Search

Large Language Models (LLMs) have exhibited strong mathematical reasoning and computational prowess, tackling tasks ranging from basic arithmetic to advanced competition-level problems. However, frequently occurring subtle errors, such as miscalculations or incorrect substitutions, limit the models' full mathematical potential. Existing studies to improve mathematical ability typically involve distilling reasoning skills from stronger LLMs or applying preference learning to step-wise response pairs. Although these methods leverage samples of varying granularity to mitigate reasoning errors, they overlook the frequently occurring subtle errors. A major reason is that sampled preference pairs involve differences unrelated to the errors, which may distract the model from focusing on subtle errors. In this work, we propose a novel preference learning framework called eRror-Injected Self-Editing (RISE), which injects predefined subtle errors into partial tokens of correct solutions to construct hard pairs for error mitigation. In detail, RISE uses the model itself to edit a small number of tokens in the solution, injecting designed subtle errors. Then, pairs composed of self-edited solutions and their corresponding correct ones, along with pairs of correct and incorrect solutions obtained through sampling, are used together for subtle error-aware DPO training. Compared with other preference learning methods, RISE further refines the training objective to focus on predefined errors and their tokens, without requiring fine-grained sampling or preference annotation. Extensive experiments validate the effectiveness of RISE, with preference learning on Qwen2-7B-Instruct yielding notable improvements of 3.0% on GSM8K and 7.9% on MATH.

Meta learning has been widely used to exploit rich-resource source tasks to improve the performance of low-resource target tasks. Unfortunately, most existing meta learning approaches treat different source tasks equally, ignoring the relatedness of source tasks to the target task in knowledge transfer. To mitigate this issue, we propose a reinforcement-based multi-source meta-transfer learning framework (Meta-RTL) for low-resource commonsense reasoning. In this framework, we present a reinforcement-based approach to dynamically estimating source task weights that measure the contribution of the corresponding tasks to the target task in the meta-transfer learning. The differences between the general loss of the meta model and task-specific losses of source-specific temporal meta models on sampled target data are fed into the policy network of the reinforcement learning module as rewards. The policy network is built upon LSTMs that capture long-term dependencies on source task weight estimation across meta learning iterations. We evaluate the proposed Meta-RTL using both BERT and ALBERT as the backbone of the meta model on three commonsense reasoning benchmark datasets. Experimental results demonstrate that Meta-RTL substantially outperforms strong baselines and previous task selection strategies and achieves larger improvements on extremely low-resource settings.

GPT-4o, an omni-modal model that enables vocal conversations with diverse emotions and tones, marks a milestone for omni-modal foundation models. However, empowering Large Language Models to perceive and generate images, texts, and speeches end-to-end with publicly available data remains challenging in the open-source community. Existing vision-language models rely on external tools for the speech processing, while speech-language models still suffer from limited or even without vision-understanding abilities. To address this gap, we propose EMOVA (EMotionally Omni-present Voice Assistant), to enable Large Language Models with end-to-end speech capabilities while maintaining the leading vision-language performance. With a semantic-acoustic disentangled speech tokenizer, we notice surprisingly that omni-modal alignment can further enhance vision-language and speech abilities compared with the corresponding bi-modal aligned counterparts. Moreover, a lightweight style module is proposed for flexible speech style controls (e.g., emotions and pitches). For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks, and meanwhile, supporting omni-modal spoken dialogue with vivid emotions., Comment: Project Page: https://emova-ollm.github.io/

With the rapid advancement of Large Language Models (LLMs), long-context information understanding and processing have become a hot topic in academia and industry. However, benchmarks for evaluating the ability of LLMs to handle long-context information do not seem to have kept pace with the development of LLMs. Despite the emergence of various long-context evaluation benchmarks, the types of capability assessed are still limited, without new capability dimensions. In this paper, we introduce DetectiveQA, a narrative reasoning benchmark featured with an average context length of over 100K tokens. DetectiveQA focuses on evaluating the long-context reasoning ability of LLMs, which not only requires a full understanding of context but also requires extracting important evidences from the context and reasoning according to extracted evidences to answer the given questions. This is a new dimension of capability evaluation, which is more in line with the current intelligence level of LLMs. We use detective novels as data sources, which naturally have various reasoning elements. Finally, we manually annotated 600 questions in Chinese and then also provided an English edition of the context information and questions. We evaluate many long-context LLMs on DetectiveQA, including commercial and open-sourced models, and the results indicate that existing long-context LLMs still require significant advancements to effectively process true long-context dependency questions.

Function calling significantly extends the application boundary of large language models, where high-quality and diverse training data is critical for unlocking this capability. However, real function-calling data is quite challenging to collect and annotate, while synthetic data generated by existing pipelines tends to lack coverage and accuracy. In this paper, we present ToolACE, an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data. ToolACE leverages a novel self-evolution synthesis process to curate a comprehensive API pool of 26,507 diverse APIs. Dialogs are further generated through the interplay among multiple agents, guided by a formalized thinking process. To ensure data accuracy, we implement a dual-layer verification system combining rule-based and model-based checks. We demonstrate that models trained on our synthesized data, even with only 8B parameters, achieve state-of-the-art performance on the Berkeley Function-Calling Leaderboard, rivaling the latest GPT-4 models. Our model and a subset of the data are publicly available at https://huggingface.co/Team-ACE., Comment: 21 pages, 22 figures

The long-context capability of the Large Language Models (LLM) has made significant breakthroughs, but the maximum supported context length remains a critical bottleneck limiting their practical applications. The constraint of context length in LLMs arises from the self-attention mechanism, which cannot effectively and efficiently capture the semantic relationships within infinitely long contexts via the limited pre-trained positional information and attention scope. In this work, we propose \textbf{ReAttention}, a training-free approach enabling LLM based on the self-attention mechanism to support an infinite context with a finite attention scope under sufficient memory resources. ReAttention performs the position-agnostic top-$k$ attention before the ordinary position-aware self-attention, freeing LLMs from the length extrapolation issue. We validate the performance of ReAttention on the LongBench, L-Eval, and InfiniteBench and demonstrate that it is on par with traditional methods. Furthermore, we also apply ReAttention on mainstream LLMs, including LLaMA3.1-8B and Mistral-v0.3-7B, enabling them to support context lengths of at least 1M and even expanding the context length of LLaMA3.2-3B-chat by 128$\times$ to 4M without any further training in Needle-In-A-Haystack tests. We also improve the efficiency of ReAttention with Triton and achieve an efficient extrapolation without additional overhead., Comment: 18 pages, 12 figures

