Recent advancements in large language models (LLMs) have demonstrated remarkable text generation capabilities. However, controlling specific attributes of generated text remains challenging without architectural modifications or extensive fine-tuning. Current methods typically toggle a single, basic attribute but struggle with precise multi-attribute control. In scenarios where attribute requirements conflict, existing methods lack coordination mechanisms, causing interference between desired attributes. Furthermore, these methods fail to incorporate iterative optimization processes in the controlled generation pipeline. To address these limitations, we propose Conflict-aware, Composite, and Collaborative Controlled Text Generation (C³TG), a two-phase framework for fine-grained, multi-dimensional text attribute control. During generation, C³TG selectively pairs the LLM with the required attribute classifiers from the 17 available dimensions and employs weighted KL-divergence to adjust token probabilities. The optimization phase then leverages an energy function combining classifier scores and penalty terms to resolve attribute conflicts through iterative feedback, enabling precise control over multiple dimensions simultaneously while preserving natural text flow. Experiments show that C³TG significantly outperforms baselines across multiple metrics including attribute accuracy, linguistic fluency, and output diversity, while simultaneously reducing toxicity. These results establish C³TG as an effective and flexible solution for multi-dimensional text attribute control that requires no costly model modifications.

Fengyuan Liu, Yi Huang, Sichun Luo, Yuqi Wang, Yazheng Yang, Xinye Li, Zefa Hu, Junlan Feng, Qi Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Existing user simulators based on prompting to role-play or SFT are generally confined to imitating users' textual utterances, without adequately considering the multi-faceted cognitive processes that underlie human decision-making during interactions.To facilitate better alignment with real human thinking patterns, we construct the LMSYS-UserThinking dataset, in which we augment 51k human-LLM conversations by reconstructing the user's inner reasoning both during and at the end of each dialogue.Furthermore, to enhance controllability and situational coherence, we introduce scenario settings that describe the global context and user goals throughout multi-turn conversations.Using this dataset, we train user simulators called ThinkingUS on different base models.We evaluate our approach from both offline and online user simulation perspectives, ultimately demonstrating its effectiveness.

In a conversational system, dynamically generating follow-up questions based on context can help users explore information and provide a better user experience. Humans are usually able to ask questions that involve some general life knowledge and demonstrate higher order cognitive skills. However, the questions generated by existing methods are often limited to shallow contextual questions that are uninspiring and have a large gap to the human level. In this paper, we propose a three-stage external knowledge-enhanced follow-up question generation method, which generates questions by identifying contextual topics, constructing a knowledge graph (KG) online, and finally combining these with a large language model to generate the final question. The model generates information-rich and exploratory follow-up questions by introducing external common sense knowledge and performing a knowledge fusion operation. Experiments show that compared to baseline models, our method generates questions that are more informative and closer to human questioning levels while maintaining contextual relevance.

Zhe Yang, Yi Huang, Yaqin Chen, XiaotingWu XiaotingWu, Junlan Feng, Chao Deng. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Continual Structured Knowledge Reasoning (CSKR) focuses on training models to handle sequential tasks, where each task involves translating natural language questions into structured queries grounded in structured knowledge. Existing general continual learning approaches face significant challenges when applied to this task, including poor generalization to heterogeneous structured knowledge and inefficient reasoning due to parameter growth as tasks increase. To address these limitations, we propose a novel CSKR framework, \textsc{K-DeCore}, which operates with a fixed number of tunable parameters. Unlike prior methods, \textsc{K-DeCore} introduces a knowledge decoupling mechanism that disentangles the reasoning process into task-specific and task-agnostic stages, effectively bridging the gaps across diverse tasks. Building on this foundation, \textsc{K-DeCore} integrates a dual-perspective memory consolidation mechanism for distinct stages and introduces a structure-guided pseudo-data synthesis strategy to further enhance the model's generalization capabilities. Extensive experiments on four benchmark datasets demonstrate the superiority of \textsc{K-DeCore} over existing continual learning methods across multiple metrics, leveraging various backbone large language models.

