Language models show sensitivity to subtle words like just and only
Not Just an Evaluation: A Case Study of Discourse Particle Sensitivity in Language Models
Computation and Language
Summary
Some small words like 'just' and 'only' change how we understand sentences without affecting their basic truth. These words can mean different things depending on context and are tricky for language models to handle. The paper shows that large language models (LLMs) respond to these words similarly to humans, especially when guided by instructions. The researchers also identified parts of the models that focus on these words and how the models represent their meanings. This helps us better understand how AI processes subtle language cues.
What this means in practice
- •For language technology developers: Improve natural language understanding by refining how models handle subtle words shaping conversational meaning.
- •For dialogue system designers: Enhance dialogue agents' responses by integrating better sensitivity to discourse particles like just and only.
Authors
William Sheffield, Jiahui Chen, Amy Zhang, Alex Warstadt, Junyi Jessy Li, Kanishka Misra
Abstract
Discourse particles subtly shape the flow of discourse by reflecting conversational context and interlocutor attitudes, often leaving the truth-conditional meaning untouched. They are also often polyfunctional, with distinct senses which partially overlap with other particles; for example, the exclusive use of English "just" and "only" ("John only/just drinks water"). Despite the frequency of these words in natural language, they have gone largely unstudied due to the subtle nature of their effects. We investigate LLM sensitivity to "just" and "only" in natural conversations using behavioral and mechanistic analyses. First, we find that LLMs and humans similarly prefer the presence of these particles in context, and both are more sensitive to "just" compared to "only". Second, when sampling from LLMs with and without these particles, we find LLMs consistently rate generations as more likely, given their source context, demonstrating their particle sensitivity. In both experiments, instruction tuning increases sensitivity. Finally, we localize particle-sensitive units within LLMs and find the overlap between "just" and "only" is asymmetric: "only" units transfer selectively to exclusive "just", while all "just" units transfer to "only" equally well regardless of sense, suggesting LLM lexical-level particle representations are partially organized by sense. We release our code, sense-labeled dialogue stimuli, and human preference annotations at https://github.com/JiahuiKChen/Not_Just_an_Evaluation.