Video Question Answering (VideoQA) has emerged as a challenging frontier in the field of multimedia processing, requiring intricate interactions between visual and textual modalities. Simply uniformly sampling frames or indiscriminately aggregating frame-level visual features often falls short in capturing the nuanced and relevant contexts of videos to well perform VideoQA. To mitigate these issues, we propose VidF4, a novel VideoQA framework equipped with tailored frame selection strategy for effective and efficient VideoQA. We propose three frame-scoring mechanisms that consider both question relevance and inter-frame similarity to evaluate the importance of each frame for a given question on the video. Furthermore, we design a differentiable adaptive frame sampling mechanism to facilitate end-to-end training for the frame selector and answer generator. The experimental results across three widely adopted benchmarks demonstrate that our model consistently outperforms existing VideoQA methods, establishing a new SOTA across NExT-QA (+0.3%), STAR (+0.9%), and TVQA (+1.0%). Furthermore, through both quantitative and qualitative analyses, we validate the effectiveness of each design choice.

Current research found the issue of Early Answering in large language models (LLMs), where the models already have an answer before generating the Chain-of-Thought (CoT). This phenomenon suggests a potential lack of necessary dependency between the predicted answer and the reasoning process. Consequently, two important questions arise: (1) Is CoT still necessary if the model already has an answer? (2) Can the correctness of the answer serve as valid evidence for the correctness of CoT? To address these questions, we propose a method, namely Chain-of-Probe (CoP), to probe changes in the mind during the model's reasoning. The probing results show that in a significant number of question-answer cases, CoT appears to be unnecessary, and this necessity correlates with the simplicity of the task, defined by reasoning steps required. Furthermore, by analyzing patterns in mind change, we examine the correctness of the model's reasoning. Our validation reveals that many responses, although correct in their final answer, contain errors in their reasoning process. To this end, we propose a strategic approach based on CoP to prioritize answers with correct reasoning among multiple candidates, thereby bolstering the reliability of the model's reasoning.

The emerging citation-based QA systems are gaining more attention especially in generative AI search applications. The importance of extracted knowledge provided to these systems is vital from both accuracy (completeness of information) and efficiency (extracting the information in a timely manner). In this regard, citation-based QA systems are suffering from two shortcomings. First, they usually rely only on web as a source of extracted knowledge and adding other external knowledge sources can hamper the efficiency of the system. Second, web-retrieved contents are usually obtained by some simple heuristics such as fixed length or breakpoints which might lead to splitting information into pieces. To mitigate these issues, we propose our enhanced web and efficient knowledge graph (KG) retrieval solution (EWEK-QA) to enrich the content of the extracted knowledge fed to the system. This has been done through designing an adaptive web retriever and incorporating KGs triples in an efficient manner. We demonstrate the effectiveness of EWEK-QA over the open-source state-of-the-art (SoTA) web-based and KG baseline models using a comprehensive set of quantitative and human evaluation experiments. Our model is able to: first, improve the web-retriever baseline in terms of extracting more relevant passages (>20\%), the coverage of answer span (>25\%) and self containment (>35\%); second, obtain and integrate KG triples into its pipeline very efficiently (by avoiding any LLM calls) to outperform the web-only and KG-only SoTA baselines significantly in 7 quantitative QA tasks and our human evaluation.

Large language models (LLMs) have attracted great attention given their strong performance on a wide range of NLP tasks. In practice, users often expect generated texts to fall within a specific length range, making length controlled generation an important topic, especially for GPT-style models. Existing length control methods mostly focus on a simple control type of "equal to" a target length. Different from them, we propose a prompt-based method to achieve length controlled generation under different control types with high accuracy. In particular, we adopt reinforcement learning (RL) and sample filtering with the reward signal given by rule-based reward models, which enhances the length control ability of models by rewarding outputs that follow certain control instructions. In addition, we introduce a standard prompt extractor to parse arbitrary users' input into standard control instructions. Experiments show that our method significantly improves the accuracy of prompt-based length control on popular summarization datasets like CNNDM and NYT under multiple control types. Moreover, both the standard prompt extractor and RL-tuned model show strong generalization to unseen control prompt templates., Comment: Accepted by ACL 2024 findings. arXiv admin note: text overlap with arXiv:2308.12030

In this paper, we study how open-source large language models (LLMs) can be effectively deployed for improving query rewriting in conversational search, especially for ambiguous queries. We introduce CHIQ, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting. This approach contrasts with prior studies that predominantly use closed-source LLMs to directly generate search queries from conversation history. We demonstrate on five well-established benchmarks that CHIQ leads to state-of-the-art results across most settings, showing highly competitive performances with systems leveraging closed-source LLMs. Our study provides a first step towards leveraging open-source LLMs in conversational search, as a competitive alternative to the prevailing reliance on commercial LLMs. Data, models, and source code will be publicly available upon acceptance at https://github.com/fengranMark/CHIQ., Comment: Accepted by EMNLP 2024

Integrating external knowledge into large language models (LLMs) presents a promising solution to overcome the limitations imposed by their antiquated and static parametric memory. Prior studies, however, have tended to over-reliance on external knowledge, underestimating the valuable contributions of an LLMs' intrinsic parametric knowledge. The efficacy of LLMs in blending external and parametric knowledge remains largely unexplored, especially in cases where external knowledge is incomplete and necessitates supplementation by their parametric knowledge. We propose to deconstruct knowledge fusion into four distinct scenarios, offering the first thorough investigation of LLM behavior across each. We develop a systematic pipeline for data construction and knowledge infusion to simulate these fusion scenarios, facilitating a series of controlled experiments. Our investigation reveals that enhancing parametric knowledge within LLMs can significantly bolster their capability for knowledge integration. Nonetheless, we identify persistent challenges in memorizing and eliciting parametric knowledge, and determining parametric knowledge boundaries. Our findings aim to steer future explorations on harmonizing external and parametric knowledge within LLMs., Comment: 15 pages, 3 figures, 3 tables

