Authors: Samuel Pfrommer, Yatong Bai, Tanmay Gautam, Somayeh Sojoudi
Abstract: Major search engine providers are rapidly incorporating Large Language Model (LLM)-generated content in response to user queries. These conversational search engines operate by loading retrieved website text into the LLM context for summarization and interpretation. Recent research demonstrates that LLMs are highly vulnerable to jailbreaking and prompt injection attacks, which disrupt the safety and quality goals of LLMs using adversarial strings. This work investigates the impact of prompt injections on the ranking order of sources referenced by conversational search engines. To this end, we introduce a focused dataset of real-world consumer product websites and formalize conversational search ranking as an adversarial problem. Experimentally, we analyze conversational search rankings in the absence of adversarial injections and show that different LLMs vary significantly in prioritizing product name, document content, and context position. We then present a tree-of-attacks-based jailbreaking technique which reliably promotes low-ranked products. Importantly, these attacks transfer effectively to state-of-the-art conversational search engines such as perplexity.ai. Given the strong financial incentive for website owners to boost their search ranking, we argue that our problem formulation is of critical importance for future robustness work.
Authors: Matteo Gabburo, Nicolaas Paul Jedema, Siddhant Garg, Leonardo F. R. Ribeiro, Alessandro Moschitti
Abstract: In this paper, we investigate which questions are challenging for retrieval-based Question Answering (QA). We (i) propose retrieval complexity (RC), a novel metric conditioned on the completeness of retrieved documents, which measures the difficulty of answering questions, and (ii) propose an unsupervised pipeline to measure RC given an arbitrary retrieval system. Our proposed pipeline measures RC more accurately than alternative estimators, including LLMs, on six challenging QA benchmarks. Further investigation reveals that RC scores strongly correlate with both QA performance and expert judgment across five of the six studied benchmarks, indicating that RC is an effective measure of question difficulty. Subsequent categorization of high-RC questions shows that they span a broad set of question shapes, including multi-hop, compositional, and temporal QA, indicating that RC scores can categorize a new subset of complex questions. Our system can also have a major impact on retrieval-based systems by helping to identify more challenging questions on existing datasets.
Authors: Yang Wu, Chenghao Wang, Ece Gumusel, Xiaozhong Liu
Abstract: The integration of generative Large Language Models (LLMs) into various applications, including the legal domain, has been accelerated by their expansive and versatile nature. However, when facing a legal case, users without a legal background often struggle to formulate professional queries and may inadvertently overlook critical legal factors when presenting their case narrative to LLMs. To address this issue, we propose the Diagnostic Legal Large Language Model (D3LM), which utilizes adaptive lawyer-like diagnostic questions to collect additional case information and then provides high-quality feedback. D3LM incorporates an innovative graph-based Positive-Unlabeled Reinforcement Learning (PURL) algorithm, enabling the generation of critical questions and enhancing user-LLM interactions. Moreover, an integrated LLM-based stopping criterion facilitates precise Court Views Generation (CVG). Our research also introduces a new English-language CVG dataset based on the US case law database, enriching the realm of LLM research and deployment with a vital dimension. D3LM surpasses classical LLMs by delivering outstanding performance and a remarkable user experience in the legal domain.
Authors: Avi Caciularu, Alon Jacovi, Eyal Ben-David, Sasha Goldshtein, Tal Schuster, Jonathan Herzig, Gal Elidan, Amir Globerson
Abstract: Large Language Models (LLMs) often do not perform well on queries that require the aggregation of information across texts. To better evaluate this setting and facilitate modeling efforts, we introduce TACT - Text And Calculations through Tables, a dataset crafted to evaluate LLMs' reasoning and computational abilities using complex instructions. TACT contains challenging instructions that demand stitching information scattered across one or more texts, and performing complex integration on this information to generate the answer. We construct this dataset by leveraging an existing dataset of texts and their associated tables. For each such tables, we formulate new queries, and gather their respective answers. We demonstrate that all contemporary LLMs perform poorly on this dataset, achieving an accuracy below 38\%. To pinpoint the difficulties and thoroughly dissect the problem, we analyze model performance across three components: table-generation, Pandas command-generation, and execution. Unexpectedly, we discover that each component presents substantial challenges for current LLMs. These insights lead us to propose a focused modeling framework, which we refer to as IE as a tool. Specifically, we propose to add "tools" for each of the above steps, and implement each such tool with few-shot prompting. This approach shows an improvement over existing prompting techniques, offering a promising direction for enhancing model capabilities in these tasks.
Authors: Dyah Adila, Changho Shin, Yijing Zhang, Frederic Sala
Abstract: Aligning pretrained language models (LMs) is a complex and resource-intensive process, often requiring access to large amounts of ground-truth preference data and substantial compute. Are these costs necessary? That is, it is possible to align using only inherent model knowledge and without additional training? We tackle this challenge with AlignEZ, a novel approach that uses (1) self-generated preference data and (2) representation editing to provide nearly cost-free alignment. During inference, AlignEZ modifies LM representations to reduce undesirable and boost desirable components using subspaces identified via self-generated preference pairs. Our experiments reveal that this nearly cost-free procedure significantly narrows the gap between base pretrained and tuned models by an average of 31.6%, observed across six datasets and three model architectures. Additionally, we explore the potential of using AlignEZ as a means of expediting more expensive alignment procedures. Our experiments show that AlignEZ improves DPO models tuned only using a small subset of ground-truth preference data. Lastly, we study the conditions under which improvement using AlignEZ is feasible, providing valuable insights into its effectiveness.
Authors: Daiki Asami, Saku Sugawara
Abstract: Psycholinguistic research suggests that humans may build a representation of linguistic input that is 'good-enough' for the task at hand. This study examines what architectural features make language models learn human-like good-enough language processing. We focus on the number of layers and self-attention heads in Transformers. We create a good-enough language processing (GELP) evaluation dataset (7,680 examples), which is designed to test the effects of two plausibility types, eight construction types, and three degrees of memory cost on language processing. To annotate GELP, we first conduct a crowdsourcing experiment whose design follows prior psycholinguistic studies. Our model evaluation against the annotated GELP then reveals that the full model as well as models with fewer layers and/or self-attention heads exhibit a good-enough performance. This result suggests that models with shallower depth and fewer heads can learn good-enough language processing.
Authors: Jingxuan Tu, Keer Xu, Liulu Yue, Bingyang Ye, Kyeongmin Rim, James Pustejovsky
Abstract: Semantic textual similarity (STS) is a fundamental NLP task that measures the semantic similarity between a pair of sentences. In order to reduce the inherent ambiguity posed from the sentences, a recent work called Conditional STS (C-STS) has been proposed to measure the sentences' similarity conditioned on a certain aspect. Despite the popularity of C-STS, we find that the current C-STS dataset suffers from various issues that could impede proper evaluation on this task. In this paper, we reannotate the C-STS validation set and observe an annotator discrepancy on 55% of the instances resulting from the annotation errors in the original label, ill-defined conditions, and the lack of clarity in the task definition. After a thorough dataset analysis, we improve the C-STS task by leveraging the models' capability to understand the conditions under a QA task setting. With the generated answers, we present an automatic error identification pipeline that is able to identify annotation errors from the C-STS data with over 80% F1 score. We also propose a new method that largely improves the performance over baselines on the C-STS data by training the models with the answers. Finally we discuss the conditionality annotation based on the typed-feature structure (TFS) of entity types. We show in examples that the TFS is able to provide a linguistic foundation for constructing C-STS data with new conditions.
Authors: Keyon Vafa, Justin Y. Chen, Jon Kleinberg, Sendhil Mullainathan, Ashesh Rambachan
Abstract: Recent work suggests that large language models may implicitly learn world models. How should we assess this possibility? We formalize this question for the case where the underlying reality is governed by a deterministic finite automaton. This includes problems as diverse as simple logical reasoning, geographic navigation, game-playing, and chemistry. We propose new evaluation metrics for world model recovery inspired by the classic Myhill-Nerode theorem from language theory. We illustrate their utility in three domains: game playing, logic puzzles, and navigation. In all domains, the generative models we consider do well on existing diagnostics for assessing world models, but our evaluation metrics reveal their world models to be far less coherent than they appear. Such incoherence creates fragility: using a generative model to solve related but subtly different tasks can lead it to fail badly. Building generative models that meaningfully capture the underlying logic of the domains they model would be immensely valuable; our results suggest new ways to assess how close a given model is to that goal.
Authors: Anand Subramanian, Viktor Schlegel, Abhinav Ramesh Kashyap, Thanh-Tung Nguyen, Vijay Prakash Dwivedi, Stefan Winkler
Abstract: There is vivid research on adapting Large Language Models (LLMs) to perform a variety of tasks in high-stakes domains such as healthcare. Despite their popularity, there is a lack of understanding of the extent and contributing factors that allow LLMs to recall relevant knowledge and combine it with presented information in the clinical and biomedical domain: a fundamental pre-requisite for success on down-stream tasks. Addressing this gap, we use Multiple Choice and Abstractive Question Answering to conduct a large-scale empirical study on 22 datasets in three generalist and three specialist biomedical sub-domains. Our multifaceted analysis of the performance of 15 LLMs, further broken down by sub-domain, source of knowledge and model architecture, uncovers success factors such as instruction tuning that lead to improved recall and comprehension. We further show that while recently proposed domain-adapted models may lack adequate knowledge, directly fine-tuning on our collected medical knowledge datasets shows encouraging results, even generalising to unseen specialist sub-domains. We complement the quantitative results with a skill-oriented manual error analysis, which reveals a significant gap between the models' capabilities to simply recall necessary knowledge and to integrate it with the presented context. To foster research and collaboration in this field we share M-QALM, our resources, standardised methodology, and evaluation results, with the research community to facilitate further advancements in clinical knowledge representation learning within language models.
