A Unified Prompt-based Framework with Label Semantic Expansion for Joint Multi-intent Detection and Slot Filling.
Yu Wang, Jinmao Xu, Miao Wang, Tianrui Li, Xuanren Qu · ACM Transactions on Asian and Low-Resource Language Information Processing · 2025
Natural Language Understanding (NLU), which includes both intent detection (ID) and slot filling (SF), is essential for extracting crucial textual information, significantly influencing subsequent tasks and applications. Traditionally, ID and SF were treated as separate tasks, but the trend has shifted toward joint models that can handle complex scenarios, including multiple intents and slots. Recent advancements in prompt learning provide a unified framework for this purpose, but current prompt-based approaches do not fully leverage the semantic richness of intent and slot labels, and the possibility of using prompt learning to model the correlations between ID and SF for multi-intent scenarios has yet to be fully explored. To overcome these limitations and align with the joint modeling approach, a novel unified prompt-based framework, LSE-NLU, is proposed, which encompasses five subtasks designed to detect multiple intents and fill slots concurrently. Additionally, experimental results and further analysis demonstrate that our proposed model achieves new state-of-the-art performance on two public multi-intent SLU datasets. Specifically, it achieves an overall accuracy improvement of 2.9% on the MixATIS dataset and 0.2% on the MixSNIPS dataset compared to previous best models, confirming its potential in enhancing NLU performance.