Question Under Discussion (QUD) is a discourse framework that uses implicit questions to reveal discourse relationships between sentences. In QUD parsing, each sentence is viewed as an answer to a question triggered by an anchor sentence in prior context. The resulting QUD structure is required to conform to several theoretical criteria like answer compatibility(how well the question is answered), making QUD parsing a challenging task. Previous works construct QUD parsers in a pipelined manner (i.e. detect the trigger sentence in context and then generate the question). However, these parsers lack a holistic view of the task and can hardly satisfy all the criteria. In this work, we introduce QUDSELECT, a joint-training framework that selectively decodes the QUD dependency structures considering the QUD criteria criteria. Using instruction-tuning, we train models to simultaneously predict the anchor sentence and generate the associated question. To explicitly incorporate the criteria, we adopt a selective decoding strategy of sampling multiple QUD candidates during inference, followed by selecting the best one with criteria scorers. Our method outperforms the state-of-the-art baseline models by 9% in human evaluation and 4% in automatic evaluation, demonstrating the effectiveness of our framework. Code and data are in https://github.com/asuvarna31/qudselect.

The applications of large language models (LLMs) have expanded well beyond the confines of text processing, signaling a new era where LLMs are envisioned as generalist agents capable of operating within complex environments. These environments are often highly expansive, making it impossible for the LLM to process them within its short-term memory. Motivated by recent research on extending the capabilities of LLMs with tools, we seek to investigate the intriguing potential of tools to augment LLMs in handling such complexity by introducing a novel class of tools, termed *middleware*, to aid in the proactive exploration within these massive environments. Such specialized tools can serve as a middleware layer shielding the LLM from environmental complexity. In two representative complex environments—knowledge bases (KBs) and databases—we demonstrate the significant potential of augmenting language agents with tools in complex environments. Notably, equipped with the middleware, GPT-4 achieves **2.8**X the performance of the best baseline in tasks requiring access to database content and **2.2**X in KB tasks. Our findings illuminate the path for advancing language agents in real-world applications.

LLMs can help humans working with long documents, but are known to hallucinate. *Attribution* can increase trust in LLM responses: The LLM provides evidence that supports its response, which enhances verifiability. Existing approaches to attribution have only been evaluated in RAG settings, where the initial retrieval confounds LLM performance. This is crucially different from the long document setting, where retrieval is not needed, but could help. Thus, a long document specific evaluation of attribution is missing. To fill this gap, we present LAB, a benchmark of 6 diverse long document tasks with attribution, and experiments with different approaches to attribution on 5 LLMs of different sizes. We find that *citation*, i.e. response generation and evidence extraction in one step, performs best for large and fine-tuned models, while additional retrieval can help for small, prompted models. We investigate whether the “Lost in the Middle” phenomenon exists for attribution, but do not find this. We also find that evidence quality can predict response quality on datasets with simple responses, but not so for complex responses, as models struggle with providing evidence for complex claims. We release code and data for further investigation. [Link](https://github.com/UKPLab/arxiv2024-attribute-or-abstain)

We introduce DA-Code, a code generation benchmark specifically designed to assess LLMs on agent-based data science tasks. This benchmark features three core elements: First, the tasks within DA-Code are inherently challenging, setting them apart from traditional code generation tasks and demanding advanced coding skills in grounding and planning. Second, examples in DA-Code are all based on real and diverse data, covering a wide range of complex data wrangling and analytics tasks. Third, to solve the tasks, the models must utilize complex data science programming languages, including Python and SQL, to perform intricate data processing and derive the answers. We set up the benchmark in a controllable and executable environment that aligns with real-world data analysis scenarios and is scalable. The annotators meticulously designed the evaluation suite to ensure the accuracy and robustness of the evaluation. We developed the DA-Agent baseline. Experiments show that although the baseline performs better than other existing frameworks, using the current best LLMs achieves only 30.5% accuracy, leaving ample room for improvement. We release our benchmark at [link](https://github.com/yiyihum/dabench)

The proliferation of Large Language Models like ChatGPT has significantly advanced language understanding and generation, impacting a broad spectrum of applications. However, these models predominantly excel in text-based tasks, overlooking the complexity of real-world multimodal information. This study introduces MultiAPI, a pioneering comprehensive large-scale API benchmark dataset aimed at expanding LLMs’ proficiency in multimodal contexts. Developed collaboratively through ChatGPT, MultiAPI consists of 187 diverse API calls and 1,799 contextual prompts, offering a unique platform evaluation of tool-augmented LLMs handling multimodal tasks. Through comprehensive experiments, our findings reveal that while LLMs demonstrate proficiency in API call decision-making, they face challenges in domain identification, function selection, and argument generation. What’s more, we surprisingly notice that auxiliary context can actually impair the performance. An in-depth error analysis paves the way for a new paradigm to address these challenges, suggesting a potential direction for future LLM research.

Open large language models (LLMs) with great performance in various tasks have significantly advanced the development of LLMs. However, they are far inferior to commercial models such as ChatGPT and GPT-4 when acting as agents to tackle complex tasks in the real world. These agent tasks employ LLMs as the central controller responsible for planning, memorization, and tool utilization, necessitating both fine-grained prompting methods and robust LLMs to achieve satisfactory performance. Though many prompting methods have been proposed to complete particular agent tasks, there is lack of research focusing on improving the agent capabilities of LLMs themselves without compromising their general abilities. In this work, we present AgentTuning, a simple and general method to enhance the agent abilities of LLMs while maintaining their general LLM capabilities. We construct AgentInstruct, a lightweight instruction-tuning dataset containing high-quality interaction trajectories. We employ a hybrid instruction-tuning strategy by combining AgentInstruct with open-source instructions from general domains. AgentTuning is used to instruction-tune the Llama 2 series, resulting in AgentLM. Our evaluations show that AgentTuning enables LLMs’ agent capabilities without compromising general abilities. The AgentLM-70B is comparable to GPT-3.5-turbo on unseen agent tasks, demonstrating generalized agent capabilities. We open source the AgentInstruct and AgentLM-7B, 13B, and 70B models at https://anonymous.4open.science/r/AgentTuning, serving open and powerful alternatives to commercial LLMs for agent tasks.

