Future superhuman models will surpass the ability of humans and humans will only be able to \textit{weakly} supervise superhuman models. To alleviate the issue of lacking high-quality data for model alignment, some works on weak-to-strong generalization (W2SG) finetune a strong pretrained model with a weak supervisor so that it can generalize beyond weak supervision. However, the invariable use of weak supervision in existing methods exposes issues in robustness, with a proportion of weak labels proving harmful to models. In this paper, we propose a selective W2SG framework to avoid using weak supervision when unnecessary. We train a binary classifier P(IK) to identify questions that a strong model can answer and use its self-generated labels for alignment. We further refine weak labels with a graph smoothing method. Extensive experiments on three benchmarks show that our method consistently outperforms competitive baselines. Further analyses show that P(IK) can generalize across tasks and difficulties, which indicates selective W2SG can help superalignment.

In-context learning (ICL) has emerged as a powerful paradigm for Large Visual Language Models (LVLMs), enabling them to leverage a few examples directly from input contexts. However, the effectiveness of this approach is heavily reliant on the selection of demonstrations, a process that is NP-hard. Traditional strategies, including random, similarity-based sampling and infoscore-based sampling, often lead to inefficiencies or suboptimal performance, struggling to balance both efficiency and effectiveness in demonstration selection. In this paper, we propose a novel demonstration selection framework named Coreset-based Dual Retrieval (CoDR). We show that samples within a diverse subset achieve a higher expected mutual information. To implement this, we introduce a cluster-pruning method to construct a diverse coreset that aligns more effectively with the query while maintaining diversity. Additionally, we develop a dual retrieval mechanism that enhances the selection process by achieving global demonstration selection while preserving efficiency. Experimental results demonstrate that our method significantly improves the ICL performance compared to the existing strategies, providing a robust solution for effective and efficient demonstration selection.

Yihong Dong, Xue Jiang, Yongding Tao, Huanyu Liu, Kechi Zhang, Lili Mou, Rongyu Cao, Yingwei MA, Jue Chen, Binhua Li, Zhi Jin, Fei Huang, Yongbin Li, Ge Li. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Xuanyu Lei, Chenliang Li, Yuning Wu, Kaiming Liu, Weizhou Shen, Peng Li, Ming Yan, Fei Huang, Ya-Qin Zhang, Yang Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Role-playing agents (RPAs) require balancing multiple objectives, such as instruction following, persona consistency, and stylistic fidelity, which are not always perfectly aligned across different dimensions.While prior work has primarily relied on supervised finetuning or reinforcement learning with scalarized rewards, these approaches do not explicitly address the coordination of multiple reward dimensions during optimization.We present MOA (Multi-Objective Alignment), a reinforcement-learning framework that enables multi-dimensional, fine-grained rubric optimization for general RPAs.MOA introduces a novel multi-objective optimization strategy that trains simultaneously on multiple fine-grained rubrics to boost optimization performance.Additionally, to improve both output diversity and generation quality, we employ thought-augmented rollouts with offpolicy guidance.Experiments on PersonaGym and RoleMRC show that MOA consistently improves multi-dimensional role-playing performance over supervised and standard RL baselines.Under identical evaluation protocols, an 8B model trained with MOA reaches performance competitive with strong closed-source models across multiple evaluation dimensions.These results suggest that MOA provides a practical framework for training more capable general-purpose role-playing agents.

With the rapid advancement of post-training techniques for reasoning and information seeking, large language models (LLMs) can incorporate a large quantity of retrieved knowledge to solve complex tasks. However, the limited context window of LLMs obstructs scaling the amount of external knowledge input, prohibiting further improvement. Existing context window extension methods inevitably cause information loss. LLM-based multi-agent methods emerge as a new paradigm to handle massive input in a distributional manner, where we identify two core bottlenecks in existing agent orchestration designs. In this work, we develop a multi-agent framework, \textbf{\ExtAgents}, to overcome the bottlenecks and enable better scalability in inference-time knowledge integration without longer-context training. Benchmarked with our enhanced multi-hop question answering test, \textbf{$\boldsymbol{\infty}$Bench+}, and other public test sets including long survey generation, \ExtAgents significantly enhances the performance over existing non-training methods with the same amount of external knowledge input, regardless of whether it falls \emph{within or exceeds the context window}. Moreover, the method maintains efficiency due to high parallelism. We believe further study in the coordination of LLM agents on increasing external knowledge input could benefit real-world applications.

Zhengxi Lu, Jiabo Ye, Fei Tang, Yongliang Shen, Haiyang Xu, Ziwei Zheng, Weiming Lu, Ming Yan, Fei Huang, Jun Xiao, Yueting Zhuang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Fuwen Luo, Shengfeng Lou, Chi Chen, Ziyue Wang, Chenliang Li, Weizhou Shen, Jiyue Guo, Peng Li, Ming Yan, Ji Zhang, Fei Huang, Yang Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Binghai Wang, Yantao Liu, Yuxuan Liu, Tianyi Tang, Shenzhi Wang, Chang Gao, Chujie Zheng, Yichang Zhang, Le Yu, Shixuan Liu, Tao Gui, Qi Zhang, Xuanjing Huang, Bowen Yu, Fei Huang, Junyang Lin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Guibin Zhang, Haiyang Yu, Kaiming Yang, Bingli Wu, Fei Huang, Yongbin Li, Shuicheng Yan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

In financial trading, factor models are widely used to price assets and capture excess returns from mispricing. Recently, we have witnessed the rise of variational autoencoder-based latent factor models, which learn latent factors self-adaptively. While these models focus on modeling overall market conditions, they often fail to effectively capture the temporal patterns of individual stocks. Additionally, representing multiple factors as single values simplifies the model but limits its ability to capture complex relationships and dependencies. As a result, the learned factors are of low quality and lack diversity, reducing their effectiveness and robustness across different trading periods. To address these issues, we propose a Spatio-Temporal factOR Model based on dual vector quantized variational autoencoders, named STORM, which extracts features of stocks from temporal and spatial perspectives, then fuses and aligns these features at the fine-grained and semantic level, and represents the factors as multi-dimensional embeddings. The discrete codebooks cluster similar factor embeddings, ensuring orthogonality and diversity, which helps distinguish between different factors and enables factor selection in financial trading. To show the performance of the proposed factor model, we apply it to two downstream experiments: portfolio management on two stock datasets and individual trading tasks on six specific stocks. The extensive experiments demonstrate STORM's flexibility in adapting to downstream tasks and superior performance over baseline models.

Complex logic errors in LLM-generated code are challenging to diagnose and repair. While existing LLM-based self-repair approaches conduct intensive static semantic analysis or rely on superficial execution logs, they miss the in-depth runtime behaviors that often expose bug root causes—lacking the interactive dynamic analysis capabilities that make human debugging effective. We present InspectCoder, the first agentic program repair system that empowers LLMs to actively conduct dynamic analysis via interactive debugger control. Our dual-agent framework enables strategic breakpoint placement, targeted state inspection, and incremental runtime experimentation within stateful debugger sessions. Unlike existing methods that follow fixed log collection procedures, InspectCoder adaptively inspects and perturbs relevant intermediate states at runtime, and leverages immediate process rewards from debugger feedback to guide multi-step reasoning, transforming LLM debugging paradigm from blind trial-and-error into systematic root cause diagnosis. We conduct comprehensive experiments on two challenging self-repair benchmarks: BigCodeBench-R and LiveCodeBench-R. InspectCoder achieves 5.10%–60.37% relative improvements in repair accuracy over the strongest baseline, while delivering 1.67x-2.24x superior bug-fix efficiency respectively.We also contribute InspectWare, an open-source middleware that abstracts debugger complexities and maintains stateful debugging sessions across mainstream Python testing frameworks. Our work provides actionable insight into the interactive LLM-debugger systems, demonstrating the significant potential of LLM-driven dynamic analysis for automated software engineering.

We propose a novel embedding-based captioning metric termed asL-CLIPScorethat can be used for efficiently evaluating caption quality and training captioning model. L-CLIPScore is calculated from a lightweight CLIP (L-CLIP), which is a dual-encoder architecture compressed and distilled from CLIP. To compress, we apply two powerful techniques which are weight multiplexing and matrix decomposition for reducing the parameters of encoders and word embedding matrix, respectively. To distill, we design a novel multi-modal Similarity Regulator (SR) loss to transfer more vision-language alignment knowledge. Specifically, SR loss amplifies the multi-modal embedding similarity if the given image-text pair is matched and diminishes the similarity if the pair is non-matched. By compressing and distilling by this novel SR loss, our L-CLIP achieves comparable multi-modal alignment ability to the original CLIP while it requires fewer computation resources and running time. We carry out exhaustive experiments to validate the efficiency and effectiveness of L-CLIPScore when using it as the judge to evaluate caption quality. We also discover that when using L-CLIPScore as the supervisor to train the captioning model, it should be mixed up by an n-gram-based metric and meanwhile analyze why using L-CLIPScore only will cause fail training

We propose a novel Transformer-based image-to-text generation model termed asACFthat adaptively clusters vision patches into object regions and language words into phrases to implicitly learn object-phrase alignments for better visual-text coherence. To achieve this, we design a novel self-attention layer that applies self-attention over the elements in a local cluster window instead of the whole sequence. The window size is softly decided by a clustering matrix that is calculated by the current input data and thus this process is adaptive. By stacking these revised self-attention layers to construct ACF, the small clusters in the lower layers can be grouped into a bigger cluster, e.g., vision/language. ACF clusters small objects/phrases into bigger ones. In this gradual clustering process, a parsing tree is generated which embeds the hierarchical knowledge of the input sequence. As a result, by using ACF to build the vision encoder and language decoder, the hierarchical object-phrase alignments are embedded and then transferred from vision to language domains in two popular image-to-text tasks: Image captioning and Visual Question Answering. The experiment results demonstrate the effectiveness of ACF, which outperforms most SOTA captioning and VQA models and achieves comparable scores compared with some large-scale pre-trained models.

Common methods for aligning already-capable models with desired behavior rely on the ability of humans to provide supervision. However, future superhuman models will surpass the capability of humans. Therefore, humans will only be able to weakly supervise superhuman models. This expected deficiency of human evaluation would weaken the safety of future AI systems. Scalable oversight and weak-to-strong generalization are two complementary approaches to tackle this issue. In this paper, we attempt to combine the strengths of these two approaches to further improve alignment. Specifically, we investigate ways of improving human supervision with a strong pretrained model and then supervise the strong model with enhanced weak human supervision. To make iterative empirical progress, we consider an analogy: can we use a strong model to improve weak model supervision and then use it to supervise the strong model? We empirically test it by finetuning a small weak model on ground truth labels with the additional help from a large strong model, and then finetuning the strong model on labels generated by the weak model. We find that debate can assist a weak model in extracting trustworthy information from an untrustworthy strong model, which provides leverage as context on samples when training a weak model. We also show that an ensemble of weak models helps exploit long arguments generated by strong model debaters and obtain a more robust supervision estimate. Extensive experiments on the OpenAI weak-to-strong NLP benchmarks show that the combination approach leads to better alignment, which indicates that debate has the potential to help weak-to-strong generalization.

Retrieval-augmented generation (RAG) demonstrates remarkable performance across tasks in open-domain question-answering. However, traditional search engines may retrieve shallow content, limiting the ability of LLMs to handle complex, multi-layered information. To address this, we introduce WebWalkerQA, a benchmark designed to assess the ability of LLMs to perform web traversal. It evaluates the capacity of LLMs to traverse a website’s subpages to extract high-quality data systematically. We propose WebWalker, which is a multi-agent framework that mimics human-like web navigation through an explore-critic paradigm. Extensive experimental results show that WebWalkerQA is challenging and demonstrates the effectiveness of RAG combined with WebWalker, through this horizontal and vertical integration in real-world scenarios.

Anwen Hu, Haiyang Xu, Liang Zhang, Jiabo Ye, Ming Yan, Ji Zhang, Qin Jin, Fei Huang, Jingren Zhou. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Social agents powered by large language models (LLMs) can simulate human social behaviors but fall short in handling complex social dialogues. Direct Preference Optimization (DPO) has proven effective in aligning LLM behavior with human preferences across various agent tasks. However, standard DPO focuses solely on individual turns, which limits its effectiveness in multi-turn social interactions. Several DPO-based multi-turn alignment methods with session-level data have shown potential in addressing this problem. While these methods consider multiple turns across entire sessions, they are often overly coarse-grained, introducing training noise, and lack robust theoretical support. To resolve these limitations, we propose Segment-Level Direct Preference Optimization (SDPO), which dynamically select key segments within interactions to optimize multi-turn agent behavior. SDPO minimizes training noise and is grounded in a rigorous theoretical framework. Evaluations on the SOTOPIA benchmark demonstrate that SDPO-tuned agents consistently outperform both existing DPO-based methods and proprietary LLMs like GPT-4o, underscoring SDPO’s potential to advance the social intelligence of LLM-based agents. We release our code and data at https://anonymous.4open.science/r/SDPO-CE8F.

Zhuoqun Li, Haiyang Yu, Xuanang Chen, Hongyu Lin, Yaojie Lu, Fei Huang, Xianpei Han, Yongbin Li, Le Sun. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Large Language Models (LLMs) have shown impressive reasoning capabilities in well-defined problems with clear solutions, such as mathematics and coding. However, they still struggle with complex real-world scenarios like business negotiations, which require strategic reasoning—an ability to navigate dynamic environments and align long-term goals amidst uncertainty.Existing methods for strategic reasoning face challenges in adaptability, scalability, and transferring strategies to new contexts.To address these issues, we propose explicit policy optimization (EPO) for strategic reasoning, featuring an LLM that provides strategies in open-ended action space and can be plugged into arbitrary LLM agents to motivate goal-directed behavior.To improve adaptability and policy transferability, we train the strategic reasoning model via multi-turn reinforcement learning (RL), utilizing process rewards and iterative self-play.Experiments across social and physical domains demonstrate EPO’s ability of long-term goal alignment through enhanced strategic reasoning, achieving state-of-the-art performance on social dialogue and web navigation tasks. Our findings reveal various collaborative reasoning mechanisms emergent in EPO and its effectiveness in generating novel strategies, underscoring its potential for strategic reasoning in real-world applications. Code and data are available at https://github.com/lxqpku/EPO.

Ziheng Qiao, Houquan Zhou, Yumeng Liu, Zhenghua Li, Min Zhang, Bo Zhang, Chen Li, Ji Zhang, Fei Huang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Shuofei Qiao, Zhisong Qiu, Baochang Ren, Xiaobin Wang, Xiangyuan Ru, Ningyu Zhang, Xiang Chen, Yong Jiang, Pengjun Xie, Fei Huang, Huajun Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Yunkun Wang, Yue Zhang, Zhen Qin, Chen Zhi, Binhua Li, Fei Huang, Yongbin Li, Shuiguang Deng. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

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.

In the realm of large language models (LLMs), the ability of models to accurately follow instructions is paramount as more agents and applications leverage LLMs for construction, where the complexity of instructions are rapidly increasing.However, on the one hand, there is only a certain amount of complex instruction evaluation data; on the other hand, there are no dedicated algorithms to improve the ability to follow complex instructions.To this end, this paper introduces TRACE, a benchmark for improving and evaluating the complex instructionfollowing ability, which consists of 120K training data and 1K evaluation data.Furthermore, we propose IOPO (Input-Output Preference Optimization) alignment method which takes both input and output preference pairs into consideration, where LLMs not only rapidly align with response preferences but also meticulously explore the instruction preferences.Extensive experiments on both in-domain and outof-domain datasets confirm the effectiveness of IOPO, showing 8.15%, 2.18% improvements on in-domain data and 5.91%, 2.83% on outof-domain data compared to SFT and DPO respectively.

Artificial intelligence has significantly revolutionized healthcare, particularly through large language models (LLMs) that demonstrate superior performance in static medical question answering benchmarks. However, evaluating the potential of LLMs for real-world clinical applications remains challenging due to the intricate nature of doctor-patient interactions. To address this, we introduce AI Hospital, a multi-agent framework emulating dynamic medical interactions between Doctor as player and NPCs including Patient and Examiner. This setup allows for more practical assessments of LLMs in simulated clinical scenarios. We develop the Multi-View Medical Evaluation (MVME) benchmark, utilizing high-quality Chinese medical records and multiple evaluation strategies to quantify the performance of LLM-driven Doctor agents on symptom collection, examination recommendations, and diagnoses. Additionally, a dispute resolution collaborative mechanism is proposed to enhance medical interaction capabilities through iterative discussions. Despite improvements, current LLMs (including GPT-4) still exhibit significant performance gaps in multi-turn interactive scenarios compared to non-interactive scenarios. Our findings highlight the need for further research to bridge these gaps and improve LLMs’ clinical decision-making capabilities. Our data, code, and experimental results are all open-sourced at https://github.com/LibertFan/AI_Hospital.

How to better evaluate the capabilities of Large Language Models (LLMs) is the focal point and hot topic in current LLMs research. Previous work has noted that due to the extremely high cost of iterative updates of LLMs, they are often unable to answer the latest dynamic questions well. To promote the improvement of Chinese LLMs’ ability to answer dynamic questions, in this paper, we introduce CDQA, a Chinese Dynamic QA benchmark containing question-answer pairs related to the latest news on the Chinese Internet. We obtain high-quality data through a pipeline that combines humans and models, and carefully classify the samples according to the frequency of answer changes to facilitate a more fine-grained observation of LLMs’ capabilities. We have also evaluated and analyzed mainstream and advanced Chinese LLMs on CDQA. Extensive experiments and valuable insights suggest that our proposed CDQA is challenging and worthy of more further study. We believe that the benchmark we provide will become one of the key data resources for improving LLMs’ Chinese question-answering ability in the future.

Recently, model merging methods have demonstrated powerful strengths in combining abilities on various tasks from multiple Large Language Models (LLMs). While previous model merging methods mainly focus on merging homogeneous models with identical architecture, they meet challenges when dealing with Multimodal Large Language Models (MLLMs) with inherent heterogeneous property, including differences in model architecture and the asymmetry in the parameter space. In this work, we propose AdaMMS1, a novel model merging method tailored for heterogeneous MLLMs. Our method tackles the challenges in three steps: mapping, merging and searching. Specifically, we first design mapping function between models to apply model merging on MLLMs with different architecture. Then we apply linear interpolation on model weights to actively adapt the asymmetry in the heterogeneous MLLMs. Finally in the hyper-parameter searching step, we propose an unsupervised hyper-parameter selection method for model merging. As the first model merging method capable of merging heterogeneous MLLMs without labeled data, extensive experiments on various model combinations demonstrated that AdaMMS outperforms previous model merging methods on various vision-language benchmarks.2

As language models continue to scale, Large Language Models (LLMs) have exhibited emerging capabilities in In-Context Learning (ICL), enabling them to solve language tasks by prefixing a few in-context demonstrations (ICDs) as context. Inspired by these advancements, researchers have extended these techniques to develop Large Multimodal Models (LMMs) with ICL capabilities. However, existing LMMs face a critical issue: they often fail to effectively leverage the visual context in multimodal demonstrations and instead simply follow textual patterns. This indicates that LMMs do not achieve effective alignment between multimodal demonstrations and model outputs. To address this problem, we propose Symbol Demonstration Direct Preference Optimization (SymDPO). Specifically, SymDPO aims to break the traditional paradigm of constructing multimodal demonstrations by using random symbols to replace text answers within instances. This forces the model to carefully understand the demonstration images and establish a relationship between the images and the symbols to answer questions correctly. We validate the effectiveness of this method on multiple benchmarks, demonstrating that with SymDPO, LMMs can more effectively understand the multimodal context within examples and utilize this knowledge to answer questions better. Code is available at https://github.com/APiaoG/SymDPO.

