Understanding rich narratives, such as dialogues and stories, often requires natural language processing systems to access relevant knowledge from commonsense knowledge graphs. However, these systems typically retrieve facts from KGs using simple heuristics that disregard the complex challenges of identifying situationally-relevant commonsense knowledge (e.g., contextualization, implicitness, ambiguity). In this work, we propose the new task of commonsense fact linking, where models are given contexts and trained to identify situationally-relevant commonsense knowledge from KGs. Our novel benchmark, ComFact, contains ~293k in-context relevance annotations for commonsense triplets across four stylistically diverse dialogue and storytelling datasets. Experimental results confirm that heuristic fact linking approaches are imprecise knowledge extractors. Learned fact linking models demonstrate across-the-board performance improvements (~34.6% F1) over these heuristics. Furthermore, improved knowledge retrieval yielded average downstream improvements of 9.8% for a dialogue response generation task. However, fact linking models still significantly underperform humans, suggesting our benchmark is a promising testbed for research in commonsense augmentation of NLP systems.

Contextual knowledge is essential in answering questions given specific observations. While recent approaches to building commonsense knowledge basesvia text mining and/or crowdsourcing are successful,contextual knowledge is largely missing. To addressthis gap, this paper presents SocialExplain, a novel approach to acquiring contextual commonsense knowledge from explanations of social content. The acquisition process is broken into two cognitively simple tasks:to identify contextual clues from the given social content, and to explain the content with the clues. An experiment was conducted to show that multiple piecesof contextual commonsense knowledge can be identi-fied from a small number of tweets. Online users verified that 92.45% of the acquired sentences are good,and 95.92% are new sentences compared with existingcrowd-sourced commonsense knowledge bases.