Reading Between the Lines: The One-Sided Conversation Problem

TL;DR

This paper introduces the one-sided conversation problem (1SC), focusing on reconstructing missing dialogue turns and generating summaries from limited transcripts, advancing privacy-aware conversational AI.

Contribution

It formalizes 1SC, evaluates prompting and finetuning methods on multiple datasets, and demonstrates promising results for privacy-preserving conversational AI tasks.

Findings

Access to future turns and utterance length improves reconstruction.

Placeholder prompting reduces hallucination in generated turns.

Large models excel with prompting; smaller models need finetuning.

Abstract

Conversational AI is constrained in many real-world settings where only one side of a dialogue can be recorded, such as telemedicine, call centers, and smart glasses. We formalize this as the one-sided conversation problem (1SC): inferring and learning from one side of a conversation. We study two tasks: (1) reconstructing the missing speaker's turns for real-time use cases, and (2) generating summaries from one-sided transcripts. Evaluating prompting and finetuned models on MultiWOZ, DailyDialog, and Candor with both human A/B testing and LLM-as-a-judge metrics, we find that access to one future turn and information about utterance length improves reconstruction, placeholder prompting helps to mitigate hallucination, and while large models generate promising reconstructions with prompting, smaller models require finetuning. Further, high-quality summaries can be generated without reconstructing missing turns. We present 1SC as a novel challenge and report promising results that mark a step toward privacy-aware conversational AI.