Recent studies have explored the use of Large Language Models (LLMs) with Retrieval Augmented Generation (RAG) for Knowledge Graph Question Answering (KGQA).They typically require rewriting retrieved subgraphs into natural language formats comprehensible to LLMs.However, when tackling complex questions, the knowledge rewritten by existing methods may include irrelevant information, omit crucial details, or fail to align with the question's semantics.To address them, we propose a novel rewriting method CoTKR, Chain-of-Thought Enhanced Knowledge Rewriting, for generating reasoning traces and corresponding knowledge in an interleaved manner, thereby mitigating the limitations of single-step knowledge rewriting.Additionally, to bridge the preference gap between the knowledge rewriter and the question answering (QA) model, we propose a training strategy PAQAF, Preference Alignment from Question Answering Feedback, for leveraging feedback from the QA model to further optimize the knowledge rewriter.We conduct experiments using various LLMs across several KGQA benchmarks.Experimental results demonstrate that, compared with previous knowledge rewriting methods, CoTKR generates the most beneficial knowledge representation for QA models, which significantly improves the performance of LLMs in KGQA 1 .* Equal contribution.†Corresponding author. 1 Our code is available at https://github.com/wuyike2000/CoTKR.Question: what is the unit of area that the measurement system that have an electric field strength units of volt per metre have?Answer: Square kilometer, Square meter KR Methods KR Results SummaryThe International System of Units includes various measurement systems for different physical quantities such as length, area, volume, conductance, resistance, current, force, and pressure.The measurement system with an electric field strength unit of Volt per meter does not directly relate to a specific area unit within the International System of Units.CoTKR Reason: I need to know the unit of area associated with the measurement system that has an electric field strength unit of Volt per meter.Summarize: The International System of Units includes various units for different physical quantities such as length, area, volume, etc. Reason: I need to find the specific unit of area in the International System of Units that corresponds to the electric field strength unit of Volt per meter.Summarize: The International System of Units includes units like Square kilometer, Square meter, etc. for measuring area.CoTKR+PA Reason: I need to know which measurement system includes the unit for electric field strength in Volt per meter.Knowledge: The International System of Units contains the unit for electric field strength in Volt per meter.Reason: I need to identify the unit of area within the International System of Units.Knowledge: The International System of Units includes the units Square kilometer and Square meter as area units.

Dialogue generation is a ambitious task that requires generating coherent and natural responses. In this paper, we propose a novel approach that utilizes a Seq2Seq generator enhanced by neural architecture search (NAS) on top of adversarial training to this end. Specifically, we apply the NAS for data native exploration, motivating the intrinsic characteristics for an adaptive structure. Further, we borrow the two-phase paradigm idea of pre-training and fine-tuning, extending the adversarial training process in terms of diversity and alignment. Experiments on the MultiWOZ-series datasets demonstrate our framework achieves superior performance in both automatic and human evaluations, and effectiveness is shown under low-resource settings. Comparative experiments indicate that the searched structure has a significantly faster convergence speed.

Recently, increasing research interests have focused on retrieval augmented generation (RAG) to mitigate hallucination for large language models (LLMs). Following this trend, we launch the FutureDial-RAG challenge at SLT 2024, which aims at promoting the study of RAG for dialog systems. The challenge builds upon the MobileCS2 dataset, a real-life customer service datasets with nearly 3000 high-quality dialogs containing annotations for knowledge base query and corresponding results. Over the dataset, we define two tasks, track 1 for knowledge retrieval and track 2 for response generation, which are core research questions in dialog systems with RAG. We build baseline systems for the two tracks and design metrics to measure whether the systems can perform accurate retrieval and generate informative and coherent response. The baseline results show that it is very challenging to perform well on the two tasks, which encourages the participating teams and the community to study how to make better use of RAG for real-life dialog systems.

The advancements in large language models (LLMs) have significantly propelled the level of artificial intelligence, further enhancing the model’s problem-solving capabilities across a variety of dialogue-oriented tasks. However, the substantial costs associated with training and inference processes for LLMs hinder their deployment across various dialogue scenarios, while small language models (SLMs) tend to perform poorly with limited samples in new settings or domains. Therefore, we propose a collaborative mechanism between LLMs and SLMs, wherein prompts are employed to bridge the gap between them. For the dialogue system, the LLM acts as a source from which SLM derives, facilitating task planning, data generating, training and optimization. Experimental results indicate that our method can significantly reduce inference overhead in new dialogue scenarios and outperforms the original pipeline architecture in terms of inference performance.