Large language models (LLMs) have manifested strong ability to generate codes for productive activities. However, current benchmarks for code synthesis, such as HumanEval, MBPP, and DS-1000, are predominantly oriented towards introductory tasks on algorithm and data science, insufficiently satisfying challenging requirements prevalent in real-world coding. To fill this gap, we propose NaturalCodeBench (NCB), a challenging code benchmark designed to mirror the complexity and variety of scenarios in real coding tasks. NCB comprises 402 high-quality problems in Python and Java, meticulously selected from natural user queries from online coding services, covering 6 different domains. Noting the extraordinary difficulty in creating testing cases for real-world queries, we also introduce a semi-automated pipeline to enhance the efficiency of test case construction. Comparing with manual solutions, it achieves an efficiency increase of more than 4 times. Our systematic experiments on 39 LLMs find that performance gaps on NCB between models with close HumanEval scores could still be significant, indicating a lack of focus on practical code synthesis scenarios or over-specified optimization on HumanEval. On the other hand, even the best-performing GPT-4 is still far from satisfying on NCB. The evaluation toolkit and development set are available at https://github.com/THUDM/NaturalCodeBench.

Existing large language models struggle to support numerous low-resource languages, particularly the extremely low-resource ones, for which there is minimal training data available for effective parameter updating. We thus investigate whether LLMs can learn a new language on the fly solely through prompting. To study this question, we collect a research suite for Zhuang, a language supported by no LLMs currently. We introduce DiPMT++, a framework for adapting LLMs to unseen languages by in-context learning. Using a dictionary and 5K parallel sentences only, DiPMT++ significantly enhances the performance of GPT-4 from 0 to 16 BLEU for Chinese-to-Zhuang translation and achieves 32 BLEU for Zhuang-to-Chinese translation. We also validate the effectiveness of our framework on Kalamang, another unseen language. Furthermore, we demonstrate the practical utility of DiPMT++ in aiding humans in translating completely unseen languages, which could contribute to the preservation of linguistic diversity.

We identify two crucial limitations in the evaluation of recent parallel-integrated method Parallel Context Windows (PCW), which extends the maximum context lengths of language models, e.g., 2048 for LLaMA, by harnessing window-wise attention and positional embedding techniques. We first show that a simple yet strong baseline, weighted sum ensemble, is missing for the in-context few-shot classification. Moreover, on more challenging Chain-of-Thought (CoT) reasoning (e.g., HotpotQA), PCW would present unexpected deterioration regarding question miscomprehension and false inference. Based on our findings, we suggest that the existing PCW design may not guarantee sufficient improvement and practicality in handling lengthy documents in real-world applications. More community efforts on enabling language models’ long context understanding ability should be paid.

Quantitative reasoning is a critical skill to analyze data, yet the assessment of such ability remains limited. To address this gap, we introduce the Quantitative Reasoning with Data (QRData) benchmark, aiming to evaluate Large Language Models’ capability in statistical and causal reasoning with real-world data. The benchmark comprises a carefully constructed dataset of 411 questions accompanied by data sheets from textbooks, online learning materials, and academic papers. To compare models’ quantitative reasoning abilities on data and text, we enrich the benchmark with an auxiliary set of 290 text-only questions, namely QRText. We evaluate natural language reasoning, program-based reasoning, and agent reasoning methods including Chain-of-Thought, Program-of-Thoughts, ReAct, and code interpreter assistants on diverse models. The strongest model GPT-4 achieves an accuracy of 58%, which has much room for improvement. Among open-source models, Deepseek-coder-instruct, a code LLM pretrained on 2T tokens, gets the highest accuracy of 37%. Analysis reveals that models encounter difficulties in data analysis and causal reasoning, and struggle in using causal knowledge and provided data simultaneously. Code and data are in https://github.com/xxxiaol/QRData.

Large language models (LLMs) have demonstrated impressive reasoning capabilities, yet there is ongoing debate about these abilities and the potential data contamination problem recently. This paper aims to evaluate the reasoning capacities of LLMs, specifically in solving recent competition-level programming problems in Codeforces, which are expert-crafted and unique, requiring deep understanding and robust reasoning skills. We first provide a comprehensive evaluation of GPT-4’s perceived zero-shot performance on this task, considering various aspects such as problems’ release time, difficulties, and types of errors encountered. Surprisingly, the perceived performance of GPT-4 has experienced a cliff like decline in problems after September 2021 consistently across all the difficulties and types of problems, which shows the potential data contamination, as well as the challenges for any existing LLM to solve unseen complex reasoning problems. We further explore various approaches such as fine-tuning, Chain-of-Thought prompting and problem description simplification. Unfortunately, none of them is able to consistently mitigate the challenges. Through our work, we emphasize the importance of this excellent data source for assessing the genuine reasoning capabilities of LLMs, and foster the development of LLMs with stronger reasoning abilities and better generalization in the future.

Although Large Language Models (LLMs) are becoming increasingly powerful, they still exhibit significant but subtle weaknesses, such as mistakes in instruction-following or coding tasks.As these unexpected errors could lead to severe consequences in practical deployments, it is crucial to investigate the limitations within LLMs systematically.Traditional benchmarking approaches cannot thoroughly pinpoint specific model deficiencies, while manual inspections are costly and not scalable. In this paper, we introduce a unified framework, AutoDetect, to automatically expose weaknesses in LLMs across various tasks. Inspired by the educational assessment process that measures students’ learning outcomes, AutoDetect consists of three LLM-powered agents: Examiner, Questioner, and Assessor.The collaboration among these three agents is designed to realize comprehensive and in-depth weakness identification. Our framework demonstrates significant success in uncovering flaws, with an identification success rate exceeding 30% in prominent models such as ChatGPT and Claude.More importantly, these identified weaknesses can guide specific model improvements, proving more effective than untargeted data augmentation methods like Self-Instruct. Our approach has led to substantial enhancements in popular LLMs, including the Llama series and Mistral-7b, boosting their performance by over 10% across several benchmarks.Code and data are publicly available at https://github.com/thu-coai/AutoDetect.

