This paper studies the exposure bias problem in task-oriented dialog systems, where the model's generated content over multiple turns drives the dialog context away from the ground-truth distribution at training time, introducing error propagation and damaging the robustness of the TOD system. To bridge the gap between training and inference for multi-turn task-oriented dialogs, we propose session-level sampling which explicitly exposes the model to sampled generated content of dialog context during training. Additionally, we employ a dropout-based consistency regularization with the masking strategy R-Mask to further improve the robustness and performance of the model. The proposed UBARv2 achieves state-of-the-art performance on the standardized evaluation benchmark MultiWOZ and extensive experiments show the effectiveness of the proposed methods.
School of Computer Science, Fudan University（复旦大学计算机科学学院）；Shanghai Key Laboratory of Intelligent Information Processing, Fudan University（复旦大学上海市智能信息处理重点实验室）
As the functionality of dialogue systems evolves, hybrid dialogue systems that accomplish user-specific goals and participate in open-topic chitchat with users are attracting growing attention. Existing research learns both tasks concurrently utilizing a multi-task fusion technique but ignores the negative transfer phenomenon induced by the unique textual style differences. Therefore, contrastive learning based on the latent variable model is used to decouple the various textual genres in the latent space. We devise supervised and self-supervised positive and negative sample constructions for …
Michigan State University（密歇根州立大学）；TAL Education Group（好未来教育集团）
Dialogue systems play an increasingly important role in various aspects of our daily life. It is evident from recent research that dialogue systems trained on human conversation data are biased. In particular, they can produce responses that reflect people's gender prejudice. Many debiasing methods have been developed for various NLP tasks, such as word embedding. However, they are not directly applicable to dialogue systems because they are likely to force dialogue models to generate similar responses for different genders. This greatly degrades the diversity of the generated responses and im…
Mitigating the Negative Impact of Over-association for Conversational Query Production
School of Information, Xiamen University（厦门大学信息学院）；Xiamen Key Laboratory of Intelligent Storage and Computing（厦门智能存储与计算重点实验室）；Tencent AI Lab（腾讯AI实验室）；College of Computer and Control Engineering, Minjiang University（闽江学院计算机与控制工程学院）
SpokenUS: A Spoken User Simulator for Task-Oriented Dialogue
Graduate School of Data Science, Seoul National University（首尔国立大学数据科学研究生院）；Department of Information Systems, Hanyang University（翰阳大学信息系统系）；Department of Computer Science and Engineering, Seoul National University（首尔国立大学计算机科学与工程系）
A Unified Data Augmentation Framework for Low-Resource Multi-Domain Dialogue Generation
Northeastern University（东北大学）；Center for Information and Language Processing, LMU Munich（慕尼黑大学信息与语言处理中心）；Institute of Informatics, LMU Munich（慕尼黑大学信息学研究所）
Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Institute of AI Industry Research, Tsinghua University（清华大学智能产业研究院）
Although exposure bias has been widely studied in some NLP tasks, it faces its unique challenges in dialogue response generation, the representative one-to-various generation scenario. In real human dialogue, there are many appropriate responses for the same context, not only with different expressions, but also with different topics. Therefore, due to the much bigger gap between various ground-truth responses and the generated synthetic response, exposure bias is more challenging in dialogue generation task. What's more, as MLE encourages the model to only learn the common words among differe…
Mitigating the Exposure Bias in Sentence-Level Grapheme-to-Phoneme (G2P) Transduction
Korea Advanced Institute of Science and Technology (KAIST)（韩国科学技术院）；Samsung Electronics（三星电子）；University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
Backtracking When It Strays: Mitigating Dual Exposure Biases in LLM Reasoning Distillation
College of Computer Science and Technology, Jilin University（吉林大学计算机科学与技术学院）；Key Laboratory of Symbolic Computation and Knowledge Engineering, MoE, Jilin University（吉林大学符号计算与知识工程重点实验室）；Tongyi Lab, Alibaba Group（阿里集团通义实验室）；College of Computer Science and Technology, Zhejiang University（浙江大学计算机科学与技术学院）；School of Artificial Intelligence, Jilin University（吉林大学人工智能学院）
From Oracle to Noisy Context: Mitigating Contextual Exposure Bias in Speech-LLMs
School of Computer Science and Technology, Xinjiang University, Urumqi, China（新疆大学计算机科学与技术学院，乌鲁木齐，中国）；Joint International Research Laboratory of Silk Road Multilingual Cognitive Computing, Urumqi, China（丝绸之路多语种认知计算联合国际研究实验室，乌鲁木齐，中国）；Timekettle
Robustness Risk of Conversational Retrieval: Identifying and Mitigating Noise Sensitivity in Qwen3-Embedding Model
Beijing University of Posts and Telecommunications（北京邮电大学）；SenseTime（商汤科技）；Beijing Key Laboratory of Network System and Network Culture（网络系统与网络文化北京市重点实验室）
Mitigating Strategy Preference Bias in Emotional Support Conversation via Uncertainty Estimations
Shanghai Institute of Artificial Intelligence for Education（上海人工智能教育研究院）；School of Computer Science and Technology（计算机科学与技术学院）；School of Psychology and Cognitive Science（心理学与认知科学学院）；School of Pharmacy（药学院）
Natural language generation (NLG) is an essential component of task-oriented dialog systems. Despite the recent success of neural approaches for NLG, they are typically developed in an offline manner for particular domains. To better fit real-life applications where new data come in a stream, we study NLG in a "continual learning" setting to expand its knowledge to new domains or functionalities incrementally. The major challenge towards this goal is catastrophic forgetting, meaning that a continually trained model tends to forget the knowledge it has learned before. To this end, we propose a …
Despite strong performance in many sequence-to-sequence tasks, autoregressive models trained with maximum likelihood estimation suffer from exposure bias, i.e. the discrepancy between the ground-truth prefixes used during training and the model-generated prefixes used at inference time. Scheduled sampling is a simple and empirically successful approach which addresses this issue by incorporating model-generated prefixes into training. However, it has been argued that it is an inconsistent training objective leading to models ignoring the prefixes altogether. In this paper, we conduct systemati…
