Recently, large pretrained language models achieve compelling performance on commonsense benchmarks. Nevertheless, it is unclear what commonsense knowledge the models learn and whether they purely exploit spurious patterns. Feature attributions are popular explainability techniques that identify important input concepts to model outputs. However, commonsense knowledge tends to be implicit and rarely explicitly presented in inputs. These methods cannot help infer models’ implicit reasoning over mentioned concepts.

In this thesis, we develop CommonsenseVIS, a visual explanatory system that utilizes external commonsense knowledge bases to contextualize model behavior for commonsense question-answering. Particularly, we extract relevant commonsense knowledge in inputs as references to align...[ Read more ]

Recently, large pretrained language models achieve compelling performance on commonsense benchmarks. Nevertheless, it is unclear what commonsense knowledge the models learn and whether they purely exploit spurious patterns. Feature attributions are popular explainability techniques that identify important input concepts to model outputs. However, commonsense knowledge tends to be implicit and rarely explicitly presented in inputs. These methods cannot help infer models’ implicit reasoning over mentioned concepts.

In this thesis, we develop CommonsenseVIS, a visual explanatory system that utilizes external commonsense knowledge bases to contextualize model behavior for commonsense question-answering. Particularly, we extract relevant commonsense knowledge in inputs as references to align model behavior with human knowledge. Our system features multi-level visualization and interactive probing of model behavior on different concepts and their underlying relations. Through case studies and 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.