Authors: Fanyou Wu, Weijie Xu, Chandan K. Reddy, Srinivasan H. Sengamedu
Abstract: In this study, we tackle the challenge of inadequate and costly training data that has hindered the development of conversational question answering (ConvQA) systems. Enterprises have a large corpus of diverse internal documents. Instead of relying on a searching engine, a more compelling approach for people to comprehend these documents is to create a dialogue system. In this paper, we propose a robust dialog synthesising method. We learn the segmentation of data for the dialog task instead of using segmenting at sentence boundaries. The synthetic dataset generated by our proposed method achieves superior quality when compared to WikiDialog, as assessed through machine and human evaluations. By employing our inpainted data for ConvQA retrieval system pre-training, we observed a notable improvement in performance across OR-QuAC benchmarks.
Authors: Lei Liu, Xiaoyan Yang, Junchi Lei, Xiaoyang Liu, Yue Shen, Zhiqiang Zhang, Peng Wei, Jinjie Gu, Zhixuan Chu, Zhan Qin, Kui Ren
Abstract: Large language models (LLMs), such as GPT series models, have received substantial attention due to their impressive capabilities for generating and understanding human-level language. More recently, LLMs have emerged as an innovative and powerful adjunct in the medical field, transforming traditional practices and heralding a new era of enhanced healthcare services. This survey provides a comprehensive overview of Medical Large Language Models (Med-LLMs), outlining their evolution from general to the medical-specific domain (i.e, Technology and Application), as well as their transformative impact on healthcare (e.g., Trustworthiness and Safety). Concretely, starting from the fundamental history and technology of LLMs, we first delve into the progressive adaptation and refinements of general LLM models in the medical domain, especially emphasizing the advanced algorithms that boost the LLMs' performance in handling complicated medical environments, including clinical reasoning, knowledge graph, retrieval-augmented generation, human alignment, and multi-modal learning. Secondly, we explore the extensive applications of Med-LLMs across domains such as clinical decision support, report generation, and medical education, illustrating their potential to streamline healthcare services and augment patient outcomes. Finally, recognizing the imperative and responsible innovation, we discuss the challenges of ensuring fairness, accountability, privacy, and robustness in Med-LLMs applications. Finally, we conduct a concise discussion for anticipating possible future trajectories of Med-LLMs, identifying avenues for the prudent expansion of Med-LLMs. By consolidating above-mentioned insights, this review seeks to provide a comprehensive investigation of the potential strengths and limitations of Med-LLMs for professionals and researchers, ensuring a responsible landscape in the healthcare setting.
Authors: Chun Liu, Hongguang Zhang, Kainan Zhao, Xinghai Ju, Lin Yang
Abstract: With the booming of Large Language Models (LLMs), prompt-learning has become a promising method mainly researched in various research areas. Recently, many attempts based on prompt-learning have been made to improve the performance of text classification. However, most of these methods are based on heuristic Chain-of-Thought (CoT), and tend to be more complex but less efficient. In this paper, we rethink the LLM-based text classification methodology, propose a simple and effective transfer learning strategy, namely LLMEmbed, to address this classical but challenging task. To illustrate, we first study how to properly extract and fuse the text embeddings via various lightweight LLMs at different network depths to improve their robustness and discrimination, then adapt such embeddings to train the classifier. We perform extensive experiments on publicly available datasets, and the results show that LLMEmbed achieves strong performance while enjoys low training overhead using lightweight LLM backbones compared to recent methods based on larger LLMs, i.e. GPT-3, and sophisticated prompt-based strategies. Our LLMEmbed achieves adequate accuracy on publicly available benchmarks without any fine-tuning while merely use 4% model parameters, 1.8% electricity consumption and 1.5% runtime compared to its counterparts. Code is available at: https://github.com/ChunLiu-cs/LLMEmbed-ACL2024.
Authors: Zhouyu Jiang, Ling Zhong, Mengshu Sun, Jun Xu, Rui Sun, Hui Cai, Shuhan Luo, Zhiqiang Zhang
Abstract: To tackle the problem of domain-specific knowledge scarcity within large language models (LLMs), knowledge graph-retrievalaugmented method has been proven to be an effective and efficient technique for knowledge infusion. However, existing approaches face two primary challenges: knowledge mismatch between public available knowledge graphs and the specific domain of the task at hand, and poor information compliance of LLMs with knowledge graphs. In this paper, we leverage a small set of labeled samples and a large-scale corpus to efficiently construct domain-specific knowledge graphs by an LLM, addressing the issue of knowledge mismatch. Additionally, we propose a three-stage KG-LLM alignment strategyto enhance the LLM's capability to utilize information from knowledge graphs. We conduct experiments with a limited-sample setting on two biomedical question-answering datasets, and the results demonstrate that our approach outperforms existing baselines.
Authors: Shuo Huang, William MacLean, Xiaoxi Kang, Anqi Wu, Lizhen Qu, Qiongkai Xu, Zhuang Li, Xingliang Yuan, Gholamreza Haffari
Abstract: Increasing concerns about privacy leakage issues in academia and industry arise when employing NLP models from third-party providers to process sensitive texts. To protect privacy before sending sensitive data to those models, we suggest sanitizing sensitive text using two common strategies used by humans: i) deleting sensitive expressions, and ii) obscuring sensitive details by abstracting them. To explore the issues and develop a tool for text rewriting, we curate the first corpus, coined NAP^2, through both crowdsourcing and the use of large language models (LLMs). Compared to the prior works based on differential privacy, which lead to a sharp drop in information utility and unnatural texts, the human-inspired approaches result in more natural rewrites and offer an improved balance between privacy protection and data utility, as demonstrated by our extensive experiments.
Authors: Yang Hou, Zhenghua Li
Abstract: Revealing the syntactic structure of sentences in Chinese poses significant challenges for word-level parsers due to the absence of clear word boundaries. To facilitate a transition from word-level to character-level Chinese dependency parsing, this paper proposes modeling latent internal structures within words. In this way, each word-level dependency tree is interpreted as a forest of character-level trees. A constrained Eisner algorithm is implemented to ensure the compatibility of character-level trees, guaranteeing a single root for intra-word structures and establishing inter-word dependencies between these roots. Experiments on Chinese treebanks demonstrate the superiority of our method over both the pipeline framework and previous joint models. A detailed analysis reveals that a coarse-to-fine parsing strategy empowers the model to predict more linguistically plausible intra-word structures.
Authors: Faisal Tareque Shohan, Mir Tafseer Nayeem, Samsul Islam, Abu Ubaida Akash, Shafiq Joty
Abstract: Millions of news articles published online daily can overwhelm readers. Headlines and entity (topic) tags are essential for guiding readers to decide if the content is worth their time. While headline generation has been extensively studied, tag generation remains largely unexplored, yet it offers readers better access to topics of interest. The need for conciseness in capturing readers' attention necessitates improved content selection strategies for identifying salient and relevant segments within lengthy articles, thereby guiding language models effectively. To address this, we propose to leverage auxiliary information such as images and captions embedded in the articles to retrieve relevant sentences and utilize instruction tuning with variations to generate both headlines and tags for news articles in a multilingual context. To make use of the auxiliary information, we have compiled a dataset named XL-HeadTags, which includes 20 languages across 6 diverse language families. Through extensive evaluation, we demonstrate the effectiveness of our plug-and-play multimodal-multilingual retrievers for both tasks. Additionally, we have developed a suite of tools for processing and evaluating multilingual texts, significantly contributing to the research community by enabling more accurate and efficient analysis across languages.
Authors: Kohei Makino, Makoto Miwa, Yutaka Sasaki
Abstract: This study addresses a crucial challenge in instance-based relation extraction using text generation models: end-to-end training in target relation extraction task is not applicable to retrievers due to the non-differentiable nature of instance selection. We propose a novel End-to-end TRAinable Soft K-nearest neighbor retriever (ETRASK) by the neural prompting method that utilizes a soft, differentiable selection of the $k$ nearest instances. This approach enables the end-to-end training of retrievers in target tasks. On the TACRED benchmark dataset with a low-resource setting where the training data was reduced to 10\%, our method achieved a state-of-the-art F1 score of 71.5\%. Moreover, ETRASK consistently improved the baseline model by adding instances for all settings. These results highlight the efficacy of our approach in enhancing relation extraction performance, especially in resource-constrained environments. Our findings offer a promising direction for future research with extraction and the broader application of text generation in natural language processing.
Authors: Naibin Gu, Peng Fu, Xiyu Liu, Bowen Shen, Zheng Lin, Weiping Wang
Abstract: Parameter-efficient fine-tuning (PEFT) has emerged as the predominant technique for fine-tuning in the era of large language models. However, existing PEFT methods still have inadequate training efficiency. Firstly, the utilization of large-scale foundation models during the training process is excessively redundant for certain fine-tuning tasks. Secondly, as the model size increases, the growth in trainable parameters of empirically added PEFT modules becomes non-negligible and redundant, leading to inefficiency. To achieve task-specific efficient fine-tuning, we propose the Light-PEFT framework, which includes two methods: Masked Early Pruning of the Foundation Model and Multi-Granularity Early Pruning of PEFT. The Light-PEFT framework allows for the simultaneous estimation of redundant parameters in both the foundation model and PEFT modules during the early stage of training. These parameters can then be pruned for more efficient fine-tuning. We validate our approach on GLUE, SuperGLUE, QA tasks, and various models. With Light-PEFT, parameters of the foundation model can be pruned by up to over 40%, while still controlling trainable parameters to be only 25% of the original PEFT method. Compared to utilizing the PEFT method directly, Light-PEFT achieves training and inference speedup, reduces memory usage, and maintains comparable performance and the plug-and-play feature of PEFT.
