Known Intents, New Combinations: Clause-Factorized Decoding for Compositional Multi-Intent Detection

Abhilash Nandy

cs.CLMar 30, 2026

Known Intents, New Combinations: Clause-Factorized Decoding for Compositional Multi-Intent Detection

Authors: Abhilash Nandy

Organizations: Microsoft Research India

Abstract

Multi-intent detection papers usually ask whether a model can recover multiple intents from one utterance. We ask a harder and, for deployment, more useful question: can it recover new combinations of familiar intents? Existing benchmarks only weakly test this, because train and test often share the same broad co-occurrence patterns. We introduce CoMIX-Shift, a controlled benchmark built to stress compositional generalization in multi-intent detection through held-out intent pairs, discourse-pattern shift, longer and noisier wrappers, held-out clause templates, and zero-shot triples. We also present ClauseCompose, a lightweight decoder trained only on singleton intents, and compare it to whole-utterance baselines including a fine-tuned tiny BERT model. Across three random seeds, ClauseCompose reaches 95.7 exact match on unseen intent pairs, 93.9 on discourse-shifted pairs, 62.5 on longer/noisier pairs, 49.8 on held-out templates, and 91.1 on unseen triples. WholeMultiLabel reaches 81.4, 55.7, 18.8, 15.5, and 0.0; the BERT baseline reaches 91.5, 77.6, 48.9, 11.0, and 0.0. We also add a 240-example manually authored SNIPS-style compositional set with five held-out pairs; there, ClauseCompose reaches 97.5 exact match on unseen pairs and 86.7 under connector shift, compared with 41.3 and 10.4 for WholeMultiLabel. The results suggest that multi-intent detection needs more compositional evaluation, and that simple factorization goes surprisingly far once evaluation asks for it.

Figures & tables

Split	Examples	Notes
Train single	2,800	280 per intent
Train pair seen	6,480	27 seen pairs
Test pair seen	1,080	familiar pairs
Test pair unseen	1,080	held-out pairs
Test pair shift	1,080	held-out pairs + new discourse
Test pair long shift	1,080	held-out pairs + longer/noisier wrappers

Table 1: Benchmark scale.

Model	Seen	Unseen	Pair shift	Long shift	Template holdout	Triple shift
AtomicSet	100.0 $\pm$ 0.0	0.0 $\pm$ 0.0	0.0 $\pm$ 0.0	0.0 $\pm$ 0.0	0.0 $\pm$ 0.0	0.0 $\pm$ 0.0
WholeMultiLabel	100.0 $\pm$ 0.0	81.4 $\pm$ 7.6	55.7 $\pm$ 11.0	18.8 $\pm$ 15.3	15.5 $\pm$ 8.6	0.0 $\pm$ 0.0
BERTTinyMultiLabel	100.0 $\pm$ 0.0	91.5 $\pm$ 10.4	77.6 $\pm$ 20.0	48.9 $\pm$ 23.4	11.0 $\pm$ 9.2	0.0 $\pm$ 0.0
ClauseCompose	95.2 $\pm$ 1.9	95.7 $\pm$ 2.9	93.9 $\pm$ 3.3	62.5 $\pm$ 9.2	49.8 $\pm$ 17.2	91.1 $\pm$ 2.6

Table 2: Main synthetic results over three seeds. All numbers are exact-match set accuracy (EM).

Model	Manual unseen pairs	Manual pair shift
WholeMultiLabel	41.3 $\pm$ 3.1	10.4 $\pm$ 1.2
BERTTinyMultiLabel	16.7 $\pm$ 7.7	10.8 $\pm$ 6.8
ClauseCompose	97.5 $\pm$ 0.0	86.7 $\pm$ 7.4

Table 3: Manual SNIPS-style compositional check over six intents and five held-out pairs. All numbers are EM.

Intent Classification

Intent classification assigns user utterances or behaviors to goals, supporting dialogue systems, search, safety decisions, and proactive assistance.

