Entity binding failures in speech LLMs: diagnosis and CoT intervention

Arxiv CS.CL (NLP/language models)

The paper reveals that entity binding failures are a key modality gap in speech LLMs, with speech-to-text reasoning matching or exceeding text in other areas. Evaluating three diverse SLLMs, the authors propose a chain-of-thought intervention to improve entity binding.

llmchain-of-thought

Arxiv CS.CL (NLP/language models)·Ming-Hao HsuXiaohai TianJun ZhangZhizheng Wu

7 days ago