Talk Less, Interact Better: Evaluating In-context Conversational Adaptation in Multimodal LLMs

TL;DR

This paper introduces ICCA, a framework to evaluate whether multimodal large language models adapt their language efficiency during interactions, revealing current models' limitations in spontaneous conversational adaptation.

Contribution

The paper presents ICCA, a novel automated framework for assessing in-context conversational adaptation in multimodal large language models, highlighting their inability to naturally develop efficient communication.

Findings

Current MLLMs do not spontaneously increase communication efficiency over time.

Only some models like GPT-4 can be prompted to adapt their language efficiency.

This behavior does not emerge from existing training regimes.

Abstract

Humans spontaneously use increasingly efficient language as interactions progress, by adapting and forming ad-hoc conventions. This phenomenon has been studied extensively using reference games, showing properties of human language that go beyond relaying intents. It remains unexplored whether multimodal large language models (MLLMs) similarly increase communication efficiency during interactions, and what mechanisms they may adopt for this purpose. We introduce ICCA, an automated framework to evaluate such conversational adaptation as an in-context behavior in MLLMs. We evaluate several state-of-the-art MLLMs, and observe that while they may understand the increasingly efficient language of their interlocutor, they do not spontaneously make their own language more efficient over time. This latter ability can only be elicited in some models (e.g., GPT-4) with heavy-handed prompting. This shows that this property of linguistic interaction does not arise from current training regimes, even though it is a common hallmark of human language. ICCA is available at https://github.com/lil-lab/ICCA.