Authors: Jiaming Zhou, Shiwan Zhao, Hui Wang, Tian-Hao Zhang, Haoqin Sun, Xuechen Wang, Yong Qin
Abstract: The kNN-CTC model has proven to be effective for monolingual automatic speech recognition (ASR). However, its direct application to multilingual scenarios like code-switching, presents challenges. Although there is potential for performance improvement, a kNN-CTC model utilizing a single bilingual datastore can inadvertently introduce undesirable noise from the alternative language. To address this, we propose a novel kNN-CTC-based code-switching ASR (CS-ASR) framework that employs dual monolingual datastores and a gated datastore selection mechanism to reduce noise interference. Our method selects the appropriate datastore for decoding each frame, ensuring the injection of language-specific information into the ASR process. We apply this framework to cutting-edge CTC-based models, developing an advanced CS-ASR system. Extensive experiments demonstrate the remarkable effectiveness of our gated datastore mechanism in enhancing the performance of zero-shot Chinese-English CS-ASR.
Authors: Dan Zhang, Sining Zhoubian, Yisong Yue, Yuxiao Dong, Jie Tang
Abstract: Recent methodologies in LLM self-training mostly rely on LLM generating responses and filtering those with correct output answers as training data. This approach often yields a low-quality fine-tuning training set (e.g., incorrect plans or intermediate reasoning). In this paper, we develop a reinforced self-training approach, called ReST-MCTS*, based on integrating process reward guidance with tree search MCTS* for collecting higher-quality reasoning traces as well as per-step value to train policy and reward models. ReST-MCTS* circumvents the per-step manual annotation typically used to train process rewards by tree-search-based reinforcement learning: Given oracle final correct answers, ReST-MCTS* is able to infer the correct process rewards by estimating the probability this step can help lead to the correct answer. These inferred rewards serve dual purposes: they act as value targets for further refining the process reward model and also facilitate the selection of high-quality traces for policy model self-training. We first show that the tree-search policy in ReST-MCTS* achieves higher accuracy compared with prior LLM reasoning baselines such as Best-of-N and Tree-of-Thought, within the same search budget. We then show that by using traces searched by this tree-search policy as training data, we can continuously enhance the three language models for multiple iterations, and outperform other self-training algorithms such as ReST$^\text{EM}$ and Self-Rewarding LM.
Authors: Aswin RRV, Nemika Tyagi, Md Nayem Uddin, Neeraj Varshney, Chitta Baral
Abstract: This study explores the sycophantic tendencies of Large Language Models (LLMs), where these models tend to provide answers that match what users want to hear, even if they are not entirely correct. The motivation behind this exploration stems from the common behavior observed in individuals searching the internet for facts with partial or misleading knowledge. Similar to using web search engines, users may recall fragments of misleading keywords and submit them to an LLM, hoping for a comprehensive response. Our empirical analysis of several LLMs shows the potential danger of these models amplifying misinformation when presented with misleading keywords. Additionally, we thoroughly assess four existing hallucination mitigation strategies to reduce LLMs sycophantic behavior. Our experiments demonstrate the effectiveness of these strategies for generating factually correct statements. Furthermore, our analyses delve into knowledge-probing experiments on factual keywords and different categories of sycophancy mitigation.
Authors: Huaiyuan Ying, Zijian Wu, Yihan Geng, Jiayu Wang, Dahua Lin, Kai Chen
Abstract: Large language models have demonstrated impressive capabilities across various natural language processing tasks, especially in solving mathematical problems. However, large language models are not good at math theorem proving using formal languages like Lean. A significant challenge in this area is the scarcity of training data available in these formal languages. To address this issue, we propose a novel pipeline that iteratively generates and filters synthetic data to translate natural language mathematical problems into Lean 4 statements, and vice versa. Our results indicate that the synthetic data pipeline can provide useful training data and improve the performance of LLMs in translating and understanding complex mathematical problems and proofs. Our final dataset contains about 57K formal-informal question pairs along with searched proof from the math contest forum and 21 new IMO questions. We open-source our code at https://github.com/InternLM/InternLM-Math and our data at https://huggingface.co/datasets/InternLM/Lean-Workbook.
URLs: https://github.com/InternLM/InternLM-Math, https://huggingface.co/datasets/InternLM/Lean-Workbook.
Authors: Jiahao Liu, Qifan Wang, Jingang Wang, Xunliang Cai
Abstract: The recent advancements in large language models (LLMs) have been extraordinary, yet the escalating inference costs associated with them present challenges in real-world applications. To address these challenges, we propose a novel approach called Early-exiting Speculative Decoding (EESD) with lossless acceleration. Specifically, EESD utilizes a segment of the LLM to generate draft tokens, incorporating Early-exiting structures after the first N layers. To enhance the quality of draft tokens, a self-distillation method is integrated. This early-exiting design not only reduces deployment and training costs but also significantly accelerates the token generation speed. Moreover, we introduce a novel sampling mechanism that leverages Thompson Sampling to regulate the generation processes, automatically determining the quantity of draft tokens in each round. The original LLM is then employed to validate these draft tokens through a single forward pass, and thus guarantees that the final output text maintains a distribution consistent with vanilla auto-regressive decoding. The experimental results on both 13B and 70B models demonstrate that our approach decodes tokens at a markedly accelerated rate compared to prior methods, showing the effectiveness of our approach.
Authors: Eden Avnat, Michal Levy, Daniel Herstain, Elia Yanko, Daniel Ben Joya, Michal Tzuchman Katz, Dafna Eshel, Sahar Laros, Yael Dagan, Shahar Barami, Joseph Mermelstein, Shahar Ovadia, Noam Shomron, Varda Shalev, Raja-Elie E. Abdulnour
Abstract: Clinical problem-solving requires processing of semantic medical knowledge such as illness scripts and numerical medical knowledge of diagnostic tests for evidence-based decision-making. As large language models (LLMs) show promising results in many aspects of language-based clinical practice, their ability to generate non-language evidence-based answers to clinical questions is inherently limited by tokenization. Therefore, we evaluated LLMs' performance on two question types: numeric (correlating findings) and semantic (differentiating entities) while examining differences within and between LLMs in medical aspects and comparing their performance to humans. To generate straightforward multi-choice questions and answers (QAs) based on evidence-based medicine (EBM), we used a comprehensive medical knowledge graph (encompassed data from more than 50,00 peer-reviewed articles) and created the "EBMQA". EBMQA contains 105,000 QAs labeled with medical and non-medical topics and classified into numerical or semantic questions. We benchmarked this dataset using more than 24,500 QAs on two state-of-the-art LLMs: Chat-GPT4 and Claude3-Opus. We evaluated the LLMs accuracy on semantic and numerical question types and according to sub-labeled topics. For validation, six medical experts were tested on 100 numerical EBMQA questions. We found that both LLMs excelled more in semantic than numerical QAs, with Claude3 surpassing GPT4 in numerical QAs. However, both LLMs showed inter and intra gaps in different medical aspects and remained inferior to humans. Thus, their medical advice should be addressed carefully.
Authors: Rachel Wicks, Matt Post, Philipp Koehn
Abstract: Data availability limits the scope of any given task. In machine translation, historical models were incapable of handling longer contexts, so the lack of document-level datasets was less noticeable. Now, despite the emergence of long-sequence methods, we remain within a sentence-level paradigm and without data to adequately approach context-aware machine translation. Most large-scale datasets have been processed through a pipeline that discards document-level metadata. In this work, we reconstruct document-level information for three (ParaCrawl, News Commentary, and Europarl) large datasets in German, French, Spanish, Italian, Polish, and Portuguese (paired with English). We then introduce a document-level filtering technique as an alternative to traditional bitext filtering. We present this filtering with analysis to show that this method prefers context-consistent translations rather than those that may have been sentence-level machine translated. Last we train models on these longer contexts and demonstrate improvement in document-level translation without degradation of sentence-level translation. We release our dataset, ParaDocs, and resulting models as a resource to the community.
Authors: Chen Wang, Minpeng Liao, Zhongqiang Huang, Junhong Wu, Chengqing Zong, Jiajun Zhang
Abstract: The recent release of GPT-4o showcased the potential of end-to-end multimodal models, not just in terms of low latency but also in their ability to understand and generate expressive speech with rich emotions. While the details are unknown to the open research community, it likely involves significant amounts of curated data and compute, neither of which is readily accessible. In this paper, we present BLSP-Emo (Bootstrapped Language-Speech Pretraining with Emotion support), a novel approach to developing an end-to-end speech-language model capable of understanding both semantics and emotions in speech and generate empathetic responses. BLSP-Emo utilizes existing speech recognition (ASR) and speech emotion recognition (SER) datasets through a two-stage process. The first stage focuses on semantic alignment, following recent work on pretraining speech-language models using ASR data. The second stage performs emotion alignment with the pretrained speech-language model on an emotion-aware continuation task constructed from SER data. Our experiments demonstrate that the BLSP-Emo model excels in comprehending speech and delivering empathetic responses, both in instruction-following tasks and conversations.
Authors: Shoutao Guo, Shaolei Zhang, Yang Feng
Abstract: Simultaneous Machine Translation (SiMT) generates translation while reading source tokens, essentially producing the target prefix based on the source prefix. To achieve good performance, it leverages the relationship between source and target prefixes to exact a policy to guide the generation of translations. Although existing SiMT methods primarily focus on the Encoder-Decoder architecture, we explore the potential of Decoder-only architecture, owing to its superior performance in various tasks and its inherent compatibility with SiMT. However, directly applying the Decoder-only architecture to SiMT poses challenges in terms of training and inference. To alleviate the above problems, we propose the first Decoder-only SiMT model, named Decoder-only Streaming Transformer (DST). Specifically, DST separately encodes the positions of the source and target prefixes, ensuring that the position of the target prefix remains unaffected by the expansion of the source prefix. Furthermore, we propose a Streaming Self-Attention (SSA) mechanism tailored for the Decoder-only architecture. It is capable of obtaining translation policy by assessing the sufficiency of input source information and integrating with the soft-attention mechanism to generate translations. Experiments demonstrate that our approach achieves state-of-the-art performance on three translation tasks.