Large language models (LLMs) have shown excellent mastering of human language but still struggle in real-world applications that require mathematical problem-solving. While many strategies and datasets to enhance LLMs’ mathematics are developed, it remains a challenge to simultaneously maintain and improve both language and mathematical capabilities in deployed LLM systems. In this work, we tailor the Self-Critique pipeline, which addresses the challenge in the feedback learning stage of LLM alignment. We first train a general Math-Critique model from the LLM itself to provide feedback signals. Then, we sequentially employ rejective fine-tuning and direct preference optimization over the LLM’s own generations for data collection. Based on ChatGLM3-32B, we conduct experiments on both academic and our newly created challenging dataset, MathUserEval. Results show that our pipeline significantly enhances the LLM’s mathematical problem-solving while still improving its language ability, outperforming LLMs that could be two times larger. Related techniques have been deployed to ChatGLM, an online serving LLM. Related evaluation datasets and scripts are released at https://github.com/THUDM/ChatGLM-Math.

The argument sufficiency assessment task aims to determine if the premises of a given argument support its conclusion.To tackle this task, existing works often train a classifier on data annotated by humans. However, annotating data is laborious, and annotations are often inconsistent due to subjective criteria. Motivated by the definition of probability of sufficiency (PS) in the causal literature, we proposeCASA, a zero-shot causality-driven argument sufficiency assessment framework. PS measures how likely introducing the premise event would lead to the conclusion when both the premise and conclusion events are absent. To estimate this probability, we propose to use large language models (LLMs) to generate contexts that are inconsistent with the premise and conclusion and revise them by injecting the premise event.Experiments on two logical fallacy detection datasets demonstrate that CASA accurately identifies insufficient arguments. We further deploy CASA in a writing assistance application, and find that suggestions generated by CASA enhance the sufficiency of student-written arguments. Code and data are available at https://github.com/xxxiaol/CASA.

Although large language models (LLMs) demonstrate impressive performance for many language tasks, most of them can only handle texts a few thousand tokens long, limiting their applications on longer sequence inputs, such as books, reports, and codebases. Recent works have proposed methods to improve LLMs’ long context capabilities by extending context windows and more sophisticated memory mechanisms. However, comprehensive benchmarks tailored for evaluating long context understanding are lacking. In this paper, we introduce LongBench, the first bilingual, multi-task benchmark for long context understanding, enabling a more rigorous evaluation of long context understanding. LongBench comprises 21 datasets across 6 task categories in both English and Chinese, with an average length of 6,711 words (English) and 13,386 characters (Chinese). These tasks cover key long-text application areas including single-doc QA, multi-doc QA, summarization, few-shot learning, synthetic tasks, and code completion. All datasets in LongBench are standardized into a unified format, allowing for effortless automatic evaluation of LLMs. Upon comprehensive evaluation of 8 LLMs on LongBench, we find that: (1) Commercial model (GPT-3.5-Turbo-16k) outperforms other open-sourced models, but still struggles on longer contexts. (2) Scaled position embedding and fine-tuning on longer sequences lead to substantial improvement on long context understanding. (3) Context compression technique such as retrieval brings improvement for model with weak ability on long contexts, but the performance still lags behind models that have strong long context understanding capability.

Large language models (LLMs) have shown impressive success in various applications. However, these models are often not well aligned with human intents, which calls for additional treatments on them; that is, the alignment problem. To make LLMs better follow user instructions, existing alignment methods primarily focus on further training them. However, the extra training of LLMs is usually expensive in terms of GPU computing; even worse, some LLMs are not accessible for user-demanded training, such as GPTs. In this work, we take a different perspective—Black-Box Prompt Optimization (BPO)—to perform alignments. The idea is to optimize user prompts to suit LLMs’ input understanding, so as to best realize users’ intents without updating LLMs’ parameters. BPO leverages human preferences to optimize prompts, thus making it superior to LLM (e.g., ChatGPT) as a prompt engineer. Moreover, BPO is model-agnostic, and the empirical results demonstrate that the BPO-aligned ChatGPT yields a 22% increase in the win rate against its original version and 10% for GPT-4. Notably, the BPO-aligned LLMs can outperform the same models aligned by PPO and DPO, and it also brings additional performance gains when combining BPO with PPO or DPO. Code and datasets are released at https://github.com/thu-coai/BPO.

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs’ alignment in Chinese. We tailor a human-in-the-loop data curation pipeline, containing 8 main categories, 683 real-scenario rooted queries and corresponding human verified references.To ensure references’ correctness, each knowledge-intensive query is accompanied with evidences collected from reliable webpages (including the url and quotation) by our annotators.For automatic evaluation, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge (CITATION) with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability.All evaluation codes and data are publicly available at https://github.com/THUDM/AlignBench

Since the natural language processing (NLP) community started to make large language models (LLMs) act as a critic to evaluate the quality of generated texts, most of the existing works train a critique generation model on the evaluation data labeled by GPT-4’s direct prompting. We observe that these models lack the ability to generate informative critiques in both pointwise grading and pairwise comparison especially without references. As a result, their generated critiques cannot provide fine-grained distinguishability on generated texts, causing unsatisfactory evaluation performance. In this paper, we propose a simple yet effective method called Eval-Instruct, which can first acquire pointwise grading critiques with pseudo references and then revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings, including pointwise grading and pairwise comparison with / without references. After fine-tuning on these data, the resulting model CritiqueLLM is empirically shown to outperform ChatGPT and all the open-source baselines and even achieve comparable evaluation performance to GPT-4 in system-level correlations of pointwise grading. We also demonstrate that our generated critiques can act as scalable feedback to further improve the generation quality of strong LLMs like ChatGPT.

