Answers are grounded in this paper and cite their sources.
Multi-Intent Detection (MID) and Slot Filling (SF) are fundamental tasks in Natural Language Understanding (NLU) for goal-oriented dialogue systems. Despite advances in various joint models, which improve performance by learning interactions between intents and slots, existing models might not adequately capture intricate and complex relationships between intents in multi-intent utterances, especially if slot-level semantic information is not adequately utilized during intent detection (ID). To this end, this paper proposes a novel framework, namely Slot-Aware Semantic Signaling for Multi-Intent Classification (S3MIC), which is a joint framework that leverages slot-level semantic information to improve intent prediction in multi-intent dialogue systems. We conduct experiments with the proposed framework on benchmark datasets, namely MixATIS and MixSNIPS, and report consistent improvements over existing benchmark models in terms of Slot F1, Intent Accuracy, and Semantic Frame Accuracy (SeFr Acc). Specifically, our proposed framework achieves 52.5% SeFr Acc on MixATIS and 86.28% SeFr Acc on MixSNIPS, thereby validating the efficacy of slot-aware semantic signaling in improving joint MID and SF performance in goal-oriented dialogue systems.
No takes yet. Share an insight, caveat, or question.
Shah et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: