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

TL;DR

CommonsenseVIS is a visual tool that leverages external knowledge bases to analyze and interpret how large language models perform commonsense reasoning, addressing the challenge of understanding implicit knowledge in model outputs.

Contribution

It introduces a novel visualization system that contextualizes model behavior with external commonsense knowledge, enabling systematic analysis of implicit reasoning in NLP models.

Findings

Helps NLP experts analyze model reasoning over concepts.

Facilitates understanding of implicit commonsense knowledge.

Enhances interpretability of model decisions through visualization.

Abstract

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.