Despite the advancements made in Vision Large Language Models (VLLMs), like text Large Language Models (LLMs), they have limitations in addressing questions that require real-time information or are knowledgeintensive.Indiscriminately adopting Retrieval Augmented Generation (RAG) techniques is an effective yet expensive way to enable models to answer queries beyond their knowledge scopes.To mitigate the dependence on retrieval and simultaneously maintain, or even improve, the performance benefits provided by retrieval, we propose a method to detect the knowledge boundary of VLLMs, allowing for more efficient use of techniques like RAG.Specifically, we propose a method with two variants that finetune a VLLM on an automatically constructed dataset for boundary identification.Experimental results on various types of Visual Question Answering datasets show that our method successfully depicts a VLLM's knowledge boundary, based on which we are able to reduce indiscriminate retrieval while maintaining or improving the performance.In addition, we show that the knowledge boundary identified by our method for one VLLM can be used as a surrogate boundary for other VLLMs.Code will be released at https://github.com/Chord-Che n-30/VLLM-KnowledgeBoundary

Yikang Liu, Wanyang Zhang, Yiming Wang, Jialong Tang, Pei Zhang, Baosong Yang, Fei Huang, Rui Wang, Hai Hu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Retrieval-Augmented Generation (RAG) systems have emerged as a pivotal methodology for enhancing Large Language Models (LLMs) through the dynamic integration of external knowledge.To further improve RAG's flexibility, Agentic RAG introduces autonomous agents into the workflow.However, Agentic RAG faces several challenges: (1) the success of each step depends on both high-quality planning and accurate search, (2) the lack of supervision for intermediate reasoning steps, and (3) the exponentially large candidate space for planning and searching.To address these challenges, we propose DecoupleSearch, a novel framework that decouples planning and search processes using dual value models, enabling independent optimization of plan reasoning and search grounding.Our approach constructs a reasoning tree, where each node represents planning and search steps.We leverage Monte Carlo Tree Search to assess the quality of each step.During inference, Hierarchical Beam Search iteratively refines planning and search candidates with dual value models.Extensive experiments across policy models of varying parameter sizes, demonstrate the effectiveness of our method.

Large language models (LLMs) exhibit remarkable multilingual capabilities despite Englishdominated pre-training, attributed to crosslingual mechanisms during pre-training.Existing methods for enhancing cross-lingual transfer remain constrained by parallel resources, suffering from limited linguistic and domain coverage.We propose Cross-lingual In-context Pre-training (CrossIC-PT), a simple and scalable approach that enhances cross-lingual transfer by leveraging semantically related bilingual texts via simple next-word prediction.We construct CrossIC-PT samples by interleaving semantic-related bilingual Wikipedia documents into a single context window.To access window size constraints, we implement a systematic segmentation policy to split long bilingual document pairs into chunks while adjusting the sliding window mechanism to preserve contextual coherence.We further extend data availability through a semantic retrieval framework to construct CrossIC-PT samples from web-crawled corpus.Experimental results demonstrate that CrossIC-PT improves multilingual performance on three models (Llama-3.1-8B,Qwen2.5-7B, and Qwen2.5-1.5B)across six target languages, yielding performance gains of 3.79%, 3.99%, and 1.95%, respectively, with additional improvements after data augmentation.

Zekun Xi, Wenbiao Yin, Jizhan Fang, Jialong Wu, Runnan Fang, Yong Jiang, Pengjun Xie, Fei Huang, Huajun Chen, Ningyu Zhang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Yidan Zhang, Yu Wan, Boyi Deng, Baosong Yang, Hao-Ran Wei, Fei Huang, Bowen Yu, Dayiheng Liu, Junyang Lin, Fei Huang, Jingren Zhou. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Ding-Chu Zhang, Yida Zhao, Jialong Wu, Liwen Zhang, Baixuan Li, Wenbiao Yin, Yong Jiang, Yu-Feng Li, Kewei Tu, Pengjun Xie, Fei Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

With the development and widespread application of large language models (LLMs), the new paradigm of "Model as Product" is rapidly evolving, and demands higher capabilities to address complex user needs, often requiring precise workflow execution which involves the accurate understanding of multiple tasks.However, existing benchmarks focusing on single-task environments with limited constraints lack the complexity required to fully reflect real-world scenarios.To bridge this gap, we present the Extremely Complex Instruction Following Benchmark (EIFBENCH), meticulously crafted to facilitate a more realistic and robust evaluation of LLMs.EIFBENCH not only includes multi-task scenarios that enable comprehensive assessment across diverse task types concurrently, but also integrates a variety of constraints, replicating complex operational environments.Furthermore, we propose the Segment Policy Optimization (SegPO) algorithm to enhance the LLM's ability to accurately fulfill multi-task workflow.Evaluations on EIFBENCH have unveiled considerable performance discrepancies in existing LLMs when challenged with these extremely complex instructions.This finding underscores the necessity for ongoing optimization to navigate the intricate challenges posed by LLM applications.

Retrieval-augmented generation (RAG) is a key means to effectively enhance large language models (LLMs) in many knowledge-based tasks. However, existing RAG methods struggle with knowledge-intensive reasoning tasks, because useful information required to these tasks are badly scattered. This characteristic makes it difficult for existing RAG methods to accurately identify key information and perform global reasoning with such noisy augmentation. In this paper, motivated by the cognitive theories that humans convert raw information into various structured knowledge when tackling knowledge-intensive reasoning, we proposes a new framework, StructRAG, which can identify the optimal structure type for the task at hand, reconstruct original documents into this structured format, and infer answers based on the resulting structure. Extensive experiments across various knowledge-intensive tasks show that StructRAG achieves state-of-the-art performance, particularly excelling in challenging scenarios, demonstrating its potential as an effective solution for enhancing LLMs in complex real-world applications.

Multimodal Retrieval Augmented Generation (mRAG) plays an important role in mitigating the “hallucination” issue inherent in multimodal large language models (MLLMs). Although promising, existing heuristic mRAGs typically predefined fixed retrieval processes, which causes two issues: (1) Non-adaptive Retrieval Queries. (2) Overloaded Retrieval Queries. However, these flaws cannot be adequately reflected by current knowledge-seeking visual question answering (VQA) datasets, since the most required knowledge can be readily obtained with a standard two-step retrieval. To bridge the dataset gap, we first construct Dyn-VQA dataset, consisting of three types of ``dynamic'' questions, which require complex knowledge retrieval strategies variable in query, tool, and time: (1) Questions with rapidly changing answers. (2) Questions requiring multi-modal knowledge. (3) Multi-hop questions. Experiments on Dyn-VQA reveal that existing heuristic mRAGs struggle to provide sufficient and precisely relevant knowledge for dynamic questions due to their rigid retrieval processes. Hence, we further propose the first self-adaptive planning agent for multimodal retrieval, **OmniSearch**. The underlying idea is to emulate the human behavior in question solution which dynamically decomposes complex multimodal questions into sub-question chains with retrieval action. Extensive experiments prove the effectiveness of our OmniSearch, also provide direction for advancing mRAG. Code and dataset will be open-sourced.

Large Language Models (LLMs), with their exceptional ability to handle a wide range of tasks, have driven significant advancements in tackling reasoning and planning tasks, wherein decomposing complex problems into executable workflows is a crucial step in this process. Existing workflow evaluation frameworks either focus solely on holistic performance or suffer from limitations such as restricted scenario coverage, simplistic workflow structures, and lax evaluation standards. To this end, we introduce WorfBench, a unified workflow generation benchmark with multi-faceted scenarios and intricate graph workflow structures. Additionally, we present WorfEval, a systemic evaluation protocol utilizing subsequence and subgraph matching algorithms to accurately quantify the LLM agent's workflow generation capabilities. Through comprehensive evaluations across different types of LLMs, we discover distinct gaps between the sequence planning capabilities and graph planning capabilities of LLM agents, with even GPT-4 exhibiting a gap of around 15%. We also train two open-source models and evaluate their generalization abilities on held-out tasks. Furthermore, we observe that the generated workflows can enhance downstream tasks, enabling them to achieve superior performance with less time during inference. Code and dataset are available at https://github.com/zjunlp/WorfBench.

Multi-modal Large Language Models have demonstrated remarkable capabilities in executing instructions for a variety of single-image tasks. Despite this progress, significant challenges remain in modeling long image sequences. In this work, we introduce the versatile multi-modal large language model, mPLUG-Owl3, which enhances the capability for long image-sequence understanding in scenarios that incorporate retrieved image-text knowledge, multimodal in-context examples, and lengthy videos. Specifically, we propose novel hyper attention blocks to efficiently integrate vision and language into a common language-guided semantic space, thereby facilitating the processing of extended multi-image scenarios. We conduct evaluations on 21 benchmarks that cover single/multi-image, and short/long video understanding. mPLUG-Owl3 achieves competitive performance with the state-of-the-art methods while reducing inference time and memory usage by 87.8\% and 48.5\% in average. Moreover, we propose a Distractor Resistance evaluation to assess the ability of models to maintain focus amidst distractions. mPLUG-Owl3 also demonstrates outstanding performance in distractor resistance on ultra-long visual sequence inputs. We hope that mPLUG-Owl3 can contribute to the development of more efficient and powerful multimodal large language models.

Large language models (LLMs) achieve state-of-the-art performance on multiple language tasks, yet their safety guardrails can be circumvented, leading to harmful generations. In light of this, recent research on safety mechanisms has emerged, revealing that when safety representations or component are suppressed, the safety capability of LLMs are compromised. However, existing research tends to overlook the safety impact of multi-head attention mechanisms, despite their crucial role in various model functionalities. Hence, in this paper, we aim to explore the connection between standard attention mechanisms and safety capability to fill this gap in the safety-related mechanistic interpretability. We propose an novel metric which tailored for multi-head attention, the Safety Head ImPortant Score (Ships), to assess the individual heads' contributions to model safety. Base on this, we generalize Ships to the dataset level and further introduce the Safety Attention Head AttRibution Algorithm (Sahara) to attribute the critical safety attention heads inside the model. Our findings show that special attention head has a significant impact on safety. Ablating a single safety head allows aligned model (e.g., Llama-2-7b-chat) to respond to **16$\times\uparrow$** more harmful queries, while only modifying **0.006\%** $\downarrow$ of the parameters, in contrast to the $\sim$ **5\%** modification required in previous studies. More importantly, we demonstrate that attention heads primarily function as feature extractors for safety and models fine-tuned from the same base model exhibit overlapping safety heads through comprehensive experiments. Together, our attribution approach and findings provide a novel perspective for unpacking the black box of safety mechanisms in large models.

Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.

Endowing large language models (LLMs) with continual learning (CL) capacities is practically important, which enables them to dynamically acquire new knowledge over time. Although many effective methods have been proposed for CL of LLMs, they did not consider online scenarios, thereby sharing a common problem: information leakage (IL), where the task-related information of learned tasks is accessed or reused again. IL not only imposes potential risks on data privacy protection but also significantly hinders the deployment of LLMs in real-world scenarios. To avoid IL while maintaining outstanding CL performance, we propose a novel CL method for LLMs, which first characterizes a parameter-efficient fine-tuning (PEFT) block by a presentative feature distribution, and then dynamically selects the appropriate PEFT blocks for each instance based on its similarity with the presentative feature distributions. Extensive experiments validate the effectiveness of our method on the CL of LLM, showcasing its potential to enhance both privacy and adaptability in practical applications.

This paper presents the first study on adapting the visual in-context learning (V-ICL) paradigm to optical character recognition tasks, specifically focusing on text removal and segmentation. Most existing V-ICL generalists employ a reasoning-as-reconstruction approach: they turn to using a straightforward image-label compositor as the prompt and query input, and then masking the query label to generate the desired output. This direct prompt confines the model to a challenging single-step reasoning process. To address this, we propose a task-chaining compositor in the form of image-removal-segmentation, providing an enhanced prompt that elicits reasoning with enriched intermediates. Additionally, we introduce context-aware aggregation, integrating the chained prompt pattern into the latent query representation, thereby strengthening the model’s in-context reasoning. We also consider the issue of visual heterogeneity, which complicates the selection of homogeneous demonstrations in text recognition. Accordingly, this is effectively addressed through a simple self-prompting strategy, preventing the model’s in-context learnability from devolving into specialist-like, context-free inference. Collectively, these insights culminate in our ConText model, which achieves new state-of-the-art across both in- and out-of-domain benchmarks. The code is available at https://github.com/Ferenas/ConText.

As Large Language Model (LLM) context windows expand, the necessity of Retrieval-Augmented Generation (RAG) for integrating external knowledge is debated. Existing RAG vs. long-context (LC) LLM comparisons are often inconclusive due to benchmark limitations. We introduce LaRA, a novel benchmark with 2326 test cases across four QA tasks and three long context types, for rigorous evaluation. Our analysis of eleven LLMs reveals the optimal choice between RAG and LC depends on a complex interplay of model capabilities, context length, task type, and retrieval characteristics, offering actionable guidelines for practitioners. Our code and dataset is provided at:https://github.com/Alibaba-NLP/LaRA

Recent advancements in software engineering agents have demonstrated promising capabilities in automating program improvements. However, their reliance on closed-source or resource-intensive models introduces significant deployment challenges in private environments, prompting a critical question: How can personally deployable open-source LLMs (e.g., 32B models running on a single GPU) achieve comparable code reasoning performanceƒ To this end, we propose a unified Test-Time Compute (TTC) scaling framework that leverages increased inference-time computation instead of larger models. Our framework incorporates two complementary strategies: internal TTC and external TTC. Internally, we introduce a development-contextualized trajectory synthesis method leveraging real-world software repositories to bootstrap multi-stage reasoning processes, such as fault localization and patch generation. We further enhance trajectory quality through rejection sampling, rigorously evaluating trajectories along accuracy and complexity. Externally, we propose a novel development-process-based search strategy guided by reward models and execution verification. This approach enables targeted computational allocation at critical development decision points, overcoming limitations of existing "end-point only" verification methods.Evaluations on SWE-bench Verified demonstrate our 32B model achieves a 46% issue resolution rate, surpassing significantly larger models such as DeepSeek R1 671B and OpenAI o1. Additionally, we provide the empirical validation of the test-time scaling phenomenon within SWE agents, revealing that models dynamically allocate more tokens to increasingly challenging problems, effectively enhancing reasoning capabilities. We publicly release all training data, models, and code to facilitate future research.1. In fact, our method has been deployed in Tongyi Lingma, an IDE-based coding assistant developed by Alibaba Cloud, where it helps developers solve real-world programming problems.

In recent years, the recommendation content on e-commerce platforms has become increasingly rich -- a single user feed may contain multiple entities, such as selling products, short videos, and content posts. To deal with the multi-entity cross-domain recommendation problem, an intuitive solution is to adopt the shared-network-based architecture for joint training. The underlying idea is to transfer the knowledge from one type of entity (source entity) to another (target entity). However, different from the conventional same-entity cross-domain recommendation, multi-entity knowledge transfer encounters several important challenges: (1) data distributions of the source entity and target entity are naturally different, making the shared-network-based joint training susceptible to the negative transfer issue(2) the corresponding feature schema of each entity is not exactly aligned (e.g., price is an essential feature for selling product while missing for content posts), making the existing approaches no longer appropriate. Recent researchers have also experimented with the pre-training and fine-tuning paradigm. Again, they only take into account the scenarios with the same entity type and feature schemas, which is inappropriate in our case. To this end, we design a pre-training & fine-tuning based Multi-entity Knowledge Transfer framework called MKT. MKT utilizes a multi-entity pre-training module to extract transferable knowledge across different entities. In particular, a feature alignment module is first applied to scale and align different feature schemas. Afterward, a couple of knowledge extractors are employed to extract the common and independent knowledge. In the end, the extracted common knowledge is adopted for target entity model training. Through extensive offline and online experiments on public and industrial datasets, we demonstrated the superiority of MKT over multiple State-Of-The-Art methods. MKT has also been deployed for content post recommendations on our production system.

Most recommendation systems typically follow a product-based paradigm utilizing user-product interactions to identify the most engaging items for users. However, this product-based paradigm has notable drawbacks for Xianyu~. Xianyu is China's largest online C2C e-commerce platform where a large portion of the product are post by individual sellers. Most of the product on Xianyu posted from individual sellers often have limited stock available for distribution, and once the product is sold, it's no longer available for distribution. This result in most items distributed product on Xianyu having relatively few interactions, affecting the effectiveness of traditional recommendation depending on accumulating user-item interactions. To address these issues, we introduce IU4Rec, an Interest Unit-based two-stage Recommendation system framework. We first group products into clusters based on attributes such as category, image, and semantics. These IUs are then integrated into the Recommendation system, delivering both product and technological innovations. IU4Rec begins by grouping products into clusters based on attributes such as category, image, and semantics, forming Interest Units (IUs). Then we redesign the recommendation process into two stages. In the first stage, the focus is on recommend these Interest Units, capturing broad-level interests. In the second stage, it guides users to find the best option among similar products within the selected Interest Unit. User-IU interactions are incorporated into our ranking models, offering the advantage of more persistent IU behaviors compared to item-specific interactions. This interest unit based recommendation can be beneficial from mitigating the side effect of limited-stock problem, since most interaction can be gathered on interest units which can persist and accumulate over time. Experimental results on the production dataset and online A/B testing demonstrate the effectiveness and superiority of our proposed IU-centric recommendation approach. This study not only advances recommendation technologies but also emphasizes the potential for co-evolution between product innovations and the technologies involved in item supply and distribution.

As large language models (LLMs) are increasingly deployed in real-world applications, ensuring the safety of their outputs during decoding has become a critical challenge. However, existing decoding-time interventions, such as Contrastive Decoding, often force a severe trade-off between safety and response quality. In this work, we propose **CARE**, a novel framework for decoding-time safety alignment that integrates three key components: (1) a guard model for real-time safety monitoring, enabling detection of potentially unsafe content; (2) a rollback mechanism with a token buffer to correct unsafe outputs efficiently at an earlier stage without disrupting the user experience; and (3) a novel introspection-based intervention strategy, where the model generates self-reflective critiques of its previous outputs and incorporates these reflections into the context to guide subsequent decoding steps. The framework achieves a superior safety-quality trade-off by using its guard model for precise interventions, its rollback mechanism for timely corrections, and our novel introspection method for effective self-correction. Experimental results demonstrate that our framework achieves a superior balance of safety, quality, and efficiency, attaining a **low harmful response rate** and **minimal disruption to the user experience** while **maintaining high response quality**.