Authors: Matthias Sperber, Ond\v{r}ej Bojar, Barry Haddow, D\'avid Javorsk\'y, Xutai Ma, Matteo Negri, Jan Niehues, Peter Pol\'ak, Elizabeth Salesky, Katsuhito Sudoh, Marco Turchi
Abstract: Human evaluation is a critical component in machine translation system development and has received much attention in text translation research. However, little prior work exists on the topic of human evaluation for speech translation, which adds additional challenges such as noisy data and segmentation mismatches. We take first steps to fill this gap by conducting a comprehensive human evaluation of the results of several shared tasks from the last International Workshop on Spoken Language Translation (IWSLT 2023). We propose an effective evaluation strategy based on automatic resegmentation and direct assessment with segment context. Our analysis revealed that: 1) the proposed evaluation strategy is robust and scores well-correlated with other types of human judgements; 2) automatic metrics are usually, but not always, well-correlated with direct assessment scores; and 3) COMET as a slightly stronger automatic metric than chrF, despite the segmentation noise introduced by the resegmentation step systems. We release the collected human-annotated data in order to encourage further investigation.
Authors: Ziyun Cui, Chang Lei, Wen Wu, Yinan Duan, Diyang Qu, Ji Wu, Runsen Chen, Chao Zhang
Abstract: The early detection of suicide risk is important since it enables the intervention to prevent potential suicide attempts. This paper studies the automatic detection of suicide risk based on spontaneous speech from adolescents, and collects a Mandarin dataset with 15 hours of suicide speech from more than a thousand adolescents aged from ten to eighteen for our experiments. To leverage the diverse acoustic and linguistic features embedded in spontaneous speech, both the Whisper speech model and textual large language models (LLMs) are used for suicide risk detection. Both all-parameter finetuning and parameter-efficient finetuning approaches are used to adapt the pre-trained models for suicide risk detection, and multiple audio-text fusion approaches are evaluated to combine the representations of Whisper and the LLM. The proposed system achieves a detection accuracy of 0.807 and an F1-score of 0.846 on the test set with 119 subjects, indicating promising potential for real suicide risk detection applications.
Authors: Anushka Singh, Ananya B. Sai, Raj Dabre, Ratish Puduppully, Anoop Kunchukuttan, Mitesh M Khapra
Abstract: While machine translation evaluation has been studied primarily for high-resource languages, there has been a recent interest in evaluation for low-resource languages due to the increasing availability of data and models. In this paper, we focus on a zero-shot evaluation setting focusing on low-resource Indian languages, namely Assamese, Kannada, Maithili, and Punjabi. We collect sufficient Multi-Dimensional Quality Metrics (MQM) and Direct Assessment (DA) annotations to create test sets and meta-evaluate a plethora of automatic evaluation metrics. We observe that even for learned metrics, which are known to exhibit zero-shot performance, the Kendall Tau and Pearson correlations with human annotations are only as high as 0.32 and 0.45. Synthetic data approaches show mixed results and overall do not help close the gap by much for these languages. This indicates that there is still a long way to go for low-resource evaluation.
Authors: Tzuf Paz-Argaman, Itai Mondshine, Asaf Achi Mordechai, Reut Tsarfaty
Abstract: While large language models (LLMs) excel in various natural language tasks in English, their performance in lower-resourced languages like Hebrew, especially for generative tasks such as abstractive summarization, remains unclear. The high morphological richness in Hebrew adds further challenges due to the ambiguity in sentence comprehension and the complexities in meaning construction. In this paper, we address this resource and evaluation gap by introducing HeSum, a novel benchmark specifically designed for abstractive text summarization in Modern Hebrew. HeSum consists of 10,000 article-summary pairs sourced from Hebrew news websites written by professionals. Linguistic analysis confirms HeSum's high abstractness and unique morphological challenges. We show that HeSum presents distinct difficulties for contemporary state-of-the-art LLMs, establishing it as a valuable testbed for generative language technology in Hebrew, and MRLs generative challenges in general.
Authors: Firoj Alam, Abul Hasnat, Fatema Ahmed, Md Arid Hasan, Maram Hasanain
Abstract: With the rise of digital communication, memes have become a significant medium for cultural and political expression that is often used to mislead audiences. Identification of such misleading and persuasive multimodal content has become more important among various stakeholders, including social media platforms, policymakers, and the broader society as they often cause harm to individuals, organizations, and/or society. While there has been effort to develop AI-based automatic systems for resource-rich languages (e.g., English), it is relatively little to none for medium to low resource languages. In this study, we focused on developing an Arabic memes dataset with manual annotations of propagandistic content. We annotated ~6K Arabic memes collected from various social media platforms, which is a first resource for Arabic multimodal research. We provide a comprehensive analysis aiming to develop computational tools for their detection. We will make them publicly available for the community.
Authors: Chen Cecilia Liu, Iryna Gurevych, Anna Korhonen
Abstract: The surge of interest in culturally aware and adapted Natural Language Processing (NLP) has inspired much recent research. However, the lack of common understanding of the concept of "culture" has made it difficult to evaluate progress in this emerging area. Drawing on prior research in NLP and related fields, we propose an extensive taxonomy of elements of culture that can provide a systematic framework for analyzing and understanding research progress. Using the taxonomy, we survey existing resources and models for culturally aware and adapted NLP, providing an overview of the state of the art and the research gaps that still need to be filled.
Authors: Kaiyan Zhang, Sihang Zeng, Ermo Hua, Ning Ding, Zhang-Ren Chen, Zhiyuan Ma, Haoxin Li, Ganqu Cui, Biqing Qi, Xuekai Zhu, Xingtai Lv, Hu Jinfang, Zhiyuan Liu, Bowen Zhou
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains and are moving towards more specialized areas. Recent advanced proprietary models such as GPT-4 and Gemini have achieved significant advancements in biomedicine, which have also raised privacy and security challenges. The construction of specialized generalists hinges largely on high-quality datasets, enhanced by techniques like supervised fine-tuning and reinforcement learning from human or AI feedback, and direct preference optimization. However, these leading technologies (e.g., preference learning) are still significantly limited in the open source community due to the scarcity of specialized data. In this paper, we present the UltraMedical collections, which consist of high-quality manual and synthetic datasets in the biomedicine domain, featuring preference annotations across multiple advanced LLMs. By utilizing these datasets, we fine-tune a suite of specialized medical models based on Llama-3 series, demonstrating breathtaking capabilities across various medical benchmarks. Moreover, we develop powerful reward models skilled in biomedical and general reward benchmark, enhancing further online preference learning within the biomedical LLM community.
Authors: Neemesh Yadav, Sarah Masud, Vikram Goyal, Vikram Goyal, Md Shad Akhtar, Tanmoy Chakraborty
Abstract: Employing language models to generate explanations for an incoming implicit hate post is an active area of research. The explanation is intended to make explicit the underlying stereotype and aid content moderators. The training often combines top-k relevant knowledge graph (KG) tuples to provide world knowledge and improve performance on standard metrics. Interestingly, our study presents conflicting evidence for the role of the quality of KG tuples in generating implicit explanations. Consequently, simpler models incorporating external toxicity signals outperform KG-infused models. Compared to the KG-based setup, we observe a comparable performance for SBIC (LatentHatred) datasets with a performance variation of +0.44 (+0.49), +1.83 (-1.56), and -4.59 (+0.77) in BLEU, ROUGE-L, and BERTScore. Further human evaluation and error analysis reveal that our proposed setup produces more precise explanations than zero-shot GPT-3.5, highlighting the intricate nature of the task.
Authors: Wei Tang, Yixin Cao, Jiahao Ying, Bo Wang, Yuyue Zhao, Yong Liao, Pengyuan Zhou
Abstract: Retrieval-Augmented Generation (RAG) is an effective solution to supplement necessary knowledge to large language models (LLMs). Targeting its bottleneck of retriever performance, "generate-then-read" pipeline is proposed to replace the retrieval stage with generation from the LLM itself. Although promising, this research direction is underexplored and still cannot work in the scenario when source knowledge is given. In this paper, we formalize a general "A + B" framework with varying combinations of foundation models and types for systematic investigation. We explore the efficacy of the base and chat versions of LLMs and found their different functionalities suitable for generator A and reader B, respectively. Their combinations consistently outperform single models, especially in complex scenarios. Furthermore, we extend the application of the "A + B" framework to scenarios involving source documents through continuous learning, enabling the direct integration of external knowledge into LLMs. This approach not only facilitates effective acquisition of new knowledge but also addresses the challenges of safety and helpfulness post-adaptation. The paper underscores the versatility of the "A + B" framework, demonstrating its potential to enhance the practical application of LLMs across various domains.
