Feiteng Fang, Dingwei Chen, Xiang Huang, Ting-En Lin, Yuchuan Wu, Xiong Liu, Jing Ye, Ziqiang Liu, Haonan Zhang, Liang Zhu, Hamid Alinejad-Rokny, Min Yang, Yongbin Li. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Haonan Zhang, Run Luo, Xiong Liu, Yuchuan Wu, Ting-En Lin, Pengpeng Zeng, Qiang Qu, Feiteng Fang, Min Yang, Lianli Gao, Jingkuan Song, Fei Huang, Yongbin Li. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The development of large language models (LLMs) has significantly advanced the emergence of large multimodal models (LMMs). While LMMs have achieved tremendous success by promoting the synergy between multimodal comprehension and creation, they often face challenges when confronted with out-of-distribution data, such as which can hardly distinguish orientation, quantity, color, structure, etc. This is primarily due to their reliance on image encoders trained to encode images into task-relevant features, which may lead them to disregard irrelevant details. Delving into the modeling capabilities of diffusion models for images naturally prompts the question: Can diffusion models serve as the eyes of large language models for image perception? In this paper, we propose DEEM, a simple but effective approach that utilizes the generative feedback of diffusion models to align the semantic distributions of the image encoder. This addresses the drawbacks of previous methods that solely relied on image encoders like CLIP-ViT, thereby enhancing the model's resilience against out-of-distribution samples and reducing visual hallucinations. Importantly, this is achieved without requiring additional training modules and with fewer training parameters. We extensively evaluated DEEM on both our newly constructed RobustVQA benchmark and other well-known benchmarks, POPE and MMVP, for visual hallucination and perception. In particular, DEEM improves LMM's visual perception performance to a large extent (e.g., 4\% ↑ on RobustVQA, 6.5\% ↑ on MMVP and 12.8 \% ↑ on POPE ). Compared to the state-of-the-art interleaved content generation models, DEEM exhibits enhanced robustness and a superior capacity to alleviate model hallucinations while utilizing fewer trainable parameters, less pre-training data (10\%), and a smaller base model size. Extensive experiments demonstrate that DEEM enhances the performance of LMMs on various downstream tasks without inferior performance in the long term, including visual question answering, image captioning, and text-conditioned image synthesis.

Recent advancements in omnimodal learning have significantly improved understanding and generation across images, text, and speech, yet these developments remain predominantly confined to proprietary models. The lack of high-quality omnimodal datasets and the challenges of real-time emotional speech synthesis have notably hindered progress in open-source research. To address these limitations, we introduce OpenOmni, a two-stage training framework that integrates omnimodal alignment and speech generation to develop a state-of-the-art omnimodal large language model. In the alignment phase, a pretrained speech model undergoes further training on image-text tasks, enabling (near) zero-shot generalization from vision to speech, outperforming models trained on tri-modal datasets. In the speech generation phase, a lightweight decoder is trained on speech tasks with direct preference optimization, which enables real-time emotional speech synthesis with high fidelity. Extensive experiments demonstrate that OpenOmni surpasses state-of-the-art models across omnimodal, vision-language, and speech-language benchmarks. It achieves a 4-point absolute improvement on OmniBench over the leading open-source model VITA, despite using 5$\times$ fewer training examples and a smaller model size (7B vs. 7$\times$8B). Besides, OpenOmni achieves real-time speech generation with less than 1 second latency at non-autoregressive mode, reducing inference time by 5$\times$ compared to autoregressive methods, and improves emotion classification accuracy by 7.7\%. The codebase is available at https://github.com/RainBowLuoCS/OpenOmni.

Abstract Recent advances in large language models (LLMs) focus on aligning models with human values to minimize harmful content. However, existing methods often rely on a single type of feedback, such as preferences, annotated labels, or critiques, which can lead to overfitting and suboptimal performance. In this paper, we propose Diverse AIFeedback (DAIF), a novel approach that integrates three types of feedback—critique, refinement, and preference—tailored to tasks of varying uncertainty levels. Through an analysis of information gain, we show that critique feedback is most effective for low-uncertainty tasks, refinement feedback for medium-uncertainty tasks, and preference feedback for high-uncertainty tasks. Training with this diversified feedback reduces overfitting and improves alignment. Experimental results across three tasks—question answering, dialog generation, and text summarization–demonstrate that DAIF outperforms traditional methods relying on a single feedback type.1

Yuhan Chen, Ang Lv, Ting-En Lin, Changyu Chen, Yuchuan Wu, Fei Huang, Yongbin Li, Rui Yan. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Changyu Chen, Xiting Wang, Ting-En Lin, Ang Lv, Yuchuan Wu, Xin Gao, Ji-Rong Wen, Rui Yan, Yongbin Li. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Task-oriented dialogue (TOD) systems facilitate users in executing various activities via multi-turn dialogues, but Large Language Models (LLMs) often struggle to comprehend these intricate contexts. In this study, we propose a novel “Self-Explanation” prompting strategy to enhance the comprehension abilities of LLMs in multi-turn dialogues. This task-agnostic approach requires the model to analyze each dialogue utterance before task execution, thereby improving performance across various dialogue-centric tasks. Experimental results from six benchmark datasets confirm that our method consistently outperforms other zero-shot prompts and matches or exceeds the efficacy of few-shot prompts, demonstrating its potential as a powerful tool in enhancing LLMs’ comprehension in complex dialogue tasks.

