Taolin Zhang, Dongyang Li, Chen Chen, Qizhou Chen, Jiuheng Wan, Xiaofeng He, Chengyu Wang, Richang Hong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Taolin Zhang, Pukun Zhao, Qizhou Chen, Jiuheng Wan, Chen Chen, Xiaofeng He, Chengyu Wang, Richang Hong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Abstract Recently, the reasoning capabilities of Large Reasoning Models (LRMs), such as DeepSeek-R1, have witnessed significant advancements through computationally intensive “slow thinking” processes. These models have demonstrated impressive performance across a variety of complex reasoning tasks. However, despite their remarkable success, LRMs come with substantial computational demands that pose considerable challenges in terms of resource consumption, scalability, and accessibility. In contrast, Small Reasoning Models (SRMs), which are often distilled from larger models, offer a more efficient alternative while still achieving competitive performance. Beyond their efficiency, SRMs frequently exhibit distinct capabilities and cognitive trajectories compared with their larger counterparts, making them particularly interesting from both practical and theoretical perspectives. In this work, we provide a timely and comprehensive survey of recently published research focused on SRMs. We first review the current landscape of SRMs. Then, we analyze diverse training paradigms and inference techniques tailored to enhance the reasoning capabilities of SRMs. Furthermore, we offer an extensive review of domain-specific applications where SRMs have been effectively leveraged. Finally, we discuss promising future research directions that aim to bridge existing gaps. By consolidating recent advances, this survey serves as an essential reference for researchers and practitioners interested in leveraging or developing SRMs to unlock advanced reasoning functionalities with improved efficiency.

Model editing aims to correct outdated or erroneous knowledge in large models without costly retraining. Recent research discovered that the mid-layer representation of the subject's final token in a prompt has a strong influence on factual predictions, and developed Large Language Model (LLM) editing techniques based on this observation. However, for Vision-LLMs (VLLMs), how visual representations impact the predictions from a decoder-only language model remains largely unexplored. To the best of our knowledge, model editing for VLLMs has not been extensively studied in the literature. In this work, we employ the contribution allocation and noise perturbation methods to measure the contributions of visual representations for token predictions. Our attribution analysis shows that visual representations in mid-to-later layers that are highly relevant to the prompt contribute significantly to predictions. Based on these insights, we propose *VisEdit*, a novel model editor for VLLMs that effectively corrects knowledge by editing intermediate visual representations in regions important to the edit prompt. We evaluated *VisEdit* using multiple VLLM backbones and public VLLM editing benchmark datasets. The results show the superiority of *VisEdit* over the strong baselines adapted from existing state-of-the-art editors for LLMs.

Multi-hop question answering (QA) involves finding multiple relevant passages and performing step-by-step reasoning to answer complex questions.Previous works on multi-hop QA employ specific methods from different modeling perspectives based on large language models (LLMs), regardless of question types.In this paper, we first conduct an in-depth analysis of public multi-hop QA benchmarks, categorizing questions into four types and evaluating five types of cutting-edge methods: Chainof-Thought (CoT), Single-step, Iterative-step, Sub-step, and Adaptive-step.We find that different types of multi-hop questions exhibit varying degrees of sensitivity to different types of methods.Thus, we propose a Bi-levEL muLti-agEnt reasoning (BELLE) framework to address multi-hop QA by specifically focusing on the correspondence between question types and methods, with each type of method regarded as an "operator" by prompting LLMs differently.The first level of BELLE includes multiple agents that debate to formulate an executable plan of combined "operators" to address the multi-hop QA task comprehensively.During the debate, in addition to the basic roles of affirmative debater, negative debater, and judge, at the second level, we further leverage fast and slow debaters to monitor whether changes in viewpoints are reasonable.Extensive experiments demonstrate that BELLE significantly outperforms strong baselines in various datasets.Additionally, the model consumption of BELLE is higher cost-effectiveness than that of single models in more complex multihop QA scenarios.(B) Single-step [Multi-Hop Question:] What was the former band of the member of Mother Love Bone who died just before the release of Apple?[Answer:] Malfunkshun Multi-Hop Question Retrieval Docs Multi-Hop Answer (C) Iterative-step (D) Sub-step Multi-Hop Question (Intermediate) K times Multi-Hop Answer Multi-Hop Question 1 2 3 A1 A2 A3 Multi-Hop Answer (E) Adaptive-step Multi-Hop Question Classifier (D) Sub-Step (B) Single-step (A) CoT ✔ (A) CoT (C) Iterative-step Let's think step-by-step Multi-Hop Question Multi-Hop Question Retrieval-augmented Reasoning Closed-book Reasoning Our Agent-Based Reasoning Operators Pool …… Multi-Hop Question Inference Comparison Temporal Null 1. Use Sub-Step to decompose query 2. Use Single-Step to retrieve subquery 3. Aggregate the sub-answer Execution Plan Agents Multi-hop QA Task Environ -ment interaction invoke solve