Authors: Ankan Mullick, Sombit Bose, Rounak Saha, Ayan Kumar Bhowmick, Pawan Goyal, Niloy Ganguly, Prasenjit Dey, Ravi Kokku
Abstract: In an ever-expanding world of domain-specific knowledge, the increasing complexity of consuming, and storing information necessitates the generation of summaries from large information repositories. However, every persona of a domain has different requirements of information and hence their summarization. For example, in the healthcare domain, a persona-based (such as Doctor, Nurse, Patient etc.) approach is imperative to deliver targeted medical information efficiently. Persona-based summarization of domain-specific information by humans is a high cognitive load task and is generally not preferred. The summaries generated by two different humans have high variability and do not scale in cost and subject matter expertise as domains and personas grow. Further, AI-generated summaries using generic Large Language Models (LLMs) may not necessarily offer satisfactory accuracy for different domains unless they have been specifically trained on domain-specific data and can also be very expensive to use in day-to-day operations. Our contribution in this paper is two-fold: 1) We present an approach to efficiently fine-tune a domain-specific small foundation LLM using a healthcare corpus and also show that we can effectively evaluate the summarization quality using AI-based critiquing. 2) We further show that AI-based critiquing has good concordance with Human-based critiquing of the summaries. Hence, such AI-based pipelines to generate domain-specific persona-based summaries can be easily scaled to other domains such as legal, enterprise documents, education etc. in a very efficient and cost-effective manner.
Authors: Hadi Askari, Anshuman Chhabra, Muhao Chen, Prasant Mohapatra
Abstract: Large Language Models (LLMs) have achieved state-of-the-art performance at zero-shot generation of abstractive summaries for given articles. However, little is known about the robustness of such a process of zero-shot summarization. To bridge this gap, we propose relevance paraphrasing, a simple strategy that can be used to measure the robustness of LLMs as summarizers. The relevance paraphrasing approach identifies the most relevant sentences that contribute to generating an ideal summary, and then paraphrases these inputs to obtain a minimally perturbed dataset. Then, by evaluating model performance for summarization on both the original and perturbed datasets, we can assess the LLM's one aspect of robustness. We conduct extensive experiments with relevance paraphrasing on 4 diverse datasets, as well as 4 LLMs of different sizes (GPT-3.5-Turbo, Llama-2-13B, Mistral-7B, and Dolly-v2-7B). Our results indicate that LLMs are not consistent summarizers for the minimally perturbed articles, necessitating further improvements.
Authors: Kayo Yin, Terry Regier, Dan Klein
Abstract: Communicative efficiency is a prominent theory in linguistics and cognitive science. While numerous studies have shown how the pressure to save energy is reflected in the form of spoken languages, few have explored this phenomenon in signed languages. In this paper, we show how handshapes in American Sign Language (ASL) reflect these efficiency pressures and we present new evidence of communicative efficiency in the visual-gestural modality. We focus on handshapes that are used in both native ASL signs and signs borrowed from English to compare efficiency pressures from both ASL and English. First, we design new methodologies to quantify the articulatory effort required to produce handshapes as well as the perceptual effort needed to recognize them. Then, we compare correlations between communicative effort and usage statistics in ASL and English. Our findings reveal that frequent ASL handshapes are easier to produce and that pressures for communicative efficiency mostly come from ASL usage, not from English lexical borrowing.
Authors: Caimei Yang, Qihang Yang, Xingzhi Su, Chenxi Fu, Xiaoyi Wang, Ying Yan, Zaijiang Man
Abstract: There have been apparently conflicting claims over the syntax-semantics relationship in child acquisition. However, few of them have assessed the child's path toward the acquisition of recursive relative clauses (RRCs). The authors of the current paper did experiments to investigate 3- to 11-year-olds' most-structured elicited production of eight Mandarin RRCs in a 4 (syntactic types)*2 (semantic conditions) design. The four syntactic types were RRCs with a subject-gapped RC embedded in an object-gapped RC (SORRCs), RRCs with an object-gapped RC embedded in another object-gapped RC (OORRCs), RRCs with an object-gapped RC embedded in a subject-gapped RC (OSRRCs), and RRCs with a subject-gapped RC embedded in another subject-gapped RC (SSRRCs). Each syntactic type was put in two conditions differing in internal semantics: irreversible internal semantics (IIS) and reversible internal semantics (RIS). For example, "the balloon that [the girl that _ eats the banana] holds _" is SORRCs in the IIS condition; "the monkey that [the dog that _ bites the pig] hits_" is SORRCs in the RIS condition. For each target, the participants were provided with a speech-visual stimulus constructing a condition of irreversible external semantics (IES). The results showed that SSRRCs, OSRRCs and SORRCs in the IIS-IES condition were produced two years earlier than their counterparts in the RIS-IES condition. Thus, a 2-stage development path is proposed: the language acquisition device starts with the interface between (irreversible) syntax and IIS, and ends with the interface between syntax and IES, both abiding by the syntax-semantic interface principle.
Authors: Jisu Shin, Hoyun Song, Huije Lee, Soyeong Jeong, Jong C. Park
Abstract: Social bias is shaped by the accumulation of social perceptions towards targets across various demographic identities. To fully understand such social bias in large language models (LLMs), it is essential to consider the composite of social perceptions from diverse perspectives among identities. Previous studies have either evaluated biases in LLMs by indirectly assessing the presence of sentiments towards demographic identities in the generated text or measuring the degree of alignment with given stereotypes. These methods have limitations in directly quantifying social biases at the level of distinct perspectives among identities. In this paper, we aim to investigate how social perceptions from various viewpoints contribute to the development of social bias in LLMs. To this end, we propose a novel strategy to intuitively quantify these social perceptions and suggest metrics that can evaluate the social biases within LLMs by aggregating diverse social perceptions. The experimental results show the quantitative demonstration of the social attitude in LLMs by examining social perception. The analysis we conducted shows that our proposed metrics capture the multi-dimensional aspects of social bias, enabling a fine-grained and comprehensive investigation of bias in LLMs.
Authors: Agostina Calabrese, Leonardo Neves, Neil Shah, Maarten W. Bos, Bj\"orn Ross, Mirella Lapata, Francesco Barbieri
Abstract: Content moderators play a key role in keeping the conversation on social media healthy. While the high volume of content they need to judge represents a bottleneck to the moderation pipeline, no studies have explored how models could support them to make faster decisions. There is, by now, a vast body of research into detecting hate speech, sometimes explicitly motivated by a desire to help improve content moderation, but published research using real content moderators is scarce. In this work we investigate the effect of explanations on the speed of real-world moderators. Our experiments show that while generic explanations do not affect their speed and are often ignored, structured explanations lower moderators' decision making time by 7.4%.
Authors: Adil Soubki, Owen Rambow
Abstract: The notion of face described by Brown and Levinson (1987) has been studied in great detail, but a critical aspect of the framework, that which focuses on how intentions mediate the planning of turns which impose upon face, has received far less attention. We present an analysis of three computational systems trained for classifying both intention and politeness, focusing on how the former influences the latter. In politeness theory, agents attend to the desire to have their wants appreciated (positive face), and a complementary desire to act unimpeded and maintain freedom (negative face). Similar to speech acts, utterances can perform so-called face acts which can either raise or threaten the positive or negative face of the speaker or hearer. We begin by using an existing corpus to train a model which classifies face acts, achieving a new SoTA in the process. We then observe that every face act has an underlying intention that motivates it and perform additional experiments integrating dialog act annotations to provide these intentions by proxy. Our analysis finds that dialog acts improve performance on face act detection for minority classes and points to a close relationship between aspects of face and intent.
Authors: Chaoxu Pang, Yixuan Cao, Chunhao Yang, Ping Luo
Abstract: Tables are recognized for their high information density and widespread usage, serving as essential sources of information. Seeking information from tables (TIS) is a crucial capability for Large Language Models (LLMs), serving as the foundation of knowledge-based Q&A systems. However, this field presently suffers from an absence of thorough and reliable evaluation. This paper introduces a more reliable benchmark for Table Information Seeking (TabIS). To avoid the unreliable evaluation caused by text similarity-based metrics, TabIS adopts a single-choice question format (with two options per question) instead of a text generation format. We establish an effective pipeline for generating options, ensuring their difficulty and quality. Experiments conducted on 12 LLMs reveal that while the performance of GPT-4-turbo is marginally satisfactory, both other proprietary and open-source models perform inadequately. Further analysis shows that LLMs exhibit a poor understanding of table structures, and struggle to balance between TIS performance and robustness against pseudo-relevant tables (common in retrieval-augmented systems). These findings uncover the limitations and potential challenges of LLMs in seeking information from tables. We release our data and code to facilitate further research in this field.
Authors: Aryo Pradipta Gema, Joshua Ong Jun Leang, Giwon Hong, Alessio Devoto, Alberto Carlo Maria Mancino, Rohit Saxena, Xuanli He, Yu Zhao, Xiaotang Du, Mohammad Reza Ghasemi Madani, Claire Barale, Robert McHardy, Joshua Harris, Jean Kaddour, Emile van Krieken, Pasquale Minervini
Abstract: Maybe not. We identify and analyse errors in the popular Massive Multitask Language Understanding (MMLU) benchmark. Even though MMLU is widely adopted, our analysis demonstrates numerous ground truth errors that obscure the true capabilities of LLMs. For example, we find that 57% of the analysed questions in the Virology subset contain errors. To address this issue, we introduce a comprehensive framework for identifying dataset errors using a novel error taxonomy. Then, we create MMLU-Redux, which is a subset of 3,000 manually re-annotated questions across 30 MMLU subjects. Using MMLU-Redux, we demonstrate significant discrepancies with the model performance metrics that were originally reported. Our results strongly advocate for revising MMLU's error-ridden questions to enhance its future utility and reliability as a benchmark. Therefore, we open up MMLU-Redux for additional annotation https://huggingface.co/datasets/edinburgh-dawg/mmlu-redux.
URLs: https://huggingface.co/datasets/edinburgh-dawg/mmlu-redux.
