Key points are not available for this paper at this time.
Understanding narratives requires reading between the lines, which in turn, interpreting the likely causes and effects of events, even when they not mentioned explicitly. In this paper, we introduce Cosmos QA, a-scale dataset of 35, 600 problems that require commonsense-based reading, formulated as multiple-choice questions. In stark contrast to existing reading comprehension datasets where the questions focus on and literal understanding of the context paragraph, our dataset focuses reading between the lines over a diverse collection of people's everyday, asking such questions as "what might be the possible reason of. . .? ", or "what would have happened if. . . " that require reasoning beyond the text spans in the context. To establish baseline performances on Cosmos, we experiment with several state-of-the-art neural architectures for comprehension, and also propose a new architecture that improves over competitive baselines. Experimental results demonstrate a significant gap machine (68. 4%) and human performance (94%), pointing to avenues for research on commonsense machine comprehension. Dataset, code and is publicly available at https: //wilburone. github. io/cosmos.
Huang et al. (Sat,) studied this question.