Extracting relational triplets aims at detecting entity pairs and their semantic relations. Compared with pipeline models, joint models can reduce error propagation and achieve better performance. However, all of these models require large amounts of training data, therefore performing poorly on many long-tail relations in reality with insufficient data. In this article, we propose a novel end-to-end model, called TGIN, for few-shot triplet extraction. The core of TGIN is a multilayer heterogeneous graph with two types of nodes (entity node and relation node) and three types of edges (relation-entity edge, entity-entity edge, and relation-relation edge). On the one hand, this heterogeneous graph with entities and relations as nodes can intuitively extract relational triplets jointly, thereby reducing error propagation. On the other hand, it enables the triplet information of limited labeled data to interact better, thus maximizing the advantage of this information for few-shot triplet extraction. Moreover, we devise a graph aggregation and update method that utilizes translation algebraic operations to mine semantic features while retaining structure features between entities and relations, thereby improving the robustness of the TGIN in a few-shot setting. After updating the node and edge features through layers, TGIN propagates the label information from a few labeled examples to unlabeled examples, thus inferring triplets from these unlabeled examples. Extensive experiments on three reconstructed datasets demonstrate that TGIN can significantly improve the accuracy of triplet extraction by 2.34%~10.74% compared with the state-of-the-art baselines. To the best of our knowledge, we are the first to introduce a heterogeneous graph for few-shot relational triplet extraction.

Multi-action dialog policy (MADP), which generates multiple atomic dialog actions per turn, has been widely applied in task-oriented dialog systems to provide expressive and efficient system responses. Existing MADP models usually imitate action combinations from the labeled multi-action dialog samples. Due to data limitations, they generalize poorly toward unseen dialog flows. While reinforcement learning-based methods are proposed to incorporate the service ratings from real users and user simulators as external supervision signals, they suffer from sparse and less credible dialog-level rewards. To cope with this problem, we explore to improve MADPL with explicit and implicit turn-level user feedback received for historical predictions (i.e., logged user feedback) that are cost-efficient to collect and faithful to real-world scenarios. The task is challenging since the logged user feedback provides only partial label feedback limited to the particular historical dialog actions predicted by the agent. To fully exploit such feedback information, we propose BanditMatch, which addresses the task from a feedback-enhanced semi-supervised learning perspective with a hybrid learning objective of SSL and bandit learning. BanditMatch integrates pseudo-labeling methods to better explore the action space through constructing full label feedback. Extensive experiments show that our BanditMatch improves MADPL over the state-of-the-art methods by generating more concise and informative responses. The source code and the appendix of this paper can be obtained from https://github.com/ShuoZhangXJTU/BanditMatch.

Continual learning is crucial for dialog state tracking (DST) in dialog systems, since requirements from users for new functionalities are often encountered. However, most of existing continual learning methods for DST require task identities during testing, which is a severe limit in real-world applications. In this paper, we aim to address continual learning of DST in the class-incremental scenario (namely the task identity is unknown in testing). Inspired by the recently emerging prompt tuning method that performs well on dialog systems, we propose to use the prompt pool method, where we maintain a pool of key-value paired prompts and select prompts from the pool according to the distance between the dialog history and the prompt keys. The proposed method can automatically identify tasks and select appropriate prompts during testing. We conduct experiments on Schema-Guided Dialog dataset (SGD) and another dataset collected from a real-world dialog application. Experiment results show that the prompt pool method achieves much higher joint goal accuracy than the baseline. After combining with a rehearsal buffer, the model performance can be further improved.

Recently, two approaches, fine-tuning large pre-trained language models and variational training, have attracted significant interests, separately, for semi-supervised end-to-end task-oriented dialog (TOD) systems. In this paper, we propose Variational Latent-State GPT model (VLS-GPT), which is the first to combine the strengths of the two approaches. Among many options of models, we propose the generative model and the inference model for variational learning of the end-to-end TOD system, both as auto-regressive language models based on GPT-2, which can be further trained over a mix of labeled and unlabeled dialog data in a semi-supervised manner. Variational training of VLS-GPT is both statistically and computationally more challenging than previous variational learning works for sequential latent variable models, which use turn-level first-order Markovian. The inference model in VLS-GPT is non-Markovian due to the use of the Transformer architecture. In this work, we establish Recursive Monte Carlo Approximation (RMCA) to the variational objective with non-Markovian inference model and prove its unbiasedness. Further, we develop the computational strategy of sampling-then-forward-computation to realize RMCA, which successfully overcomes the memory explosion issue of using GPT in variational learning and speeds up training. Semi-supervised TOD experiments are conducted on two benchmark multi-domain datasets of different languages - MultiWOZ2.1 and CrossWOZ. VLS-GPT is shown to significantly outperform both supervised-only and semi-supervised self-training baselines.

