EMNLP Findings2022
WordTies: Measuring Word Associations in Language Models via Constrained Sampling
Abstract
Word associations are widely used in psychology to provide insights on how humans perceive and understand concepts. Comparing word associations in language models (LMs) to those generated by human subjects can serve as a proxy to uncover embedded lexical and commonsense knowledge in language models. While much helpful work has been done applying direct metrics, such as cosine similarity, to help understand latent spaces, these metrics are symmetric, while human word associativity is asymmetric. We propose WordTies, an algorithm based on constrained sampling from LMs, which allows an asymmetric measurement of associated words, given a cue word as the input. Comparing to existing methods, word associations found by this method share more overlap with associations provided by humans, and observe the asymmetric property of human associations. To examine possible reasons behind associations, we analyze the knowledge and reasoning behind the word pairings as they are linked to lexical and commonsense knowledge graphs. When the knowledge about the nature of the word pairings is combined with a probability that the LM has learned that information, we have a new way to examine what information is captured in LMs.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics cites Plug and Play Language Models: A Simple Approach to Controlled Text Generation
- COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics cites CommonGen: A Constrained Text Generation Challenge for Generative Commonsense Reasoning
- COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics cites Abductive Commonsense Reasoning
Builds on7
- Extracting Training Data from Large Language ModelsNicholas Carlini, Florian Tramèr, Eric Wallace, Matthew Jagielski et al.USENIX Security 2021 · 2,866 citations
- Plug and Play Language Models: A Simple Approach to Controlled Text GenerationSumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung et al.ICLR 2020 · 1,166 citations
- Abductive Commonsense ReasoningChandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi et al.ICLR 2020 · 521 citations
- CommonGen: A Constrained Text Generation Challenge for Generative Commonsense ReasoningBill Yuchen Lin, Wangchunshu Zhou, Ming Shen, Pei Zhou et al.EMNLP Findings 2020 · 239 citations
- COLD Decoding: Energy-based Constrained Text Generation with Langevin DynamicsLianhui Qin, Sean Welleck, Daniel Khashabi, Yejin ChoiNeurIPS 2022 · 217 citations
Related papers
- On the Consistency of Commonsense in Large Language ModelsGuozheng Li, Peng Wang, Wenjun Ke, Zijie Xu et al.ACL Findings 2025
- Towards Faithful Knowledge Graph Explanation Through Deep Alignment in Commonsense Question AnsweringWeihe Zhai, Arkaitz Zubiaga, Bingquan Liu, Chengjie Sun et al.EMNLP 2024
- Fusing Context Into Knowledge Graph for Commonsense Question AnsweringYichong Xu, Chenguang Zhu, Ruochen Xu, Yang Liu et al.ACL Findings 2021 · 62 citations
- Comparing Moral Values in Western English-speaking societies and LLMs with Word AssociationsChaoyi Xiang, Chunhua Liu, Simon De Deyne, Lea FrermannACL 2025 · 5 citations
- Data-Efficient Concept Extraction from Pre-trained Language Models for Commonsense Explanation GenerationYanbo Fang, Yongfeng ZhangEMNLP Findings 2022 · 5 citations
