Testing Pre-trained Language Models' Understanding of Distributivity via Causal Mediation Analysis

2022/09/11 by Pangbo Ban, Yifan Jiang, Ban, Pangbo +5
Computer Science · Mathematics · Psychology · #68T50 #Artificial Intelligence (cs.AI) #Artificial intelligence #Cognitive psychology #Computation and Language (cs.CL) #Computer science #Distributive property #Distributivity #ENCODE #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #I.2.7 #Inference #Linguistics #Machine Learning in Healthcare #Mathematics #Mediation #Natural language #Natural language processing #Psychology #Task (project management) #Topic Modeling #Vocabulary

paper · pdf · doi:10.48550/arxiv.2209.04761

openalex publication_date 2022/09/11 · openalex created_date 2022/09/14 · openalex updated_date 2026/07/28

Abstract

To what extent do pre-trained language models grasp semantic knowledge regarding the phenomenon of distributivity? In this paper, we introduce DistNLI, a new diagnostic dataset for natural language inference that targets the semantic difference arising from distributivity, and employ the causal mediation analysis framework to quantify the model behavior and explore the underlying mechanism in this semantically-related task. We find that the extent of models' understanding is associated with model size and vocabulary size. We also provide insights into how models encode such high-level semantic knowledge.

Related

• Lost in Inference: Rediscovering the Role of Natural Language Inference for Large Language Models
• Probing Natural Language Inference Models through Semantic Fragments
• Assessing the Limits of the Distributional Hypothesis in Semantic Spaces: Trait-based Relational Knowledge and the Impact of Co-occurrences
• CausalGym: Benchmarking causal interpretability methods on linguistic tasks
• CLEAR-3K: Assessing Causal Explanatory Capabilities in Language Models