GLUCOSE: GeneraLized and COntextualized Story Explanations

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 4569–4586, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 4569 GLUCOSE: GeneraLized and COntextualized Story Explanations Nasrin Mostafazadeh∗ Aditya Kalyanpur Lori Moon David Buchanan† Lauren Berkowitz Or Biran Elemental CognitionJennifer Chu-Carroll New York, NY, USA nasrin@verneek.com {adityak, lorim, orb, jenniferc}@elementalcognition.com david.buchanan@quillbot.com Abstract When humans read or listen, they make im- plicit commonsense inferences that frame their understanding of what happened and why. As a step toward AI systems that can build sim- ilar mental models, we introduce GLUCOSE, a large-scale dataset of implicit commonsense causal knowledge, encoded as causal mini- theories about the world, each grounded in a narrative context. To construct GLUCOSE, we drew on cognitive psychology to identify ten dimensions of causal explanation, focus- ing on events, states, motivations, and emo- tions. Each GLUCOSE entry includes a story- specific causal statement paired with an infer- ence rule generalized from the statement. This paper details two concrete contributions. First, we present our platform for effectively crowd- sourcing GLUCOSE data at scale, which uses semi-structured templates to elicit causal ex- planations. Using this platform, we collected a total of ˜670K specific statements and gen- eral rules that capture implicit commonsense knowledge about everyday situations. Second, we show that existing knowledge resources and pretrained language models do not include or readily predict GLUCOSE’s rich inferential content. However, when state-of-the-art neural models are trained on this knowledge, they can start to make commonsense inferences on un- seen stories that match humans’ mental mod- els. 1 Introduction Humans make countless implicit commonsense in- ferences about everyday situations. For example, consider the following short story from the ROC- Stories corpus (Mostafazadeh et al., 2016): Gage was riding his bike. A car turned in front of him. Gage turned his bike sharply. He fell off of his ∗Current affiliation Verneek, Inc. †Current affiliation QuillBot Inc. bike. Gage skinned his knee. When even young children read this story, they construct a coherent representation of what happened and why, combin- ing information from the text with relevant back- ground knowledge (Kintsch and Van Dijk, 1978). For example, they can construct the causal chain that explains how the car’s unexpected turn ulti- mately led to Gage falling, describe how Gage’s emotion and location changed throughout the story, and even hypothesize that he likely shouted for help after falling. Though humans build such mental models with ease (Zwaan et al., 1995), AI systems for tasks such as reading comprehension and dialogue remain far from exhibiting similar commonsense reasoning capabilities. Two major bottlenecks have been ac- quiring commonsense knowledge and successfully incorporating it into state-of-the-art AI systems. To address the first bottleneck, we have built an ef- fective platform to acquire causal commonsense knowledge at scale. To address the second, we show that pre-trained neural models can start to make similar inferences when trained on such rich curated data. We introduce the GLUCOSE1 (GeneraLized and COntextualized Story Explanations) dataset. Given a short story and a sentence X …

Content truncated to 500 words