Measuring and Improving Consistency in Pretrained Language Models

• Published in: Transactions of the Association for Computational Linguistics Vol. 9; pp. 1012 - 1031
• Main Authors: Elazar, Yanai, Kassner, Nora, Ravfogel, Shauli, Ravichander, Abhilasha, Hovy, Eduard, Schütze, Hinrich, Goldberg, Yoav
• Format: Journal Article
• Language: English
• Published: One Rogers Street, Cambridge, MA 02142-1209, USA MIT Press 01.01.2021; MIT Press Journals, The; The MIT Press
• Subjects: Accuracy; Alternations; Consistency; Knowledge; Knowledge representation; Language; Language modeling; Linguistics; Natural language; Natural language processing; Paraphrase; Queries; Variance analysis
• ISSN: 2307-387X; 2307-387X
• DOI: 10.1162/tacl_a_00410

of a model—that is, the invariance of its behavior under meaning-preserving alternations in its input—is a highly desirable property in natural language processing. In this paper we study the question: Are Pretrained Language Models (PLMs) consistent with respect to factual knowledge? To this end, we create 🤘, a high-quality resource of cloze-style query English paraphrases. It contains a total of 328 paraphrases for 38 relations. Using 🤘, we show that the consistency of all PLMs we experiment with is poor— though with high variance between relations. Our analysis of the representational spaces of PLMs suggests that they have a poor structure and are currently not suitable for representing knowledge robustly. Finally, we propose a method for improving model consistency and experimentally demonstrate its effectiveness.