Model editing aims to correct inaccurate knowledge, update outdated information, and incorporate new data into Large Language Models (LLMs) without the need for retraining. This task poses challenges in lifelong scenarios where edits must be continuously applied for real-world applications. While some editors demonstrate strong robustness for lifelong editing in pure LLMs, Vision LLMs (VLLMs), which incorporate an additional vision modality, are not directly adaptable to existing LLM editors. In this paper, we propose LiveEdit, a Lifelong vision language model Edit to bridge the gap between lifelong LLM editing and VLLMs. We begin by training an editing expert generator to independently produce low-rank experts for each editing instance, with the goal of correcting the relevant responses of the VLLM. A hard filtering mechanism is developed to utilize visual semantic knowledge, thereby coarsely eliminating visually irrelevant experts for input queries during the inference stage of the post-edited model. Finally, to integrate visually relevant experts, we introduce a soft routing mechanism based on textual semantic relevance to achieve multi-expert fusion. For evaluation, we establish a benchmark for lifelong VLLM editing. Extensive experiments demonstrate that LiveEdit offers significant advantages in lifelong VLLM editing scenarios. Further experiments validate the rationality and effectiveness of each module design in LiveEdit.1

Model editing aims to efficiently revise incorrect or outdated knowledge within LLMs without incurring the high cost of full retraining and risking catastrophic forgetting. Currently, most LLM editing datasets are confined to narrow knowledge domains and cover a limited range of editing evaluation. They often overlook the broad scope of editing demands and the diversity of ripple effects resulting from edits. In this context, we introduce \uniedit, a unified benchmark for LLM editing grounded in open-domain knowledge. First, we construct editing samples by selecting entities from 25 common domains across five major categories, utilizing the extensive triple knowledge available in open-domain knowledge graphs to ensure comprehensive coverage of the knowledge domains. To address the issues of generality and locality in editing, we design an Neighborhood Multi-hop Chain Sampling (NMCS) algorithm to sample subgraphs based on a given knowledge piece to entail comprehensive ripple effects to evaluate. Finally, we employ proprietary LLMs to convert the sampled knowledge subgraphs into natural language text, guaranteeing grammatical accuracy and syntactical diversity. Extensive statistical analysis confirms the scale, comprehensiveness, and diversity of our \uniedit benchmark. We conduct comprehensive experiments across multiple LLMs and editors, analyzing their performance to highlight strengths and weaknesses in editing across open knowledge domains and various evaluation criteria, thereby offering valuable insights for future research endeavors.

The minimal feature removal problem in the post-hoc explanation area aims to identify the minimal feature set (MFS). Prior studies using the greedy algorithm to calculate the minimal feature set lack the exploration of feature interactions under a monotonic assumption which cannot be satisfied in general scenarios. In order to address the above limitations, we propose a Cooperative Integrated Dynamic Refining method (CIDR) to efficiently discover minimal feature sets. Specifically, we design Cooperative Integrated Gradients (CIG) to detect interactions between features. By incorporating CIG and characteristics of the minimal feature set, we transform the minimal feature removal problem into a knapsack problem. Additionally, we devise an auxiliary Minimal Feature Refinement algorithm to determine the minimal feature set from numerous candidate sets. To the best of our knowledge, our work is the first to address the minimal feature removal problem in the field of natural language processing. Extensive experiments demonstrate that CIDR is capable of tracing representative minimal feature sets with improved interpretability across various models and datasets.