With the rapid development of Large Language Models (LLMs), increasing attention has been paid to their safety concerns. Consequently, evaluating the safety of LLMs has become an essential task for facilitating the broad applications of LLMs. Nevertheless, the absence of comprehensive safety evaluation benchmarks poses a significant impediment to effectively assess and enhance the safety of LLMs. In this work, we present SafetyBench, a comprehensive benchmark for evaluating the safety of LLMs, which comprises 11,435 diverse multiple choice questions spanning across 7 distinct categories of safety concerns. Notably, SafetyBench also incorporates both Chinese and English data, facilitating the evaluation in both languages. Our extensive tests over 25 popular Chinese and English LLMs in both zero-shot and few-shot settings reveal a substantial performance advantage for GPT-4 over its counterparts, and there is still significant room for improving the safety of current LLMs. We also demonstrate that the measured safety understanding abilities in SafetyBench are correlated with safety generation abilities. Data and evaluation guidelines are available at https://github.com/thu-coai/SafetyBench. Submission entrance and leaderboard are available at https://llmbench.ai/safety.

We introduce OpenWebAgent, an open toolkit designed to optimize web automation by integrating both large language models (LLMs) and large multimodal models (LMMs). This toolkit focuses on enhancing human-computer interactions on the web, simplifying complex tasks through an advanced HTML parser, a rapid action generation module, and an intuitive user interface. At the core of OpenWebAgent is an innovative web agent framework that uses a modular design to allow developers to seamlessly integrate a variety of models and tools to process web information and automate tasks on the web. This enables the development of powerful, task-oriented web agents, significantly enhancing user experience and operational efficiency on the web. The OpenWebAgent framework, Chrome plugin, and demo video are available at https://github.com/THUDM/OpenWebAgent/.

Multiple-choice questions (MCQs) are widely used in the evaluation of large language models (LLMs) due to their simplicity and efficiency. However, there are concerns about whether MCQs can truly measure LLM’s capabilities, particularly in knowledge-intensive scenarios where long-form generation (LFG) answers are required. The misalignment between the task and the evaluation method demands a thoughtful analysis of MCQ’s efficacy, which we undertake in this paper by evaluating nine LLMs on four question-answering (QA) datasets in two languages: Chinese and English. We identify a significant issue: LLMs exhibit an order sensitivity in bilingual MCQs, favoring answers located at specific positions, i.e., the first position. We further quantify the gap between MCQs and long-form generation questions (LFGQs) by comparing their direct outputs, token logits, and embeddings. Our results reveal a relatively low correlation between answers from MCQs and LFGQs for identical questions. Additionally, we propose two methods to quantify the consistency and confidence of LLMs’ output, which can be generalized to other QA evaluation benchmarks. Notably, our analysis challenges the idea that the higher the consistency, the greater the accuracy. We also find MCQs to be less reliable than LFGQs in terms of expected calibration error. Finally, the misalignment between MCQs and LFGQs is not only reflected in the evaluation performance but also in the embedding space. Our code and models can be accessed at https://github.com/Meetyou-AI-Lab/Can-MC-Evaluate-LLMs.

The multi-answer phenomenon, where a question may have multiple answers scattered in the document, can be well handled by humans but is challenging enough for machine reading comprehension (MRC) systems. Despite recent progress in multi-answer MRC, there lacks a systematic analysis of how this phenomenon arises and how to better address it. In this work, we design a taxonomy to categorize commonly-seen multi-answer MRC instances, with which we inspect three multi-answer datasets and analyze where the multi-answer challenge comes from. We further analyze how well different paradigms of current multi-answer MRC models deal with different types of multi-answer instances. We find that some paradigms capture well the key information in the questions while others better model the relation between questions and contexts. We thus explore strategies to make the best of the strengths of different paradigms. Experiments show that generation models can be a promising platform to incorporate different paradigms. Our annotations and code are released for further research.

Causal reasoning, the ability to identify cause-and-effect relationship, is crucial in human thinking. Although large language models (LLMs) succeed in many NLP tasks, it is still challenging for them to conduct complex causal reasoning like abductive reasoning and counterfactual reasoning. Given the fact that programming code may express causal relations more often and explicitly with conditional statements like “if“, we want to explore whether Code-LLMs acquire better causal reasoning abilities. Our experiments show that compared to text-only LLMs, Code-LLMs with code prompts are better causal reasoners. We further intervene on the prompts from different aspects, and discover that the key point is the programming structure. Code and data are available at https://github.com/xxxiaol/magic-if.

Event extraction aims to recognize pre-defined event triggers and arguments from texts, which suffer from the lack of high-quality annotations. In most NLP applications, involving a large scale of synthetic training data is a practical and effective approach to alleviate the problem of data scarcity. However, when applying to the task of event extraction, recent data augmentation methods often neglect the problem of grammatical incorrectness, structure misalignment, and semantic drifting, leading to unsatisfactory performances. In order to solve these problems, we propose a denoised structure-to-text augmentation framework for event extraction (DAEE), which generates additional training data through the knowledge-based structure-to-text generation model and selects the effective subset from the generated data iteratively with a deep reinforcement learning agent. Experimental results on several datasets demonstrate that the proposed method generates more diverse text representations for event extraction and achieves comparable results with the state-of-the-art.

