"Don't Say It!": Constraints, Compliance, and Communication when Language Models Play Taboo
2026-07-01 • Computation and Language
Computation and Language
AI summaryⓘ
The authors studied how well large language models (LLMs) can play the game Taboo, where you describe a word without using certain forbidden words. They tested two models by changing how the models generate their descriptions and checked if they followed the rules and if people or machines could guess the target words. Their results show that while the models try to balance following the rules and being clear, they are still much worse than humans at this task. This suggests that current models struggle to use words correctly when there are strict limits.
Taboo gamelarge language modelslexical constraintspromptinggeneration-time constraintsinternal representationforbidden word violationcommunicative effectivenesslexical grounding
Authors
Sara Candussio, Francesca Padovani, Daniel Scalena, Malvina Nissim
Abstract
The game of Taboo requires describing a target word without using a set of forbidden words, so that other players can guess it. This deceptively simple task combines strict lexical constraints with the need for communicatively effective descriptions, making it a compelling playground for examining how LLMs navigate competing demands at inference time. We evaluate two open-weight models under conditions that intervene at progressively deeper levels of the generative process, from prompting to generation-time constraints to internal representations manipulations. We assess their outputs through forbidden word violation detection, LLM-as-a-judge measuring the degree to which generated descriptions successfully evoke the target concept for both human and machine guessers, and examining whether the strategies models adopt under constraint align with those of human players. Our results show that compliance with the rules of the game and communicative effectiveness trade off differently across conditions, and that models remain substantially weaker than humans as guessers, suggesting that lexical grounding under constraint is an open challenge for current language models.