Tianshu Yu, Haoyu Gao, Ting-En Lin, Min Yang, Yuchuan Wu, Wentao Ma, Chao Wang, Fei Huang, Yongbin Li. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Empathetic dialogue is a human-like behavior that requires the perception of both affective factors (e.g., emotion status) and cognitive factors (e.g., cause of the emotion). Besides concerning emotion status in early work, the latest approaches study emotion causes in empathetic dialogue. These approaches focus on understanding and duplicating emotion causes in the context to show empathy for the speaker. However, instead of only repeating the contextual causes, the real empathic response often demonstrate a logical and emotion-centered transition from the causes in the context to those in the responses. In this work, we propose an emotion cause transition graph to explicitly model the natural transition of emotion causes between two adjacent turns in empathetic dialogue. With this graph, the concept words of the emotion causes in the next turn can be predicted and used by a specifically designed concept-aware decoder to generate the empathic response. Automatic and human experimental results on the benchmark dataset demonstrate that our method produces more empathetic, coherent, informative, and specific responses than existing models.

Despite the remarkable progress of Fine-grained visual classification (FGVC) with years of history, it is still limited to recognizing 2D images. Recognizing objects in the physical world (i.e., 3D environment) poses a unique challenge – discriminative information is not only present in visible local regions but also in other unseen views. Therefore, in addition to finding the distinguishable part from the current view, efficient and accurate recognition requires inferring the critical perspective with minimal glances. E.g., a person might recognize a "Ford sedan" with a glance at its side and then know that looking at the front can help tell which model it is. In this paper, towards FGVC in the real physical world, we put forward the problem of multi-view active fine-grained visual recognition (MAFR) and complete this study in three steps: (i) a multi-view, fine-grained vehicle dataset is collected as the testbed, (ii) a pilot experiment is designed to validate the need and research value of MAFR, (iii) a policy-gradient-based framework along with a dynamic exiting strategy is proposed to achieve efficient recognition with active view selection. Our comprehensive experiments demonstrate that the proposed method outperforms previous multi-view recognition works and can extend existing state-of-the-art FGVC methods and advanced neural networks to become "FGVC experts" in the 3D environment. Our code is available at https://github.com/PRIS-CV/MAFR.

Sentiment analysis is a crucial task that aims to understand people's emotional states and predict emotional categories based on multimodal information. It consists of several subtasks, such as emotion recognition in conversation (ERC), aspect-based sentiment analysis (ABSA), and multimodal sentiment analysis (MSA). However, unifying all subtasks in sentiment analysis presents numerous challenges, including modality alignment, unified input/output forms, and dataset bias. To address these challenges, we propose a Task-Specific Prompt method to jointly model subtasks and introduce a multimodal generative framework called UniSA. Additionally, we organize the benchmark datasets of main subtasks into a new Sentiment Analysis Evaluation benchmark, SAEval. We design novel pre-training tasks and training methods to enable the model to learn generic sentiment knowledge among subtasks to improve the model's multimodal sentiment perception ability. Our experimental results show that UniSA performs comparably to the state-of-the-art on all subtasks and generalizes well to various subtasks in sentiment analysis.

Task-oriented dialogue (TOD) models have made significant progress in recent years. However, previous studies primarily focus on datasets written by annotators, which has resulted in a gap between academic research and real-world spoken con- versation scenarios. While several small-scale spoken TOD datasets are proposed to address robustness issues such as ASR errors, they ignore the unique challenges in spoken conversation. To tackle the limitations, we introduce SpokenWOZ, a large-scale speech-text dataset for spoken TOD, containing 8 domains, 203k turns, 5.7k dialogues and 249 hours of audios from human-to-human spoken conversations. SpokenWOZ further incorporates common spoken characteristics such as word-by-word processing and reasoning in spoken language. Based on these characteristics, we present cross-turn slot and reasoning slot detection as new challenges. We conduct experiments on various baselines, including text-modal models, newly proposed dual-modal models, and LLMs, e.g., ChatGPT. The results show that the current models still have substantial room for improvement in spoken conversation, where the most advanced dialogue state tracker only achieves 25.65% in joint goal accuracy and the SOTA end-to-end model only correctly completes the user request in 52.1% of dialogues. Our dataset, code, and leaderboard are available at https://spokenwoz.github.io/SpokenWOZ-github.io/.

Dialogue Topic Segmentation (DTS) plays an essential role in a variety of dialogue modeling tasks. Previous DTS methods either focus on semantic similarity or dialogue coherence to assess topic similarity for unsupervised dialogue segmentation. However, the topic similarity cannot be fully identified via semantic similarity or dialogue coherence. In addition, the unlabeled dialogue data, which contains useful clues of utterance relationships, remains underexploited. In this paper, we propose a novel unsupervised DTS framework, which learns topic-aware utterance representations from unlabeled dialogue data through neighboring utterance matching and pseudo-segmentation. Extensive experiments on two benchmark datasets (i.e., DialSeg711 and Doc2Dial) demonstrate that our method significantly outperforms the strong baseline methods. For reproducibility, we provide our code and data at: https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/dial-start.