Extractive summarization aims to form a summary by directly extracting sentences from the source document. Existing works mostly formulate it as a sequence labeling problem by making individual sentence label predictions. This paper proposes DiffuSum, a novel paradigm for extractive summarization, by directly generating the desired summary sentence representations with diffusion models and extracting sentences based on sentence representation matching. In addition, DiffuSum jointly optimizes a contrastive sentence encoder with a matching loss for sentence representation alignment and a multi-class contrastive loss for representation diversity. Experimental results show that DiffuSum achieves the new state-of-the-art extractive results on CNN/DailyMail with ROUGE scores of 44.83/22.56/40.56. Experiments on the other two datasets with different summary lengths and cross-dataset evaluation also demonstrate the effectiveness of DiffuSum. The strong performance of our framework shows the great potential of adapting generative models for extractive summarization.

Extractive summarization is a crucial task in natural language processing that aims to condense long documents into shorter versions by directly extracting sentences. The recent introduction of large language models has attracted significant interest in the NLP community due to its remarkable performance on a wide range of downstream tasks. This paper first presents a thorough evaluation of ChatGPT’s performance on extractive summarization and compares it with traditional fine-tuning methods on various benchmark datasets. Our experimental analysis reveals that ChatGPT exhibits inferior extractive summarization performance in terms of ROUGE scores compared to existing supervised systems, while achieving higher performance based on LLM-based evaluation metrics. In addition, we explore the effectiveness of in-context learning and chain-of-thought reasoning for enhancing its performance. Furthermore, we find that applying an extract-then-generate pipeline with ChatGPT yields significant performance improvements over abstractive baselines in terms of summary faithfulness. These observations highlight potential directions for enhancing ChatGPT’s capabilities in faithful summarization using two-stage approaches.

With the advance of large language models (LLMs), the research field of LLM applications becomes more and more popular and the idea of constructing pipelines to accomplish complex tasks by stacking LLM API calls come true. However, this kind of methods face two limitations: narrow information coverage and low fault tolerance. In this work, we propose a novel method called ALLIES. Given an input query, ALLIES leverages LLMs to iteratively generate new queries related to the original query, enabling an iterative reasoning process. By iteratively refining and expanding the scope of the original query, ALLIES captures and utilizes hidden knowledge that may not be directly obtainable through retrieval. We take zero-shot open-domain question answering (ODQA) as an application scene and evaluate ALLIES on the widely-used benchmarks, such as NQ, WebQ and TriviaQA. The experimental results demonstrate that ALLIES significantly outperforms other zero-shot baselines, indicating its effectiveness in tackling those challenges. Our code is available in https://github.com/microsoft/SimXNS/tree/main/ALLIES.

Existing text summarization systems have made significant progress in recent years, but typically generate summaries in a single step. The one-shot summarization setting is sometimes inadequate, however, as the generated summary may contain hallucinations or overlook important details related to the reader’s interests. In this paper, we address this limitation by proposing SummIt, an iterative text summarization framework based on large language models like ChatGPT. Our framework enables the model to refine the generated summary iteratively through self-evaluation and feedback, closely resembling the iterative process humans undertake when drafting and revising summaries. Furthermore, we explore the potential benefits of integrating knowledge and topic extractors into the framework to enhance summary faithfulness and controllability. We evaluate the performance of our framework on three benchmark summarization datasets through empirical and qualitative analyses. We also conduct a human evaluation to validate the effectiveness of the model’s refinements and find a potential issue of over-correction.

Prompt tuning attempts to update few task-specific parameters in pre-trained models. It has achieved comparable performance to fine-tuning of the full parameter set on both language understanding and generation tasks. In this work, we study the problem of prompt tuning for neural text retrievers. We introduce parameter-efficient prompt tuning for text retrieval across in-domain, cross-domain, and cross-topic settings. Through an extensive analysis, we show that the strategy can mitigate the two issues—parameter-inefficiency and weak generalizability—faced by fine-tuning based retrieval methods. Notably, it can significantly improve the out-of-domain zero-shot generalization of the retrieval models. By updating only 0.1% of the model parameters, the prompt tuning strategy can help retrieval models achieve better generalization performance than traditional methods in which all parameters are updated. Finally, to facilitate research on retrievers’ cross-topic generalizability, we curate and release an academic retrieval dataset with 18K query-results pairs in 87 topics, making it the largest topic-specific one to date.

Instruction tuning has emerged to enhance the capabilities of large language models (LLMs) to comprehend instructions and generate appropriate responses. Existing methods either manually annotate or employ LLM (e.g., GPT-series) to generate data for instruction tuning. However, they often overlook associating instructions with existing annotated datasets. In this paper, we propose Dynosaur, a dynamic growth paradigm for the automatic curation of instruction-tuning data. Based on the metadata of existing datasets, we use LLMs to automatically construct instruction-tuning data by identifying relevant data fields and generating appropriate instructions. By leveraging the existing annotated datasets, Dynosaur offers several advantages: 1) it reduces the API cost for generating instructions (e.g., it costs less than $12 USD by calling GPT-3.5-turbo for generating 800K instruction tuning samples; 2) it provides high-quality data for instruction tuning (e.g., it performs better than Alpaca and Flan on Super-NI and Longform with comparable data sizes); and 3) it supports the continuous improvement of models by generating instruction-tuning data when a new annotated dataset becomes available. We further investigate a continual learning scheme for learning with the ever-growing instruction-tuning dataset, and demonstrate that replaying tasks with diverse instruction embeddings not only helps mitigate forgetting issues but generalizes to unseen tasks better. Code and data are available at https://github.com/WadeYin9712/Dynosaur.

Most of the existing compositional generalization datasets are synthetically-generated, resulting in a lack of natural language variation. While there have been recent attempts to introduce non-synthetic datasets for compositional generalization, they suffer from either limited data scale or a lack of diversity in the forms of combinations. To better investigate compositional generalization with more linguistic phenomena and compositional diversity, we propose the DIsh NamE Recognition (DiNeR) task and create a large realistic Chinese dataset. Given a recipe instruction, models are required to recognize the dish name composed of diverse combinations of food, actions, and flavors. Our dataset consists of 3,811 dishes and 228,114 recipes, and involves plenty of linguistic phenomena such as anaphora, omission and ambiguity. We provide two strong baselines based on T5 and large language models (LLMs). This work contributes a challenging task, baseline methods to tackle the task, and insights into compositional generalization in the context of dish name recognition.

