourced dataset containing more than 24K span-selection questions that require resolving coreference among entities in over 4.7K English paragraphs from Wikipedia. Obtaining questions focused on such phenomena is challenging, because it is hard to avoid lexical cues that shortcut complex reasoning. We deal with this issue by using a strong baseline model as an adversary in the crowdsourcing loop, which helps crowdworkers avoid writing questions with exploitable surface cues. We show that state-of-the-art reading comprehension models perform significantly worse than humans on this benchmark---the best model performance is 70.5 F1, while the estimated human performance is 93.4 F1.