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

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.

arXiv commentThis paper is accepted by IEEE VIS, 2023. To appear in IEEE Transactions on Visualization and Computer Graphics (IEEE TVCG). 14 pages, 11 figures

References in corpus (14)

• UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
• Training language models to follow instructions with human feedback
• Axiomatic Attribution for Deep Networks
• The What-If Tool: Interactive Probing of Machine Learning Models
• FACE: Feasible and Actionable Counterfactual Explanations
• Learning the Difference that Makes a Difference with Counterfactually-Augmented Data
• Visualizing and Measuring the Geometry of BERT
• Mass-Editing Memory in a Transformer
• ChatGPT is a Knowledgeable but Inexperienced Solver: An Investigation of Commonsense Problem in Large Language Models
• XNLI: Explaining and Diagnosing NLI-based Visual Data Analysis
• UnifiedQA-v2: Stronger Generalization via Broader Cross-Format Training
• MultiViz: Towards Visualizing and Understanding Multimodal Models
• ShortcutLens: A Visual Analytics Approach for Exploring Shortcuts in Natural Language Understanding Dataset
• Mediators: Conversational Agents Explaining NLP Model Behavior

Cited by in corpus (3)

• JailbreakLens: Visual Analysis of Jailbreak Attacks Against Large Language Models
• RAGTrace: Understanding and Refining Retrieval-Generation Dynamics in Retrieval-Augmented Generation
• SciDaSynth: Interactive Structured Data Extraction from Scientific Literature with Large Language Model