Reasoning over Object Descriptions Improves Coreference Resolution in Task-Based Dialogue Systems

2026-04-30

AI Summary

This work addresses the challenges of coreference resolution in task-oriented dialogue systems under cross-domain settings, where existing approaches suffer from poor generalization and reliance on supervised models prone to overfitting. The study introduces, for the first time, test-time step-by-step reasoning to this task and proposes a unimodal method that requires no additional training. By leveraging large language models to integrate structured object metadata with dialogue history and employing prompt engineering for few-shot inference, the approach achieves substantial performance gains over encoder-based supervised models on the SIMMC 2.1 dataset. Moreover, it demonstrates superior robustness and generalization capabilities in cross-domain evaluations.

Abstract

Task-based dialogue systems assist users in achieving specific goals, such as executing actions or retrieving information, through natural language interactions. Accurate coreference resolution is essential, as it involves identifying object references within the dialogue - a task that becomes increasingly challenging in visually grounded environments characterized by complex scenes and diverse object metadata. However, coreference resolution in task-based dialogue remains limited by poor generalization across domains and heavy reliance on supervised models that often overfit to dataset-specific artifacts. In this work, we propose a unimodal test-time reasoning approach that enables large language models (LLMs) to reason over detailed object metadata and dialogue history to improve coreference resolution. Empirical results on the SIMMC 2.1 dataset demonstrate that LLMs can generate step-by-step reasoning processes that effectively align dialogue context with objects present in the scene. Extensive experiments highlight the models' ability to link conversations and objects accurately. Moreover, we show that test-time reasoning under few-shot settings generalizes effectively to unseen scenarios and novel objects, outperforming encoder-based supervised methods in cross-domain evaluations. These findings underscore the critical role of structured metadata and careful prompt engineering in enhancing the robustness and generalization of task-oriented dialogue systems.

Problem

• coreference resolution
• task-based dialogue systems
• visually grounded environments
• cross-domain generalization
• object metadata

Innovation

• test-time reasoning
• coreference resolution
• large language models
• structured metadata
• few-shot generalization