Recently, reasoning-based MLLMs have achieved a degree of success in generating long-form textual reasoning chains. However, they still struggle with complex tasks that necessitate dynamic and iterative focusing on and revisiting of visual regions to achieve precise grounding of textual reasoning in visual evidence. We introduce VLM-R³ (Visual Language Model with Region Recognition, Reasoning, and Refinement ), a framework that equips an MLLM with the ability to (i) decide when additional visual evidence is needed, (ii) determine where to ground within the image, and (iii) seamlessly weave the relevant sub-image content back into an interleaved chain-of-thought. The core of our method is \textbf{Region-Conditioned Reinforcement Policy Optimization (R-GRPO)}, a training paradigm that rewards the model for selecting informative regions, formulating appropriate transformations (e.g. crop, zoom), and integrating the resulting visual context into subsequent reasoning steps. To bootstrap this policy, we compile a modest but carefully curated Visuo-Lingual Interleaved Rationale (VLIR) corpus that provides step-level supervision on region selection and textual justification. Extensive experiments on MathVista, ScienceQA, and other benchmarks show that VLM-R$^3$ sets a new state of the art in zero-shot and few-shot settings, with the largest gains appearing on questions demanding subtle spatial reasoning or fine-grained visual cue extraction.

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.

Effectively retrieving, reasoning and understanding visually rich information remains a challenge for traditional Retrieval-Augmented Generation (RAG) methods. On the one hand, traditional text-based methods cannot handle visual-related information. On the other hand, current vision-based RAG approaches are often limited by fixed pipelines and frequently struggle to reason effectively due to the insufficient activation of the fundamental capabilities of models. As reinforcement learning (RL) has been proven to be beneficial for model reasoning, we introduce VRAG-RL, a novel RL framework tailored for complex reasoning across visually rich information. With this framework, VLMs interact with search engines, autonomously sampling single-turn or multi-turn reasoning trajectories with the help of visual perception tokens and undergoing continual optimization based on these samples. Our approach highlights key limitations of RL in RAG domains: (i) Prior Multi-modal RAG approaches tend to merely incorporate images into the context, leading to insufficient reasoning token allocation and neglecting visual-specific perception; and (ii) When models interact with search engines, their queries often fail to retrieve relevant information due to the inability to articulate requirements, thereby leading to suboptimal performance. To address these challenges, we define an action space tailored for visually rich inputs, with actions including cropping and scaling, allowing the model to gather information from a coarse-to-fine perspective. Furthermore, to bridge the gap between users' original inquiries and the retriever, we employ a simple yet effective reward that integrates query rewriting and retrieval performance with a model-based reward. Our VRAG-RL optimizes VLMs for RAG tasks using specially designed RL strategies, aligning the model with real-world applications. Extensive experiments on diverse and challenging benchmarks show that our VRAG-RL outperforms existing methods by 20\% (Qwen2.5-VL-7B) and 30\% (Qwen2.5-VL-3B), demonstrating the effectiveness of our approach. The code is available at https://github.com/Alibaba-NLP/VRAG.

Test-time scaling enhances large language model performance by allocating additional compute resources during decoding. Best-of-$N$ (BoN) sampling serves as a common sampling-based scaling technique, broadening the search space in parallel to find better solutions from the model distribution. However, its cost–performance trade-off is still underexplored. Two main challenges limit the efficiency of BoN sampling: (1) Generating $N$ full samples consumes substantial GPU memory, reducing inference capacity under limited resources. (2) Reward models add extra memory and latency overhead, and training strong reward models introduces potential training data costs. Although some studies have explored efficiency improvements, none have addressed both challenges at once. To address this gap, we propose **Self-Truncation Best-of-$N$ (ST-BoN)**, a decoding method that avoids fully generating all $N$ samples and eliminates the need for reward models. It leverages early sampling consistency in the model’s internal states to identify the most promising path and truncate suboptimal ones. In terms of cost, ST-BoN reduces dynamic GPU memory usage by over 80% and inference latency by 50%. In terms of cost–performance trade-off, ST-BoN achieves the same performance as Full-BoN while saving computational cost by 70%–80%, and under the same cost, it can improve accuracy by 3–4 points.

In this paper, we introduce PolyMath, a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs.We conduct a comprehensive evaluation for advanced LLMs and find that even Qwen-3-235B-A22B-Thinking and Gemini-2.5-pro, achieve only 54.6 and 52.2 benchmark scores, with about 40% accuracy under the highest level.From a language perspective, our benchmark reveals several key challenges of LLMs in multilingual reasoning:(1) Reasoning performance varies widely across languages for current LLMs;(2) Input-output language consistency is low in reasoning LLMs and may be correlated with performance;(3) The thinking length differs significantly by language for current LLMs.Additionally, we demonstrate that controlling the output language in the instructions has the potential to affect reasoning performance, especially for some low-resource languages, suggesting a promising direction for improving multilingual capabilities in LLMs.

In recent years, Multimodal Large Language Models (MLLMs) have been extensively utilized for multimodal reasoning tasks, including Graphical User Interface (GUI) automation. Unlike general offline multimodal tasks, GUI automation is executed in online interactive environments, necessitating step-by-step decision-making based on the real-time status of the environment. This task has a lower tolerance for decision-making errors at each step, as any mistakes may cumulatively disrupt the process and potentially lead to irreversible outcomes like deletions or payments. To address these issues, we introduce a pre-operative critic mechanism that provides effective feedback prior to the actual execution, by reasoning about the potential outcome and correctness of actions. Specifically, we propose a Suggestion-aware Group Relative Policy Optimization (S-GRPO) strategy to construct our pre-operative critic model GUI-Critic-R1, incorporating a novel suggestion reward to enhance the reliability of the model's feedback. Furthermore, we develop a reasoning-bootstrapping based data collection pipeline to create a GUI-Critic-Train and a GUI-Critic-Test, filling existing gaps in GUI critic data. Static experiments on the GUI-Critic-Test across both mobile and web domains reveal that our GUI-Critic-R1 offers significant advantages in critic accuracy compared to current MLLMs. Dynamic evaluation on GUI automation benchmark further highlights the effectiveness and superiority of our model, as evidenced by improved success rates and operational efficiency. The code is available at https://github.com/X-PLUG/MobileAgent/tree/main/GUI-Critic-R1.

Addressing intricate real-world problems necessitates in-depth information seeking and multi-step reasoning. Recent progress in agentic systems, exemplified by Deep Research, underscores the potential for autonomous multi-step research. In this work, we present a cohesive paradigm for building end-to-end agentic information seeking agents from a data-centric and training-stage perspective. Our approach consists of four key stages: (1) browsing data construction, (2) trajectories sampling, (3) supervised fine-tuning for effective cold start, and (4) reinforcement learning for enhanced generalisation. We instantiate this framework in a web agent based on the ReAct format, WebDancer. Empirical evaluations on the challenging GAIA and WebWalkerQA benchmarks demonstrate the strong performance of WebDancer, achieving considerable results and highlighting the efficacy of our training paradigm. Further analysis of agent training provides valuable insights and actionable, systematic pathways for developing more capable agentic models.

Recent advancements in large language models (LLMs) have significantly enhanced text generation capabilities, yet evaluating their performance in generative writing remains a challenge. Existing benchmarks primarily focus on generic text generation or limited in writing tasks, failing to capture the diverse requirements of high-quality written contents across various domains. To bridge this gap, we present WritingBench, a comprehensive benchmark designed to evaluate LLMs across 6 core writing domains and 100 subdomains. We further propose a query-dependent evaluation framework that empowers LLMs to dynamically generate instance-specific assessment criteria. This framework is complemented by a fine-tuned critic model for criteria-aware scoring, enabling evaluations in style, format and length. The framework's validity is further demonstrated by its data curation capability, which enables a 7B-parameter model to outperform the performance of GPT-4o in writing. We open-source the benchmark, along with evaluation tools and modular framework components, to advance the development of LLMs in writing.

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

Large language models (LLMs) often contain misleading content, emphasizing the need to align them with human values to ensure secure AI systems. Reinforcement learning from human feedback (RLHF) has been employed to achieve this alignment. However, it encompasses two main drawbacks: (1) RLHF exhibits complexity, instability, and sensitivity to hyperparameters in contrast to SFT. (2) Despite massive trial-and-error, multiple sampling is reduced to pair-wise contrast, thus lacking contrasts from a macro perspective. In this paper, we propose Preference Ranking Optimization (PRO) as an efficient SFT algorithm to directly fine-tune LLMs for human alignment. PRO extends the pair-wise contrast to accommodate preference rankings of any length. By iteratively contrasting candidates, PRO instructs the LLM to prioritize the best response while progressively ranking the rest responses. In this manner, PRO effectively transforms human alignment into aligning the probability ranking of n responses generated by LLM with the preference ranking of humans towards these responses. Experiments have shown that PRO outperforms baseline algorithms, achieving comparable results to ChatGPT and human responses through automatic-based, reward-based, GPT-4, and human evaluations.

Recently, instruction-following Large Language Models (LLMs) , represented by ChatGPT, have exhibited exceptional performance in general Natural Language Processing (NLP) tasks. However, the unique characteristics of E-commerce data pose significant challenges to general LLMs. An LLM tailored specifically for E-commerce scenarios, possessing robust cross-dataset/task generalization capabilities, is a pressing necessity. To solve this issue, in this work, we proposed the first E-commerce instruction dataset EcomInstruct, with a total of 2.5 million instruction data. EcomInstruct scales up the data size and task diversity by constructing atomic tasks with E-commerce basic data types, such as product information, user reviews. Atomic tasks are defined as intermediate tasks implicitly involved in solving a final task, which we also call Chain-of-Task tasks. We developed EcomGPT with different parameter scales by training the backbone model BLOOMZ with the EcomInstruct. Benefiting from the fundamental semantic understanding capabilities acquired from the Chain-of-Task tasks, EcomGPT exhibits excellent zero-shot generalization capabilities. Extensive experiments and human evaluations demonstrate that EcomGPT outperforms ChatGPT in term of cross-dataset/task generalization on E-commerce tasks. The EcomGPT will be public at https://github.com/Alibaba-NLP/EcomGPT.

Large language models (LLMs) have shown impressive abilities for open-domain NLP tasks. However, LLMs are sometimes too footloose for natural language understanding (NLU) tasks which always have restricted output and input format. Their performances on NLU tasks are highly related to prompts or demonstrations and are shown to be poor at performing several representative NLU tasks, such as event extraction and entity typing. To this end, we present SeqGPT, a bilingual (i.e., English and Chinese) open-source autoregressive model specially enhanced for open-domain natural language understanding. We express all NLU tasks with two atomic tasks, which define fixed instructions to restrict the input and output format but still ``open'' for arbitrarily varied label sets. The model is first instruction-tuned with extremely fine-grained labeled data synthesized by ChatGPT and then further fine-tuned by 233 different atomic tasks from 152 datasets across various domains. The experimental results show that SeqGPT has decent classification and extraction ability, and is capable of performing language understanding tasks on unseen domains. We also conduct empirical studies on the scaling of data and model size as well as on the transfer across tasks. Our models are accessible at https://github.com/Alibaba-NLP/SeqGPT.

Chi Chen, Yiyang Du, Zheng Fang, Ziyue Wang, Fuwen Luo, Peng Li, Ming Yan, Ji Zhang, Fei Huang, Maosong Sun, Yang Liu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

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.

Yunshui Li, Binyuan Hui, Xiaobo Xia, Jiaxi Yang, Min Yang, Lei Zhang, Shuzheng Si, Ling-Hao Chen, Junhao Liu, Tongliang Liu, Fei Huang, Yongbin Li. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Ziyue Wang, Chi Chen, Yiqi Zhu, Fuwen Luo, Peng Li, Ming Yan, Ji Zhang, Fei Huang, Maosong Sun, Yang Liu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Jiaxi Yang, Binyuan Hui, Min Yang, Bailin Wang, Bowen Li, Binhua Li, Fei Huang, Yongbin Li. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

With the rapid growth of geographic applications, automatable and intelligent models are essential to be designed to handle the large volume of information. However, few researchers focus on geographic natural language processing, and there has never been a benchmark to build a unified standard. In this work, we propose a GeoGraphic Language Understanding Evaluation benchmark, named GeoGLUE. We collect data from open-released geographic resources and introduce six natural language understanding tasks, including geographic textual similarity on recall, geographic textual similarity on rerank, geographic elements tagging, geographic composition analysis, geographic where what cut, and geographic entity alignment. We also provide evaluation experiments and analysis of general baselines, indicating the effectiveness and significance of the GeoGLUE benchmark ( https://modelscope.cn/datasets/iic/GeoGLUE/summary ).

Alignment with human preference prevents large language models (LLMs) from generating misleading or toxic content while requiring high-cost human feedback. Assuming resources of human annotation are limited, there are two different ways of allocating considered: more diverse PROMPTS or more diverse RESPONSES to be labeled. Nonetheless, a straightforward comparison between their impact is absent. In this work, we first control the diversity of both sides according to the number of samples for fine-tuning, which can directly reflect their influence. We find that instead of numerous prompts, more responses but fewer prompts better trigger LLMs for human alignment. Additionally, the concept of diversity for prompts can be more complex than responses that are typically quantified by single digits. Consequently, a new formulation of prompt diversity is proposed, further implying a linear correlation with the final performance of LLMs after fine-tuning. We also leverage it on data augmentation and conduct experiments to show its effect on different algorithms.

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.

Out-of-Domain (OOD) intent detection is vital for practical dialogue systems, and it usually requires considering multi-turn dialogue contexts. However, most previous OOD intent detection approaches are limited to single dialogue turns. In this paper, we introduce a context-aware OOD intent detection (Caro) framework to model multi-turn contexts in OOD intent detection tasks. Specifically, we follow the information bottleneck principle to extract robust representations from multi-turn dialogue contexts. Two different views are constructed for each input sample and the superfluous information not related to intent detection is removed using a multi-view information bottleneck loss. Moreover, we also explore utilizing unlabeled data in Caro. A two-stage training process is introduced to mine OOD samples from these unlabeled data, and these OOD samples are used to train the resulting model with a bootstrapping approach. Comprehensive experiments demonstrate that Caro establishes state-of-the-art performances on multi-turn OOD detection tasks by improving the F1-OOD score of over 29% compared to the previous best method.

In vision-language pre-training (VLP), masked image modeling (MIM) has recently been introduced for fine-grained cross-modal alignment. However, in most existing methods, the reconstruction targets for MIM lack high-level semantics, and text is not sufficiently involved in masked modeling. These two drawbacks limit the effect of MIM in facilitating cross-modal semantic alignment. In this work, we propose a semantics-enhanced cross-modal MIM framework (SemMIM) for vision-language representation learning. Specifically, to provide more semantically meaningful supervision for MIM, we propose a local semantics enhancing approach, which harvest high-level semantics from global image features via self-supervised agreement learning and transfer them to local patch encodings by sharing the encoding space. Moreover, to achieve deep involvement of text during the entire MIM process, we propose a text-guided masking strategy and devise an efficient way of injecting textual information in both masked modeling and reconstruction target acquisition. Experimental results validate that our method improves the effectiveness of the MIM task in facilitating cross-modal semantic alignment. Compared to previous VLP models with similar model size and data scale, our SemMIM model achieves state-of-the-art or competitive performance on multiple downstream vision-language tasks.

In video-text retrieval, most existing methods adopt the dual-encoder architecture for fast retrieval, which employs two individual encoders to extract global latent representations for videos and texts. However, they face challenges in capturing fine-grained semantic concepts. In this work, we propose the UNIFY framework, which learns lexicon representations to capture fine-grained semantics and combines the strengths of latent and lexicon representations for video-text retrieval. Specifically, we map videos and texts into a pre-defined lexicon space, where each dimension corresponds to a semantic concept. A two-stage semantics grounding approach is proposed to activate semantically relevant dimensions and suppress irrelevant dimensions. The learned lexicon representations can thus reflect fine-grained semantics of videos and texts. Furthermore, to leverage the complementarity between latent and lexicon representations, we propose a unified learning scheme to facilitate mutual learning via structure sharing and self-distillation. Experimental results show our UNIFY framework largely outperforms previous video-text retrieval methods, with 4.8% and 8.2% Recall@1 improvement on MSR-VTT and DiDeMo respectively.

Training large language models (LLMs) with open-domain instruction data has yielded remarkable success in aligning to end tasks and human preferences. Extensive research has highlighted the importance of the quality and diversity of instruction data. However, the impact of data complexity, as a crucial metric, remains relatively unexplored from three aspects: (1)where the sustainability of performance improvements with increasing complexity is uncertain; (2)whether the improvement brought by complexity merely comes from introducing more training tokens; and (3)where the potential benefits of incorporating instructions from easy to difficult are not yet fully understood. In this paper, we propose Tree-Instruct to systematically enhance the instruction complexity in a controllable manner. By adding a specified number of nodes to instructions’ semantic trees, this approach not only yields new instruction data from the modified tree but also allows us to control the difficulty level of modified instructions. Our preliminary experiments reveal the following insights: (1)Increasing complexity consistently leads to sustained performance improvements of LLMs. (2)Under the same token budget, a few complex instructions outperform diverse yet simple instructions. (3)Curriculum instruction tuning might not yield the anticipated results; focusing on increasing complexity appears to be the key.

Multi-modal large language models (MLLMs) have been shown to efficiently integrate natural language with visual information to handle multi-modal tasks. However, MLLMs still face a fundamental limitation of hallucinations, where they tend to generate erroneous or fabricated information. In this paper, we address hallucinations in MLLMs from a novel perspective of representation learning. We first analyzed the representation distribution of textual and visual tokens in MLLM, revealing two important findings: 1) there is a significant gap between textual and visual representations, indicating unsatisfactory cross-modal representation alignment; 2) representations of texts that contain and do not contain hallucinations are entangled, making it challenging to distinguish them. These two observations inspire us with a simple yet effective method to mitigate hallucinations. Specifically, we introduce contrastive learning into MLLMs and use text with hallucination as hard negative examples, naturally bringing representations of non-hallucinative text and visual samples closer while pushing way representations of non-hallucinating and hallucinative text. We evaluate our method quantitatively and qualitatively, showing its effectiveness in reducing hallucination occurrences and improving performance across multiple benchmarks. On the MMhal-Bench benchmark, our method obtains a 34.66% /29.5% improvement over the baseline MiniGPT-4/LLaVA. Our code is available on https://github.com/X-PLUG/mPLUG-HalOwl/tree/main/hacl.

Recently, visually-situated text parsing (VsTP) has experienced notable advancements, driven by the increasing demand for automated document understanding and the emergence of Generative Large Language Models (LLMs) capable of processing document-based questions. Various methods have been proposed to address the challenging problem of VsTP. However, due to the diversified targets and heterogeneous schemas, previous works usually design task-specific architectures and objectives for individual tasks, which in- advertently leads to modal isolation and complex workflow. In this paper, we propose a unified paradigm for parsing visually-situated text across diverse scenarios. Specifically, we devise a universal model, called OmniParser, which can simultaneously handle three typical visually-situated text parsing tasks: text spotting, key information extraction, and table recognition. In OmniParser, all tasks share the unified encoder-decoder architecture, the unified objective: point- conditioned text generation, and the unified input&output representation: prompt & structured sequences. Extensive experiments demonstrate that the proposed OmniParser achieves state-of-the-art (SOTA) or highly competitive performances on 7 datasets for the three visually-situated text parsing tasks, despite its unified, concise design. The code is available at AdvancedLiterateMachinery.