Cross-lingual representation learning transfers knowledge from resource-rich data to resource-scarce ones to improve the semantic understanding abilities of different languages. However, previous works rely on shallow unsupervised data generated by token surface matching, regardless of the global context-aware semantics of the surrounding text tokens. In this paper, we propose an Unsupervised Pseudo Semantic Data Augmentation (UniPSDA) mechanism for cross-lingual natural language understanding to enrich the training data without human interventions. Specifically, to retrieve the tokens with similar meanings for the semantic data augmentation across different languages, we propose a sequential clustering process in 3 stages: within a single language, across multiple languages of a language family, and across languages from multiple language families. Meanwhile, considering the multi-lingual knowledge infusion with context-aware semantics while alleviating computation burden, we directly replace the key constituents of the sentences with the above-learned multi-lingual family knowledge, viewed as pseudo-semantic. The infusion process is further optimized via three de-biasing techniques without introducing any neural parameters. Extensive experiments demonstrate that our model consistently improves the performance on general zero-shot cross-lingual natural language understanding tasks, including sequence classification, information extraction, and question answering.

Knowledge-enhanced pre-trained language models (KEPLMs) leverage relation triples from knowledge graphs (KGs) and integrate these external data sources into language models via self-supervised learning. Previous works treat knowledge enhancement as two independent operations, i.e., knowledge injection and knowledge integration. In this paper, we propose to learn Knowledge-Enhanced language representations with Hierarchical Reinforcement Learning (KEHRL), which jointly addresses the problems of detecting positions for knowledge injection and integrating external knowledge into the model in order to avoid injecting inaccurate or irrelevant knowledge. Specifically, a high-level reinforcement learning (RL) agent utilizes both internal and prior knowledge to iteratively detect essential positions in texts for knowledge injection, which filters out less meaningful entities to avoid diverting the knowledge learning direction. Once the entity positions are selected, a relevant triple filtration module is triggered to perform low-level RL to dynamically refine the triples associated with polysemic entities through binary-valued actions. Experiments validate KEHRL’s effectiveness in probing factual knowledge and enhancing the model’s performance on various natural language understanding tasks.

KEPLMs are pre-trained models that utilize external knowledge to enhance language understanding. Previous language models facilitated knowledge acquisition by incorporating knowledge-related pre-training tasks learned from relation triples in knowledge graphs. However, these models do not prioritize learning embeddings for entity-related tokens. Updating all parameters in KEPLM is computationally demanding. This paper introduces TRELM, a Robust and Efficient Pre-training framework for Knowledge-Enhanced Language Models. We observe that text corpora contain entities that follow a long-tail distribution, where some are suboptimally optimized and hinder the pre-training process. To tackle this, we employ a robust approach to inject knowledge triples and employ a knowledge-augmented memory bank to capture valuable information. Moreover, updating a small subset of neurons in the feed-forward networks (FFNs) that store factual knowledge is both sufficient and efficient. Specifically, we utilize dynamic knowledge routing to identify knowledge paths in FFNs and selectively update parameters during pre-training. Experimental results show that TRELM achieves at least a 50% reduction in pre-training time and outperforms other KEPLMs in knowledge probing tasks and multiple knowledge-aware language understanding tasks.

Retrieval-augmented large language models (LLMs) leverage relevant content retrieved by information retrieval systems to generate correct responses, aiming to alleviate the hallucination problem. However, existing retriever-responder methods typically append relevant documents to the prompt of LLMs to perform text generation tasks without considering the interaction of fine-grained structural semantics between the retrieved documents and the LLMs. This issue is particularly important for accurate response generation as LLMs tend to “lose in the middle” when dealing with input prompts augmented with lengthy documents. In this work, we propose a new pipeline named “Reinforced Retriever-Reorder-Responder” (R4) to learn document orderings for retrieval-augmented LLMs, thereby further enhancing their generation abilities while the large numbers of parameters of LLMs remain frozen. The reordering learning process is divided into two steps according to the quality of the generated responses: document order adjustment and document representation enhancement. Specifically, document order adjustment aims to organize retrieved document orderings into beginning, middle, and end positions based on graph attention learning, which maximizes the reinforced reward of response quality. Document representation enhancement further refines the representations of retrieved documents for responses of poor quality via document-level gradient adversarial learning. Extensive experiments demonstrate that our proposed pipeline achieves better factual question-answering performance on knowledge-intensive tasks compared to strong baselines across various public datasets. The source codes and trained models will be released upon paper acceptance.