“Our research paper delves into the Adversarial Robustness Evaluation for Chinese Gaokao Read-ing Comprehension (GCRC advRobust). While Chinese reading comprehension tasks havegained significant attention in recent years, previous methods have not proven effective for thischallenging dataset. We focus on exploring how prompt engineering can impact a model’s read-ing comprehension ability. Through our experiments using ChatGLM, GPT3.5, and GPT4, wediscovered a correlation between prompt and LLM reading comprehension ability, and found thatprompt engineering improves the performance of each model. Our team submitted the results ofour system evaluation, which ranked first in three indexes and total scores. Keywords— LLM, Prompt, Chinese Reading Comprehension”

The extended structural context has made scientific paper summarization a challenging task. This paper proposes CHANGES, a contrastive hierarchical graph neural network for extractive scientific paper summarization. CHANGES represents a scientific paper with a hierarchical discourse graph and learns effective sentence representations with dedicated designed hierarchical graph information aggregation. We also propose a graph contrastive learning module to learn global theme-aware sentence representations. Extensive experiments on the PubMed and arXiv benchmark datasets prove the effectiveness of CHANGES and the importance of capturing hierarchical structure information in modeling scientific papers.

Different from previous fact extraction and verification tasks that only consider evidence of a single format, FEVEROUS brings further challenges by extending the evidence format to both plain text and tables. Existing works convert all candidate evidence into either sentences or tables, thus often failing to fully capture the rich context in their original format from the converted evidence, let alone the context information lost during conversion. In this paper, we propose a Dual Channel Unified Format fact verification model (DCUF), which unifies various evidence into parallel streams, i.e., natural language sentences and a global evidence table, simultaneously. With carefully-designed evidence conversion and organization methods, DCUF makes the most of pre-trained table/language models to encourage each evidence piece to perform early and thorough interactions with other pieces in its original format. Experiments show that our model can make better use of existing pre-trained models to absorb evidence of two formats, thus outperforming previous works by a large margin. Our code and models are publicly available.

There have been various types of pretraining architectures including autoencoding models (e.g., BERT), autoregressive models (e.g., GPT), and encoder-decoder models (e.g., T5). However, none of the pretraining frameworks performs the best for all tasks of three main categories including natural language understanding (NLU), unconditional generation, and conditional generation. We propose a General Language Model (GLM) based on autoregressive blank infilling to address this challenge. GLM improves blank filling pretraining by adding 2D positional encodings and allowing an arbitrary order to predict spans, which results in performance gains over BERT and T5 on NLU tasks. Meanwhile, GLM can be pretrained for different types of tasks by varying the number and lengths of blanks. On a wide range of tasks across NLU, conditional and unconditional generation, GLM outperforms BERT, T5, and GPT given the same model sizes and data, and achieves the best performance from a single pretrained model with 1.25× parameters of BERT Large , demonstrating its generalizability to different downstream tasks.

Spatial commonsense, the knowledge about spatial position and relationship between objects (like the relative size of a lion and a girl, and the position of a boy relative to a bicycle when cycling), is an important part of commonsense knowledge. Although pretrained language models (PLMs) succeed in many NLP tasks, they are shown to be ineffective in spatial commonsense reasoning. Starting from the observation that images are more likely to exhibit spatial commonsense than texts, we explore whether models with visual signals learn more spatial commonsense than text-based PLMs. We propose a spatial commonsense benchmark that focuses on the relative scales of objects, and the positional relationship between people and objects under different actions. We probe PLMs and models with visual signals, including vision-language pretrained models and image synthesis models, on this benchmark, and find that image synthesis models are more capable of learning accurate and consistent spatial knowledge than other models. The spatial knowledge from image synthesis models also helps in natural language understanding tasks that require spatial commonsense.

We consider event extraction in a generative manner with template-based conditional generation. Although there is a rising trend of casting the task of event extraction as a sequence generation problem with prompts, these generation-based methods have two significant challenges, including using suboptimal prompts and static event type information. In this paper, we propose a generative template-based event extraction method with dynamic prefix (GTEE-DynPref) by integrating context information with type-specific prefixes to learn a context-specific prefix for each context. Experimental results show that our model achieves competitive results with the state-of-the-art classification-based model OneIE on ACE 2005 and achieves the best performances on ERE.Additionally, our model is proven to be portable to new types of events effectively.

Prompt tuning, which only tunes continuous prompts with a frozen language model, substantially reduces per-task storage and memory usage at training. However, in the context of NLU, prior work reveals that prompt tuning does not perform well for normal-sized pretrained models. We also find that existing methods of prompt tuning cannot handle hard sequence labeling tasks, indicating a lack of universality. We present a novel empirical finding that properly optimized prompt tuning can be universally effective across a wide range of model scales and NLU tasks. It matches the performance of finetuning while having only 0.1%-3% tuned parameters. Our method P-Tuning v2 is an implementation of Deep Prompt Tuning (CITATION) optimized and adapted for NLU. Given the universality and simplicity of P-Tuning v2, we believe it can serve as an alternative to finetuning and a strong baseline for future research.