The rapid evolution of large language models (LLMs) has ushered in the need for comprehensive assessments of their performance across various dimensions. In this paper, we propose LFED, a Literary Fiction Evaluation Dataset, which aims to evaluate the capability of LLMs on the long fiction comprehension and reasoning. We collect 95 literary fictions that are either originally written in Chinese or translated into Chinese, covering a wide range of topics across several centuries. We define a question taxonomy with 8 question categories to guide the creation of 1,304 questions. Additionally, we conduct an in-depth analysis to ascertain how specific attributes of literary fictions (e.g., novel types, character numbers, the year of publication) impact LLM performance in evaluations. Through a series of experiments with various state-of-the-art LLMs, we demonstrate that these models face considerable challenges in effectively addressing questions related to literary fictions, with ChatGPT reaching only 57.08% under the zero-shot setting. The dataset will be publicly available at https://github.com/tjunlp-lab/LFED.git

As the capabilities of large language models (LLMs) have expanded dramatically, aligning these models with human values presents a significant challenge. Traditional alignment strategies rely heavily on human intervention, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), or on the self-alignment capacities of LLMs, which usually require a strong LLM's emergent ability to improve its original bad answer. To address these challenges, we propose a novel self-alignment method that utilizes a Chain of Thought (CoT) approach, termed AlignCoT. This method encompasses stages of Question Analysis, Answer Guidance, and Safe Answer production. It is designed to enable LLMs to generate high-quality, safe responses throughout various stages of their development. Furthermore, we introduce the Mixture of insighTful Experts (MoTE) architecture, which applies mixture of experts to enhance each component of the AlignCoT process, markedly increasing alignment efficiency. The MoTE approach not only outperforms existing methods in aligning LLMs with human values but also highlights the benefits of using self-generated data, revealing the dual benefits of improved alignment and training efficiency.

Event reasoning is a fundamental ability that underlies many applications. It requires event schema knowledge to perform global reasoning and needs to deal with the diversity of the inter-event relations and the reasoning paradigms. How well LLMs accomplish event reasoning on various relations and reasoning paradigms remains unknown. To mitigate this disparity, we comprehensively evaluate the abilities of event reasoning of LLMs. We introduce a novel benchmark EV2 for EValuation of EVent reasoning. EV2 consists of two levels of evaluation of schema and instance and is comprehensive in relations and reasoning paradigms. We conduct extensive experiments on EV2. We find that LLMs have abilities to accomplish event reasoning but their performances are far from satisfactory. We also notice the imbalance of event reasoning abilities in LLMs. Besides, LLMs have event schema knowledge, however, they're not aligned with humans on how to utilize the knowledge. Based on these findings, we guide the LLMs in utilizing the event schema knowledge as memory leading to improvements on event reasoning.

Prior study shows that pre-training techniques can boost the performance of visual document understanding (VDU), which typically requires models to gain abilities to perceive and reason both document texts and layouts (e.g., locations of texts and table-cells). To this end, we propose visually guided generative text-layout pre-training, named ViTLP. Given a document image, the model optimizes hierarchical language and layout modeling objectives to generate the interleaved text and layout sequence. In addition, to address the limitation of processing long documents by Transformers, we introduce a straightforward yet effective multi-segment generative pre-training scheme, facilitating ViTLP to process word-intensive documents of any length. ViTLP can function as a native OCR model to localize and recognize texts of document images. Besides, ViTLP can be effectively applied to various downstream VDU tasks. Extensive experiments show that ViTLP achieves competitive performance over existing baselines on benchmark VDU tasks, including information extraction, document classification, and document question answering., Comment: Accepted to NAACL 2024 main conference. The first version of this paper was submitted to OpenReview (https://openreview.net/forum?id=ARtBIBAmNR) in June 2023

Video-text Large Language Models (video-text LLMs) have shown remarkable performance in answering questions and holding conversations on simple videos. However, they perform almost the same as random on grounding text queries in long and complicated videos, having little ability to understand and reason about temporal information, which is the most fundamental difference between videos and images. In this paper, we propose HawkEye, one of the first video-text LLMs that can perform temporal video grounding in a fully text-to-text manner. To collect training data that is applicable for temporal video grounding, we construct InternVid-G, a large-scale video-text corpus with segment-level captions and negative spans, with which we introduce two new time-aware training objectives to video-text LLMs. We also propose a coarse-grained method of representing segments in videos, which is more robust and easier for LLMs to learn and follow than other alternatives. Extensive experiments show that HawkEye is better at temporal video grounding and comparable on other video-text tasks with existing video-text LLMs, which verifies its superior video-text multi-modal understanding abilities.

In today's fast-paced world, the growing demand to quickly generate comprehensive and accurate Wikipedia documents for emerging events is both crucial and challenging. However, previous efforts in Wikipedia generation have often fallen short of meeting real-world requirements. Some approaches focus solely on generating segments of a complete Wikipedia document, while others overlook the importance of faithfulness in generation or fail to consider the influence of the pre-training corpus. In this paper, we simulate a real-world scenario where structured full-length Wikipedia documents are generated for emergent events using input retrieved from web sources. To ensure that Large Language Models (LLMs) are not trained on corpora related to recently occurred events, we select events that have taken place recently and introduce a new benchmark Wiki-GenBen, which consists of 309 events paired with their corresponding retrieved web pages for generating evidence. Additionally, we design a comprehensive set of systematic evaluation metrics and baseline methods, to evaluate the capability of LLMs in generating factual full-length Wikipedia documents. The data and code are open-sourced at WikiGenBench.

Generative search engines have the potential to transform how people seek information online, but generated responses from existing large language models (LLMs)-backed generative search engines may not always be accurate. Nonetheless, retrieval-augmented generation exacerbates safety concerns, since adversaries may successfully evade the entire system by subtly manipulating the most vulnerable part of a claim. To this end, we propose evaluating the robustness of generative search engines in the realistic and high-risk setting, where adversaries have only black-box system access and seek to deceive the model into returning incorrect responses. Through a comprehensive human evaluation of various generative search engines, such as Bing Chat, PerplexityAI, and YouChat across diverse queries, we demonstrate the effectiveness of adversarial factual questions in inducing incorrect responses. Moreover, retrieval-augmented generation exhibits a higher susceptibility to factual errors compared to LLMs without retrieval. These findings highlight the potential security risks of these systems and emphasize the need for rigorous evaluation before deployment., Comment: 21 pages, 7 figures, 4 tables