Multimodal sentiment analysis (MSA) and emotion recognition in conversation (ERC) are key research topics for computers to understand human behaviors.From a psychological perspective, emotions are the expression of affect or feelings during a short period, while sentiments are formed and held for a longer period.However, most existing works study sentiment and emotion separately and do not fully exploit the complementary knowledge behind the two.In this paper, we propose a multimodal sentiment knowledge-sharing framework (UniMSE) that unifies MSA and ERC tasks from features, labels, and models.We perform modality fusion at the syntactic and semantic levels and introduce contrastive learning between modalities and samples to better capture the difference and consistency between sentiments and emotions.Experiments on four public benchmark datasets, MOSI, MOSEI, MELD, and IEMO-CAP, demonstrate the effectiveness of the proposed method and achieve consistent improvements compared with state-of-the-art methods.

In this paper, we present Duplex Conversation, a multi-turn, multimodal spoken dialogue system that enables telephone-based agents to interact with customers like a human. We use the concept of full-duplex in telecommunication to demonstrate what a human-like interactive experience should be and how to achieve smooth turn-taking through three subtasks: user state detection, backchannel selection, and barge-in detection. Besides, we propose semi-supervised learning with multimodal data augmentation to leverage unlabeled data to increase model generalization. Experimental results on three sub-tasks show that the proposed method achieves consistent improvements compared with baselines. We deploy the Duplex Conversation to Alibaba intelligent customer service and share lessons learned in production. Online A/B experiments show that the proposed system can significantly reduce response latency by 50%.

Open intent classification is a challenging task in dialogue systems. On the one hand, it should ensure the quality of known intent identification. On the other hand, it needs to detect the open (unknown) intent without prior knowledge. Current models are limited in finding the appropriate decision boundary to balance the performances of both known intents and the open intent. In this paper, we propose a post-processing method to learn the adaptive decision boundary (ADB) for open intent classification. We first utilize the labeled known intent samples to pre-train the model. Then, we automatically learn the adaptive spherical decision boundary for each known class with the aid of well-trained features. Specifically, we propose a new loss function to balance both the empirical risk and the open space risk. Our method does not need open intent samples and is free from modifying the model architecture. Moreover, our approach is surprisingly insensitive with less labeled data and fewer known intents. Extensive experiments on three benchmark datasets show that our method yields significant improvements compared with the state-of-the-art methods.

Discovering new intents is a crucial task in dialogue systems. Most existing methods are limited in transferring the prior knowledge from known intents to new intents. These methods also have difficulties in providing high-quality supervised signals to learn clustering-friendly features for grouping unlabeled intents. In this work, we propose an effective method (Deep Aligned Clustering) to discover new intents with the aid of limited known intent data. Firstly, we leverage a few labeled known intent samples as prior knowledge to pre-train the model. Then, we perform k-means to produce cluster assignments as pseudo-labels. Moreover, we propose an alignment strategy to tackle the label inconsistency problem during clustering assignments. Finally, we learn the intent representations under the supervision of the aligned pseudo-labels. With an unknown number of new intents, we predict the number of intent categories by eliminating low-confidence intent-wise clusters. Extensive experiments on two benchmark datasets show that our method is more robust and achieves substantial improvements over the state-of-the-art methods.

Discovering new user intents is an emerging task in the dialogue system. In this paper, we propose a self-supervised clustering method that can naturally incorporate pairwise constraints as prior knowledge to guide the clustering process and does not require intensive feature engineering. Extensive experiments on three benchmark datasets show that our method can yield significant improvements over strong baselines.

Identifying new user intents is an essential task in the dialogue system. However, it is hard to get satisfying clustering results since the definition of intents is strongly guided by prior knowledge. Existing methods incorporate prior knowledge by intensive feature engineering, which not only leads to overfitting but also makes it sensitive to the number of clusters. In this paper, we propose constrained deep adaptive clustering with cluster refinement (CDAC+), an end-to-end clustering method that can naturally incorporate pairwise constraints as prior knowledge to guide the clustering process. Moreover, we refine the clusters by forcing the model to learn from the high confidence assignments. After eliminating low confidence assignments, our approach is surprisingly insensitive to the number of clusters. Experimental results on the three benchmark datasets show that our method can yield significant improvements over strong baselines. 1

Identifying the unknown (novel) user intents that have never appeared in the training set is a challenging task in the dialogue system.In this paper, we present a two-stage method for detecting unknown intents.We use bidirectional long short-term memory (BiLSTM) network with the margin loss as the feature extractor.With margin loss, we can learn discriminative deep features by forcing the network to maximize inter-class variance and to minimize intra-class variance.Then, we feed the feature vectors to the density-based novelty detection algorithm, local outlier factor (LOF), to detect unknown intents.Experiments on two benchmark datasets show that our method can yield consistent improvements compared with the baseline methods.