People can acquire knowledge in an unsupervised manner by reading, and compose the knowledge to make novel combinations. In this paper, we investigate whether pretrained language models can perform compositional generalization in a realistic setting: recipe generation. We design the counterfactual recipe generation task, which asks models to modify a base recipe according to the change of an ingredient. This task requires compositional generalization at two levels: the surface level of incorporating the new ingredient into the base recipe, and the deeper level of adjusting actions related to the changing ingredient. We collect a large-scale recipe dataset in Chinese for models to learn culinary knowledge, and a subset of action-level fine-grained annotations for evaluation.We finetune pretrained language models on the recipe corpus, and use unsupervised counterfactual generation methods to generate modified recipes.Results show that existing models have difficulties in modifying the ingredients while preserving the original text style, and often miss actions that need to be adjusted. Although pretrained language models can generate fluent recipe texts, they fail to truly learn and use the culinary knowledge in a compositional way. Code and data are available at https://github.com/xxxiaol/counterfactual-recipe-generation.

We introduce a new open information extraction (OIE) benchmark for pre-trained language models (LM). Recent studies have demonstrated that pre-trained LMs, such as BERT and GPT, may store linguistic and relational knowledge. In particular, LMs are able to answer “fill-in-the-blank” questions when given a pre-defined relation category. Instead of focusing on pre-defined relations, we create an OIE benchmark aiming to fully examine the open relational information present in the pre-trained LMs. We accomplish this by turning pre-trained LMs into zero-shot OIE systems. Surprisingly, pre-trained LMs are able to obtain competitive performance on both standard OIE datasets (CaRB and Re-OIE2016) and two new large-scale factual OIE datasets (TAC KBP-OIE and Wikidata-OIE) that we establish via distant supervision. For instance, the zero-shot pre-trained LMs outperform the F1 score of the state-of-the-art supervised OIE methods on our factual OIE datasets without needing to use any training sets.

Extractive summarization for long documents is challenging due to the extended structured input context. The long-distance sentence dependency hinders cross-sentence relations modeling, the critical step of extractive summarization. This paper proposes HEGEL, a hypergraph neural network for long document summarization by capturing high-order cross-sentence relations. HEGEL updates and learns effective sentence representations with hypergraph transformer layers and fuses different types of sentence dependencies, including latent topics, keywords coreference, and section structure. We validate HEGEL by conducting extensive experiments on two benchmark datasets, and experimental results demonstrate the effectiveness and efficiency of HEGEL.

Sampling proper negatives from a large document pool is vital to effectively train a dense retrieval model. However, existing negative sampling strategies suffer from the uninformative or false negative problem. In this work, we empirically show that according to the measured relevance scores, the negatives ranked around the positives are generally more informative and less likely to be false negatives. Intuitively, these negatives are not too hard (may be false negatives) or too easy (uninformative). They are the ambiguous negatives and need more attention during training.Thus, we propose a simple ambiguous negatives sampling method, SimANS, which incorporates a new sampling probability distribution to sample more ambiguous negatives.Extensive experiments on four public and one industry datasets show the effectiveness of our approach.We made the code and models publicly available in https://github.com/microsoft/SimXNS.

We introduce a method for improving the structural understanding abilities of language models. Unlike previous approaches that finetune the models with task-specific augmentation, we pretrain language models to generate structures from the text on a collection of task-agnostic corpora. Our structure pretraining enables zero-shot transfer of the learned knowledge that models have about the structure tasks. We study the performance of this approach on 28 datasets, spanning 10 structure prediction tasks including open information extraction, joint entity and relation extraction, named entity recognition, relation classification, semantic role labeling, event extraction, coreference resolution, factual probe, intent detection, and dialogue state tracking. We further enhance the pretraining with the task-specific training sets. We show that a 10B parameter language model transfers non-trivially to most tasks and obtains state-of-the-art performance on 21 of 28 datasets that we evaluate. Our code and datasets will be made publicly available.

Causal inference is the process of capturing cause-effect relationship among variables. Most existing works focus on dealing with structured data, while mining causal relationship among factors from unstructured data, like text, has been less examined, but is of great importance, especially in the legal domain. In this paper, we propose a novel Graph-based Causal Inference (GCI) framework, which builds causal graphs from fact descriptions without much human involvement and enables causal inference to facilitate legal practitioners to make proper decisions. We evaluate the framework on a challenging similar charge disambiguation task. Experimental results show that GCI can capture the nuance from fact descriptions among multiple confusing charges and provide explainable discrimination, especially in few-shot settings. We also observe that the causal knowledge contained in GCI can be effectively injected into powerful neural networks for better performance and interpretability.

The key challenge of the visual dialog task is how to fuse features from multimodal sources and extract relevant information from dialog history to answer the current query. In this work, we formulate a visual dialog as an information flow in which each piece of information is encoded with the joint visual-linguistic representation of a single dialog round. Based on this formulation, we consider the visual dialog task as a sequence problem consisting of ordered visual-linguistic vectors. For featurization, we use a Dense SymmetricCo-Attention network (Nguyen and Okatani,2018) as a lightweight vison-language joint representation generator to fuse multimodal features (i.e., image and text), yielding better computation and data efficiencies. For inference, we propose two Sequential Dialog Networks (SeqDialN): the first uses LSTM(Hochreiter and Schmidhuber,1997) for information propagation (IP) and the second uses a modified Transformer (Vaswani et al.,2017) for multi-step reasoning (MR). Our architecture separates the complexity of multimodal feature fusion from that of inference, which allows simpler design of the inference engine. On VisDial v1.0 test-std dataset, our best single generative SeqDialN achieves 62.54% NDCG and 48.63% MRR; our ensemble generative SeqDialN achieves 63.78% NDCG and 49.98% MRR, which set a new state-of-the-art generative visual dialog model. We fine-tune discriminative SeqDialN with dense annotations and boost the performance up to 72.41% NDCG and 55.11% MRR. In this work, we discuss the extensive experiments we have conducted to demonstrate the effectiveness of our model components. We also provide visualization for the reasoning process from the relevant conversation rounds and discuss our fine-tuning methods. The code is available at https://github.com/xiaoxiaoheimei/SeqDialN.