E-commerce companies deal with a high volume of customer service requests daily. While a simple annotation system is often used to summarize the topics of customer contacts, thoroughly exploring each specific issue can be challenging. This presents a critical concern, especially during an emerging outbreak where companies must quickly identify and address specific issues. To tackle this challenge, we propose a novel machine learning algorithm that leverages natural language techniques and topological data analysis to monitor emerging and trending customer issues. Our approach involves an end-to-end deep learning framework that simultaneously tags the primary question sentence of each customer's transcript and generates sentence embedding vectors. We then whiten the embedding vectors and use them to construct an undirected graph. From there, we define trending and emerging issues based on the topological properties of each transcript. We have validated our results through various methods and found that they are highly consistent with news sources., Comment: Accepted in KDD 2023 Workshop on Decision Intelligence and Analytics for Online Marketplaces

Building machine learning models can be a time-consuming process that often takes several months to implement in typical business scenarios. To ensure consistent model performance and account for variations in data distribution, regular retraining is necessary. This paper introduces a solution for improving online customer service in e-commerce by presenting a universal model for predict-ing labels based on customer questions, without requiring training. Our novel approach involves using machine learning techniques to tag customer questions in transcripts and create a repository of questions and corresponding labels. When a customer requests assistance, an information retrieval model searches the repository for similar questions, and statistical analysis is used to predict the corresponding label. By eliminating the need for individual model training and maintenance, our approach reduces both the model development cycle and costs. The repository only requires periodic updating to maintain accuracy.

Customer service is often the most time-consuming aspect for e-commerce websites, with each contact typically taking 10-15 minutes. Effectively routing customers to appropriate agents without transfers is therefore crucial for e-commerce success. To this end, we have developed a machine learning framework that predicts the complexity of customer contacts and routes them to appropriate agents accordingly. The framework consists of two parts. First, we train a teacher model to score the complexity of a contact based on the post-contact transcripts. Then, we use the teacher model as a data annotator to provide labels to train a student model that predicts the complexity based on pre-contact data only. Our experiments show that such a framework is successful and can significantly improve customer experience. We also propose a useful metric called complexity AUC that evaluates the effectiveness of customer service at a statistical level., Comment: KDD 2023 Workshop on End-End Customer Journey Optimization

Customers who reach out for customer service support may face a range of issues that vary in complexity. Routing high-complexity contacts to junior agents can lead to multiple transfers or repeated contacts, while directing low-complexity contacts to senior agents can strain their capacity to assist customers who need professional help. To tackle this, a machine learning model that accurately predicts the complexity of customer issues is highly desirable. However, defining the complexity of a contact is a difficult task as it is a highly abstract concept. While consensus-based data annotation by experienced agents is a possible solution, it is time-consuming and costly. To overcome these challenges, we have developed a novel machine learning approach to define contact complexity. Instead of relying on human annotation, we trained an AI expert model to mimic the behavior of agents and evaluate each contact's complexity based on how the AI expert responds. If the AI expert is uncertain or lacks the skills to comprehend the contact transcript, it is considered a high-complexity contact. Our method has proven to be reliable, scalable, and cost-effective based on the collected data., Comment: Accepted in KDD 2023 Workshop on Decision Intelligence and Analytics for Online Marketplaces

Knowledge editing techniques, aiming to efficiently modify a minor proportion of knowledge in large language models (LLMs) without negatively impacting performance across other inputs, have garnered widespread attention. However, existing methods predominantly rely on memorizing the updated knowledge, impeding LLMs from effectively combining the new knowledge with their inherent knowledge when answering questions. To this end, we propose a Learning to Edit (LTE) framework, focusing on teaching LLMs to apply updated knowledge into input questions, inspired by the philosophy of "Teach a man to fish." LTE features a two-phase process: (i) the Alignment Phase, which fine-tunes LLMs on a meticulously curated parallel dataset to make reliable, in-scope edits while preserving out-of-scope information and linguistic proficiency; and (ii) the Inference Phase, which employs a retrieval-based mechanism for real-time and mass knowledge editing. By comparing our approach with seven advanced baselines across four popular knowledge editing benchmarks and two LLM architectures, we demonstrate LTE's superiority in knowledge editing performance, robustness in both batch and sequential editing, minimal interference on general tasks, and rapid editing speeds. The data and code are available at https://github.com/YJiangcm/LTE., Comment: 17 pages, 8 figures, 9 tables. ACL 2024 main camera-ready version

The aim of this workshop is to bring together experts working on open-domain dialogue research. In this speedily advancing research area many challenges still exist, such as learning information from conversations, engaging in realistic and convincing simulation of human intelligence and reasoning. SCI-CHAT follows previous workshops on open domain dialogue but with a focus on the simulation of intelligent conversation as judged in a live human evaluation. Models aim to include the ability to follow a challenging topic over a multi-turn conversation, while positing, refuting and reasoning over arguments. The workshop included both a research track and shared task. The main goal of this paper is to provide an overview of the shared task and a link to an additional paper that will include an in depth analysis of the shared task results following presentation at the workshop.

The recent trend of using Large Language Models (LLMs) as tool agents in real-world applications underscores the necessity for comprehensive evaluations of their capabilities, particularly in complex scenarios involving planning, creating, and using tools. However, existing benchmarks typically focus on simple synthesized queries that do not reflect real-world complexity, thereby offering limited perspectives in evaluating tool utilization. To address this issue, we present UltraTool, a novel benchmark designed to improve and evaluate LLMs' ability in tool utilization within real-world scenarios. UltraTool focuses on the entire process of using tools - from planning and creating to applying them in complex tasks. It emphasizes real-world complexities, demanding accurate, multi-step planning for effective problem-solving. A key feature of UltraTool is its independent evaluation of planning with natural language, which happens before tool usage and simplifies the task solving by mapping out the intermediate steps. Thus, unlike previous work, it eliminates the restriction of pre-defined toolset. Through extensive experiments on various LLMs, we offer novel insights into the evaluation of capabilities of LLMs in tool utilization, thereby contributing a fresh perspective to this rapidly evolving field. The benchmark is publicly available at https://github.com/JoeYing1019/UltraTool., Comment: Accepted by ACL2024 Findings