Multi-modal Large Language Models (MLLMs) have demonstrated impressive instruction abilities across various open-ended tasks. However, previous methods primarily fo-cus on enhancing multi-modal capabilities. In this work, we introduce a versatile multi-modal large language model, mPLUG-Owl2, which effectively leverages modality collab-oration to improve performance in both text and multi-modal tasks. mPLUG-Owl2 utilizes a modularized network design, with the language decoder acting as a universal interface for managing different modalities. Specifically, mPLUG-Owl2 incorporates shared functional modules to facilitate modal-ity collaboration and introduces a modality-adaptive module that preserves modality-specific features. Extensive experi-ments reveal that mPLUG-Owl2 is capable of generalizing both text tasks and multi-modal tasks and achieving state-of-the-art performances with a single generic model. Notably, mPLUG-Owl2 is the first MLLM model that demonstrates the modality collaboration phenomenon in both pure-text and multi-modal scenarios, setting a pioneering path in the development of future multi-modal foundation models.

Yong Cao, Ruixue Ding, Boli Chen, Xianzhi Li, Min Chen, Daniel Hershcovich, Pengjun Xie, Fei Huang. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

This work proposes a simple training-free prompt-free approach to leverage large language models (LLMs) for the Chinese spelling correction (CSC) task, which is totally different from all previous CSC approaches.The key idea is to use an LLM as a pure language model in a conventional manner.The LLM goes through the input sentence from the beginning, and at each inference step, produces a distribution over its vocabulary for deciding the next token, given a partial sentence.To ensure that the output sentence remains faithful to the input sentence, we design a minimal distortion model that utilizes pronunciation or shape similarities between the original and replaced characters.Furthermore, we propose two useful reward strategies to address practical challenges specific to the CSC task.Experiments on five public datasets demonstrate that our approach significantly improves LLM performance, enabling them to compete with state-of-the-art domain-general CSC models.

Haowei Liu, Xi Zhang, Haiyang Xu, Yaya Shi, Chaoya Jiang, Ming Yan, Ji Zhang, Fei Huang, Chunfeng Yuan, Bing Li, Weiming Hu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Large Language Model (LLM) agents significantly extend the capabilities of standalone LLMs, empowering them to interact with external tools (e.g., APIs, functions) and complete various tasks in a self-directed fashion.The challenge of tool use demands that LLMs not only understand user queries and generate answers accurately but also excel in task planning, tool invocation, and result summarization.While traditional works focus on training a single LLM with all these capabilities, performance limitations become apparent, particularly with smaller models.To overcome these challenges, we propose a novel approach that decomposes the aforementioned capabilities into a planner, caller, and summarizer.Each component is implemented by a single LLM that focuses on a specific capability and collaborates with others to accomplish the task.This modular framework facilitates individual updates and the potential use of smaller LLMs for building each capability.To effectively train this framework, we introduce a two-stage training paradigm.First, we fine-tune a backbone LLM on the entire dataset without discriminating sub-tasks, providing the model with a comprehensive understanding of the task.Second, the fine-tuned LLM is used to instantiate the planner, caller, and summarizer respectively, which are continually fine-tuned on respective sub-tasks.Evaluation across various tool-use benchmarks illustrates that our proposed multi-LLM framework surpasses the traditional single-LLM approach, highlighting its efficacy and advantages in tool learning.

In-context learning (ICL) has been instrumental in adapting large language models (LLMs) to downstream tasks using correct input-output examples.Recent advances have attempted to improve model performance through principles derived from mistakes, yet these approaches suffer from lack of customization and inadequate error coverage.To address these limitations, we propose Retrieved In-Context Principles (RICP), a novel teacherstudent framework.In RICP, the teacher model analyzes mistakes from the student model to generate reasons and insights for preventing similar mistakes.These mistakes are clustered based on their underlying reasons for developing task-level principles, enhancing the error coverage of principles.During inference, the most relevant mistakes for each question are retrieved to create question-level principles, improving the customization of the provided guidance.RICP is orthogonal to existing prompting methods and does not require intervention from the teacher model during inference.Experimental results across seven reasoning benchmarks reveal that RICP effectively enhances performance when applied to various prompting strategies.

Minzheng Wang, Longze Chen, Fu Cheng, Shengyi Liao, Xinghua Zhang, Bingli Wu, Haiyang Yu, Nan Xu, Lei Zhang, Run Luo, Yunshui Li, Min Yang, Fei Huang, Yongbin Li. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Charts are important for presenting and explaining complex data relationships.Recently, multimodal large language models (MLLMs) have shown remarkable capabilities in chart understanding.However, the sheer size of these models limits their use in resource-constrained environments.In this paper, we present Tiny-Chart, an efficient MLLM for chart understanding with only 3B parameters.TinyChart overcomes two key challenges in efficient chart understanding: (1) reduce the burden of learning numerical computations through Programof-Thoughts (PoT) learning, which trains the model to generate Python programs for numerical calculations, and (2) reduce lengthy vision feature sequences through Vision Token Merging, which gradually merges most similar vision tokens.Extensive experiments demonstrate that our 3B TinyChart achieves SOTA performance on various chart understanding benchmarks including ChartQA, Chart-to-Text, Chart-to-Table, OpenCQA, and ChartX.It outperforms several chart-understanding MLLMs with up to 13B parameters, and close-sourced MLLM GPT-4V on ChartQA, with higher throughput during inference due to a smaller model scale and more efficient vision encoding.

Recently, substantial advancements in pre-trained visionlanguage models have greatly enhanced the capabilities of multi-modal dialog systems. These models have demonstrated significant improvements by fine-tuning on downstream tasks. However, the existing pre-trained models primarily focus on effectively capturing the alignment between vision and language modalities, often ignoring the intricate nature of dialog context. In this paper, we propose a parameter-efficient prompt-tuning method named DialCLIP for multi-modal dialog retrieval. Specifically, our approach introduces a multi-modal context prompt generator to learn context features which are subsequently distilled into prompts within the pre-trained vision-language model CLIP. Besides, we introduce domain prompt to mitigate the disc repancy from the downstream dialog data. To facilitate various types of retrieval, we also design multiple experts to learn mappings from CLIP outputs to multi-modal representation space, with each expert being responsible to one specific retrieval type. Extensive experiments show that DialCLIP achieves stateof-the-art performance on two widely recognized benchmark datasets (i.e., PhotoChat and MMDialog) by tuning a mere 0.04% of the total parameters. These results highlight the efficacy and efficiency of our proposed approach, underscoring its potential to advance the field of multi-modal dialog retrieval.

In this paper, we unveil that Language Models (LMs) can acquire new capabilities by assimilating parameters from homologous models without retraining or GPUs. We first introduce DARE to set most delta parameters (i.e., the disparity between fine-tuned and pre-trained parameters) to zeros without affecting the abilities of Supervised Fine-Tuning (SFT) LMs, which randomly **D**rops delta parameters with a ratio $p$ **A**nd **RE**scales the remaining ones by $1 / (1 - p)$ to approximate the original embeddings. Then, we use DARE as a versatile plug-in to sparsify delta parameters of multiple SFT homologous models for mitigating parameter interference and merge them into a single model by parameter fusing. We experiment with encoder- and decoder-based LMs, showing that: (1) SFT delta parameter value ranges are typically small (within 0.002) with extreme redundancy, and DARE can effortlessly eliminate 90% or even 99% of them; (2) DARE can merge multiple task-specific LMs into one LM with diverse capabilities. Notably, this phenomenon is more pronounced in large-scale LMs, where the merged LM reveals the potential to surpass the performance of any source LM, providing a new discovery. We also utilize DARE to create a merged LM that ranks first among models with 7 billion parameters on the Open LLM Leaderboard.

Weak diagram analysis abilities of LLMs or Multimodal LLMs greatly limit their application scenarios for scientific academic paper writing. In this work, towards a more versatile copilot for academic paper writing, we mainly focus on strengthening the multi-modal diagram analysis ability of Multimodal LLMs. By parsing Latex source files of academic papers, we carefully build a multi-modal diagram understanding dataset M-Paper. By aligning diagrams in the paper with related paragraphs, we construct professional diagram analysis samples for training and evaluation. M-Paper is the first dataset to support joint comprehension of multiple scientific diagrams, including figures and tables in the format of images or Latex codes. Besides, to better align the copilot with the user's intention, we introduce the 'outline' as the control signal, which could be directly given by the user or revised based on auto-generated ones. Comprehensive experiments with a state-of-the-art Multimodal LLM demonstrate that training on our dataset shows stronger scientific diagram understanding performance. The dataset, code, and model are publicly available at https://github.com/X-PLUG/mPLUG-DocOwl/tree/main/PaperOwl.

Hao Lang, Fei Huang, Yongbin Li. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Xiao Li, Yong Jiang, Shen Huang, Pengjun Xie, Gong Cheng, Fei Huang. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Mobile device operation tasks are increasingly becoming a popular multi-modal AI application scenario. Current Multi-modal Large Language Models (MLLMs), constrained by their training data, lack the capability to function effectively as operation assistants. Instead, MLLM-based agents, which enhance capabilities through tool invocation, are gradually being applied to this scenario. However, the two major navigation challenges in mobile device operation tasks — task progress navigation and focus content navigation — are difficult to effectively solve under the single-agent architecture of existing work. This is due to the overly long token sequences and the interleaved text-image data format, which limit performance. To address these navigation challenges effectively, we propose Mobile-Agent-v2, a multi-agent architecture for mobile device operation assistance. The architecture comprises three agents: planning agent, decision agent, and reflection agent. The planning agent condenses lengthy, interleaved image-text history operations and screens summaries into a pure-text task progress, which is then passed on to the decision agent. This reduction in context length makes it easier for decision agent to navigate the task progress. To retain focus content, we design a memory unit that updates with task progress by decision agent. Additionally, to correct erroneous operations, the reflection agent observes the outcomes of each operation and handles any mistake accordingly. Experimental results indicate that Mobile-Agent-v2 achieves over a 30% improvement in task completion compared to the single-agent architecture of Mobile-Agent. The code is open-sourced at https://github.com/X-PLUG/MobileAgent.

How to evaluate Large Language Models (LLMs) in code generation remains an open question. Many benchmarks have been proposed, but they have two limitations, i.e., data leakage and lack of domain-specific evaluation.The former hurts the fairness of benchmarks, and the latter hinders practitioners from selecting superior LLMs for specific programming domains.To address these two limitations, we propose a new benchmark - EvoCodeBench, which has the following advances: (1) Evolving data. EvoCodeBench will be dynamically updated every period (e.g., 6 months) to avoid data leakage. This paper releases the first version - EvoCodeBench-2403, containing 275 samples from 25 repositories.(2) A domain taxonomy and domain labels. Based on the statistics of open-source communities, we design a programming domain taxonomy consisting of 10 popular domains. Based on the taxonomy, we annotate each sample in EvoCodeBench with a domain label. EvoCodeBench provides a broad platform for domain-specific evaluations.(3) Domain-specific evaluations. Besides the Pass@k, we compute the Domain-Specific Improvement (DSI) and define LLMs' comfort and strange domains. These evaluations help practitioners select superior LLMs in specific domains and discover the shortcomings of existing LLMs.Besides, EvoCodeBench is collected by a rigorous pipeline and aligns with real-world repositories in multiple aspects (e.g., code distributions).We evaluate 8 popular LLMs (e.g., gpt-4, DeepSeek Coder, StarCoder 2) on EvoCodeBench and summarize some insights. EvoCodeBench reveals the actual abilities of these LLMs in real-world repositories. For example, the highest Pass@1 of gpt-4 on EvoCodeBench-2403 is only 20.74%. Besides, we evaluate LLMs in different domains and discover their comfort and strange domains. For example, gpt-4 performs best in most domains but falls behind others in the Internet domain. StarCoder 2-15B unexpectedly performs well in the Database domain and even outperforms 33B LLMs. We release EvoCodeBench, all prompts, and LLMs' completions for further community analysis.

Large language models (LLMs) need knowledge updates to meet the ever-growing world facts and correct the hallucinated responses, facilitating the methods of lifelong model editing. Where the updated knowledge resides in memories is a fundamental question for model editing. In this paper, we find that editing either long-term memory (direct model parameters) or working memory (non-parametric knowledge of neural network activations/representations by retrieval) will result in an impossible triangle---reliability, generalization, and locality can not be realized together in the lifelong editing settings. For long-term memory, directly editing the parameters will cause conflicts with irrelevant pretrained knowledge or previous edits (poor reliability and locality). For working memory, retrieval-based activations can hardly make the model understand the edits and generalize (poor generalization). Therefore, we propose WISE to bridge the gap between memories. In WISE, we design a dual parametric memory scheme, which consists of the main memory for the pretrained knowledge and a side memory for the edited knowledge. We only edit the knowledge in the side memory and train a router to decide which memory to go through when given a query. For continual editing, we devise a knowledge-sharding mechanism where different sets of edits reside in distinct subspaces of parameters, and are subsequently merged into a shared memory without conflicts. Extensive experiments show that WISE can outperform previous model editing methods and overcome the impossible triangle under lifelong model editing of question answering, hallucination, and out-of-distribution settings across trending LLM architectures, e.g., GPT, LLaMA, and Mistral.

This paper presents MaVEn, an innovative Multi-granularity Visual Encoding framework designed to enhance the capabilities of Multimodal Large Language Models (MLLMs) in multi-image reasoning. Current MLLMs primarily focus on single-image visual understanding, limiting their ability to interpret and integrate information across multiple images. MaVEn addresses this limitation by combining discrete visual symbol sequences, which abstract coarse-grained semantic concepts, with traditional continuous representation sequences that model fine-grained features. This dual approach bridges the semantic gap between visual and textual data, thereby improving the model's ability to process and interpret information from multiple images effectively. Additionally, we design a dynamic reduction mechanism by for long-sequence continuous features to enhance multi-image processing efficiency. Experimental results demonstrate that MaVEn significantly enhances MLLMs' understanding in complex multi-image scenarios, while also improving performance in single-image contexts.

Recent endeavors towards directly using large language models (LLMs) as agent models to execute interactive planning tasks have shown commendable results. Despite their achievements, however, they still struggle with brainless trial-and-error in global planning and generating hallucinatory actions in local planning due to their poor understanding of the "real" physical world. Imitating humans' mental world knowledge model which provides global prior knowledge before the task and maintains local dynamic knowledge during the task, in this paper, we introduce parametric World Knowledge Model (WKM) to facilitate agent planning. Concretely, we steer the agent model to self-synthesize knowledge from both expert and sampled trajectories. Then we develop WKM, providing prior task knowledge to guide the global planning and dynamic state knowledge to assist the local planning. Experimental results on three real-world simulated datasets with Mistral-7B, Gemma-7B, and Llama-3-8B demonstrate that our method can achieve superior performance compared to various strong baselines. Besides, we analyze to illustrate that our WKM can effectively alleviate the blind trial-and-error and hallucinatory action issues, providing strong support for the agent's understanding of the world. Other interesting findings include: 1) our instance-level task knowledge can generalize better to unseen tasks, 2) weak WKM can guide strong agent model planning, and 3) unified WKM training has promising potential for further development.

The rise of large language models (LLMs) has significantly transformed both the construction and application of information retrieval (IR) systems. However, current interactions between IR systems and LLMs remain limited, with LLMs merely serving as part of components within IR systems, and IR systems being constructed independently of LLMs. This separated architecture restricts knowledge sharing and deep collaboration between them. In this paper, we introduce Self-Retrieval, a novel end-to-end LLM-driven information retrieval architecture. Self-Retrieval unifies all essential IR functions within a single LLM, leveraging the inherent capabilities of LLMs throughout the IR process. Specifically, Self-Retrieval internalizes the retrieval corpus through self-supervised learning, transforms the retrieval process into sequential passage generation, and performs relevance assessment for reranking. Experimental results demonstrate that Self-Retrieval not only outperforms existing retrieval approaches by a significant margin, but also substantially enhances the performance of LLM-driven downstream applications like retrieval-augmented generation.

Abstract Data-to-text (D2T) generation aims to transform structured data into natural language text. Data-to-text pre-training has proved to be powerful in enhancing D2T generation and yields impressive performance. However, previous pre-training methods either oversimplified structured data into a sequence without considering input structures or designed training objectives tailored for a specific data structure (e.g., table or knowledge graph). In this paper, we unify different types of structured data (i.e., table, key-value data, knowledge graph) into the graph format and cast different D2T generation tasks as graph-to-text generation. To effectively exploit the structural information of the input graph, we propose a structure-enhanced pre-training method for D2T generation by designing a structure-enhanced Transformer. Concretely, we devise a position matrix for the Transformer, encoding relative positional information of connected nodes in the input graph. In addition, we propose a new attention matrix to incorporate graph structures into the original Transformer by taking the available explicit connectivity structure into account. Extensive experiments on six benchmark datasets show the effectiveness of our model. Our source codes are available at https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/unid2t.

Sequence labeling is a crucial task in the NLP community that aims at identifying and assigning spans within the input sentence. It has wide applications in various fields such as information extraction, dialogue system, and sentiment analysis. However, previously proposed span-based or sequence-to-sequence models conduct locating and assigning in order, resulting in problems of error propagation and unnecessary training loss, respectively. This paper addresses the problem by reformulating the sequence labeling as a non-autoregressive set generation to realize locating and assigning in parallel. Herein, we propose aDual-QuerySetGeneration (DQSetGen) model for unified sequence labeling tasks. Specifically, the dual-query set, including a prompted type query and a positional query with anchor span, is fed into the non-autoregressive decoder to probe the spans which correspond to the positional query and have similar patterns with the type query. By avoiding the autoregressive nature of previous approaches, our method significantly improves efficiency and reduces error propagation. Experimental results illustrate that our approach can obtain superior performance on 5 sub-tasks across 11 benchmark datasets. The non-autoregressive nature of our method allows for parallel computation, achieving faster inference speed than compared baselines. In conclusion, our proposed non-autoregressive dual-query set generation method offers a more efficient and accurate approach to sequence labeling tasks in NLP. Its advantages in terms of performance and efficiency make it a promising solution for various applications in data mining and other related fields.

The task of text-to-SQL parsing, which aims at converting natural language questions into executable SQL queries, has garnered increasing attention in recent years. One of the major challenges in text-to-SQL parsing is domain generalization, i.e., how to generalize well to unseen databases. Recently, the pre-trained text-to-text transformer model, namely T5, though not specialized for text-to-SQL parsing, has achieved state-of-the-art performance on standard benchmarks targeting domain generalization. In this work, we explore ways to further augment the pre-trained T5 model with specialized components for text-to-SQL parsing. Such components are expected to introduce structural inductive bias into text-to-SQL parsers thus improving the model’s capacity on (potentially multi-hop) reasoning, which is critical for generating structure-rich SQLs. To this end, we propose a new architecture GRAPHIX-T5, a mixed model with the standard pre-trained transformer model augmented by specially-designed graph-aware layers. Extensive experiments and analysis demonstrate the effectiveness of GRAPHIX-T5 across four text-to-SQL benchmarks: SPIDER, SYN, REALISTIC and DK. GRAPHIX-T5 surpasses all other T5-based parsers with a significant margin, achieving new state-of-the-art performance. Notably, GRAPHIX-T5-large reaches performance superior to the original T5-large by 5.7% on exact match (EM) accuracy and 6.6% on execution accuracy (EX). This even outperforms the T5-3B by 1.2% on EM and 1.5% on EX

Deep neural models (e.g. Transformer) naturally learn spurious features, which create a ``shortcut'' between the labels and inputs, thus impairing the generalization and robustness. This paper advances self-attention mechanism to its robust variant for Transformer-based pre-trained language models (e.g. BERT). We propose Adversarial Self-Attention mechanism (ASA), which adversarially biases the attentions to effectively suppress the model reliance on features (e.g. specific keywords) and encourage its exploration of broader semantics. We conduct comprehensive evaluation across a wide range of tasks for both pre-training and fine-tuning stages. For pre-training, ASA unfolds remarkable performance gain compared to naive training for longer steps. For fine-tuning, ASA-empowered models outweigh naive models by a large margin considering both generalization and robustness.