Query Rewriting (QR) is proposed to solve the problem of the word mismatch between queries and documents in Web search. Existing approaches usually model QR with an end-to-end sequence-to-sequence (seq2seq) model. The state-of-the-art Transformer-based models can effectively learn textual semantics from user session logs, but they often ignore users’ geographic location information that is crucial for the Point-of-Interest (POI) search of map services. In this paper, we proposed a pre-training model, called Geo-BERT, to integrate semantics and geographic information in the pre-trained representations of POIs. Firstly, we simulate POI distribution in the real world as a graph, in which nodes represent POIs and multiple geographic granularities. Then we use graph representation learning methods to get geographic representations. Finally, we train a BERT-like pre-training model with text and POIs’ graph embeddings to get an integrated representation of both geographic and semantic information, and apply it in the QR of POI search. The proposed model achieves excellent accuracy on a wide range of real-world datasets of map services.

We cast a suite of information extraction tasks into a text-to-triple translation framework. Instead of solving each task relying on task-specific datasets and models, we formalize the task as a translation between task-specific input text and output triples. By taking the task-specific input, we enable a task-agnostic translation by leveraging the latent knowledge that a pre-trained language model has about the task. We further demonstrate that a simple pre-training task of predicting which relational information corresponds to which input text is an effective way to produce task-specific outputs. This enables the zero-shot transfer of our framework to downstream tasks. We study the zero-shot performance of this framework on open information extraction (OIE2016, NYT, WEB, PENN), relation classification (FewRel and TACRED), and factual probe (Google-RE and T-REx). The model transfers non-trivially to most tasks and is often competitive with a fully supervised method without the need for any task-specific training. For instance, we significantly outperform the F1 score of the supervised open information extraction without needing to use its training set.

Computational social science studies often contextualize content analysis within standard demographics. Since demographics are unavailable on many social media platforms (e.g. Twitter), numerous studies have inferred demographics automatically. Despite many studies presenting proof-of-concept inference of race and ethnicity, training of practical systems remains elusive since there are few annotated datasets. Existing datasets are small, inaccurate, or fail to cover the four most common racial and ethnic groups in the United States. We present a method to identify self-reports of race and ethnicity from Twitter profile descriptions. Despite the noise of automated supervision, our self-report datasets enable improvements in classification performance on gold standard self-report survey data. The result is a reproducible method for creating large-scale training resources for race and ethnicity.

Structural heterogeneity between knowledge graphs is an outstanding challenge for entity alignment. This paper presents Neighborhood Matching Network (NMN), a novel entity alignment framework for tackling the structural heterogeneity challenge. NMN estimates the similarities between entities to capture both the topological structure and the neighborhood difference. It provides two innovative components for better learning representations for entity alignment. It first uses a novel graph sampling method to distill a discriminative neighborhood for each entity. It then adopts a cross-graph neighborhood matching module to jointly encode the neighborhood difference for a given entity pair. Such strategies allow NMN to effectively construct matching-oriented entity representations while ignoring noisy neighbors that have a negative impact on the alignment task. Extensive experiments performed on three entity alignment datasets show that NMN can well estimate the neighborhood similarity in more tough cases and significantly outperforms 12 previous state-of-the-art methods.

We consider open domain event extraction, the task of extracting unconstraint types of events from news clusters. A novel latent variable neural model is constructed, which is scalable to very large corpus. A dataset is collected and manually annotated, with task-specific evaluation metrics being designed. Results show that the proposed unsupervised model gives better performance compared to the state-of-the-art method for event schema induction.

Entity alignment is a viable means for integrating heterogeneous knowledge among different knowledge graphs (KGs). Recent developments in the field often take an embedding-based approach to model the structural information of KGs so that entity alignment can be easily performed in the embedding space. However, most existing works do not explicitly utilize useful relation representations to assist in entity alignment, which, as we will show in the paper, is a simple yet effective way for improving entity alignment. This paper presents a novel joint learning framework for entity alignment. At the core of our approach is a Graph Convolutional Network (GCN) based framework for learning both entity and relation representations. Rather than relying on pre-aligned relation seeds to learn relation representations, we first approximate them using entity embeddings learned by the GCN. We then incorporate the relation approximation into entities to iteratively learn better representations for both. Experiments performed on three real-world cross-lingual datasets show that our approach substantially outperforms state-of-the-art entity alignment methods.

In this paper, we propose a soft label approach to target-level sentiment classification task, in which a history-based soft labeling model is proposed to measure the possibility of a context word as an opinion word. We also apply a convolution layer to extract local active features, and introduce positional weights to take relative distance information into consideration. In addition, we obtain more informative target representation by training with context tokens together to make deeper interaction between target and context tokens. We conduct experiments on SemEval 2014 datasets and the experimental results show that our approach significantly outperforms previous models and gives state-of-the-art results on these datasets.

Twitter has become one of the most import channels to spread latest scholarly information because of its fast information spread speed. How to predict whether a scholarly tweet will be retweeted is a key task in understanding the message propagation within large user communities. Hence, we present the real-time scholarly retweeting prediction system that retrieves scholarly tweets which will be retweeted. First, we filter scholarly tweets from tracking a tweet stream. Then, we extract Tweet Scholar Blocks indicating metadata of papers. At last, we combine scholarly features with the Tweet Scholar Blocks to predict whether a scholarly tweet will be retweeted. Our system outperforms chosen baseline systems. Additionally, our system has the potential to predict scientific impact in real-time.

Event extraction is of practical utility in natural language processing. In the real world, it is a common phenomenon that multiple events existing in the same sentence, where extracting them are more difficult than extracting a single event. Previous works on modeling the associations between events by sequential modeling methods suffer a lot from the low efficiency in capturing very long-range dependencies. In this paper, we propose a novel Jointly Multiple Events Extraction (JMEE) framework to jointly extract multiple event triggers and arguments by introducing syntactic shortcut arcs to enhance information flow and attention-based graph convolution networks to model graph information. The experiment results demonstrate that our proposed framework achieves competitive results compared with state-of-the-art methods.