Large language models (LLMs) are increasingly relied upon for complex multi-turn conversations across diverse real-world applications. However, existing benchmarks predominantly focus on single-turn evaluations, overlooking the models' capabilities in multi-turn interactions. To address this gap, we introduce MT-Eval, a comprehensive benchmark designed to evaluate multi-turn conversational abilities. By analyzing human-LLM conversations, we categorize interaction patterns into four types: recollection, expansion, refinement, and follow-up. We construct multi-turn queries for each category either by augmenting existing datasets or by creating new examples with GPT-4 to avoid data leakage. To study the factors impacting multi-turn abilities, we create single-turn versions of the 1170 multi-turn queries and compare performance. Our evaluation of 11 well-known LLMs shows that while closed-source models generally surpass open-source ones, certain open-source models exceed GPT-3.5-Turbo in specific tasks. We observe significant performance degradation in multi-turn settings compared to single-turn settings in most models, which is not correlated with the models' fundamental capabilities. Moreover, we identify the distance to relevant content and susceptibility to error propagation as the key factors influencing multi-turn performance. MT-Eval is released publicly to encourage future research towards more robust conversational models., Comment: Code and data are available at https://github.com/KwanWaiChung/MT-Eval

Although large language models (LLMs) have demonstrated adeptness in a range of tasks, they still lag behind human learning efficiency. This disparity is often linked to the inherent human capacity to learn from basic examples, gradually generalize and handle more complex problems, and refine their skills with continuous feedback. Inspired by this, this paper introduces YODA, a novel teacher-student progressive learning framework that emulates the teacher-student education process to improve the efficacy of model fine-tuning. The framework operates on an interactive \textit{basic-generalized-harder} loop. The teacher agent provides tailored feedback on the student's answers, and systematically organizes the education process. This process unfolds by teaching the student basic examples, reinforcing understanding through generalized questions, and then enhancing learning by posing questions with progressively enhanced complexity. With the teacher's guidance, the student learns to iteratively refine its answer with feedback, and forms a robust and comprehensive understanding of the posed questions. The systematic procedural data, which reflects the progressive learning process of humans, is then utilized for model training. Taking math reasoning as a testbed, experiments show that training LLaMA2 with data from YODA improves SFT with significant performance gain (+17.01\% on GSM8K and +9.98\% on MATH). In addition, we find that training with curriculum learning further improves learning robustness., Comment: 14 pages, 4 figures, 3 tables

Large Language Models (LLMs) have succeeded remarkably in understanding long-form contents. However, exploring their capability for generating long-form contents, such as reports and articles, has been relatively unexplored and inadequately assessed by existing benchmarks. The prevalent evaluation methods, which predominantly rely on crowdsourcing, are recognized for their labor-intensive nature and lack of efficiency, whereas automated metrics, such as the ROUGE score, demonstrate discordance with human judgment criteria. In this paper, we propose ProxyQA, an innovative framework dedicated to assessing long-text generation. ProxyQA comprises in-depth human-curated meta-questions spanning various domains, each accompanied by specific proxy-questions with pre-annotated answers. LLMs are tasked to generate extensive content in response to these meta-questions, by engaging an evaluator and incorporating the generated texts as contextual background, ProxyQA assesses the generated content's quality through the evaluator's accuracy in addressing the proxy-questions. We examine multiple LLMs, emphasizing ProxyQA's demanding nature as a high-quality assessment tool. Human evaluation demonstrates that the proxy-question method is notably self-consistent and aligns closely with human evaluative standards. The dataset and leaderboard is available at \url{https://proxy-qa.com}., Comment: Accepted to ACL 2024 main conference

The rapid progress of Transformers in artificial intelligence has come at the cost of increased resource consumption and greenhouse gas emissions due to growing model sizes. Prior work suggests using pretrained small models to improve training efficiency, but this approach may not be suitable for new model structures. On the other hand, training from scratch can be slow, and progressively stacking layers often fails to achieve significant acceleration. To address these challenges, we propose a novel method called Apollo, which prep\textbf{a}res lessons for ex\textbf{p}anding \textbf{o}perations by \textbf{l}earning high-\textbf{l}ayer functi\textbf{o}nality during training of low layers. Our approach involves low-value-prioritized sampling (LVPS) to train different depths and weight sharing to facilitate efficient expansion. We also introduce an interpolation method for stable model depth extension. Experiments demonstrate that Apollo achieves state-of-the-art acceleration ratios, even rivaling methods using pretrained models, making it a universal and efficient solution for training deep models while reducing time, financial, and environmental costs.

The recent trend of large language models (LLMs) is to increase the scale of both model size (\aka the number of parameters) and dataset to achieve better generative ability, which is definitely proved by a lot of work such as the famous GPT and Llama. However, large models often involve massive computational costs, and practical applications cannot afford such high prices. However, the method of constructing a strong model architecture for LLMs is rarely discussed. We first analyze the state-of-the-art language model architectures and observe the feature collapse problem. Based on the theoretical analysis, we propose that the nonlinearity is also very important for language models, which is usually studied in convolutional neural networks for vision tasks. The series informed activation function is then introduced with tiny calculations that can be ignored, and an augmented shortcut is further used to enhance the model nonlinearity. We then demonstrate that the proposed approach is significantly effective for enhancing the model nonlinearity through carefully designed ablations; thus, we present a new efficient model architecture for establishing modern, namely, PanGu-$\pi$. Experiments are then conducted using the same dataset and training strategy to compare PanGu-$\pi$ with state-of-the-art LLMs. The results show that PanGu-$\pi$-7B can achieve a comparable performance to that of benchmarks with about 10\% inference speed-up, and PanGu-$\pi$-1B can achieve state-of-the-art performance in terms of accuracy and efficiency. In addition, we have deployed PanGu-$\pi$-7B in the high-value domains of finance and law, developing an LLM named YunShan for practical application. The results show that YunShan can surpass other models with similar scales on benchmarks.

Retrieval-augmented generation (RAG) grounds large language model (LLM) output by leveraging external knowledge sources to reduce factual hallucinations. However, prior works lack a comprehensive evaluation of different language families, making it challenging to evaluate LLM robustness against errors in external retrieved knowledge. To overcome this, we establish NoMIRACL, a human-annotated dataset for evaluating LLM robustness in RAG across 18 typologically diverse languages. NoMIRACL includes both a non-relevant and a relevant subset. Queries in the non-relevant subset contain passages judged as non-relevant, whereas queries in the relevant subset include at least a single judged relevant passage. We measure LLM robustness using two metrics: (i) hallucination rate, measuring model tendency to hallucinate an answer, when the answer is not present in passages in the non-relevant subset, and (ii) error rate, measuring model inaccuracy to recognize relevant passages in the relevant subset. In our work, we measure robustness for a wide variety of multilingual-focused LLMs and observe that most of the models struggle to balance the two capacities. Models such as LLAMA-2, Orca-2, and FLAN-T5 observe more than an 88% hallucination rate on the non-relevant subset, whereas, Mistral overall hallucinates less, but can achieve up to a 74.9% error rate on the relevant subset. Overall, GPT-4 is observed to provide the best tradeoff on both subsets, highlighting future work necessary to improve LLM robustness.

