CommonsenseVIS: Visualizing and Understanding Commonsense Reasoning Capabilities of Natural Language Models

Xingbo Wang, Renfei Huang, Zhihua Jin, Tianqing Fang, Huamin Qu

Room: 106

2023-10-25T00:21:00ZGMT-0600Change your timezone on the schedule page
Exemplar figure, described by caption below
Since commonsense knowledge is not explicitly stated, it is challenging to conduct a scalable analysis of what commonsense knowledge NLP models do (not) learn. We employ a knowledge graph to derive implicit commonsense in the model input as context information. Then, we use it to align model behavior with human reasoning through multi-level interactive visualizations. Thereafter, users can understand, diagnose, and edit specific knowledge areas where models do not perform well.
Fast forward
Full Video

Commonsense reasoning, visual analytics, XAI, natural language processing


Recently, large pretrained language models have achieved compelling performance on commonsense benchmarks. Nevertheless, it is unclear what commonsense knowledge the models learn and whether they solely exploit spurious patterns. Feature attributions are popular explainability techniques that identify important input concepts for model outputs. However, commonsense knowledge tends to be implicit and rarely explicitly presented in inputs. These methods cannot infer models’ implicit reasoning over mentioned concepts. We present CommonsenseVIS, a visual explanatory system that utilizes external commonsense knowledge bases to contextualize model behavior for commonsense question-answering. Specifically, we extract relevant commonsense knowledge in inputs as references to align model behavior with human knowledge. Our system features multi-level visualization and interactive model probing and editing for different concepts and their underlying relations. Through a user study, we show that CommonsenseVIS helps NLP experts conduct a systematic and scalable visual analysis of models’ relational reasoning over concepts in different situations.