Authors: Shubham Kumar Nigam, Anurag Sharma, Danush Khanna, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya
Abstract: In the era of Large Language Models (LLMs), predicting judicial outcomes poses significant challenges due to the complexity of legal proceedings and the scarcity of expert-annotated datasets. Addressing this, we introduce \textbf{Pred}iction with \textbf{Ex}planation (\texttt{PredEx}), the largest expert-annotated dataset for legal judgment prediction and explanation in the Indian context, featuring over 15,000 annotations. This groundbreaking corpus significantly enhances the training and evaluation of AI models in legal analysis, with innovations including the application of instruction tuning to LLMs. This method has markedly improved the predictive accuracy and explanatory depth of these models for legal judgments. We employed various transformer-based models, tailored for both general and Indian legal contexts. Through rigorous lexical, semantic, and expert assessments, our models effectively leverage \texttt{PredEx} to provide precise predictions and meaningful explanations, establishing it as a valuable benchmark for both the legal profession and the NLP community.
Authors: Luana Bulla, Aldo Gangemi, Misael Mongiov\`i
Abstract: The task of detecting moral values in text has significant implications in various fields, including natural language processing, social sciences, and ethical decision-making. Previously proposed supervised models often suffer from overfitting, leading to hyper-specialized moral classifiers that struggle to perform well on data from different domains. To address this issue, we introduce novel systems that leverage abstract concepts and common-sense knowledge acquired from Large Language Models and Natural Language Inference models during previous stages of training on multiple data sources. By doing so, we aim to develop versatile and robust methods for detecting moral values in real-world scenarios. Our approach uses the GPT 3.5 model as a zero-shot ready-made unsupervised multi-label classifier for moral values detection, eliminating the need for explicit training on labeled data. We compare it with a smaller NLI-based zero-shot model. The results show that the NLI approach achieves competitive results compared to the Davinci model. Furthermore, we conduct an in-depth investigation of the performance of supervised systems in the context of cross-domain multi-label moral value detection. This involves training supervised models on different domains to explore their effectiveness in handling data from different sources and comparing their performance with the unsupervised methods. Our contributions encompass a thorough analysis of both supervised and unsupervised methodologies for cross-domain value detection. We introduce the Davinci model as a state-of-the-art zero-shot unsupervised moral values classifier, pushing the boundaries of moral value detection without the need for explicit training on labeled data. Additionally, we perform a comparative evaluation of our approach with the supervised models, shedding light on their respective strengths and weaknesses.
Authors: Qi Cheng, Michael Boratko, Pranay Kumar Yelugam, Tim O'Gorman, Nalini Singh, Andrew McCallum, Xiang Lorraine Li
Abstract: Large language models have demonstrated impressive performance on commonsense tasks; however, these tasks are often posed as multiple-choice questions, allowing models to exploit systematic biases. Commonsense is also inherently probabilistic with multiple correct answers. The purpose of "boiling water" could be making tea and cooking, but it also could be killing germs. Existing tasks do not capture the probabilistic nature of common sense. To this end, we present commonsense frame completion (CFC), a new generative task that evaluates common sense via multiple open-ended generations. We also propose a method of probabilistic evaluation that strongly correlates with human judgments. Humans drastically outperform strong language model baselines on our dataset, indicating this approach is both a challenging and useful evaluation of machine common sense.
Authors: Guangliang Liu, Milad Afshari, Xitong Zhang, Zhiyu Xue, Avrajit Ghosh, Bidhan Bashyal, Rongrong Wang, Kristen Johnson
Abstract: While task-agnostic debiasing provides notable generalizability and reduced reliance on downstream data, its impact on language modeling ability and the risk of relearning social biases from downstream task-specific data remain as the two most significant challenges when debiasing Pretrained Language Models (PLMs). The impact on language modeling ability can be alleviated given a high-quality and long-contextualized debiasing corpus, but there remains a deficiency in understanding the specifics of relearning biases. We empirically ascertain that the effectiveness of task-agnostic debiasing hinges on the quantitative bias level of both the task-specific data used for downstream applications and the debiased model. We empirically show that the lower bound of the bias level of the downstream fine-tuned model can be approximated by the bias level of the debiased model, in most practical cases. To gain more in-depth understanding about how the parameters of PLMs change during fine-tuning due to the forgetting issue of PLMs, we propose a novel framework which can Propagate Socially-fair Debiasing to Downstream Fine-tuning, ProSocialTuning. Our proposed framework can push the fine-tuned model to approach the bias lower bound during downstream fine-tuning, indicating that the ineffectiveness of debiasing can be alleviated by overcoming the forgetting issue through regularizing successfully debiased attention heads based on the PLMs' bias levels from stages of pretraining and debiasing.
Authors: Lars Hillebrand, Prabhupad Pradhan, Christian Bauckhage, Rafet Sifa
Abstract: We introduce "pointer-guided segment ordering" (SO), a novel pre-training technique aimed at enhancing the contextual understanding of paragraph-level text representations in large language models. Our methodology leverages a self-attention-driven pointer network to restore the original sequence of shuffled text segments, addressing the challenge of capturing the structural coherence and contextual dependencies within documents. This pre-training approach is complemented by a fine-tuning methodology that incorporates dynamic sampling, augmenting the diversity of training instances and improving sample efficiency for various downstream applications. We evaluate our method on a diverse set of datasets, demonstrating its efficacy in tasks requiring sequential text classification across scientific literature and financial reporting domains. Our experiments show that pointer-guided pre-training significantly enhances the model's ability to understand complex document structures, leading to state-of-the-art performance in downstream classification tasks.
Authors: Peiqi Sui, Eamon Duede, Sophie Wu, Richard Jean So
Abstract: This paper presents a systematic defense of large language model (LLM) hallucinations or 'confabulations' as a potential resource instead of a categorically negative pitfall. The standard view is that confabulations are inherently problematic and AI research should eliminate this flaw. In this paper, we argue and empirically demonstrate that measurable semantic characteristics of LLM confabulations mirror a human propensity to utilize increased narrativity as a cognitive resource for sense-making and communication. In other words, it has potential value. Specifically, we analyze popular hallucination benchmarks and reveal that hallucinated outputs display increased levels of narrativity and semantic coherence relative to veridical outputs. This finding reveals a tension in our usually dismissive understandings of confabulation. It suggests, counter-intuitively, that the tendency for LLMs to confabulate may be intimately associated with a positive capacity for coherent narrative-text generation.
Authors: Shangqing Tu, Kejian Zhu, Yushi Bai, Zijun Yao, Lei Hou, Juanzi Li
Abstract: The advancement of large language models (LLMs) relies on evaluation using public benchmarks, but data contamination can lead to overestimated performance. Previous researches focus on detecting contamination by determining whether the model has seen the exact same data during training. In this work, we argue that even training on data similar to benchmark data inflates performance on in-distribution tasks without improving overall capacity, which we called In-distribution contamination. To effectively detect in-distribution contamination, we propose DICE, a novel method that leverages the internal states of LLMs to locate-then-detect the contamination. DICE first identifies the most sensitive layer to contamination, then trains a classifier based on the internal states of that layer. Experiments reveal DICE's high accuracy in detecting in-distribution contamination across various LLMs and math reasoning datasets. We also show the generalization capability of the trained DICE detector, which is able to detect contamination across multiple benchmarks with similar distributions. Additionally, we find that the DICE detection scores are positively correlated with the performance of ten LLMs fine-tuned by either us or other organizations on four math reasoning datasets (with $R^2$ values between 0.6 and 0.75). This indicates that the in-distribution contamination problem potentially lead to an overestimation of the true capabilities of many existing models. The code and data are available at https://github.com/THU-KEG/DICE.
Authors: Chun-Hsien Lin, Pu-Jen Cheng
Abstract: With the development of large-scale Language Models (LLM), fine-tuning pre-trained LLM has become a mainstream paradigm for solving downstream tasks of natural language processing. However, training a language model in the legal field requires a large number of legal documents so that the language model can learn legal terminology and the particularity of the format of legal documents. The typical NLP approaches usually rely on many manually annotated data sets for training. However, in the legal field application, it is difficult to obtain a large number of manually annotated data sets, which restricts the typical method applied to the task of drafting legal documents. The experimental results of this paper show that not only can we leverage a large number of annotation-free legal documents without Chinese word segmentation to fine-tune a large-scale language model, but more importantly, it can fine-tune a pre-trained LLM on the local computer to achieve the generating legal document drafts task, and at the same time achieve the protection of information privacy and to improve information security issues.
Authors: Yuanyi Ren, Haoran Ye, Hanjun Fang, Xin Zhang, Guojie Song
Abstract: Large Language Models (LLMs) are transforming diverse fields and gaining increasing influence as human proxies. This development underscores the urgent need for evaluating value orientations and understanding of LLMs to ensure their responsible integration into public-facing applications. This work introduces ValueBench, the first comprehensive psychometric benchmark for evaluating value orientations and value understanding in LLMs. ValueBench collects data from 44 established psychometric inventories, encompassing 453 multifaceted value dimensions. We propose an evaluation pipeline grounded in realistic human-AI interactions to probe value orientations, along with novel tasks for evaluating value understanding in an open-ended value space. With extensive experiments conducted on six representative LLMs, we unveil their shared and distinctive value orientations and exhibit their ability to approximate expert conclusions in value-related extraction and generation tasks. ValueBench is openly accessible at https://github.com/Value4AI/ValueBench.
Authors: Yusuke Sakai, Hidetaka Kamigaito, Taro Watanabe
Abstract: It is very challenging to curate a dataset for language-specific knowledge and common sense in order to evaluate natural language understanding capabilities of language models. Due to the limitation in the availability of annotators, most current multilingual datasets are created through translation, which cannot evaluate such language-specific aspects. Therefore, we propose Multilingual CommonsenseQA (mCSQA) based on the construction process of CSQA but leveraging language models for a more efficient construction, e.g., by asking LM to generate questions/answers, refine answers and verify QAs followed by reduced human efforts for verification. Constructed dataset is a benchmark for cross-lingual language-transfer capabilities of multilingual LMs, and experimental results showed high language-transfer capabilities for questions that LMs could easily solve, but lower transfer capabilities for questions requiring deep knowledge or commonsense. This highlights the necessity of language-specific datasets for evaluation and training. Finally, our method demonstrated that multilingual LMs could create QA including language-specific knowledge, significantly reducing the dataset creation cost compared to manual creation. The datasets are available at https://huggingface.co/datasets/yusuke1997/mCSQA.
