Xiaoyuan Li, Keqin Bao, Yubo Ma, Moxin Li, Wenjie Wang, Rui Men, Yichang Zhang, Fuli Feng, Dayiheng Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Yuzhen Shi, Huanghai Liu, Yiran HU, Song Gaojie, Xu Xinran, Yubo Ma, Tianyi Tang, Li Zhang, Qingjing Chen, Feng Di, Wenbo Lv, Weiheng Wu, Kexin Yang, Sen Yang, Wei Wang, Rongyao Shi, Qiu Yuanyang, Yuemeng Qi, Zhang Jingwen, Sui Xiaoyu, Yifan Chen, Zhang Yi, An Yang, Bowen Yu, Dayiheng Liu, Junyang Lin, Weixing Shen, Bing Zhao, Charles L. A. Clarke, HU Wei. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Current Large Language Models excel in general reasoning yet struggle with specialized tasks requiring proprietary or domain-specific knowledge. Fine-tuning large models for every niche application is often infeasible due to black-box constraints and high computational overhead. To address this, we propose a collaborative framework that pairs a specialized weak model with a general strong model. The weak model, tailored to specific domains, produces initial drafts and background information, while the strong model leverages its advanced reasoning to refine these drafts, extending LLMs’ capabilities to critical yet specialized tasks. To optimize this collaboration, we introduce a collaborative feedback to fine-tunes the weak model, which quantifies the influence of the weak model’s contributions in the collaboration procedure and establishes preference pairs to guide preference tuning of the weak model. We validate our framework through experiments on three domains. We find that the collaboration significantly outperforms each model alone by leveraging complementary strengths. Moreover, aligning the weak model with the collaborative preference further enhances overall performance.

Xiaobao Wu, Liangming Pan, Yuxi Xie, Ruiwen Zhou, Shuai Zhao, Yubo Ma, Mingzhe Du, Rui Mao, Anh Tuan Luu, William Yang Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Single hyperspectral image (HSI) super-resolution (SR), which is limited by the lack of exterior information, has always been a challenging task. A lot of effort has gone into fully mining spectral information or adopting pre-trained models to enhance spatial resolution. However, few SR approaches take into account structural features from the perspective of multi-dimensional segmentation of image. Therefore, a novel spectral–spatial segmentation-based local bicubic interpolation (S3LBI) is proposed to implement segmented and blocked interpolation according to the characteristics of HSI. Specifically, the bands of an HSI are clustered into several spectral segments. Then, super-pixel segmentation is carried out in each spectral segments. After that, the bicubic interpolations are separately conducted on different spectral–spatial segments. Experiments demonstrate the superiority of our S3LBI over the compared HSI SR approaches.

Event relations are crucial for narrative understanding and reasoning.Governed by nuanced logic, event relation extraction (ERE) is a challenging task that demands thorough semantic understanding and rigorous logical reasoning.In this paper, we conduct an in-depth investigation to systematically explore the capability of LLMs in understanding and applying event relation logic.More in detail, we first investigate the deficiencies of LLMs in logical reasoning across different tasks.Our study reveals that LLMs are not logically consistent reasoners, which results in their suboptimal performance on tasks that need rigorous reasoning.To address this, we explore three different approaches to endow LLMs with event relation logic, and thus enable them to generate more coherent answers across various scenarios.Based on our approach, we also contribute a synthesized dataset (LLM-ERL) involving high-order reasoning for evaluation and fine-tuning.Extensive quantitative and qualitative analyses on different tasks also validate the effectiveness of our approaches and provide insights for solving practical tasks with LLMs in future work.

Yubo Ma, Zhibin Gou, Junheng Hao, Ruochen Xu, Shuohang Wang, Liangming Pan, Yujiu Yang, Yixin Cao, Aixin Sun. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Understanding documents with rich layouts and multi-modal components is a long-standing and practical task. Recent Large Vision-Language Models (LVLMs) have made remarkable strides in various tasks, particularly in single-page document understanding (DU). However, their abilities on long-context DU remain an open problem. This work presents MMLONGBENCH-DOC, a long-context, multi- modal benchmark comprising 1,082 expert-annotated questions. Distinct from previous datasets, it is constructed upon 135 lengthy PDF-formatted documents with an average of 47.5 pages and 21,214 textual tokens. Towards comprehensive evaluation, answers to these questions rely on pieces of evidence from (1) different sources (text, image, chart, table, and layout structure) and (2) various locations (i.e., page number). Moreover, 33.7\% of the questions are cross-page questions requiring evidence across multiple pages. 20.6\% of the questions are designed to be unanswerable for detecting potential hallucinations. Experiments on 14 LVLMs demonstrate that long-context DU greatly challenges current models. Notably, the best-performing model, GPT-4o, achieves an F1 score of only 44.9\%, while the second-best, GPT-4V, scores 30.5\%. Furthermore, 12 LVLMs (all except GPT-4o and GPT-4V) even present worse performance than their LLM counterparts which are fed with lossy-parsed OCR documents. These results validate the necessity of future research toward more capable long-context LVLMs.