Self-supervised entity alignment (EA) aims to link equivalent entities across different knowledge graphs (KGs) without the use of pre-aligned entity pairs. The current state-of-the-art (SOTA) self-supervised EA approach draws inspiration from contrastive learning, originally designed in computer vision based on instance discrimination and contrastive loss, and suffers from two shortcomings. Firstly, it puts unidirectional emphasis on pushing sampled negative entities far away rather than pulling positively aligned pairs close, as is done in the well-established supervised EA. Secondly, it advocates the minimum information requirement for self-supervised EA, while we argue that self-described KG's side information (e.g., entity name, relation name, entity description) shall preferably be explored to the maximum extent for the self-supervised EA task. In this work, we propose an interactive contrastive learning model for self-supervised EA. It conducts bidirectional contrastive learning via building pseudo-aligned entity pairs as pivots to achieve direct cross-KG information interaction. It further exploits the integration of entity textual and structural information and elaborately designs encoders for better utilization in the self-supervised setting. Experimental results show that our approach outperforms the previous best self-supervised method by a large margin (over 9% [email protected] absolute improvement on average) and performs on par with previous SOTA supervised counterparts, demonstrating the effectiveness of the interactive contrastive learning for self-supervised EA. The code and data are available at https://github.com/THU-KEG/ICLEA.

Most existing slot filling models tend to memorize inherent patterns of entities and corresponding contexts from training data. However, these models can lead to system failure or undesirable outputs when being exposed to spoken language perturbation or variation in practice. We propose a perturbed semantic structure awareness transferring method for training perturbation-robust slot filling models. Specifically, we introduce two MLM-based training strategies to respectively learn contextual semantic structure and word distribution from unsupervised language perturbation corpus. Then, we transfer semantic knowledge learned from upstream training procedure into the original samples and filter generated data by consistency processing. These procedures aims to enhance the robustness of slot filling models. Experimental results show that our method consistently outperforms the previous basic methods and gains strong generalization while preventing the model from memorizing inherent patterns of entities and contexts.

Traditional intent classification models are based on a pre-defined intent set and only recognize limited in-domain (IND) intent classes. But users may input out-of-domain (OOD) queries in a practical dialogue system. Such OOD queries can provide directions for future improvement. In this paper, we define a new task, Generalized Intent Discovery (GID), which aims to extend an IND intent classifier to an open-world intent set including IND and OOD intents. We hope to simultaneously classify a set of labeled IND intent classes while discovering and recognizing new unlabeled OOD types incrementally. We construct three public datasets for different application scenarios and propose two kinds of frameworks, pipeline-based and end-to-end for future work. Further, we conduct exhaustive experiments and qualitative analysis to comprehend key challenges and provide new guidance for future GID research.

Despite the success of text-to-text pre-trained models in various natural language generation (NLG) tasks, the generation performance is largely restricted by the number of labeled data in downstream tasks, particularly in data-to-text generation tasks. Existing works mostly utilize abundant unlabeled structured data to conduct unsupervised pre-training for task adaption, which fail to model the complex relationship between source structured data and target texts. Thus, we introduce self-training as a better few-shot learner than task-adaptive pre-training, which explicitly captures this relationship via pseudo-labeled data generated by the pre-trained model. To alleviate the side-effect of low-quality pseudo-labeled data during self-training, we propose a novel method called Curriculum-Based Self-Training (CBST) to effectively leverage unlabeled data in a rearranged order determined by the difficulty of text generation. Experimental results show that our method can outperform fine-tuning and task-adaptive pre-training methods, and achieve state-of-the-art performance in the few-shot setting of data-to-text generation.