Reasoning, a crucial ability for complex problem-solving, plays a pivotal role in various real-world settings such as negotiation, medical diagnosis, and criminal investigation. It serves as a fundamental methodology in the field of Artificial General Intelligence (AGI). With the ongoing development of foundation models, e.g., Large Language Models (LLMs), there is a growing interest in exploring their abilities in reasoning tasks. In this paper, we introduce seminal foundation models proposed or adaptable for reasoning, highlighting the latest advancements in various reasoning tasks, methods, and benchmarks. We then delve into the potential future directions behind the emergence of reasoning abilities within foundation models. We also discuss the relevance of multimodal learning, autonomous agents, and super alignment in the context of reasoning. By discussing these future research directions, we hope to inspire researchers in their exploration of this field, stimulate further advancements in reasoning with foundation models, and contribute to the development of AGI., Comment: 20 Figures, 160 Pages, 750+ References, Project Page https://github.com/reasoning-survey/Awesome-Reasoning-Foundation-Models

Unsupervised extractive summarization is an important technique in information extraction and retrieval. Compared with supervised method, it does not require high-quality human-labelled summaries for training and thus can be easily applied for documents with different types, domains or languages. Most of existing unsupervised methods including TextRank and PACSUM rely on graph-based ranking on sentence centrality. However, this scorer can not be directly applied in end-to-end training, and the positional-related prior assumption is often needed for achieving good summaries. In addition, less attention is paid to length-controllable extractor, where users can decide to summarize texts under particular length constraint. This paper introduces an unsupervised extractive summarization model based on a siamese network, for which we develop a trainable bidirectional prediction objective between the selected summary and the original document. Different from the centrality-based ranking methods, our extractive scorer can be trained in an end-to-end manner, with no other requirement of positional assumption. In addition, we introduce a differentiable length control module by approximating 0-1 knapsack solver for end-to-end length-controllable extracting. Experiments show that our unsupervised method largely outperforms the centrality-based baseline using a same sentence encoder. In terms of length control ability, via our trainable knapsack module, the performance consistently outperforms the strong baseline without utilizing end-to-end training. Human evaluation further evidences that our method performs the best among baselines in terms of relevance and consistency., Comment: accepted by AAAI2024

Data plays a fundamental role in training Large Language Models (LLMs). Efficient data management, particularly in formulating a well-suited training dataset, is significant for enhancing model performance and improving training efficiency during pretraining and supervised fine-tuning stages. Despite the considerable importance of data management, the underlying mechanism of current prominent practices are still unknown. Consequently, the exploration of data management has attracted more and more attention among the research community. This survey aims to provide a comprehensive overview of current research in data management within both the pretraining and supervised fine-tuning stages of LLMs, covering various aspects of data management strategy design. Looking into the future, we extrapolate existing challenges and outline promising directions for development in this field. Therefore, this survey serves as a guiding resource for practitioners aspiring to construct powerful LLMs through efficient data management practices. The collection of the latest papers is available at https://github.com/ZigeW/data_management_LLM., Comment: Work in progress

The ability to follow instructions is crucial for Large Language Models (LLMs) to handle various real-world applications. Existing benchmarks primarily focus on evaluating pure response quality, rather than assessing whether the response follows constraints stated in the instruction. To fill this research gap, in this paper, we propose FollowBench, a Multi-level Fine-grained Constraints Following Benchmark for LLMs. FollowBench comprehensively includes five different types (i.e., Content, Situation, Style, Format, and Example) of fine-grained constraints. To enable a precise constraint following estimation on diverse difficulties, we introduce a Multi-level mechanism that incrementally adds a single constraint to the initial instruction at each increased level. To assess whether LLMs' outputs have satisfied every individual constraint, we propose to prompt strong LLMs with constraint-evolution paths to handle challenging open-ended instructions. By evaluating 13 closed-source and open-source popular LLMs on FollowBench, we highlight the weaknesses of LLMs in instruction following and point towards potential avenues for future work. The data and code are publicly available at https://github.com/YJiangcm/FollowBench., Comment: 22 pages, 11 figures, 16 tables. ACL 2024 main camera-ready version

Managing long sequences has become an important and necessary feature for large language models (LLMs). However, it is still an open question of how to comprehensively and systematically evaluate the long-sequence capability of LLMs. One of the reasons is that conventional and widely-used benchmarks mainly consist of short sequences. In this paper, we propose M4LE, a Multi-ability, Multi-range, Multi-task, Multi-domain benchmark for Long-context Evaluation. M4LE is based on a diverse NLP task pool comprising 36 NLP datasets, 11 task types and 12 domains. To alleviate the scarcity of tasks with naturally long sequences and incorporate multiple-ability assessment, we propose an automatic approach (but with negligible human annotations) to convert short-sequence tasks into a unified long-sequence scenario where LLMs have to identify single or multiple relevant spans in long contexts based on explicit or semantic hints. Specifically, the scenario includes five different types of abilities: (1) explicit single-span; (2) semantic single-span; (3) explicit multiple-span; (4) semantic multiple-span; and (5) global context understanding. The resulting samples in M4LE are evenly distributed from 1k to 8k input length. We conducted a systematic evaluation on 11 well-established LLMs, especially those optimized for long-sequence inputs. Our results reveal that: 1) Current LLMs struggle to understand long context, particularly when tasks require multiple-span attention. 2) Semantic retrieval task is more difficult for competent LLMs. 3) Models fine-tuned on longer text with position interpolation have comparable performance to those using Neural Tangent Kernel (NTK) aware scaling methods without fine-tuning. We make our benchmark publicly available to encourage future research in this challenging area., Comment: Code and data are available at https://github.com/KwanWaiChung/M4LE