Real-world data often have an open long-tailed distribution, and building a unified QA model supporting various tasks is vital for practical QA applications.However, it is non-trivial to extend previous QA approaches since they either require access to seen tasks of adequate samples or do not explicitly model samples from unseen tasks.In this paper, we define Open Long-Tailed QA (OLTQA) as learning from long-tailed distributed data and optimizing performance over seen and unseen QA tasks.We propose an OLTQA model that encourages knowledge sharing between head, tail and unseen tasks, and explicitly mines knowledge from a large pre-trained language model (LM).Specifically, we organize our model through a pool of fine-grained components and dynamically combine these components for an input to facilitate knowledge sharing.A retrieve-then-rerank frame is further introduced to select in-context examples, which guild the LM to generate text that express knowledge for QA tasks.Moreover, a twostage training approach is introduced to pretrain the framework by knowledge distillation (KD) from the LM and then jointly train the frame and a QA model through an adaptive mutual KD method.On a large-scale OLTQA dataset we curate from 43 existing QA datasets, our model consistently outperforms the stateof-the-art.We release the code and data

This paper focuses on the task of cross domain few-shot named entity recognition (NER), which aims to adapt the knowledge learned from source domain to recognize named entities in target domain with only a few labeled examples.To address this challenging task, we propose MANNER, a variational memoryaugmented few-shot NER model.Specifically, MANNER uses a memory module to store information from the source domain and then retrieve relevant information from the memory to augment few-shot tasks in the target domain.In order to effectively utilize the information from memory, MANNER uses optimal transport to retrieve and process information from memory, which can explicitly adapt the retrieved information from source domain to target domain and improve the performance in the cross domain few-shot setting.We conduct experiments on both English and Chinese cross domain fewshot NER datasets, and the experimental results demonstrate that MANNER can achieve superior performance 1 .

Cross-modal contrastive learning in vision language pretraining (VLP) faces the challenge of (partial) false negatives.In this paper, we study this problem from the perspective of Mutual Information (MI) optimization.It is common sense that InfoNCE loss used in contrastive learning will maximize the lower bound of MI between anchors and their positives, while we theoretically prove that MI involving negatives also matters when noises commonly exist.Guided by a more general lower bound form for optimization, we propose a contrastive learning strategy regulated by progressively refined cross-modal similarity, to more accurately optimize MI between an image/text anchor and its negative texts/images instead of improperly minimizing it.Our method performs competitively on four downstream cross-modal tasks and systematically balances the beneficial and harmful effects of (partial) false negative samples under theoretical guidance.

Liang Li, Ruiying Geng, Chengyang Fang, Bing Li, Can Ma, Rongyu Cao, Binhua Li, Fei Huang, Yongbin Li. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Perceiving multi-modal information and fulfilling dialogues with humans is a long-term goal of artificial intelligence.Pre-training is commonly regarded as an effective approach for multi-modal dialogue.However, due to the limited availability of multi-modal dialogue data, there is still scarce research on multi-modal dialogue pre-training.Yet another intriguing challenge emerges from the encompassing nature of multi-modal dialogue, which involves various modalities and tasks.Moreover, new forms of tasks may arise at unpredictable points in the future.Hence, it is essential for designed multi-modal dialogue models to possess sufficient flexibility to adapt to such scenarios.This paper proposes PaCE, a unified, structured, compositional multi-modal dialogue pretraining framework.It utilizes a combination of several fundamental experts to accommodate multiple dialogue-related tasks and can be pre-trained using limited dialogue and extensive non-dialogue multi-modal data.Furthermore, we propose a progressive training method where old experts from the past can assist new experts, facilitating the expansion of their capabilities.Experimental results demonstrate that PaCE achieves state-of-the-art results on eight multi-modal dialog benchmarks.

Shuofei Qiao, Yixin Ou, Ningyu Zhang, Xiang Chen, Yunzhi Yao, Shumin Deng, Chuanqi Tan, Fei Huang, Huajun Chen. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Xu Yang, Jiawei Peng, Zihua Wang, Haiyang Xu, Qinghao Ye, Chenliang Li, Songfang Huang, Fei Huang, Zhangzikang Li, Yu Zhang. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

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.

Language models with the Transformers structure have shown great performance in natural language processing.However, there still poses problems when fine-tuning pre-trained language models on downstream tasks, such as over-fitting or representation collapse.In this work, we propose HyPe, a simple yet effective fine-tuning technique to alleviate such problems by perturbing hidden representations of Transformers layers.Unlike previous works that only add noise to inputs or parameters, we argue that the hidden representations of Transformers layers convey more diverse and meaningful language information.Therefore, making the Transformers layers more robust to hidden representation perturbations can further benefit the fine-tuning of PLMs en bloc.We conduct extensive experiments and analyses on GLUE and other natural language inference datasets.Results demonstrate that HyPe outperforms vanilla fine-tuning and enhances generalization of hidden representations from different layers.In addition, HyPe acquires negligible computational overheads, and is better than and compatible with previous state-of-theart fine-tuning techniques.Codes are released at https://github.com/Yuanhy1997/HyPe.

Given a textual passage and an answer, humans are able to ask questions with various expressions, but this ability is still challenging for most question generation (QG) systems.Existing solutions mainly focus on the internal knowledge within the given passage or the semantic word space for diverse content planning.These methods, however, have not considered the potential of external knowledge for expression diversity.To bridge this gap, we propose RAST, a framework for Retrieval-Augmented Style Transfer, where the objective is to utilize the style of diverse templates for question generation.For training RAST, we develop a novel Reinforcement Learning (RL) based approach that maximizes a weighted combination of diversity reward and consistency reward.Here, the consistency reward is computed by a Question-Answering (QA) model, whereas the diversity reward measures how much the final output mimics the retrieved template.Experimental results show that our method outperforms previous diversity-driven baselines on diversity while being comparable in terms of consistency scores.Our code is available at https://github.com/gouqi666/RAST.

Recent research has demonstrated that Large Language Models (LLMs) can enhance their capabilities by utilizing external tools.However, three pivotal questions remain unanswered: (1) How effective are current LLMs in utilizing tools?(2) How can we enhance LLMs' ability to utilize tools?(3) What obstacles need to be overcome to leverage tools?To address these questions, we introduce API-Bank, a groundbreaking benchmark, specifically designed for tool-augmented LLMs.For the first question, we develop a runnable evaluation system consisting of 73 API tools.We annotate 314 tool-use dialogues with 753 API calls to assess the existing LLMs' capabilities in planning, retrieving, and calling APIs.For the second question, we construct a comprehensive training set containing 1,888 tool-use dialogues from 2,138 APIs spanning 1,000 distinct domains.Using this dataset, we train Lynx, a tool-augmented LLM initialized from Alpaca.Experimental results demonstrate that GPT-3.5 exhibits improved tool utilization compared to GPT-3, while GPT-4 excels in planning.However, there is still significant potential for further improvement.Moreover, Lynx surpasses Alpaca's tool utilization performance by more than 26 pts and approaches the effectiveness of GPT-3.5.Through error analysis, we highlight the key challenges for future research in this field to answer the third question 1 .

Previous studies have revealed that vanilla pre-trained language models (PLMs) lack the capacity to handle knowledge-intensive NLP tasks alone; thus, several works have attempted to integrate external knowledge into PLMs.However, despite the promising outcome, we empirically observe that PLMs may have already encoded rich knowledge in their pre-trained parameters but fail to fully utilize them when applying them to knowledgeintensive tasks.In this paper, we propose a new paradigm dubbed Knowledge Rumination to help the pre-trained language model utilize that related latent knowledge without retrieving it from the external corpus.By simply adding a prompt like "As far as I know" to the PLMs, we try to review related latent knowledge and inject them back into the model for knowledge consolidation.We apply the proposed knowledge rumination to various language models, including RoBERTa, De-BERTa, and GPT-3.Experimental results on six commonsense reasoning tasks and GLUE benchmarks demonstrate the effectiveness of our proposed approach, which proves that the knowledge stored in PLMs can be better exploited to enhance performance 1 .

The goal of document-grounded dialogue (DocGD) is to generate a response by anchoring the evidence in a supporting document in accordance with the dialogue context.This entails four causally interconnected variables.While task-specific pre-training has significantly enhanced performances on numerous downstream tasks, existing DocGD methods still rely on general pre-trained language models without a specifically tailored pre-training approach that explicitly captures the causal relationships.To address this, we present the first causallycomplete dataset construction strategy for developing million-scale DocGD pre-training corpora.Additionally, we propose a causallyperturbed pre-training strategy to better capture causality by introducing perturbations on the variables and optimizing the overall causal effect.Experiments conducted on three benchmark datasets demonstrate that our causal pretraining yields substantial and consistent improvements in fully-supervised, low-resource, few-shot, and zero-shot settings 1 .

Vision Transformer (ViT) based Vision-Language Pre-training (VLP) models have demonstrated impressive performance in various tasks. However, the lengthy visual token sequences fed into ViT can lead to training inefficiency and ineffectiveness. Existing efforts address the challenge by either bottom-level patch extraction in the ViT backbone or top-level patch abstraction outside, not balancing training efficiency and effectiveness well. Inspired by text summarization in natural language processing, we propose a Bottom-Up Patch Summarization approach named BUS, coordinating bottom-level extraction and top-level abstraction to learn a concise summary of lengthy visual token sequences efficiently. Specifically, We incorporate a Text-Semantics-Aware Patch Selector (TSPS) into the ViT backbone to perform a coarse-grained visual token extraction and then attach a flexible Transformer-based Patch Abstraction Decoder (PAD) upon the backbone for top-level visual abstraction. This bottom-up collaboration enables our BUS to yield high training efficiency while maintaining or even improving effectiveness. We evaluate our approach on various visual-language understanding and generation tasks and show competitive downstream task performance while boosting the training efficiency by 50%. Additionally, our model achieves state-of-the-art performance on many downstream tasks by increasing input image resolution without increasing computational costs over baselines.

In a video, an object usually appears as the trajectory, i.e., it spans over a few spatial but longer temporal patches, that contains abundant spatiotemporal contexts. However, modern Video-Language BERTs (VDL-BERTs) neglect this trajectory characteristic that they usually follow image-language BERTs (IL-BERTs) to deploy the patch-to-word (P2W) attention that may over-exploit trivial spatial contexts and neglect significant temporal contexts. To amend this, we propose a novel TW-BERT to learn Trajectory-Word alignment by a newly designed trajectory-to-word (T2W) attention for solving video-language tasks. Moreover, previous VDL-BERTs usually uniformly sample a few frames into the model while different trajectories have diverse graininess, i.e., some trajectories span longer frames and some span shorter, and using a few frames will lose certain useful temporal contexts. However, simply sampling more frames will also make pre-training infeasible due to the largely increased training burdens. To alleviate the problem, during the fine-tuning stage, we insert a novel Hierarchical Frame-Selector (HFS) module into the video encoder. HFS gradually selects the suitable frames conditioned on the text context for the later cross-modal encoder to learn better trajectory-word alignments. By the proposed T2W attention and HFS, our TW-BERT achieves SOTA performances on text-to-video retrieval tasks, and comparable performances on video question-answering tasks with some VDL-BERTs trained on much more data. The code will be available in the supplementary material.

Video-language pre-training has advanced the performance of various downstream video-language tasks. However, most previous methods directly inherit or adapt typical image-language pre-training paradigms to video-language pre-training, thus not fully exploiting the unique characteristic of video, i.e., temporal. In this paper, we propose a Hierarchical Temporal-Aware video-language pre-training framework, HiTeA, with two novel pre-training tasks for yielding temporal-aware multi-modal representation with cross-modal fine-grained temporal moment information and temporal contextual relations between video-text multi-modal pairs. First, we propose a cross-modal moment exploration task to explore moments in videos by mining the paired texts, which results in detailed video moment representation. Then, based on the learned detailed moment representations, the inherent temporal contextual relations are captured by aligning video-text pairs as a whole in different time resolutions with multi-modal temporal relation exploration task. Furthermore, we introduce the shuffling test to evaluate the temporal reliance of datasets and video-language pre-training models. We achieve state-of-the-art results on 15 well-established video-language understanding and generation tasks, especially on temporal-oriented datasets (e.g., SSv2-Template and SSv2-Label) with 8.6% and 11.1% improvement respectively. HiTeA also demonstrates strong generalization ability when directly transferred to downstream tasks in a zero-shot manner.

Recent years have witnessed a big convergence of language, vision, and multi-modal pretraining. In this work, we present mPLUG-2, a new unified paradigm with modularized design for multi-modal pretraining, which can benefit from modality collaboration while addressing the problem of modality entanglement. In contrast to predominant paradigms of solely relying on sequence-to-sequence generation or encoder-based instance discrimination, mPLUG-2 introduces a multi-module composition network by sharing common universal modules for modality collaboration and disentangling different modality modules to deal with modality entanglement. It is flexible to select different modules for different understanding and generation tasks across all modalities including text, image, and video. Empirical study shows that mPLUG-2 achieves state-of-the-art or competitive results on a broad range of over 30 downstream tasks, spanning multi-modal tasks of image-text and video-text understanding and generation, and uni-modal tasks of text-only, image-only, and video-only understanding. Notably, mPLUG-2 shows new state-of-the-art results of 48.0 top-1 accuracy and 80.3 CIDEr on the challenging MSRVTT video QA and video caption tasks with a far smaller model size and data scale. It also demonstrates strong zero-shot transferability on vision-language and video-language tasks. Code and models will be released in https://github.com/X-PLUG/mPLUG-2.

Cross-domain NER is a challenging task to address the low-resource problem in practical scenarios. Previous typical solutions mainly obtain a NER model by pre-trained language models (PLMs) with data from a rich-resource domain and adapt it to the target domain. Owing to the mismatch issue among entity types in different domains, previous approaches normally tune all parameters of PLMs, ending up with an entirely new NER model for each domain. Moreover, current models only focus on leveraging knowledge in one general source domain while failing to successfully transfer knowledge from multiple sources to the target. To address these issues, we introduce Collaborative Domain-Prefix Tuning for cross-domain NER (CP-NER) based on text-to-text generative PLMs. Specifically, we present text-to-text generation grounding domain-related instructors to transfer knowledge to new domain NER tasks without structural modifications. We utilize frozen PLMs and conduct collaborative domain-prefix tuning to stimulate the potential of PLMs to handle NER tasks across various domains. Experimental results on the Cross-NER benchmark show that the proposed approach has flexible transfer ability and performs better on both one-source and multiple-source cross-domain NER tasks.

Vision-and-language multi-modal pretraining and fine-tuning have shown great success in visual question answering (VQA). Compared to general domain VQA, the performance of biomedical VQA suffers from limited data. In this paper, we propose a retrieval-augmented pretrain-and-finetune paradigm named RAMM for biomedical VQA to overcome the data limitation issue. Specifically, we collect a new biomedical dataset named PMCPM which offers patient-based image-text pairs containing diverse patient situations from PubMed. Then, we pretrain the biomedical multi-modal model to learn visual and textual representation for image-text pairs and align these representations with image-text contrastive objective (ITC). Finally, we propose a retrieval-augmented method to better use the limited data. We propose to retrieve similar image-text pairs based on ITC from pretraining datasets and introduce a novel retrieval-attention module to fuse the representation of the image and the question with the retrieved images and texts. Experiments demonstrate that our retrieval-augmented pretrain-and-finetune paradigm obtains state-of-the-art performance on Med-VQA2019, Med-VQA2021, VQARAD, and SLAKE datasets. Further analysis shows that the proposed RAMM and PMCPM can enhance biomedical VQA performance compared with previous resources and methods. The pre-trained models and codes are published at https://github.com/GanjinZero/RAMM.

Vision-Language Pre-training (VLP) methods based on object detection enjoy the rich knowledge of fine-grained object-text alignment but at the cost of computationally expensive inference. Recent Visual-Transformer (ViT)-based approaches circumvent this issue while struggling with long visual sequences without detailed cross-modal alignment information. This paper introduces a ViT-based VLP technique that efficiently incorporates object information through a novel patch-text alignment mechanism. Specifically, we convert object-level signals into patch-level ones and devise a Patch-Text Alignment pre-training task (PTA) to learn a text-aware patch detector. By using off-the-shelf delicate object annotations in 5% training images, we jointly train PTA with other conventional VLP objectives in an end-to-end manner, bypassing the high computational cost of object detection and yielding an effective patch detector that accurately detects text-relevant patches, thus considerably reducing patch sequences and accelerating computation within the ViT backbone. Our experiments on a variety of widely-used benchmarks reveal that our method achieves a speedup of nearly 88% compared to prior VLP models while maintaining competitive or superior performance on downstream tasks with similar model size and data scale.

Previous dual-encoder pre-training methods for video-text retrieval employ contrastive learning for cross-modal alignment in a latent space. However, such learned latent spaces often result in modality gap problem [26]. In this paper, we introduce a novel SemVTR framework designed to learn semantics-grounded video-text representations in a vocabulary space, in which each dimension corresponds to a semantic concept represented by a word. The representation is obtained by grounding video and text into semantically-related dimensions with high activation values. As video-text pairs share grounded dimensions, their vocabulary representations are expected to cluster together and thus alleviate modality gap problem. So, the crux of our method lies in grounding video and text into vocabulary space. Specifically, we propose a Multi-Granularity Video Semantics Grounding approach and a Textual Semantics Preserving training strategy. The visualization illustrates that SemVTR obtains semantics-gronded vocabulary representation and also alleviates the modality gap problem. SemVTR significantly outperforms existing methods on four video-text retrieval benchmarks.

Inspired by the recent developments of large language models (LLMs), we propose mPLUG-Octopus, a versatile conversational assistant designed to provide users with coherent, engaging, and helpful interaction experiences in both text-only and multi-modal scenarios. Unlike traditional pipeline chatting systems, mPLUG-Octopus offers a diverse range of creative capabilities including open-domain QA, multi-turn chatting, and multi-modal creation, all built with a unified multimodal LLM without relying on any external API. With the modularized end-to-end multimodal LLM technology, mPLUG-Octopus efficiently facilitates engaging and open-domain conversation experience. It exhibits a wide range of uni/multi-modal elemental capabilities, enabling it to seamlessly communicate with users on open-domain topics and engage in multi-turn conversations. It also assists users in accomplishing various content creation and application tasks. Our conversational assistant can also be deployed on smart hardware to drive advanced AIGC applications.

