Know your audience: specializing grounded language models with listener subtraction

TL;DR

This paper introduces a method for adapting grounded language models to specific audiences by exploiting differences in listener knowledge, enabling context-dependent language specialization without direct supervision.

Contribution

The paper presents a novel contrastive multi-agent training approach that fine-tunes a speaker model to adapt to different listeners' knowledge using rewards only.

Findings

Speaker can adapt to different listeners' knowledge through training.

Zero-shot transfer of listener-specific language to real-world data.

Method enables language models to specialize without explicit supervision.

Abstract

Effective communication requires adapting to the idiosyncrasies of each communicative context--such as the common ground shared with each partner. Humans demonstrate this ability to specialize to their audience in many contexts, such as the popular game Dixit. We take inspiration from Dixit to formulate a multi-agent image reference game where a (trained) speaker model is rewarded for describing a target image such that one (pretrained) listener model can correctly identify it among distractors, but another listener cannot. To adapt, the speaker must exploit differences in the knowledge it shares with the different listeners. We show that finetuning an attention-based adapter between a CLIP vision encoder and a large language model in this contrastive, multi-agent setting gives rise to context-dependent natural language specialization from rewards only, without direct supervision. Through controlled experiments, we show that training a speaker with two listeners that perceive differently, using our method, allows the speaker to adapt to the idiosyncracies of the listeners. Furthermore, we show zero-shot transfer of the specialization to real-world data. Our experiments demonstrate a method for specializing grounded language models without direct supervision and highlight the interesting research challenges posed by complex multi-agent communication.