See all 60 papers

Jun 27, 2026

COMPASS: Grounding Composition-Intent Guidance in Unified Multimodal Models

Composition is a high-level visual intent that governs where subjects are placed and how a scene is organized, yet current unified multimodal models remain unreliable at fine-grained composition recognition and struggle to turn such intent into controllable generation. We present COMPASS, the first unified multimodal framework that grounds composition-intent control in a single system spanning both composition perception and composition-guided generation, with a shared expert token

τ_c

as the central intent anchor. On the perception side, COMPASS injects composition expertise into an MoE backbone in a minimally invasive manner and distills the inferred intent into

τ_c

. On the generation side, COMPASS reuses

τ_c

as a global conditioning signal that steers the denoising trajectory, effectively converting passive composition analysis into explicit layout control. To support systematic instruction-following composition learning and evaluation at scale, we construct Comp-11, a large-scale dataset with an 11-class taxonomy and reasoning-augmented annotations. Extensive experiments show that COMPASS substantially improves category-level composition understanding and delivers more composition-consistent, prompt-faithful generation than strong baselines.

Ziqi Zhou, Weize Quan, Mining Tan +6

University of Edinburgh, United Kingdom · State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of Automation, Chinese Academy of Sciences, China · University of Chinese Academy of Sciences, China +1

Multimodal Grounding Unified Multimodal Models Compositional Image Generation Compositional T2I Generation Multimodal Foundation Models Multimodal Understanding

Ask PDF

Jul 8, 2026

A Multi-cluster Boundary Learning Method for Out-of-Scope Intent Detection via MiniLM Embedding

Intent detection is a critical task that bridges human intents and system actions in human-machine interaction systems. However, there still exist challenges for detecting out-of-scope (OOS) intents. (i) The traditional methods view the OOS intent detection as a multi-class classification, then the detection accuracy decreases as the class number of the known intents increases; (ii) LLM-embedding methods require large parameters, that makes them difficult to train and practically deploy. Thus, this work proposes a multi-cluster boundary learning method to detect OOS intents via MiniLM embedding (i.e., all-MiniLM-L6-v2) in an one-class classification workflow. The method learns the boundaries of multi-cluster embeddings generated by MiniLM from the training utterances, and then rejects the out-of-domain utterances as OOS intents. Experiments are conducted on public CLINC150, StackOverflow and Banking77 datasets. The results show that the method achieves the state-of-the-art OOS intent detection performance compared the other baselines. Ablation studies are also conducted and the results show that the used MiniLM can better adapt to the workflow and utterance embedding requirements. The code is available at supplementary materials.

Yihong Xu, Mingyu Kang, Linyuan Lü

University of Science and Technology of China, Hefei, China

Intent Classification OOD Detection

Ask PDF

Aug 28, 2026

Leveraging Turn-taking Dynamics for Intent Recognition in Multi-party Conversations

We propose a multi-task learning approach for multi-party dialogue intent recognition that leverages an auxiliary task that models turn-taking dynamics. Specifically, we introduce turn-transition entropy, a self-supervised target computed from the sequence of speaker transitions, which quantifies the predictability of interaction patterns. Experiments on multiple pre-trained models demonstrate that incorporating this auxiliary task improves intent recognition performance, outperforming existing approaches which ignore multi-party interaction dynamics. We find that our proposed continuous target can be learned as a single-task objective, suggesting that it is an actual signal carrying useful information.

Galo Castillo-López, Alexis Lombard, Gaël de Chalendar +1

Université Paris-Saclay, CEA, List, Palaiseau, France

Intent Classification

Ask PDF

Abstract

Intent Classification

Explore similar work

COMPASS: Grounding Composition-Intent Guidance in Unified Multimodal Models

A Multi-cluster Boundary Learning Method for Out-of-Scope Intent Detection via MiniLM Embedding

Leveraging Turn-taking Dynamics for Intent Recognition in Multi-party Conversations