Test-time reasoning with LLMs over object descriptions and dialogue history improves coreference resolution on the SIMMC 2.1 dataset and generalizes better to unseen scenarios and novel objects than supervised encoder-based methods.
The dataset features complex scenes with an average of 19.7 objects arranged realistically, making it an ideal testbed for robust coreference resolution
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Reasoning over Object Descriptions Improves Coreference Resolution in Task-Based Dialogue Systems
Test-time reasoning with LLMs over object descriptions and dialogue history improves coreference resolution on the SIMMC 2.1 dataset and generalizes better to unseen scenarios and novel objects than supervised encoder-based methods.