Correct-Detect: Balancing Performance and Ambiguity Through the Lens of Coreference Resolution in LLMs

2025-09-17

AI Summary

This paper identifies an inherent performance trade-off in large language models (LLMs) between coreference resolution and ambiguity detection: while LLMs can be individually optimized for either task, they struggle to excel at both simultaneously. To formalize this tension, we introduce the “CORRECT-DETECT trade-off”—the first systematic characterization of the intrinsic conflict between referential accuracy and ambiguity awareness in LLMs. Using a minimal-prompt paradigm, we evaluate both tasks jointly within a unified framework and conduct empirical analysis on human-annotated data. Experiments across mainstream LLMs reveal a significant negative correlation between coreference resolution and ambiguity detection performance, with no model achieving Pareto-optimality on both. This finding exposes a fundamental limitation in LLMs’ deep semantic modeling of referential structure and provides critical theoretical insight and empirical evidence for designing robust coreference resolution systems.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs) are intended to reflect human linguistic competencies. But humans have access to a broad and embodied context, which is key in detecting and resolving linguistic ambiguities, even in isolated text spans. A foundational case of semantic ambiguity is found in the task of coreference resolution: how is a pronoun related to an earlier person mention? This capability is implicit in nearly every downstream task, and the presence of ambiguity at this level can alter performance significantly. We show that LLMs can achieve good performance with minimal prompting in both coreference disambiguation and the detection of ambiguity in coreference, however, they cannot do both at the same time. We present the CORRECT-DETECT trade-off: though models have both capabilities and deploy them implicitly, successful performance balancing these two abilities remains elusive.

Problem

• LLMs struggle with coreference resolution ambiguity
• Models cannot simultaneously detect and resolve ambiguities
• Balancing disambiguation and ambiguity detection remains challenging

Innovation

• Coreference resolution balancing performance and ambiguity
• LLMs detect ambiguity with minimal prompting
• CORRECT-DETECT trade-off manages dual capabilities implicitly