Training large models from scratch usually costs a substantial amount of resources. Towards this problem, recent studies such as bert2BERT and LiGO have reused small pretrained models to initialize a large model (termed the ``target model''), leading to a considerable acceleration in training. Despite the successes of these previous studies, they grew pretrained models by mapping partial weights only, ignoring potential correlations across the entire model. As we show in this paper, there are inter- and intra-interactions among the weights of both the pretrained and the target models. As a result, the partial mapping may not capture the complete information and lead to inadequate growth. In this paper, we propose a method that linearly correlates each weight of the target model to all the weights of the pretrained model to further enhance acceleration ability. We utilize multi-linear operators to reduce computational and spacial complexity, enabling acceptable resource requirements. Experiments demonstrate that our method can save 76\% computational costs on DeiT-base transferred from DeiT-small, which outperforms bert2BERT by +12.0\% and LiGO by +20.7\%, respectively., Comment: Accepted in NeurIPS 2023

The rapid development of large language models (LLMs) has not only provided numerous opportunities but also presented significant challenges. This becomes particularly evident when LLMs inadvertently generate harmful or toxic content, either unintentionally or because of intentional inducement. Existing alignment methods usually direct LLMs toward the favorable outcomes by utilizing human-annotated, flawless instruction-response pairs. Conversely, this study proposes a novel alignment technique based on mistake analysis, which deliberately exposes LLMs to erroneous content to learn the reasons for mistakes and how to avoid them. In this case, mistakes are repurposed into valuable data for alignment, effectively helping to avoid the production of erroneous responses. Without external models or human annotations, our method leverages a model's intrinsic ability to discern undesirable mistakes and improves the safety of its generated responses. Experimental results reveal that our method outperforms existing alignment approaches in enhancing model safety while maintaining the overall utility., Comment: Accepted by ICLR 2024

Automated theorem proving (ATP) has become an appealing domain for exploring the reasoning ability of the recent successful generative language models. However, current ATP benchmarks mainly focus on symbolic inference, but rarely involve the understanding of complex number combination reasoning. In this work, we propose TRIGO, an ATP benchmark that not only requires a model to reduce a trigonometric expression with step-by-step proofs but also evaluates a generative LM's reasoning ability on formulas and its capability to manipulate, group, and factor number terms. We gather trigonometric expressions and their reduced forms from the web, annotate the simplification process manually, and translate it into the Lean formal language system. We then automatically generate additional examples from the annotated samples to expand the dataset. Furthermore, we develop an automatic generator based on Lean-Gym to create dataset splits of varying difficulties and distributions in order to thoroughly analyze the model's generalization ability. Our extensive experiments show our proposed TRIGO poses a new challenge for advanced generative LM's including GPT-4 which is pre-trained on a considerable amount of open-source formal theorem-proving language data, and provide a new tool to study the generative LM's ability on both formal and mathematical reasoning., Comment: Accepted by EMNLP 2023. Code is available at https://github.com/menik1126/TRIGO

Pretrained language models (PLMs) based knowledge-grounded dialogue systems are prone to generate responses that are factually inconsistent with the provided knowledge source. In such inconsistent responses, the dialogue models fail to accurately express the external knowledge they rely upon. Inspired by previous work which identified that feed-forward networks (FFNs) within Transformers are responsible for factual knowledge expressions, we investigate two methods to efficiently improve the factual expression capability {of FFNs} by knowledge enhancement and alignment respectively. We first propose \textsc{K-Dial}, which {explicitly} introduces {extended FFNs in Transformers to enhance factual knowledge expressions} given the specific patterns of knowledge-grounded dialogue inputs. Additionally, we apply the reinforcement learning for factual consistency (RLFC) method to implicitly adjust FFNs' expressions in responses by aligning with gold knowledge for the factual consistency preference. To comprehensively assess the factual consistency and dialogue quality of responses, we employ extensive automatic measures and human evaluations including sophisticated fine-grained NLI-based metrics. Experimental results on WoW and CMU\_DoG datasets demonstrate that our methods efficiently enhance the ability of the FFN module to convey factual knowledge, validating the efficacy of improving factual consistency for knowledge-grounded dialogue systems., Comment: EMNLP2023 Findings

Large Language Models (LLMs) have demonstrated remarkable versatility across various domains. To further advance LLMs, we propose 'SELF' (Self-Evolution with Language Feedback), a novel approach that enables LLMs to self-improve through self-reflection, akin to human learning processes. SELF initiates with a meta-skill learning process that equips the LLMs with capabilities for self-feedback and self-refinement. Subsequently, the model undergoes an iterative process of self-evolution. In each iteration, it utilizes an unlabeled dataset of instructions to generate initial responses. These responses are enhanced through self-feedback and self-refinement. The model is then fine-tuned using this enhanced data. The model undergoes progressive improvement through this iterative self-evolution process. Moreover, the SELF framework enables the model to apply self-refinement during inference, which further improves response quality. Our experiments in mathematics and general tasks demonstrate that SELF can enhance the capabilities of LLMs without human intervention. The SELF framework indicates a promising direction for the autonomous evolution of LLMs, transitioning them from passive information receivers to active participants in their development., Comment: 20 pages, 4 figures, 11 tables

We present FIMO, an innovative dataset comprising formal mathematical problem statements sourced from the International Mathematical Olympiad (IMO) Shortlisted Problems. Designed to facilitate advanced automated theorem proving at the IMO level, FIMO is currently tailored for the Lean formal language. It comprises 149 formal problem statements, accompanied by both informal problem descriptions and their corresponding LaTeX-based informal proofs. Through initial experiments involving GPT-4, our findings underscore the existing limitations in current methodologies, indicating a substantial journey ahead before achieving satisfactory IMO-level automated theorem proving outcomes., Comment: Added a hyperlink to the dataset made accessible on GitHub

Large language models (LLMs) like ChatGPT and GPT-4 have attracted great attention given their surprising performance on a wide range of NLP tasks. Length controlled generation of LLMs emerges as an important topic, which enables users to fully leverage the capability of LLMs in more real-world scenarios like generating a proper answer or essay of a desired length. In addition, the autoregressive generation in LLMs is extremely time-consuming, while the ability of controlling this generated length can reduce the inference cost by limiting the length. Therefore, we propose a prompt-based length control method to achieve high-accuracy length controlled generation. In particular, we adopt reinforcement learning with the reward signal given by either trainable or rule-based reward models, which further enhances the length-control ability of LLMs by rewarding outputs that follows pre-defined control instruction. To enable rule-based inference, we also introduce standard prompt extractor to collect the standard control information from users' input. Experiments show that our method significantly improves the accuracy of prompt-based length control for summarization task on popular datasets like CNNDM and NYT. Both the standard prompt extractor and the RL-tuned model have show strong generalization ability to unseen control prompt templates.

