Possible Stories: Evaluating Situated Commonsense Reasoning under Multiple Possible Scenarios

Abstract

The possible consequences for the same context may vary depending on the situation we refer to. However, current studies in natural language processing do not focus on situated commonsense reasoning under multiple possible scenarios. This study frames this task by asking multiple questions with the same set of possible endings as candidate answers, given a short story text. Our resulting dataset, Possible Stories, consists of more than 4.5K questions over 1.3K story texts in English. We discover that even current strong pretrained language models struggle to answer the questions consistently, highlighting that the highest accuracy in an unsupervised setting (60.2%) is far behind human accuracy (92.5%). Through a comparison with existing datasets, we observe that the questions in our dataset contain minimal annotation artifacts in the answer options. In addition, our dataset includes examples that require counterfactual reasoning, as well as those requiring readers' reactions and fictional information, suggesting that our dataset can serve as a challenging testbed for future studies on situated commonsense reasoning.

Benchmarks for Automated Commonsense Reasoning: A Survey

“…identifying the rhyme scheme in a poem or increasing the contrast in an image. 6 In the other direction, there can be evidence of various kinds, beyond performance on a commonsense benchmark, that an AI system does or does not use commonsense reasoning in carrying out a task. In a knowledge-based system, one can often see what knowledge is being used and how; if commonsense knowledge is being used in a significant way in carrying out a particular task, then presumably this is to some extent a commonsense task.…”

“…Size Construction PIQA [12] Physical interaction QA 20,000 questions Crowd sourcing Possible Stories [6] Counterfactual 1313 texts Crowd sourcing narratives 4533 questions PROST [5] Physical reasoning 18,736 questions Expert-written cloze task template. ProtoQA [15] Reasoning about 9700 questions Crowd sourcing prototypical situations ReCoRD [159] Cloze question 120,000 questions Extracted from about news stories online news source.…”