Detecting individualPopulus euphratica (P. euphratica)trees in desert forest areas is crucial for monitoring their ecophysiological characteristics and ecological conservation. However, the presence of the spectral-similarTamarix chinensis (T. chinensis)in the habitats, along with the densely overlapping crowns in clusteredP. euphratica, presents a challenge for the task. This paper proposes a method to detect individualPopulus euphraticain very high spatial resolution (VHR) images. First, the deep learning-based semantic segmentation model is used to differentiate betweenP. euphraticaandT. chinensis. Second, the individual tree detection is converted into a constrained 2D bin packing model and solved by a heuristic template matching and filling algorithm. The experimental data consists of a World View-2 image capturing sparse desert forests of the lower reaches of the Tarim River. 22296 individualP. euphraticatrees were detected, achieving F1 scores of 0.885, 0.869, and 0.902 on three datasets with varying difficulty levels. Furthermore, experiments were conducted to compare with other methods, and the results showed that the proposed method achieved the best performance on all three datasets. The proposed method can be applied to map the distribution of individualP. euphraticatrees in sparse desert forests and can provide methodological references for similar tasks related to individual tree detection in natural forests.

Sub-pixel target detection is a challenging task in hyperspectral image processing. Most statistical detectors rely on the estimation of the background covariance matrix. In contrast to the traditional approaches by adopting global covariance matrix estimation from all image pixels, the local estimation within image segments can significantly improve detection performance for complex backgrounds in many scenarios. However, the local covariance matrix estimate may be unstable due to the high spectral dimension of the hyperspectral image, especially when the size of the local sample representing background pixels, is relatively small. In this work, a random projection (RP) is employed to reduce the spectral dimension, and a spectral similarity-based dual-window (SSDW) strategy is suggested to appropriately extract the local background statistical properties. So, a kind of sub-pixel target detector is developed in a statistical hypothesis testing framework for different observation models under t-distribution background. Especially, the projection dimension is determined without any extra experiment or training and ensures asymptotic optimality of detection power under some conditions. The superior performance of the proposed detectors is demonstrated by some synthetic data and real hyperspectral images.

Few-shot event detection (ED) has been widely studied, while this brings noticeable discrepancies, e.g., various motivations, tasks, and experimental settings, that hinder the understanding of models for future progress.This paper presents a thorough empirical study, a unified view of ED models, and a better unified baseline.For fair evaluation, we compare 12 representative methods on three datasets, which are roughly grouped into prompt-based and prototype-based models for detailed analysis.Experiments consistently demonstrate that prompt-based methods, including Chat-GPT, still significantly trail prototype-based methods in terms of overall performance.To investigate their superior performance, we break down their design elements along several dimensions and build a unified framework on prototype-based methods.Under such unified view, each prototype-method can be viewed a combination of different modules from these design elements.We further combine all advantageous modules and propose a simple yet effective baseline, which outperforms existing methods by a large margin (e.g., 2.7% F 1 gains under low-resource setting).1

Graph embedding (GE) provides an effective way to reveal the intrinsic feature of high-dimensional data on the foundation of preserving topological properties. Under the framework of GE, the hyperspectral image can be represented by a weighted graph, where pixels and similarities among them are treated as vertices and edge weights, respectively. In this article, an adaptive reference-related GE (ARGE) method is proposed to efficaciously obtain the low-dimensional feature and improve computational efficiency. The ARGE method is composed of two primary processes. The key to connecting these two processes is the reference vertices set, which is the abstraction of graph topological features. First, the reference vertices are adaptively selected through a three-step adaptive reference set selection (ARSS) algorithm. Second, the original high-dimensional graph is embedded as a low-dimensional graph through preserving the reference-related structure. Specifically, the pairwise similarities between vertices and reference vertices are preserved in embedding space. In addition, a new hybrid dissimilarity measure of Rao distance and spectral information divergence (RD-SID) is designed to depict the spectral difference between pixels. To evaluate the effectiveness of the proposed method, the obtained low-dimensional feature is fed into the anomaly detector to detect anomalous pixels. The experimental results on five real and one synthetic hyperspectral datasets demonstrate the superiority of the proposed ARGE method over the compared feature extraction methods.

In this paper, we propose an effective yet efficient model PAIE for both sentence-level and document-level Event Argument Extraction (EAE), which also generalizes well when there is a lack of training data.On the one hand, PAIE utilizes prompt tuning for extractive objectives to take the best advantages of Pre-trained Language Models (PLMs).It introduces two span selectors based on the prompt to select start/end tokens among input texts for each role.On the other hand, it captures argument interactions via multi-role prompts and conducts joint optimization with optimal span assignments via a bipartite matching loss.Also, with a flexible prompt design, PAIE can extract multiple arguments with the same role instead of conventional heuristic threshold tuning.We have conducted extensive experiments on three benchmarks, including both sentenceand document-level EAE.The results present promising improvements from PAIE (3.5% and 2.3% F1 gains in average on three benchmarks, for PAIE-base and PAIE-large respectively).Further analysis demonstrates the efficiency, generalization to few-shot settings, and effectiveness of different extractive prompt tuning strategies.Our code is available at https: //github.com/mayubo2333/PAIE.