Expressing universal semantics common to all languages is helpful to understand the meanings of complex and culture-specific sentences. The research theme underlying this scenario focuses on learning universal representations across languages with the usage of massive parallel corpora. However, due to the sparsity and scarcity of parallel data, there is still a big challenge in learning authentic ``universals'' for any two languages. In this paper, we propose Emma-X: an EM-like Multilingual pre-training Algorithm, to learn Cross-lingual universals with the aid of excessive multilingual non-parallel data. Emma-X unifies the cross-lingual representation learning task and an extra semantic relation prediction task within an EM framework. Both the extra semantic classifier and the cross-lingual sentence encoder approximate the semantic relation of two sentences, and supervise each other until convergence. To evaluate Emma-X, we conduct experiments on xrete, a newly introduced benchmark containing 12 widely studied cross-lingual tasks that fully depend on sentence-level representations. Results reveal that Emma-X achieves state-of-the-art performance. Further geometric analysis of the built representation space with three requirements demonstrates the superiority of Emma-X over advanced models.

Text-to-SQL parsing, which aims at converting natural language instructions into executable SQLs, has gained increasing attention in recent years. In particular, GPT-4 and Claude-2 have shown impressive results in this task. However, most of the prevalent benchmarks, i.e., Spider, and WikiSQL, focus on database schema with few rows of database contents leaving the gap between academic study and real-world applications. To mitigate this gap, we present BIRD, a BIg benchmark for laRge-scale Database grounded in text-to-SQL tasks, containing 12,751 pairs of text-to-SQL data and 95 databases with a total size of 33.4 GB, spanning 37 professional domains. Our emphasis on database values highlights the new challenges of dirty database contents, external knowledge between NL questions and database contents, and SQL efficiency, particularly in the context of massive databases. To solve these problems, text-to-SQL models must feature database value comprehension in addition to semantic parsing. The experimental results demonstrate the significance of database values in generating accurate text-to-SQLs for big databases. Furthermore, even the most popular and effective text-to-SQL models, i.e. GPT-4, only achieve 54.89% in execution accuracy, which is still far from the human result of 92.96%, proving that challenges still stand. We also provide an efficiency analysis to offer insights into generating text-to-efficient-SQLs that are beneficial to industries. We believe that BIRD will contribute to advancing real-world applications of text-to-SQL research.The leaderboard and source code are available: https://bird-bench.github.io/.

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/.

While distant supervision has been extensively explored and exploited in NLP tasks like named entity recognition, a major obstacle stems from the inevitable noisy distant labels tagged unsupervisedly. A few past works approach this problem by adopting a self-training framework with a sample-selection mechanism. In this work, we innovatively identify two types of biases that were omitted by prior work, and these biases lead to inferior performance of the distant-supervised NER setup. First, we characterize the noise concealed in the distant labels as highly structural rather than fully randomized. Second, the self-training framework would ubiquitously introduce an inherent bias that causes erroneous behavior in both sample selection and eventually prediction. To cope with these problems, we propose a novel self-training framework, dubbed DesERT. This framework augments the conventional NER predicative pathway to a dual form that effectively adapts the sample-selection process to conform to its innate distributional-bias structure. The other crucial component of DesERT composes a debiased module aiming to enhance the token representations, hence the quality of the pseudo-labels. Extensive experiments are conducted to validate the DesERT. The results show that our framework establishes a new state-of-art performance, it achieves a +2.22% average F1 score improvement on five standardized benchmarking datasets. Lastly, DesERT demonstrates its effectiveness under a new DSNER benchmark where additional distant supervision comes from the ChatGPT model.

Unsupervised domain adaptation (UDA) is a pivotal form in machine learning to extend the in-domain model to the distinctive target domains where the data distributions differ. Most prior works focus on capturing the inter-domain transferability but largely overlook rich intra-domain structures, which empirically results in even worse discriminability. In this work, we introduce a novel graph SPectral Alignment (SPA) framework to tackle the tradeoff. The core of our method is briefly condensed as follows: (i)-by casting the DA problem to graph primitives, SPA composes a coarse graph alignment mechanism with a novel spectral regularizer towards aligning the domain graphs in eigenspaces; (ii)-we further develop a fine-grained message propagation module --- upon a novel neighbor-aware self-training mechanism --- in order for enhanced discriminability in the target domain. On standardized benchmarks, the extensive experiments of SPA demonstrate that its performance has surpassed the existing cutting-edge DA methods. Coupled with dense model analysis, we conclude that our approach indeed possesses superior efficacy, robustness, discriminability, and transferability. Code and data are available at: https://github.com/CrownX/SPA.

Query and point of interest (POI) matching is a core task in location-based services~(LBS), e.g., navigation maps. It connects users' intent with real-world geographic information. Lately, pre-trained language models (PLMs) have made notable advancements in many natural language processing (NLP) tasks. To overcome the limitation that generic PLMs lack geographic knowledge for query-POI matching, related literature attempts to employ continued pre-training based on domain-specific corpus. However, a query generally describes the geographic context (GC) about its destination and contains mentions of multiple geographic objects like nearby roads and regions of interest (ROIs). These diverse geographic objects and their correlations are pivotal to retrieving the most relevant POI. Text-based single-modal PLMs can barely make use of the important GC and are therefore limited. In this work, we propose a novel method for query-POI matching, namely Multi-modal Geographic language model (MGeo), which comprises a geographic encoder and a multi-modal interaction module. Representing GC as a new modality, MGeo is able to fully extract multi-modal correlations to perform accurate query-POI matching. Moreover, there exists no publicly available query-POI matching benchmark. Intending to facilitate further research, we build a new open-source large-scale benchmark for this topic, i.e., Geographic TExtual Similarity (GeoTES). The POIs come from an open-source geographic information system (GIS) and the queries are manually generated by annotators to prevent privacy issues. Compared with several strong baselines, the extensive experiment results and detailed ablation analyses demonstrate that our proposed multi-modal geographic pre-training method can significantly improve the query-POI matching capability of PLMs with or without users' locations. Our code and benchmark are publicly available at https://github.com/PhantomGrapes/MGeo.

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.

Table-based reasoning has shown remarkable progress in a wide range of table-based tasks. It is a challenging task, which requires reasoning over both free-form natural language (NL) questions and (semi-)structured tabular data. However, previous table-based reasoning solutions usually suffer from significant performance degradation on ''huge'' evidence (tables). In addition, most existing methods struggle to reason over complex questions since the essential information is scattered in different places. To alleviate the above challenges, we exploit large language models (LLMs) as decomposers for effective table-based reasoning, which (i) decompose huge evidence (a huge table) into sub-evidence (a small table) to mitigate the interference of useless information for table reasoning, and (ii) decompose a complex question into simpler sub-questions for text reasoning. First, we use a powerful LLM to decompose the evidence involved in the current question into the sub-evidence that retains the relevant information and excludes the remaining irrelevant information from the ''huge'' evidence. Second, we propose a novel ''parsing-execution-filling'' strategy to decompose a complex question into simper step-by-step sub-questions by generating intermediate SQL queries as a bridge to produce numerical and logical sub-questions with a powerful LLM. Finally, we leverage the decomposed sub-evidence and sub-questions to get the final answer with a few in-context prompting examples. Extensive experiments on three benchmark datasets (TabFact, WikiTableQuestion, and FetaQA) demonstrate that our method achieves significantly better results than competitive baselines for table-based reasoning. Notably, our method outperforms human performance for the first time on the TabFact dataset. In addition to impressive overall performance, our method also has the advantage of interpretability, where the returned results are to some extent tractable with the generated sub-evidence and sub-questions. For reproducibility, we release our source code and data at: https://github.com/AlibabaResearch/DAMO-ConvAI.

Natural language generation from structured data mainly focuses on surface-level descriptions, suffering from uncontrollable content selection and low fidelity. Previous works leverage logical forms to facilitate logical knowledge-conditioned text generation. Though achieving remarkable progress, they are data-hungry, which makes the adoption for real-world applications challenging with limited data. To this end, this paper proposes a unified framework for logical knowledge-conditioned text generation in the few-shot setting. With only a few seeds logical forms (e.g., 20/100 shot), our approach leverages self-training and samples pseudo logical forms based on content and structure consistency. Experimental results demonstrate that our approach can obtain better few-shot performance than baselines.

The Visual Question Answering (VQA) task utilizes both visual image and language analysis to answer a textual question with respect to an image. It has been a popular research topic with an increasing number of real-world applications in the last decade. This paper introduces a novel hierarchical integration of vision and language AliceMind-MMU (ALIbaba’s Collection of Encoder-decoders from Machine IntelligeNce lab of Damo academy - MultiMedia Understanding) , which leads to similar or even slightly better results than a human being does on VQA. A hierarchical framework is designed to tackle the practical problems of VQA in a cascade manner including: (1) diverse visual semantics learning for comprehensive image content understanding; (2) enhanced multi-modal pre-training with modality adaptive attention; and (3) a knowledge-guided model integration with three specialized expert modules for the complex VQA task. Treating different types of visual questions with corresponding expertise needed plays an important role in boosting the performance of our VQA architecture up to the human level. An extensive set of experiments and analysis are conducted to demonstrate the effectiveness of the new research work.

Pre-trained models have proved to be powerful in enhancing task-oriented dialog systems. However, current pre-training methods mainly focus on enhancing dialog understanding and generation tasks while neglecting the exploitation of dialog policy. In this paper, we propose GALAXY, a novel pre-trained dialog model that explicitly learns dialog policy from limited labeled dialogs and large-scale unlabeled dialog corpora via semi-supervised learning. Specifically, we introduce a dialog act prediction task for policy optimization during pre-training and employ a consistency regularization term to refine the learned representation with the help of unlabeled dialogs. We also implement a gating mechanism to weigh suitable unlabeled dialog samples. Empirical results show that GALAXY substantially improves the performance of task-oriented dialog systems, and achieves new state-of-the-art results on benchmark datasets: In-Car, MultiWOZ2.0 and MultiWOZ2.1, improving their end-to-end combined scores by 2.5, 5.3 and 5.5 points, respectively. We also show that GALAXY has a stronger few-shot ability than existing models under various low-resource settings. For reproducibility, we release the code and data at https://github.com/siat-nlp/GALAXY.

Pre-trained Language Models (PLMs) have achieved great success in various Natural Language Processing (NLP) tasks under the pre-training and fine-tuning paradigm. With large quantities of parameters, PLMs are computation-intensive and resource-hungry. Hence, model pruning has been introduced to compress large-scale PLMs. However, most prior approaches only consider task-specific knowledge towards downstream tasks, but ignore the essential task-agnostic knowledge during pruning, which may cause catastrophic forgetting problem and lead to poor generalization ability. To maintain both task-agnostic and task-specific knowledge in our pruned model, we propose ContrAstive Pruning (CAP) under the paradigm of pre-training and fine-tuning. It is designed as a general framework, compatible with both structured and unstructured pruning. Unified in contrastive learn- ing, CAP enables the pruned model to learn from the pre-trained model for task-agnostic knowledge, and fine-tuned model for task-specific knowledge. Besides, to better retain the performance of the pruned model, the snapshots (i.e., the intermediate models at each pruning iteration) also serve as effective supervisions for pruning. Our extensive experiments show that adopting CAP consistently yields significant improvements, especially in extremely high sparsity scenarios. With only 3% model parameters reserved (i.e., 97% sparsity), CAP successfully achieves 99.2% and 96.3% of the original BERT performance in QQP and MNLI tasks. In addition, our probing experiments demonstrate that the model pruned by CAP tends to achieve better generalization ability.

Event argument extraction (EAE) is an important task for information extraction to discover specific argument roles. In this study, we cast EAE as a question-based cloze task and empirically analyze fixed discrete token template performance. As generating human-annotated question templates is often time-consuming and labor-intensive, we further propose a novel approach called “Learning to Ask,” which can learn optimized question templates for EAE without human annotations. Experiments using the ACE-2005 dataset demonstrate that our method based on optimized questions achieves state-of-the-art performance in both the few-shot and supervised settings.

Yongliang Shen, Xiaobin Wang, Zeqi Tan, Guangwei Xu, Pengjun Xie, Fei Huang, Weiming Lu, Yueting Zhuang. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Structured pruning has been extensively studied on monolingual pre-trained language models and is yet to be fully evaluated on their multilingual counterparts.This work investigates three aspects of structured pruning on multilingual pre-trained language models: settings, algorithms, and efficiency.Experiments on nine downstream tasks show several counterintuitive phenomena: for settings, individually pruning for each language does not induce a better result; for algorithms, the simplest method performs the best; for efficiency, a fast model does not imply that it is also small.To facilitate the comparison on all sparsity levels, we present Dynamic Sparsification, a simple approach that allows training the model once and adapting to different model sizes at inference.We hope this work fills the gap in the study of structured pruning on multilingual pre-trained models and sheds light on future research.

Ningyu Zhang, Mosha Chen, Zhen Bi, Xiaozhuan Liang, Lei Li, Xin Shang, Kangping Yin, Chuanqi Tan, Jian Xu, Fei Huang, Luo Si, Yuan Ni, Guotong Xie, Zhifang Sui, Baobao Chang, Hui Zong, Zheng Yuan, Linfeng Li, Jun Yan, Hongying Zan, Kunli Zhang, Buzhou Tang, Qingcai Chen. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Most NER methods rely on extensive labeled data for model training, which struggles in the low-resource scenarios with limited training data. Existing dominant approaches usually suffer from the challenge that the target domain has different label sets compared with a resource-rich source domain, which can be concluded as class transfer and domain transfer. In this paper, we propose a lightweight tuning paradigm for low-resource NER via pluggable prompting (LightNER). Specifically, we construct the unified learnable verbalizer of entity categories to generate the entity span sequence and entity categories without any label-specific classifiers, thus addressing the class transfer issue. We further propose a pluggable guidance module by incorporating learnable parameters into the self-attention layer as guidance, which can re-modulate the attention and adapt pre-trained weights. Note that we only tune those inserted module with the whole parameter of the pre-trained language model fixed, thus, making our approach lightweight and flexible for low-resource scenarios and can better transfer knowledge across domains. Experimental results show that LightNER can obtain comparable performance in the standard supervised setting and outperform strong baselines in low-resource settings.

Pre-training methods with contrastive learning objectives have shown remarkable success in dialog understanding tasks. However, current contrastive learning solely considers the self-augmented dialog samples as positive samples and treats all other dialog samples as negative ones, which enforces dissimilar representations even for dialogs that are semantically related. In this paper, we propose SPACE-2, a tree-structured pre-trained conversation model, which learns dialog representations from limited labeled dialogs and large-scale unlabeled dialog corpora via semi-supervised contrastive pre-training. Concretely, we first define a general semantic tree structure (STS) to unify the inconsistent annotation schema across different dialog datasets, so that the rich structural information stored in all labeled data can be exploited. Then we propose a novel multi-view score function to increase the relevance of all possible dialogs that share similar STSs and only push away other completely different dialogs during supervised contrastive pre-training. To fully exploit unlabeled dialogs, a basic self-supervised contrastive loss is also added to refine the learned representations. Experiments show that our method can achieve new state-of-the-art results on the DialoGLUE benchmark consisting of seven datasets and four popular dialog understanding tasks.

This paper aims to improve the performance of text-to-SQL parsing by exploring the intrinsic uncertainties in the neural network based approaches (called SUN). From the data uncertainty perspective, it is indisputable that a single SQL can be learned from multiple semantically-equivalent questions. Different from previous methods that are limited to one-to-one mapping, we propose a data uncertainty constraint to explore the underlying complementary semantic information among multiple semantically-equivalent questions (many-to-one) and learn the robust feature representations with reduced spurious associations. In this way, we can reduce the sensitivity of the learned representations and improve the robustness of the parser. From the model uncertainty perspective, there is often structural information (dependence) among the weights of neural networks. To improve the generalizability and stability of neural text-to-SQL parsers, we propose a model uncertainty constraint to refine the query representations by enforcing the output representations of different perturbed encoding networks to be consistent with each other. Extensive experiments on five benchmark datasets demonstrate that our method significantly outperforms strong competitors and achieves new state-of-the-art results.

Out-of-Domain (OOD) intent detection is important for practical dialog systems.To alleviate the issue of lacking OOD training samples, some works propose synthesizing pseudo OOD samples and directly assigning one-hot OOD labels to these pseudo samples.However, these one-hot labels introduce noises to the training process because some "hard" pseudo OOD samples may coincide with In-Domain (IND) intents.In this paper, we propose an adaptive soft pseudo labeling (ASoul) method that can estimate soft labels for pseudo OOD samples when training OOD detectors.Semantic connections between pseudo OOD samples and IND intents are captured using an embedding graph.A co-training framework is further introduced to produce resulting soft labels following the smoothness assumption, i.e., close samples are likely to have similar labels.Extensive experiments on three benchmark datasets show that ASoul consistently improves the OOD detection performance and outperforms various competitive baselines.

Chenliang Li, Haiyang Xu, Junfeng Tian, Wei Wang, Ming Yan, Bin Bi, Jiabo Ye, He Chen, Guohai Xu, Zheng Cao, Ji Zhang, Songfang Huang, Fei Huang, Jingren Zhou, Luo Si. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

In this paper, we introduce the task of learning unsupervised dialogue embeddings.Trivial approaches such as combining pre-trained word or sentence embeddings and encoding through pre-trained language models (PLMs) have been shown to be feasible for this task.However, these approaches typically ignore the conversational interactions between interlocutors, resulting in poor performance.To address this issue, we proposed a selfguided contrastive learning approach named dial2vec.Dial2vec considers a dialogue as an information exchange process.It captures the conversational interaction patterns between interlocutors and leverages them to guide the learning of the embeddings corresponding to each interlocutor.The dialogue embedding is obtained by an aggregation of the embeddings from all interlocutors.To verify our approach, we establish a comprehensive benchmark consisting of six widely-used dialogue datasets.We consider three evaluation tasks: domain categorization, semantic relatedness, and dialogue retrieval.Dial2vec achieves on average 8.7, 9.0, and 13.8 points absolute improvements in terms of purity, Spearman's correlation, and mean average precision (MAP) over the strongest baseline on the three tasks respectively.Further analysis shows that dial2vec obtains informative and discriminative embeddings for both interlocutors under the guidance of the conversational interactions and achieves the best performance when aggregating them through the interlocutor-level pooling strategy.

Named Entity Recognition (NER) from speech is among Spoken Language Understanding (SLU) tasks, aiming to extract semantic information from the speech signal. NER from speech is usually made through a two-step pipeline that consists of (1) processing the audio using an Automatic Speech Recognition (ASR) system and (2) applying an NER tagger to the ASR outputs. Recent works have shown the capability of the End-to-End (E2E) approach for NER from English and French speech, which is essentially entity-aware ASR. However, due to the many homophones and polyphones that exist in Chinese, NER from Chinese speech is effectively a more challenging task. In this paper, we introduce a new dataset AISEHLL-NER for NER from Chinese speech. Extensive experiments are conducted to explore the performance of several state-of-the-art methods. The results demonstrate that the performance could be improved by combining entity-aware ASR and pretrained NER tagger, which can be easily applied to the modern SLU pipeline. The dataset is publicly available at github.com/Alibaba-NLP/AISHELL-NER.

We study the knowledge extrapolation problem to embed new components (i.e., entities and relations) that come with emerging knowledge graphs (KGs) in the federated setting. In this problem, a model trained on an existing KG needs to embed an emerging KG with unseen entities and relations. To solve this problem, we introduce the meta-learning setting, where a set of tasks are sampled on the existing KG to mimic the link prediction task on the emerging KG. Based on sampled tasks, we meta-train a graph neural network framework that can construct features for unseen components based on structural information and output embeddings for them. Experimental results show that our proposed method can effectively embed unseen components and outperforms models that consider inductive settings for KGs and baselines that directly use conventional KG embedding 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%.