Information-seeking conversation, which aims to help users gather information through conversation, has achieved great progress in recent years. However, the research is still stymied by the scarcity of training data. To alleviate this problem, we propose AutoConv for synthetic conversation generation, which takes advantage of the few-shot learning ability and generation capacity of large language models (LLM). Specifically, we formulate the conversation generation problem as a language modeling task, then finetune an LLM with a few human conversations to capture the characteristics of the information-seeking process and use it for generating synthetic conversations with high quality. Experimental results on two frequently-used datasets verify that AutoConv has substantial improvements over strong baselines and alleviates the dependence on human annotation. In addition, we also provide several analysis studies to promote future research., Comment: Accepted to ACL 2023 Main Conference (Short)

Hot news is one of the most popular topics in daily conversations. However, news grounded conversation has long been stymied by the lack of well-designed task definition and scarce data. In this paper, we propose a novel task, Proactive News Grounded Conversation, in which a dialogue system can proactively lead the conversation based on some key topics of the news. In addition, both information-seeking and chit-chat scenarios are included realistically, where the user may ask a series of questions about the news details or express their opinions and be eager to chat. To further develop this novel task, we collect a human-to-human Chinese dialogue dataset \ts{NewsDialogues}, which includes 1K conversations with a total of 14.6K utterances and detailed annotations for target topics and knowledge spans. Furthermore, we propose a method named Predict-Generate-Rank, consisting of a generator for grounded knowledge prediction and response generation, and a ranker for the ranking of multiple responses to alleviate the exposure bias. We conduct comprehensive experiments to demonstrate the effectiveness of the proposed method and further present several key findings and challenges to prompt future research., Comment: Accepted to ACL 2023 Conference (Long Paper; Findings)

Large Language Models (LLMs) trained on extensive textual corpora have emerged as leading solutions for a broad array of Natural Language Processing (NLP) tasks. Despite their notable performance, these models are prone to certain limitations such as misunderstanding human instructions, generating potentially biased content, or factually incorrect (hallucinated) information. Hence, aligning LLMs with human expectations has become an active area of interest within the research community. This survey presents a comprehensive overview of these alignment technologies, including the following aspects. (1) Data collection: the methods for effectively collecting high-quality instructions for LLM alignment, including the use of NLP benchmarks, human annotations, and leveraging strong LLMs. (2) Training methodologies: a detailed review of the prevailing training methods employed for LLM alignment. Our exploration encompasses Supervised Fine-tuning, both Online and Offline human preference training, along with parameter-efficient training mechanisms. (3) Model Evaluation: the methods for evaluating the effectiveness of these human-aligned LLMs, presenting a multifaceted approach towards their assessment. In conclusion, we collate and distill our findings, shedding light on several promising future research avenues in the field. This survey, therefore, serves as a valuable resource for anyone invested in understanding and advancing the alignment of LLMs to better suit human-oriented tasks and expectations. An associated GitHub link collecting the latest papers is available at https://github.com/GaryYufei/AlignLLMHumanSurvey., Comment: work in progress

Existing evaluation metrics for natural language generation (NLG) tasks face the challenges on generalization ability and interpretability. Specifically, most of the well-performed metrics are required to train on evaluation datasets of specific NLG tasks and evaluation dimensions, which may cause over-fitting to task-specific datasets. Furthermore, existing metrics only provide an evaluation score for each dimension without revealing the evidence to interpret how this score is obtained. To deal with these challenges, we propose a simple yet effective metric called DecompEval. This metric formulates NLG evaluation as an instruction-style question answering task and utilizes instruction-tuned pre-trained language models (PLMs) without training on evaluation datasets, aiming to enhance the generalization ability. To make the evaluation process more interpretable, we decompose our devised instruction-style question about the quality of generated texts into the subquestions that measure the quality of each sentence. The subquestions with their answers generated by PLMs are then recomposed as evidence to obtain the evaluation result. Experimental results show that DecompEval achieves state-of-the-art performance in untrained metrics for evaluating text summarization and dialogue generation, which also exhibits strong dimension-level / task-level generalization ability and interpretability., Comment: Accepted by ACL 2023 (Main Conference)

In classification problems, the datasets are usually imbalanced, noisy or complex. Most sampling algorithms only make some improvements to the linear sampling mechanism of the synthetic minority oversampling technique (SMOTE). Nevertheless, linear oversampling has several unavoidable drawbacks. Linear oversampling is susceptible to overfitting, and the synthetic samples lack diversity and rarely account for the original distribution characteristics. An informed nonlinear oversampling framework with the granular ball (INGB) as a new direction of oversampling is proposed in this paper. It uses granular balls to simulate the spatial distribution characteristics of datasets, and informed entropy is utilized to further optimize the granular-ball space. Then, nonlinear oversampling is performed by following high-dimensional sparsity and the isotropic Gaussian distribution. Furthermore, INGB has good compatibility. Not only can it be combined with most SMOTE-based sampling algorithms to improve their performance, but it can also be easily extended to noisy imbalanced multi-classification problems. The mathematical model and theoretical proof of INGB are given in this work. Extensive experiments demonstrate that INGB outperforms the traditional linear sampling frameworks and algorithms in oversampling on complex datasets., Comment: 15 pages, 6 figures

This study proposes a multitask learning architecture for extractive summarization with coherence boosting. The architecture contains an extractive summarizer and coherent discriminator module. The coherent discriminator is trained online on the sentence vectors of the augmented textual input, thus improving its general ability of judging whether the input sentences are coherent. Meanwhile, we maximize the coherent scores from the coherent discriminator by updating the parameters of the summarizer. To make the extractive sentences trainable in a differentiable manner, we introduce two strategies, including pre-trained converting model (model-based) and converting matrix (MAT-based) that merge sentence representations. Experiments show that our proposed method significantly improves the proportion of consecutive sentences in the extracted summaries based on their positions in the original article (i.e., automatic sentence-level coherence metric), while the goodness in terms of other automatic metrics (i.e., Rouge scores and BertScores) are preserved. Human evaluation also evidences the improvement of coherence and consistency of the extracted summaries given by our method., Comment: 11 pages, 4 figures