Authors: Jiaoda Li, Yifan Hou, Mrinmaya Sachan, Ryan Cotterell
Abstract: Large language models (LLMs) exhibit an intriguing ability to learn a novel task from in-context examples presented in a demonstration, termed in-context learning (ICL). Understandably, a swath of research has been dedicated to uncovering the theories underpinning ICL. One popular hypothesis explains ICL by task selection. LLMs identify the task based on the demonstration and generalize it to the prompt. Another popular hypothesis is that ICL is a form of meta-learning, i.e., the models learn a learning algorithm at pre-training time and apply it to the demonstration. Finally, a third hypothesis argues that LLMs use the demonstration to select a composition of tasks learned during pre-training to perform ICL. In this paper, we empirically explore these three hypotheses that explain LLMs' ability to learn in context with a suite of experiments derived from common text classification tasks. We invalidate the first two hypotheses with counterexamples and provide evidence in support of the last hypothesis. Our results suggest an LLM could learn a novel task in context via composing tasks learned during pre-training.
Authors: Yifan Tang, Yihao Wang, Ru Zhang, Jianyi Liu
Abstract: To detect stego (steganographic text) in complex scenarios, linguistic steganalysis (LS) with various motivations has been proposed and achieved excellent performance. However, with the development of generative steganography, some stegos have strong concealment, especially after the emergence of LLMs-based steganography, the existing LS has low detection or even cannot detect them. We designed a novel LS with two modes called LSGC. In the generation mode, we created an LS-task "description" and used the generation ability of LLM to explain whether texts to be detected are stegos. On this basis, we rethought the principle of LS and LLMs, and proposed the classification mode. In this mode, LSGC deleted the LS-task "description" and changed the "causalLM" LLMs to the "sequenceClassification" architecture. The LS features can be extracted by only one pass of the model, and a linear layer with initialization weights is added to obtain the classification probability. Experiments on strongly concealed stegos show that LSGC significantly improves detection and reaches SOTA performance. Additionally, LSGC in classification mode greatly reduces training time while maintaining high performance.
Authors: Shaina Raza, Mizanur Rahman, Michael R. Zhang
Abstract: Recent improvements in large language models (LLMs) have significantly enhanced natural language processing (NLP) applications. However, these models can also inherit and perpetuate biases from their training data. Addressing this issue is crucial, yet many existing datasets do not offer evaluation across diverse NLP tasks. To tackle this, we introduce the Bias Evaluations Across Domains (BEADs) dataset, designed to support a wide range of NLP tasks, including text classification, bias entity recognition, bias quantification, and benign language generation. BEADs uses AI-driven annotation combined with experts' verification to provide reliable labels. This method overcomes the limitations of existing datasets that typically depend on crowd-sourcing, expert-only annotations with limited bias evaluations, or unverified AI labeling. Our empirical analysis shows that BEADs is effective in detecting and reducing biases across different language models, with smaller models fine-tuned on BEADs often outperforming LLMs in bias classification tasks. However, these models may still exhibit biases towards certain demographics. Fine-tuning LLMs with our benign language data also reduces biases while preserving the models' knowledge. Our findings highlight the importance of comprehensive bias evaluation and the potential of targeted fine-tuning for reducing the bias of LLMs. We are making BEADs publicly available at https://huggingface.co/datasets/shainar/BEAD Warning: This paper contains examples that may be considered offensive.
Authors: Bernardo Leite, Tom\'as Freitas Os\'orio, Henrique Lopes Cardoso
Abstract: Question Answering (QA) datasets are crucial in assessing reading comprehension skills for both machines and humans. While numerous datasets have been developed in English for this purpose, a noticeable void exists in less-resourced languages. To alleviate this gap, our paper introduces machine-translated versions of FairytaleQA, a renowned QA dataset designed to assess and enhance narrative comprehension skills in young children. By employing fine-tuned, modest-scale models, we establish benchmarks for both Question Generation (QG) and QA tasks within the translated datasets. In addition, we present a case study proposing a model for generating question-answer pairs, with an evaluation incorporating quality metrics such as question well-formedness, answerability, relevance, and children suitability. Our evaluation prioritizes quantifying and describing error cases, along with providing directions for future work. This paper contributes to the advancement of QA and QG research in less-resourced languages, promoting accessibility and inclusivity in the development of these models for reading comprehension. The code and data is publicly available at github.com/bernardoleite/fairytaleqa-translated.
Authors: Cheng Xu, Shuhao Guan, Derek Greene, M-Tahar Kechadi
Abstract: The rapid development of Large Language Models (LLMs) like GPT-4, Claude-3, and Gemini has transformed the field of natural language processing. However, it has also resulted in a significant issue known as Benchmark Data Contamination (BDC). This occurs when language models inadvertently incorporate evaluation benchmark information from their training data, leading to inaccurate or unreliable performance during the evaluation phase of the process. This paper reviews the complex challenge of BDC in LLM evaluation and explores alternative assessment methods to mitigate the risks associated with traditional benchmarks. The paper also examines challenges and future directions in mitigating BDC risks, highlighting the complexity of the issue and the need for innovative solutions to ensure the reliability of LLM evaluation in real-world applications.
Authors: Federico Barbero, Andrea Banino, Steven Kapturowski, Dharshan Kumaran, Jo\~ao G. M. Ara\'ujo, Alex Vitvitskyi, Razvan Pascanu, Petar Veli\v{c}kovi\'c
Abstract: We study how information propagates in decoder-only Transformers, which are the architectural backbone of most existing frontier large language models (LLMs). We rely on a theoretical signal propagation analysis -- specifically, we analyse the representations of the last token in the final layer of the Transformer, as this is the representation used for next-token prediction. Our analysis reveals a representational collapse phenomenon: we prove that certain distinct sequences of inputs to the Transformer can yield arbitrarily close representations in the final token. This effect is exacerbated by the low-precision floating-point formats frequently used in modern LLMs. As a result, the model is provably unable to respond to these sequences in different ways -- leading to errors in, e.g., tasks involving counting or copying. Further, we show that decoder-only Transformer language models can lose sensitivity to specific tokens in the input, which relates to the well-known phenomenon of over-squashing in graph neural networks. We provide empirical evidence supporting our claims on contemporary LLMs. Our theory also points to simple solutions towards ameliorating these issues.
Authors: Ling Yang, Zhaochen Yu, Tianjun Zhang, Shiyi Cao, Minkai Xu, Wentao Zhang, Joseph E. Gonzalez, Bin Cui
Abstract: We introduce Buffer of Thoughts (BoT), a novel and versatile thought-augmented reasoning approach for enhancing accuracy, efficiency and robustness of large language models (LLMs). Specifically, we propose meta-buffer to store a series of informative high-level thoughts, namely thought-template, distilled from the problem-solving processes across various tasks. Then for each problem, we retrieve a relevant thought-template and adaptively instantiate it with specific reasoning structures to conduct efficient reasoning. To guarantee the scalability and stability, we further propose buffer-manager to dynamically update the meta-buffer, thus enhancing the capacity of meta-buffer as more tasks are solved. We conduct extensive experiments on 10 challenging reasoning-intensive tasks, and achieve significant performance improvements over previous SOTA methods: 11% on Game of 24, 20% on Geometric Shapes and 51% on Checkmate-in-One. Further analysis demonstrate the superior generalization ability and model robustness of our BoT, while requiring only 12% of the cost of multi-query prompting methods (e.g., tree/graph of thoughts) on average. Notably, we find that our Llama3-8B+BoT has the potential to surpass Llama3-70B model. Our project is available at: https://github.com/YangLing0818/buffer-of-thought-llm
Authors: Dun-Ming Huang, Pol Van Rijn, Ilia Sucholutsky, Raja Marjieh, Nori Jacoby
Abstract: Conversational tones -- the manners and attitudes in which speakers communicate -- are essential to effective communication. Amidst the increasing popularization of Large Language Models (LLMs) over recent years, it becomes necessary to characterize the divergences in their conversational tones relative to humans. However, existing investigations of conversational modalities rely on pre-existing taxonomies or text corpora, which suffer from experimenter bias and may not be representative of real-world distributions for the studies' psycholinguistic domains. Inspired by methods from cognitive science, we propose an iterative method for simultaneously eliciting conversational tones and sentences, where participants alternate between two tasks: (1) one participant identifies the tone of a given sentence and (2) a different participant generates a sentence based on that tone. We run 100 iterations of this process with human participants and GPT-4, then obtain a dataset of sentences and frequent conversational tones. In an additional experiment, humans and GPT-4 annotated all sentences with all tones. With data from 1,339 human participants, 33,370 human judgments, and 29,900 GPT-4 queries, we show how our approach can be used to create an interpretable geometric representation of relations between conversational tones in humans and GPT-4. This work demonstrates how combining ideas from machine learning and cognitive science can address challenges in human-computer interactions.