Multi-action dialog policy (MADP), which generates multiple atomic dialog actions per turn, has been widely applied in task-oriented dialog systems to provide expressive and efficient system responses. Existing MADP models usually imitate action combinations from the labeled multi-action dialog samples. Due to data limitations, they generalize poorly toward unseen dialog flows. While interactive learning and reinforcement learning algorithms can be applied to incorporate external data sources of real users and user simulators, they take significant manual effort to build and suffer from instability. To address these issues, we propose Planning Enhanced Dialog Policy (PEDP), a novel multi-task learning framework that learns single-action dialog dynamics to enhance multi-action prediction. Our PEDP method employs model-based planning for conceiving what to express before deciding the current response through simulating single-action dialogs. Experimental results on the MultiWOZ dataset demonstrate that our fully supervised learning-based method achieves a solid task success rate of 90.6%, improving 3% compared to the state-of-the-art methods. The source code and the appendix of this paper can be obtained from https://github.com/ShuoZhangXJTU/PEDP.

Developing semi-supervised task-oriented dialog (TOD) systems by leveraging unlabeled dialog data has attracted increasing interests.For semi-supervised learning of latent state TOD models, variational learning is often used, but suffers from the annoying high-variance of the gradients propagated through discrete latent variables and the drawback of indirectly optimizing the target log-likelihood.Recently, an alternative algorithm, called joint stochastic approximation (JSA), has emerged for learning discrete latent variable models with impressive performances.In this paper, we propose to apply JSA to semi-supervised learning of the latent state TOD models, which is referred to as JSA-TOD.To our knowledge, JSA-TOD represents the first work in developing JSA based semi-supervised learning of discrete latent variable conditional models for such long sequential generation problems like in TOD systems.Extensive experiments show that JSA-TOD significantly outperforms its variational learning counterpart.Remarkably, semi-supervised JSA-TOD using 20% labels performs close to the full-supervised baseline on MultiWOZ2.1.

Recently, Transformer based pretrained language models (PLMs), such as GPT2 and T5, have been leveraged to build generative task-oriented dialog (TOD) systems. A drawback of existing PLM-based models is their non-Markov architectures across turns, i.e., the whole history is used as the conditioning input at each turn. First, this brings inefficiencies in memory and computation. Furthermore, using the whole history increases model complexity and may hurt the training efficiency, especially when facing small amounts of labeled training data (the low-resource setting). In this paper, motivated by the observation that dialog states could be viewed as Markov states, we propose to build Markovian Generative Architectures (MGA) over PLM backbones for efficient TOD systems. Experiments on MultiWOZ2.1 show that in the rich-resource setting, the proposed Markov models reduce memory and time costs without performance degradation; in the low-resource setting, the training efficiency of the Markov models is more significant.

Contract consistency is important in ensuring the legal validity of the contract. In many scenarios, a contract is written by filling the blanks in a precompiled form. Due to carelessness, two blanks that should be filled with the same (or different) content may be incorrectly filled with different (or same) content. This will result in the issue of contract inconsistencies, which may severely impair the legal validity of the contract. Traditional methods to address this issue mainly rely on manual contract review, which is labor-intensive and costly. In this work, we formulate a novel Contract Inconsistency Checking (CIC) problem, and design an end-to-end framework, called Pair-wise Blank Resolution (PBR), to solve the CIC problem with high accuracy. Our PBR model contains a novel BlankCoder to address the challenge of modeling meaningless blanks. BlankCoder adopts a two-stage attention mechanism that adequately associates a meaningless blank with its relevant descriptions while avoiding the incorporation of irrelevant context words. Experiments conducted on real-world datasets show the promising performance of our method with a balanced accuracy of 94.05% and an F1 score of 90.90% in the CIC problem.

A Dialogue State Tracker (DST) is a core component of a modular task-oriented dialogue system.Tremendous progress has been made in recent years.However, the major challenges remain.The state-of-the-art accuracy for DST is below 50% for a multi-domain dialogue task.A learnable DST for any new domain requires a large amount of labeled indomain data and training from scratch.In this paper, we propose a Meta-Reinforced Multi-Domain State Generator (MERET).Our first contribution is to improve the DST accuracy.We enhance a neural model based DST generator with a reward manager, which is built on policy gradient reinforcement learning (R-L) to fine-tune the generator.With this change, we are able to improve the joint accuracy of DST from 48.79% to 50.91% on the Multi-WOZ corpus.Second, we explore to train a DST meta-learning model with a few domains as source domains and a new domain as target domain.We apply the model-agnostic metalearning (MAML) algorithm to DST and the obtained meta-learning model is used for new domain adaptation.Our experimental results show this solution is able to outperform the traditional training approach with extremely less training data in target domain.