Model editing aims to correct outdated or erroneous knowledge in large language models (LLMs) without the need for costly retraining.Lifelong model editing is the most challenging task that caters to the continuous editing requirements of LLMs.Prior works primarily focus on single or batch editing; nevertheless, these methods fall short in lifelong editing scenarios due to catastrophic knowledge forgetting and the degradation of model performance.Although retrieval-based methods alleviate these issues, they are impeded by slow and cumbersome processes of integrating the retrieved knowledge into the model.In this work, we introduce RECIPE, a RetriEval-augmented ContInuous Prompt lEarning method, to boost editing efficacy and inference efficiency in lifelong learning.RECIPE first converts knowledge statements into short and informative continuous prompts, prefixed to the LLM's input query embedding, to efficiently refine the response grounded on the knowledge.It further integrates the Knowledge Sentinel (KS) that acts as an intermediary to calculate a dynamic threshold, determining whether the retrieval repository contains relevant knowledge.Our retriever and prompt encoder are jointly trained to achieve editing properties, i.e., reliability, generality, and locality.In our experiments, RECIPE is assessed extensively across multiple LLMs and editing datasets, where it achieves superior editing performance.RECIPE also demonstrates its capability to maintain the overall performance of LLMs alongside showcasing fast editing and inference speed.

Taolin Zhang, Ruyao Xu, Chengyu Wang, Zhongjie Duan, Cen Chen, Minghui Qiu, Dawei Cheng, Xiaofeng He, Weining Qian. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Knowledge-Enhanced Pre-trained Language Models (KEPLMs) are pre-trained models with relation triples injecting from knowledge graphs to improve language understanding abilities.Experiments show that our model outperforms other KEPLMs significantly over zero-shot knowledge probing tasks and multiple knowledge-aware language understanding tasks. To guarantee effective knowledge injection, previous studies integrate models with knowledge encoders for representing knowledge retrieved from knowledge graphs. The operations for knowledge retrieval and encoding bring significant computational burdens, restricting the usage of such models in real-world applications that require high inference speed. In this paper, we propose a novel KEPLM named DKPLM that decomposes knowledge injection process of the pre-trained language models in pre-training, fine-tuning and inference stages, which facilitates the applications of KEPLMs in real-world scenarios. Specifically, we first detect knowledge-aware long-tail entities as the target for knowledge injection, enhancing the KEPLMs' semantic understanding abilities and avoiding injecting redundant information. The embeddings of long-tail entities are replaced by ``pseudo token representations'' formed by relevant knowledge triples. We further design the relational knowledge decoding task for pre-training to force the models to truly understand the injected knowledge by relation triple reconstruction. Experiments show that our model outperforms other KEPLMs significantly over zero-shot knowledge probing tasks and multiple knowledge-aware language understanding tasks. We further show that DKPLM has a higher inference speed than other competing models due to the decomposing mechanism.

Taolin Zhang, Zerui Cai, Chengyu Wang, Minghui Qiu, Bite Yang, Xiaofeng He. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Knowledge-Enhanced Pre-trained Language Models (KEPLMs) improve the language understanding abilities of deep language models by leveraging the rich semantic knowledge from knowledge graphs, other than plain pre-training texts. However, previous efforts mostly use homogeneous knowledge (especially structured relation triples in knowledge graphs) to enhance the context-aware representations of entity mentions, whose performance may be limited by the coverage of knowledge graphs. Also, it is unclear whether these KEPLMs truly understand the injected semantic knowledge due to the "black-box'' training mechanism. In this paper, we propose a novel KEPLM named HORNET, which integrates Heterogeneous knowledge from various structured and unstructured sources into the Roberta NETwork and hence takes full advantage of both linguistic and factual knowledge simultaneously. Specifically, we design a hybrid attention heterogeneous graph convolution network (HaHGCN) to learn heterogeneous knowledge representations based on the structured relation triplets from knowledge graphs and the unstructured entity description texts. Meanwhile, we propose the explicit dual knowledge understanding tasks to help induce a more effective infusion of the heterogeneous knowledge, promoting our model for learning the complicated mappings from the knowledge graph embedding space to the deep context-aware embedding space and vice versa. Experiments show that our HORNET model outperforms various KEPLM baselines on knowledge-aware tasks including knowledge probing, entity typing and relation extraction. Our model also achieves substantial improvement over several GLUE benchmark datasets, compared to other KEPLMs.