The importance of building text-to-SQL parsers which can be applied to new databases has long been acknowledged, and a critical step to achieve this goal is schema linking, i.e., properly recognizing mentions of unseen columns or tables when generating SQLs. In this work, we propose a novel framework to elicit relational structures from large-scale pre-trained language models (PLMs) via a probing procedure based on Poincaré distance metric, and use the induced relations to augment current graph-based parsers for better schema linking. Compared with commonly-used rule-based methods for schema linking, we found that probing relations can robustly capture semantic correspondences, even when surface forms of mentions and entities differ. Moreover, our probing procedure is entirely unsupervised and requires no additional parameters. Extensive experiments show that our framework sets new state-of-the-art performance on three benchmarks. We empirically verify that our probing procedure can indeed find desired relational structures through qualitative analysis.

Yue Zhang, Zhenghua Li, Zuyi Bao, Jiacheng Li, Bo Zhang, Chen Li, Fei Huang, Min Zhang. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Prompt learning approaches have made waves in natural language processing by inducing better few-shot performance while they still follow a parametric-based learning paradigm; the oblivion and rote memorization problems in learning may encounter unstable generalization issues. Specifically, vanilla prompt learning may struggle to utilize atypical instances by rote during fully-supervised training or overfit shallow patterns with low-shot data. To alleviate such limitations, we develop RetroPrompt with the motivation of decoupling knowledge from memorization to help the model strike a balance between generalization and memorization. In contrast with vanilla prompt learning, RetroPrompt constructs an open-book knowledge-store from training instances and implements a retrieval mechanism during the process of input, training and inference, thus equipping the model with the ability to retrieve related contexts from the training corpus as cues for enhancement. Extensive experiments demonstrate that RetroPrompt can obtain better performance in both few-shot and zero-shot settings. Besides, we further illustrate that our proposed RetroPrompt can yield better generalization abilities with new datasets. Detailed analysis of memorization indeed reveals RetroPrompt can reduce the reliance of language models on memorization; thus, improving generalization for downstream tasks. Code is available in https://github.com/zjunlp/PromptKG/tree/main/research/RetroPrompt.

Pre-trained language models have contributed significantly to relation extraction by demonstrating remarkable few-shot learning abilities. However, prompt tuning methods for relation extraction may still fail to generalize to those rare or hard patterns. Note that the previous parametric learning paradigm can be viewed as memorization regarding training data as a book and inference as the close-book test. Those long-tailed or hard patterns can hardly be memorized in parameters given few-shot instances. To this end, we regard RE as an open-book examination and propose a new semiparametric paradigm of retrieval-enhanced prompt tuning for relation extraction. We construct an open-book datastore for retrieval regarding prompt-based instance representations and corresponding relation labels as memorized key-value pairs. During inference, the model can infer relations by linearly interpolating the base output of PLM with the non-parametric nearest neighbor distribution over the datastore. In this way, our model not only infers relation through knowledge stored in the weights during training but also assists decision-making by unwinding and querying examples in the open-book datastore. Extensive experiments on benchmark datasets show that our method can achieve state-of-the-art in both standard supervised and few-shot settings

Multimodal Knowledge Graphs (MKGs), which organize visual-text factual knowledge, have recently been successfully applied to tasks such as information retrieval, question answering, and recommendation system. Since most MKGs are far from complete, extensive knowledge graph completion studies have been proposed focusing on the multimodal entity, relation extraction and link prediction. However, different tasks and modalities require changes to the model architecture, and not all images/objects are relevant to text input, which hinders the applicability to diverse real-world scenarios. In this paper, we propose a hybrid transformer with multi-level fusion to address those issues. Specifically, we leverage a hybrid transformer architecture with unified input-output for diverse multimodal knowledge graph completion tasks. Moreover, we propose multi-level fusion, which integrates visual and text representation via coarse-grained prefix-guided interaction and fine-grained correlation-aware fusion modules. We conduct extensive experiments to validate that our MKGformer can obtain SOTA performance on four datasets of multimodal link prediction, multimodal RE, and multimodal NER1. https://github.com/zjunlp/MKGformer.

Recently, pre-training methods have shown remarkable success in task-oriented dialog (TOD) systems. However, most existing pre-trained models for TOD focus on either dialog understanding or dialog generation, but not both. In this paper, we propose SPACE, a novel unified pre-trained dialog model learning from large-scale dialog corpora with limited annotations, which can be effectively fine-tuned on a wide range of downstream dialog tasks. Specifically, SPACE consists of four successive components in a single transformer to maintain a task-flow in TOD systems: (i) a dialog encoding module to encode dialog history, (ii) a dialog understanding module to extract semantic vectors from either user queries or system responses, (iii) a dialog policy module to generate a policy vector that contains high-level semantics of the response, and (iv) a dialog generation module to produce appropriate responses. We design a dedicated pre-training objective for each component. Concretely, we pre-train the dialog encoding module with span mask language modeling to learn contextualized dialog information. To capture the structured dialog semantics, we pre-train the dialog understanding module via a novel tree-induced semi-supervised contrastive learning objective with the help of extra dialog annotations. In addition, we pre-train the dialog policy module by minimizing the ℒ2 distance between its output policy vector and the semantic vector of the response for policy optimization. Finally, the dialog generation model is pre-trained by language modeling. Results show that SPACE achieves state-of-the-art performance on eight downstream dialog benchmarks, including intent prediction, dialog state tracking, and end-to-end dialog modeling. We also show that SPACE has a stronger few-shot ability than existing models under the low-resource setting.

Recently, prompt-tuning has achieved promising results for specific few-shot classification tasks. The core idea of prompt-tuning is to insert text pieces (i.e., templates) into the input and transform a classification task into a masked language modeling problem. However, for relation extraction, determining an appropriate prompt template requires domain expertise, and it is cumbersome and time-consuming to obtain a suitable label word. Furthermore, there exists abundant semantic and prior knowledge among the relation labels that cannot be ignored. To this end, we focus on incorporating knowledge among relation labels into prompt-tuning for relation extraction and propose a Knowledge-aware Prompt-tuning approach with synergistic optimization (KnowPrompt). Specifically, we inject latent knowledge contained in relation labels into prompt construction with learnable virtual type words and answer words. Then, we synergistically optimize their representation with structured constraints. Extensive experimental results on five datasets with standard and low-resource settings demonstrate the effectiveness of our approach. Our code and datasets are available in GitHub1 for reproducibility.

Multimodal machine translation (MMT) aims to translate a sentence in the source language into the target language with the context of an associated image. According to where the visual information is employed, previous approaches can be categorized into two types: directly injecting the visual information at the input side or exploiting it as a visual constraint at the objective side. In this work, we propose an IO-MMT model which exploits the visual assistance in dual sides to fully exploit the visual information for multimodal machine translation. It contains a relation-aware multimodal transformer to simultaneously exploit the objects and their spatial relationships in the image with a graph at the input-side and a novel visual assistance structure to further improve visual consistency of the translation at the objective-side. Experimental results under both normal setting and input degradation settings on the Multi30k benchmark dataset show that combining the visual assistance in dual sides consistently outperforms single-side MMT models and achieves the state-of-the-art results on EN-DE and EN-FR translation tasks. We will release the codes and models at https://github.com/syuqings/MMT.

Named entity recognition (NER) is a well-studied task in natural language processing. However, the widely-used sequence labeling framework is difficult to detect entities with nested structures. In this work, we view nested NER as constituency parsing with partially-observed trees and model it with partially-observed TreeCRFs. Specifically, we view all labeled entity spans as observed nodes in a constituency tree, and other spans as latent nodes. With the TreeCRF we achieve a uniform way to jointly model the observed and the latent nodes. To compute the probability of partial trees with partial marginalization, we propose a variant of the Inside algorithm, the Masked Inside algorithm, that supports different inference operations for different nodes (evaluation for the observed, marginalization for the latent, and rejection for nodes incompatible with the observed) with efficient parallelized implementation, thus significantly speeding up training and inference. Experiments show that our approach achieves the state-of-the-art (SOTA) F1 scores on the ACE2004, ACE2005 dataset, and shows comparable performance to SOTA models on the GENIA dataset. We release the code at https://github.com/FranxYao/Partially-Observed-TreeCRFs.

Semantic parsing has long been a fundamental problem in natural language processing. Recently, cross-domain context-dependent semantic parsing has become a new focus of research. Central to the problem is the challenge of leveraging contextual information of both natural language queries and database schemas in the interaction history. In this paper, we present a dynamic graph framework that is capable of effectively modelling contextual utterances, tokens, database schemas, and their complicated interaction as the conversation proceeds. The framework employs a dynamic memory decay mechanism that incorporates inductive bias to integrate enriched contextual relation representation, which is further enhanced with a powerful reranking model. At the time of writing, we demonstrate that the proposed framework outperforms all existing models by large margins, achieving new state-of-the-art performance on two large-scale benchmarks, the SParC and CoSQL datasets. Specifically, the model attains a 55.8% question-match and 30.8% interaction-match accuracy on SParC, and a 46.8% question-match and 17.0% interaction-match accuracy on CoSQL.

Named Entity Recognition (NER) is a fundamental and important research topic for many downstream NLP tasks, aiming at detecting and classifying named entities (NEs) mentioned in unstructured text into pre-defined categories. Learning from labeled data only is far from enough when it comes to domain-specific or temporally-evolving entities (medical terminologies or restaurant names). Luckily, open-source Knowledge Bases (KBs) (Wikidata and Freebase) contain NEs that are manually labeled with predefined types in different domains, which is potentially beneficial to identify entity boundaries and recognize entity types more accurately. However, the type system of a domain-specific NER task is typically independent of that of current KBs and thus exhibits heterogeneity issue inevitably, which makes matching between the original NER and KB types (Person in NER potentially matches President in KBs) less likely, or introduces unintended noises without considering domain-specific knowledge (Band in NER should be mapped to Out_of_Entity_Types in the restaurant-related task). To better incorporate and denoise the abundant knowledge in KBs, we propose a new KB-aware NER framework (KaNa), which utilizes type-heterogeneous knowledge to improve NER. Specifically, for an entity mention along with a set of candidate entities that are linked from KBs, KaNa first uses a type projection mechanism that maps the mention type and entity types into a shared space to homogenize the heterogeneous entity types. Then, based on projected types, a noise detector filters out certain less-confident candidate entities in an unsupervised manner. Finally, the filtered mention-entity pairs are injected into a NER model as a graph to predict answers. The experimental results demonstrate KaNa's state-of-the-art performance on five public benchmark datasets from different domains.

It is important for task-oriented dialogue systems to discover the dialogue structure (i.e. the general dialogue flow) from dialogue corpora automatically. Previous work models dialogue structure by extracting latent states for each utterance first and then calculating the transition probabilities among states. These two-stage methods ignore the contextual information when calculating the probabilities, which makes the transitions between the states ambiguous. This paper proposes a conversational graph (CG) to represent deterministic dialogue structure where nodes and edges represent the utterance and context information respectively. An unsupervised Edge-Enhanced Graph Auto-Encoder (EGAE) architecture is designed to model local-contextual and global-structural information for conversational graph learning. Furthermore, a self-supervised objective is introduced with the response selection task to guide the unsupervised learning of the dialogue structure. Experimental results on several public datasets demonstrate that the novel model outperforms several alternatives in aggregating utterances with similar semantics. The effectiveness of the learned dialogue structured is also verified by more than 5\% joint accuracy improvement in the downstream task of low resource dialogue state tracking.

Despite the near-human performances already achieved on formal texts such as news articles, neural machine translation still has difficulty in dealing with "user-generated" texts that have diverse linguistic phenomena but lack large-scale high-quality parallel corpora. To address this problem, we propose a counterfactual domain adaptation method to better leverage both large-scale source-domain data (formal texts) and small-scale target-domain data (informal texts). Specifically, by considering effective counterfactual conditions (the concatenations of source-domain texts and the target-domain tag), we construct the counterfactual representations to fill the sparse latent space of the target domain caused by a small amount of data, that is, bridging the gap between the source-domain data and the target-domain data. Experiments on English-to-Chinese and Chinese-to-English translation tasks show that our method outperforms the base model that is trained only on the informal corpus by a large margin, and consistently surpasses different baseline methods by +1.12 ~ 4.34 BLEU points on different datasets. Furthermore, we also show that our method achieves competitive performances on cross-domain language translation on four language pairs.

Triple extraction is an essential task in information extraction for natural language processing and knowledge graph construction. In this paper, we revisit the end-to-end triple extraction task for sequence generation. Since generative triple extraction may struggle to capture long-term dependencies and generate unfaithful triples, we introduce a novel model, contrastive triple extraction with a generative transformer. Specifically, we introduce a single shared transformer module for encoder-decoder-based generation. To generate faithful results, we propose a novel triplet contrastive training object. Moreover, we introduce two mechanisms to further improve model performance (i.e., batch-wise dynamic attention-masking and triple-wise calibration). Experimental results on three datasets (i.e., NYT, WebNLG, and MIE) show that our approach achieves better performance than that of baselines.

Shumin Deng, Ningyu Zhang, Luoqiu Li, Chen Hui, Tou Huaixiao, Mosha Chen, Fei Huang, Huajun Chen. 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.

Zechuan Hu, Yong Jiang, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu. 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.

Zechuan Hu, Yong Jiang, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu. 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.

Chenliang Li, Bin Bi, Ming Yan, Wei Wang, Songfang Huang, Fei Huang, Luo Si. 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.

Fuli Luo, Wei Wang, Jiahao Liu, Yijia Liu, Bin Bi, Songfang Huang, Fei Huang, Luo Si. 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.

Xinyu Wang, Yong Jiang, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu. 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.

Xinyu Wang, Yong Jiang, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu. 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.

Xinyu Wang, Yong Jiang, Zhaohui Yan, Zixia Jia, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu. 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.

Haiyang Xu, Ming Yan, Chenliang Li, Bin Bi, Songfang Huang, Wenming Xiao, Fei Huang. 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.

Adapting word order from one language to another is a key problem in cross-lingual structured prediction.Current sentence encoders (e.g., RNN, Transformer with position embeddings) are usually word order sensitive.Even with uniform word form representations (MUSE, mBERT), word order discrepancies may hurt the adaptation of models.This paper builds structured prediction models with bag-of-words inputs.It introduces a new reordering module to organize words following the source language order, which learns taskspecific reordering strategies from a generalpurpose order predictor model.Experiments on zero-shot cross-lingual dependency parsing, POS tagging, and morphological tagging show that our model can significantly improve target language performances, especially for languages that are distant from the source language.1

Transition systems usually contain various dynamic structures (e.g., stacks, buffers).An ideal transition-based model should encode these structures completely and efficiently.Previous works relying on templates or neural network structures either only encode partial structure information or suffer from computation efficiency.In this paper, we propose a novel attention-based encoder unifying representation of all structures in a transition system.Specifically, we separate two views of items on structures, namely structure-invariant view and structure-dependent view.With the help of parallel-friendly attention network, we are able to encoding transition states with O(1) additional complexity (with respect to basic feature extractors).Experiments on the PTB and UD show that our proposed method significantly improves the test speed and achieves the best transition-based model, and is comparable to state-of-the-art methods. 1

Learning sentence embeddings from dialogues has drawn increasing attention due to its low annotation cost and high domain adaptability.Conventional approaches employ the siamese-network for this task, which obtains the sentence embeddings through modeling the context-response semantic relevance by applying a feed-forward network on top of the sentence encoders.However, as the semantic textual similarity is commonly measured through the element-wise distance metrics (e.g.cosine and L2 distance), such architecture yields a large gap between training and evaluating.In this paper, we propose DialogueCSE, a dialogue-based contrastive learning approach to tackle this issue.DialogueCSE first introduces a novel matching-guided embedding (MGE) mechanism, which generates a contextaware embedding for each candidate response embedding (i.e. the context-free embedding) according to the guidance of the multi-turn context-response matching matrices.Then it pairs each context-aware embedding with its corresponding context-free embedding and finally minimizes the contrastive loss across all pairs.We evaluate our model on three multi-turn dialogue datasets: the Microsoft Dialogue Corpus, the Jing Dong Dialogue Corpus, and the E-commerce Dialogue Corpus.Evaluation results show that our approach significantly outperforms the baselines across all three datasets in terms of MAP and Spearman's correlation measures, demonstrating its effectiveness.Further quantitative experiments show that our approach achieves better performance when leveraging more dialogue context and remains robust when less training data is provided.

Pre-trained self-supervised models such as BERT have achieved striking success in learning sequence representations, especially for natural language processing.These models typically corrupt the given sequences with certain types of noise, such as masking, shuffling, or substitution, and then try to recover the original input.However, such pre-training approaches are prone to learning representations that are covariant with the noise, leading to the discrepancy between the pre-training and finetuning stage.To remedy this, we present Con-trAstive Pre-Training (CAPT) to learn noise invariant sequence representations.The proposed CAPT encourages the consistency between representations of the original sequence and its corrupted version via unsupervised instance-wise training signals.In this way, it not only alleviates the pretrain-finetune discrepancy induced by the noise of pre-training, but also aids the pre-trained model in better capturing global semantics of the input via more effective sentence-level supervision.Different from most prior work that focuses on a particular modality, comprehensive empirical evidence on 11 natural language understanding and cross-modal tasks illustrates that CAPT is applicable for both language and vision-language tasks, and obtains surprisingly consistent improvement, including 0.6% absolute gain on GLUE benchmarks and 0.8% absolute increment on NLVR 2 . * Equal Contribution.Models Noise types BERT (Devlin et al., 2019) Mask tokens SpanBERT (Joshi et al., 2019) Mask spans RoBERTa (Liu et al., 2019) Mask token XLNet (Yang et al., 2019) Shuffle token ELECTRA (Clark et al., 2019) Replace tokens StructBERT (Wang et al., 2019b) Mask + Shuffle tokens BART (Lewis et al., 2019) Mask + Shuffle + Replace.UNITER (Chen et al., 2019) Mask tokens/regions LXMERT (Tan and Bansal, 2019) Mask tokens/regions

Entity retrieval, which aims at disambiguating mentions to canonical entities from massive KBs, is essential for many tasks in natural language processing.Recent progress in entity retrieval shows that the dual-encoder structure is a powerful and efficient framework to nominate candidates if entities are only identified by descriptions.However, they ignore the property that meanings of entity mentions diverge in different contexts and are related to various portions of descriptions, which are treated equally in previous works.In this work, we propose Multi-View Entity Representations (MuVER), a novel approach for entity retrieval that constructs multi-view representations for entity descriptions and approximates the optimal view for mentions via a heuristic searching method.Our method achieves the state-ofthe-art performance on ZESHEL and improves the quality of candidates on three standard Entity Linking datasets 1 .

