Measuring and Improving Consistency in Pretrained Language Models-Reference-Cited by-同舟云学术

Measuring and Improving Consistency in Pretrained Language Models

Published:2021 Issue: Volume:9 Page:1012-1031
ISSN:2307-387X
Container-title:Transactions of the Association for Computational Linguistics
language:en
Short-container-title:

Author:

Elazar Yanai¹²,Kassner Nora³,Ravfogel Shauli¹⁴,Ravichander Abhilasha⁵,Hovy Eduard⁶,Schütze Hinrich⁷,Goldberg Yoav¹⁸

Affiliation:

1. Computer Science Department, Bar Ilan University, Israel

2. Allen Institute for Artificial Intelligence, United States. yanaiela@gmail.com

3. Center for Information and Language Processing (CIS), LMU Munich, Germany. kassner@cis.lmu.de

4. Allen Institute for Artificial Intelligence, United States. shauli.ravfogel@;gmail.com

5. Language Technologies Institute, Carnegie Mellon University, United States. aravicha@cs.cmu.edu

6. Language Technologies Institute, Carnegie Mellon University, United States. hovy@cs.cmu.edu

7. Center for Information and Language Processing (CIS), LMU Munich, Germany

8. Allen Institute for Artificial Intelligence, United States. yoav.goldberg@gmail.com

Abstract

Abstract Consistency 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 ParaRel🤘, a high-quality resource of cloze-style query English paraphrases. It contains a total of 328 paraphrases for 38 relations. Using ParaRel🤘, 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.1

Publisher

MIT Press - Journals

Subject

Artificial Intelligence,Computer Science Applications,Linguistics and Language,Human-Computer Interaction,Communication

Link

https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00410/1975957/tacl_a_00410.pdf

Reference79 articles.

1. Synthetic QA corpora generation with roundtrip consistency;Alberti,2019

2. Easy cases of probabilistic satisfiability;Andersen;Annals of Mathematics and Artificial Intelligence,2001

3. VQA: Visual question answering;Antol,2015

4. Logic-guided data augmentation and regularization for consistent question answering;Asai,2020

5. What do neural machine translation models learn about morphology?;Belinkov,2017

Cited by 24 articles. 订阅此论文施引文献订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献

1. Scarecrows in Oz: The Use of Large Language Models in HRI;ACM Transactions on Human-Robot Interaction;2024-01-30

2. The Life Cycle of Knowledge in Big Language Models: A Survey;Machine Intelligence Research;2024-01-12

3. Leveraging Large Language Models for Structure Learning in Prompted Weak Supervision;2023 IEEE International Conference on Big Data (BigData);2023-12-15

4. Improved Convolutional Neural Network based Approach for Analyzing the ChatGPT and Different Fields Impact;2023 2nd International Conference on Automation, Computing and Renewable Systems (ICACRS);2023-12-11

5. Thus Spake ChatGPT;Communications of the ACM;2023-11-17