Recent pretrained language models extend from millions to billions of parameters.Thus the need to fine-tune an extremely large pretrained model with a limited training corpus arises in various downstream tasks.In this paper, we propose a straightforward yet effective fine-tuning technique, CHILD-TUNING, which updates a subset of parameters (called child network) of large pretrained models via strategically masking out the gradients of the non-child network during the backward process.Experiments on various downstream tasks in GLUE benchmark show that CHILD-TUNING consistently outperforms the vanilla fine-tuning by 1.5 ∼ 8.6 average score among four different pretrained models, and surpasses the prior fine-tuning techniques by 0.6 ∼ 1.3 points.Furthermore, empirical results on domain transfer and task transfer show that CHILD-TUNING can obtain better generalization performance by large margins.

Paraphrase generation plays key roles in NLP tasks such as question answering, machine translation, and information retrieval. In this paper, we propose a novel framework for paraphrase generation. It simultaneously decodes the output sentence using a pretrained wordset-to-sequence model and a round-trip translation model. We evaluate this framework on Quora, WikiAnswers, MSCOCO and Twitter, and show its advantage over previous state-of-the-art unsupervised methods and distantly-supervised methods by significant margins on all datasets. For Quora and WikiAnswers, our framework even performs better than some strongly supervised methods with domain adaptation. Further, we show that the generated paraphrases can be used to augment the training data for machine translation to achieve substantial improvements.

Document-level relation extraction aims to extract relations among multiple entity pairs from a document. Previously proposed graph-based or transformer-based models utilize the entities independently, regardless of global information among relational triples. This paper approaches the problem by predicting an entity-level relation matrix to capture local and global information, parallel to the semantic segmentation task in computer vision. Herein, we propose a Document U-shaped Network for document-level relation extraction. Specifically, we leverage an encoder module to capture the context information of entities and a U-shaped segmentation module over the image-style feature map to capture global interdependency among triples. Experimental results show that our approach can obtain state-of-the-art performance on three benchmark datasets DocRED, CDR, and GDA.

Translating e-commercial product descriptions, a.k.a product-oriented machine translation (PMT), is essential to serve e-shoppers all over the world. However, due to the domain specialty, the PMT task is more challenging than traditional machine translation problems. Firstly, there are many specialized jargons in the product description, which are ambiguous to translate without the product image. Secondly, product descriptions are related to the image in more complicated ways than standard image descriptions, involving various visual aspects such as objects, shapes, colors or even subjective styles. Moreover, existing PMT datasets are small in scale to support the research. In this paper, we first construct a large-scale bilingual product description dataset called Fashion-MMT, which contains over 114k noisy and 40k manually cleaned description translations with multiple product images. To effectively learn semantic alignments among product images and bilingual texts in translation, we design a unified product-oriented cross-modal cross-lingual model for pre-training and fine-tuning. Experiments on the Fashion-MMT and Multi30k datasets show that our model significantly outperforms the state-of-the-art models even pre-trained on the same dataset. It is also shown to benefit more from large-scale noisy data to improve the translation quality. We will release the dataset and codes at https://github.com/syuqings/Fashion-MMT.

Qingrong Xia, Bo Zhang, Rui Wang, Zhenghua Li, Yue Zhang, Fei Huang, Luo Si, Min Zhang. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Aiming at expanding few-shot relations' coverage in knowledge graphs (KGs), few-shot knowledge graph completion (FKGC) has recently gained more research interests. Some existing models employ a few-shot relation's multi-hop neighbor information to enhance its semantic representation. However, noise neighbor information might be amplified when the neighborhood is excessively sparse and no neighbor is available to represent the few-shot relation. Moreover, modeling and inferring complex relations of one-to-many (1-N), many-to-one (N-1), and many-to-many (N-N) by previous knowledge graph completion approaches requires high model complexity and a large amount of training instances. Thus, inferring complex relations in the few-shot scenario is difficult for FKGC models due to limited training instances. In this paper, we propose a few-shot relational learning with global-local framework to address the above issues. At the global stage, a novel gated and attentive neighbor aggregator is built for accurately integrating the semantics of a few-shot relation's neighborhood, which helps filtering the noise neighbors even if a KG contains extremely sparse neighborhoods. For the local stage, a meta-learning based TransH (MTransH) method is designed to model complex relations and train our model in a few-shot learning fashion. Extensive experiments show that our model outperforms the state-of-the-art FKGC approaches on the frequently-used benchmark datasets NELL-One and Wiki-One. Compared with the strong baseline model MetaR, our model achieves 5-shot FKGC performance improvements of 8.0% on NELL-One and 2.8% on Wiki-One by the metric [email protected]

Information extraction tasks such as triple extraction and event extraction are of great importance for natural language processing and knowledge graph construction. In this paper, we revisit the end-to-end information extraction task for sequence generation. Since generative information extraction may struggle to capture long-term dependencies and generate unfaithful triples, we introduce a novel model, contrastive information extraction with a generative transformer. Specifically, we introduce a single shared transformer module for an encoder-decoder-based generation. To generate faithful results, we propose a novel triplet contrastive training object. Moreover, we introduce two mechanisms to further improve model performance (i.e., batch-wise dynamic attention-masking and triple-wise calibration). Experimental results on five datasets (i.e., NYT, WebNLG, MIE, ACE-2005, and MUC-4) show that our approach achieves better performance than baselines.

Target-based sentiment analysis or aspect-based sentiment analysis (ABSA) refers to addressing various sentiment analysis tasks at a fine-grained level, which includes but is not limited to aspect extraction, aspect sentiment classification, and opinion extraction. There exist many solvers of the above individual subtasks or a combination of two subtasks, and they can work together to tell a complete story, i.e. the discussed aspect, the sentiment on it, and the cause of the sentiment. However, no previous ABSA research tried to provide a complete solution in one shot. In this paper, we introduce a new subtask under ABSA, named aspect sentiment triplet extraction (ASTE). Particularly, a solver of this task needs to extract triplets (What, How, Why) from the inputs, which show WHAT the targeted aspects are, HOW their sentiment polarities are and WHY they have such polarities (i.e. opinion reasons). For instance, one triplet from “Waiters are very friendly and the pasta is simply average” could be (‘Waiters’, positive, ‘friendly’). We propose a two-stage framework to address this task. The first stage predicts what, how and why in a unified model, and then the second stage pairs up the predicted what (how) and why from the first stage to output triplets. In the experiments, our framework has set a benchmark performance in this novel triplet extraction task. Meanwhile, it outperforms a few strong baselines adapted from state-of-the-art related methods.

Named entity recognition (NER) is a well-studied task in natural language processing. However, the widely-used sequence labeling framework is usually difficult to detect entities with nested structures. The span-based method that can easily detect nested entities in different subsequences is naturally suitable for the nested NER problem. However, previous span-based methods have two main issues. First, classifying all subsequences is computationally expensive and very inefficient at inference. Second, the span-based methods mainly focus on learning span representations but lack of explicit boundary supervision. To tackle the above two issues, we propose a boundary enhanced neural span classification model. In addition to classifying the span, we propose incorporating an additional boundary detection task to predict those words that are boundaries of entities. The two tasks are jointly trained under a multitask learning framework, which enhances the span representation with additional boundary supervision. In addition, the boundary detection model has the ability to generate high-quality candidate spans, which greatly reduces the time complexity during inference. Experiments show that our approach outperforms all existing methods and achieves 85.3, 83.9, and 78.3 scores in terms of F1 on the ACE2004, ACE2005, and GENIA datasets, respectively.

Most existing joint neural models for Information Extraction (IE) use local task-specific classifiers to predict labels for individual instances (e.g., trigger, relation) regardless of their interactions.For example, a VICTIM of a DIE event is likely to be a VICTIM of an AT-TACK event in the same sentence.In order to capture such cross-subtask and cross-instance inter-dependencies, we propose a joint neural framework, ONEIE, that aims to extract the globally optimal IE result as a graph from an input sentence.ONEIE performs end-to-end IE in four stages: (1) Encoding a given sentence as contextualized word representations;(2) Identifying entity mentions and event triggers as nodes; (3) Computing label scores for all nodes and their pairwise links using local classifiers; (4) Searching for the globally optimal graph with a beam decoder.At the decoding stage, we incorporate global features to capture the cross-subtask and cross-instance interactions.Experiments show that adding global features improves the performance of our model and achieves new state-of-the-art on all subtasks.As ONEIE does not use any language-specific feature, we prove it can be easily applied to new languages or trained in a multilingual manner.Our code and models for English, Spanish and Chinese are publicly available for research purpose 1 .

Multilingual sequence labeling is a task of predicting label sequences using a single unified model for multiple languages.Compared with relying on multiple monolingual models, using a multilingual model has the benefit of a smaller model size, easier in online serving, and generalizability to low-resource languages.However, current multilingual models still underperform individual monolingual models significantly due to model capacity limitations.In this paper, we propose to reduce the gap between monolingual models and the unified multilingual model by distilling the structural knowledge of several monolingual models (teachers) to the unified multilingual model (student).We propose two novel KD methods based on structure-level information:(1) approximately minimizes the distance between the student's and the teachers' structurelevel probability distributions, (2) aggregates the structure-level knowledge to local distributions and minimizes the distance between two local probability distributions.Our experiments on 4 multilingual tasks with 25 datasets show that our approaches outperform several strong baselines and have stronger zero-shot generalizability than both the baseline model and teacher models.

Self-supervised pre-training, such as BERT (Devlin et al., 2018), MASS (Song et al., 2019) and BART (Lewis et al., 2019), has emerged as a powerful technique for natural language understanding and generation.Existing pre-training techniques employ autoencoding and/or autoregressive objectives to train Transformer-based models by recovering original word tokens from corrupted text with some masked tokens.The training goals of existing techniques are often inconsistent with the goals of many language generation tasks, such as generative question answering and conversational response generation, for producing new text given context.This work presents PALM with a novel scheme that jointly pre-trains an autoencoding and autoregressive language model on a large unlabeled corpus, specifically designed for generating new text conditioned on context.The new scheme alleviates the mismatch introduced by the existing denoising scheme between pre-training and fine-tuning where generation is more than reconstructing original text.An extensive set of experiments show that PALM achieves new state-of-theart results on a variety of language generation benchmarks covering generative question answering (Rank 1 on the official MARCO leaderboard), abstractive summarization on CNN/DailyMail as well as Gigaword, question generation on SQuAD, and conversational response generation on Cornell Movie Dialogues.

The linear-chain Conditional Random Field (CRF) model is one of the most widely-used neural sequence labeling approaches.Exact probabilistic inference algorithms such as the forward-backward and Viterbi algorithms are typically applied in training and prediction stages of the CRF model.However, these algorithms require sequential computation that makes parallelization impossible.In this paper, we propose to employ a parallelizable approximate variational inference algorithm for the CRF model.Based on this algorithm, we design an approximate inference network that can be connected with the encoder of the neural CRF model to form an end-to-end network, which is amenable to parallelization for faster training and prediction.The empirical results show that our proposed approaches achieve a 12.7-fold improvement in decoding speed with long sentences and a competitive accuracy compared with the traditional CRF approach.

Aspect sentiment classification, predicting the sentiment polarity of given aspects, has drawn extensive attention.Previous attention-based models emphasize using aspect semantics to help extract opinion features for classification.However, these works are either not able to capture opinion spans as a whole or capture variable-length opinion spans.In this paper, we present a neat and effective multiple CRFs based structured attention model that is capable of extracting aspect-specific opinion spans.The sentiment polarity of the target is then classified based on the extracted opinion features and contextual information.The experimental results on four datasets demonstrate the effectiveness of the proposed model, and our further analysis shows that our model can capture aspect-specific opinion spans. 1

Fine-grained entity typing (FET) is a fundamental task for various entity-leveraging applications. Although great success has been made, existing systems still have challenges in handling noisy samples in training data introduced by distant supervision methods. To address these noise, previous studies either focus on processing the clean samples (i,e., have only one label) and noisy samples (i,e., have multiple labels) with different strategies or filtering the noisy labels based on the assumption that the distantly-supervised label set certainly contains the correct type label. In this paper, we propose a probabilistic automatic relabeling method which treats all training samples uniformly. Our method aims to estimate the pseudo-truth label distribution of each sample, and the pseudo-truth distribution will be treated as part of trainable parameters which are jointly updated during the training process. The proposed approach does not rely on any prerequisite or extra supervision, making it effective on real applications. Experiments on several benchmarks show that our method outperforms previous approaches and alleviates the noisy labeling problem.

Unsupervised neural machine translation (UNMT) has recently achieved remarkable results \cite{lample2018phrase} with only large monolingual corpora in each language. However, the uncertainty of associating target with source sentences makes UNMT theoretically an ill-posed problem. This work investigates the possibility of utilizing images for disambiguation to improve the performance of UNMT. Our assumption is intuitively based on the invariant property of image, i.e., the description of the same visual content by different languages should be approximately similar. We propose an unsupervised multi-modal machine translation (UMNMT) framework based on the language translation cycle consistency loss conditional on the image, targeting to learn the bidirectional multi-modal translation simultaneously. Through an alternate training between multi-modal and uni-modal, our inference model can translate with or without the image. On the widely used Multi30K dataset, the experimental results of our approach are significantly better than those of the text-only UNMT on the 2016 test dataset.

In this paper, we propose a method which uses semi-supervised convolutional neural networks (CNNs) to select in-domain training data for statistical machine translation. This approach is particularly effective when only tiny amounts of in-domain data are available. The in-domain data and randomly sampled general-domain data are used to train a data selection model with semi-supervised CNN, then this model computes domain relevance scores for all the sentences in the generaldomain data set. The sentence pairs with top scores are selected to train the system. We carry out experiments on 4 language directions with three test domains. Compared with strong baseline systems trained with large amount of data, this method can improve the performance up to 3.1 BLEU. Its performances are significant better than three state-of-the-art language model based data selection methods. We also show that the in-domain data used to train the selection model could be as fewas 100sentences, whichmakesfinegrained topic-dependent translation adaptation possible.

We present an adaptive translation quality estimation (QE) method to predict the human-targeted translation error rate (HTER) for a document-specific machine translation model.We first introduce features derived internal to the translation decoding process as well as externally from the source sentence analysis.We show the effectiveness of such features in both classification and regression of MT quality.By dynamically training the QE model for the document-specific MT model, we are able to achieve consistency and prediction quality across multiple documents, demonstrated by the higher correlation coefficient and F-scores in finding Good sentences.Additionally, the proposed method is applied to IBM English-to-Japanese MT post editing field study and we observe strong correlation with human preference, with a 10% increase in human translators' productivity.

Traditional isolated monolingual name taggers tend to yield inconsistent results across two languages. In this paper, we propose two novel approaches to jointly and consistently extract names from parallel corpora. The first approach uses standard linear-chain Conditional Random Fields (CRFs) as the learning framework, incorporating cross-lingual features propagated between two languages. The second approach is based on a joint CRFs model to jointly decode sentence pairs, incorporating bilingual factors based on word alignment. Experiments on Chinese-English parallel corpora demonstrated that the proposed methods significantly outperformed monolingual name taggers, were robust to automatic alignment noise and achieved state-of-the-art performance. With only 20%of the training data, our proposed methods can already achieve better performance compared to the baseline learned from the whole training set.1

In this paper, the systematic design of a space-time block code is considered for a wireless communication system with multiple transmitter-receiver antennas and flat fading, in which channel state information is completely unknown. From the viewpoint of blind signal processing, a necessary and sufficient condition is given for the unique identification of the multi-input multi-output flat fading channel and transmitted signal. Then, some novel unique factorizations for a pair of coprimeP-ary andQ-ary phase shift keying (PSK) constellations are established. With this and currently available coherent space-time block code designs, a method is developed to systematically construct full diversity blind nonunitary space-time block codes as well as unitary codes by just performing the QR decomposition of the nonunitary codes. It is proved that under a noise-free environment, the code design proposed in this paper can guarantee that the transmitted signals and the channel coefficients are uniquely identified, and that under a complex Gaussian noise environment in whichPth-order andQth-order statistics (PandQare co-prime) of the received signals are available, the channel coefficients can be still uniquely identified. In addition, a closed-form solution to determine the channel coefficients is obtained.

In this letter, a necessary and sufficient condition is developed to check whether a set of polynomial equations is solvable. Furthermore, under the sufficient condition, a closed-form solution is attained using linear Diophantine equation theory. With this, a novel signal design and a row-circular transmission scheme of phase shift keying (PSK) constellations for MIMO flat fading channels are proposed so that the channel coefficients can be uniquely identified. When certain high-order moments of the received signals are available, a closed-form solution to uniquely determine the channel coefficients is given. When only finite received data are available, a numerical algorithm is provided to efficiently and effectively estimate the channel.

Translating out-of-vocabulary (OOV) terms is a great challenge for the Cross-lingual Information Retrieval and Data-driven Machine Translation systems. Several approaches have been proposed to mine translations for OOV terms from the web, especially from pages containing mixed languages. In this paper, we propose a novel approach to automatically translate OOV terms on the fly through cross-lingual query expansion. The proposed approach does not require any web crawling and has achieved an inclusion rate of 95% and overall translation accuracy of 90%, outperforming state-of-the-art OOV translation techniques.

In this paper we describe the components of our statistical machine translation system. This system combines phrase-to-phrase translations extracted from a bilingual corpus using different alignment approaches. Special methods to extract and align named entities are used. We show how a manual lexicon can be incorporated into the statistical system in an optimized way. Experiments on Chinese-to-English and Arabic-to-English translation tasks are presented.

In this article we present an automatic approach to extracting Hindi-English (H-E) Named Entity (NE) translingual equivalences from bilingual parallel corpora. In the absence of a Hindi NE tagger or H-E translation dictionary, this approach adapts a Chinese-English (C-E) surface string transliteration model for H-E NE extraction. The model is initially trained using automatically extracted C-E NE pairs, then iteratively updated based on newly extracted H-E NE pairs. For each English person and location NE in each sentence pair, this approach searches for its Hindi correspondence with minimum transliteration cost and constructs an H-E NE list from the bilingual corpus. Experiments show that this approach extracted 1000 H-E NE pairs with a precision of 91.8%.

Translation of named entities (NE), including proper names, temporal and numerical expressions, is very important in multilingual natural language processing, like crosslingual information retrieval and statistical machine translation. We present an integrated approach to extract a named entity translation dictionary from a bilingual corpus while at the same time improving the named entity annotation quality. Starting from a bilingual corpus where the named entities are extracted independently for each language, a statistical alignment model is used to align the named entities. An iterative process is applied to extract named entity pairs with higher alignment probability. This leads to a smaller but cleaner named entity translation dictionary and also to a significant improvement of the monolingual named entity annotation quality for both languages. Experimental result shows that the dictionary size is reduced by 51.8% and the annotation quality is improved from 70.03 to 78.15 for Chinese and 73.38 to 81.46 in terms of F-score.

... information with a user. It is widely used in hospitals, hotels and conferences. In this paper, we propose an approach to interactive user registration by combining face recognition, speech recognition and speech synthesis technologies together through an efficient dialogue manager. In order to minimize a user's effort, we employ a new dialogue management model based on a finite state automaton (FSA), which uses a Baysian network to fuse the user's information from multiple channels (e.g., face image, speech, records stored in a pre-constructed database) to reliably estimate the confidence about user identity. Instead of fixing weights, the FSA adjusts its weights dynamically by integrating partial information from multiple information sources. This is achieved by maximizing an objective function to determine an optimal action at each succeeding state according to current confidence and information cues. Thus the transition between states can be done along the shortest path from the initial state to the goal state. We have developed a multimodal user registration system to demonstrate the feasibility of the proposed approach.