Authors: Sreyan Ghosh, Utkarsh Tyagi, Sonal Kumar, C. K. Evuru, S Ramaneswaran, S Sakshi, Dinesh Manocha
Abstract: We present ABEX, a novel and effective generative data augmentation methodology for low-resource Natural Language Understanding (NLU) tasks. ABEX is based on ABstract-and-EXpand, a novel paradigm for generating diverse forms of an input document -- we first convert a document into its concise, abstract description and then generate new documents based on expanding the resultant abstraction. To learn the task of expanding abstract descriptions, we first train BART on a large-scale synthetic dataset with abstract-document pairs. Next, to generate abstract descriptions for a document, we propose a simple, controllable, and training-free method based on editing AMR graphs. ABEX brings the best of both worlds: by expanding from abstract representations, it preserves the original semantic properties of the documents, like style and meaning, thereby maintaining alignment with the original label and data distribution. At the same time, the fundamental process of elaborating on abstract descriptions facilitates diverse generations. We demonstrate the effectiveness of ABEX on 4 NLU tasks spanning 12 datasets and 4 low-resource settings. ABEX outperforms all our baselines qualitatively with improvements of 0.04% - 38.8%. Qualitatively, ABEX outperforms all prior methods from literature in terms of context and length diversity.
Authors: Nadav Borenstein, Anej Svete, Robin Chan, Josef Valvoda, Franz Nowak, Isabelle Augenstein, Eleanor Chodroff, Ryan Cotterell
Abstract: What can large language models learn? By definition, language models (LM) are distributions over strings. Therefore, an intuitive way of addressing the above question is to formalize it as a matter of learnability of classes of distributions over strings. While prior work in this direction focused on assessing the theoretical limits, in contrast, we seek to understand the empirical learnability. Unlike prior empirical work, we evaluate neural LMs on their home turf-learning probabilistic languages-rather than as classifiers of formal languages. In particular, we investigate the learnability of regular LMs (RLMs) by RNN and Transformer LMs. We empirically test the learnability of RLMs as a function of various complexity parameters of the RLM and the hidden state size of the neural LM. We find that the RLM rank, which corresponds to the size of linear space spanned by the logits of its conditional distributions, and the expected length of sampled strings are strong and significant predictors of learnability for both RNNs and Transformers. Several other predictors also reach significance, but with differing patterns between RNNs and Transformers.
Authors: Jinqi Luo, Tianjiao Ding, Kwan Ho Ryan Chan, Darshan Thaker, Aditya Chattopadhyay, Chris Callison-Burch, Ren\'e Vidal
Abstract: Large Language Models (LLMs) are being used for a wide variety of tasks. While they are capable of generating human-like responses, they can also produce undesirable output including potentially harmful information, racist or sexist language, and hallucinations. Alignment methods are designed to reduce such undesirable output, via techniques such as fine-tuning, prompt engineering, and representation engineering. However, existing methods face several challenges: some require costly fine-tuning for every alignment task; some do not adequately remove undesirable concepts, failing alignment; some remove benign concepts, lowering the linguistic capabilities of LLMs. To address these issues, we propose Parsimonious Concept Engineering (PaCE), a novel activation engineering framework for alignment. First, to sufficiently model the concepts, we construct a large-scale concept dictionary in the activation space, in which each atom corresponds to a semantic concept. Then, given any alignment task, we instruct a concept partitioner to efficiently annotate the concepts as benign or undesirable. Finally, at inference time, we decompose the LLM activations along the concept dictionary via sparse coding, to accurately represent the activation as a linear combination of the benign and undesirable components. By removing the latter ones from the activation, we reorient the behavior of LLMs towards alignment goals. We conduct experiments on tasks such as response detoxification, faithfulness enhancement, and sentiment revising, and show that PaCE achieves state-of-the-art alignment performance while maintaining linguistic capabilities.
Authors: Alexander Bakumenko (Clemson University, USA), Kate\v{r}ina Hlav\'a\v{c}kov\'a-Schindler (University of Vienna, Austria), Claudia Plant (University of Vienna, Austria), Nina C. Hubig (Clemson University, USA)
Abstract: Detecting anomalies in general ledger data is of utmost importance to ensure trustworthiness of financial records. Financial audits increasingly rely on machine learning (ML) algorithms to identify irregular or potentially fraudulent journal entries, each characterized by a varying number of transactions. In machine learning, heterogeneity in feature dimensions adds significant complexity to data analysis. In this paper, we introduce a novel approach to anomaly detection in financial data using Large Language Models (LLMs) embeddings. To encode non-semantic categorical data from real-world financial records, we tested 3 pre-trained general purpose sentence-transformer models. For the downstream classification task, we implemented and evaluated 5 optimized ML models including Logistic Regression, Random Forest, Gradient Boosting Machines, Support Vector Machines, and Neural Networks. Our experiments demonstrate that LLMs contribute valuable information to anomaly detection as our models outperform the baselines, in selected settings even by a large margin. The findings further underscore the effectiveness of LLMs in enhancing anomaly detection in financial journal entries, particularly by tackling feature sparsity. We discuss a promising perspective on using LLM embeddings for non-semantic data in the financial context and beyond.
Authors: Ahad Jawaid, Shreeram Suresh Chandra, Junchen Lu, Berrak Sisman
Abstract: Recent advances in style transfer text-to-speech (TTS) have improved the expressiveness of synthesized speech. Despite these advancements, encoding stylistic information from diverse and unseen reference speech remains challenging. This paper introduces StyleMoE, an approach that divides the embedding space, modeled by the style encoder, into tractable subsets handled by style experts. The proposed method replaces the style encoder in a TTS system with a Mixture of Experts (MoE) layer. By utilizing a gating network to route reference speeches to different style experts, each expert specializes in aspects of the style space during optimization. Our experiments objectively and subjectively demonstrate the effectiveness of our proposed method in increasing the coverage of the style space for diverse and unseen styles. This approach can enhance the performance of existing state-of-the-art style transfer TTS models, marking the first study of MoE in style transfer TTS to our knowledge.
Authors: Jinlong Xue, Yayue Deng, Yicheng Han, Yingming Gao, Ya Li
Abstract: Recent advances in large language models (LLMs) and development of audio codecs greatly propel the zero-shot TTS. They can synthesize personalized speech with only a 3-second speech of an unseen speaker as acoustic prompt. However, they only support short speech prompts and cannot leverage longer context information, as required in audiobook and conversational TTS scenarios. In this paper, we introduce a novel audio codec-based TTS model to adapt context features with multiple enhancements. Inspired by the success of Qformer, we propose a multi-modal context-enhanced Qformer (MMCE-Qformer) to utilize additional multi-modal context information. Besides, we adapt a pretrained LLM to leverage its understanding ability to predict semantic tokens, and use a SoundStorm to generate acoustic tokens thereby enhancing audio quality and speaker similarity. The extensive objective and subjective evaluations show that our proposed method outperforms baselines across various context TTS scenarios.
Authors: Liyi Zhang, Michael Y. Li, Thomas L. Griffiths
Abstract: Autoregressive language models have demonstrated a remarkable ability to extract latent structure from text. The embeddings from large language models have been shown to capture aspects of the syntax and semantics of language. But what {\em should} embeddings represent? We connect the autoregressive prediction objective to the idea of constructing predictive sufficient statistics to summarize the information contained in a sequence of observations, and use this connection to identify three settings where the optimal content of embeddings can be identified: independent identically distributed data, where the embedding should capture the sufficient statistics of the data; latent state models, where the embedding should encode the posterior distribution over states given the data; and discrete hypothesis spaces, where the embedding should reflect the posterior distribution over hypotheses given the data. We then conduct empirical probing studies to show that transformers encode these three kinds of latent generating distributions, and that they perform well in out-of-distribution cases and without token memorization in these settings.
Authors: Xiaohu Du, Ming Wen, Jiahao Zhu, Zifan Xie, Bin Ji, Huijun Liu, Xuanhua Shi, Hai Jin
Abstract: Code Pre-trained Models (CodePTMs) based vulnerability detection have achieved promising results over recent years. However, these models struggle to generalize as they typically learn superficial mapping from source code to labels instead of understanding the root causes of code vulnerabilities, resulting in poor performance in real-world scenarios beyond the training instances. To tackle this challenge, we introduce VulLLM, a novel framework that integrates multi-task learning with Large Language Models (LLMs) to effectively mine deep-seated vulnerability features. Specifically, we construct two auxiliary tasks beyond the vulnerability detection task. First, we utilize the vulnerability patches to construct a vulnerability localization task. Second, based on the vulnerability features extracted from patches, we leverage GPT-4 to construct a vulnerability interpretation task. VulLLM innovatively augments vulnerability classification by leveraging generative LLMs to understand complex vulnerability patterns, thus compelling the model to capture the root causes of vulnerabilities rather than overfitting to spurious features of a single task. The experiments conducted on six large datasets demonstrate that VulLLM surpasses seven state-of-the-art models in terms of effectiveness, generalization, and robustness.
Authors: Jingyang Ou, Shen Nie, Kaiwen Xue, Fengqi Zhu, Jiacheng Sun, Zhenguo Li, Chongxuan Li
Abstract: Discrete diffusion models with absorbing processes have shown promise in language modeling. The key quantities to be estimated are the ratios between the marginal probabilities of two transitive states at all timesteps, called the concrete score. In this paper, we reveal that the concrete score in absorbing diffusion can be expressed as conditional probabilities of clean data, multiplied by a time-dependent scalar in an analytic form. Motivated by the finding, we propose reparameterized absorbing discrete diffusion (RADD), a dedicated diffusion model that characterizes the time-independent conditional probabilities. Besides its simplicity, RADD can reduce the number of function evaluations (NFEs) by caching the output of the time-independent network when the noisy sample remains unchanged in a sampling interval. Empirically, RADD is up to 3.5 times faster while consistently achieving a better performance than the strongest baseline. Built upon the new factorization of the concrete score, we further prove a surprising result that the exact likelihood of absorbing diffusion can be rewritten to a simple form (named denoising cross-entropy) and then estimated efficiently by the Monte Carlo method. The resulting approach also applies to the original parameterization of the concrete score. It significantly advances the state-of-the-art discrete diffusion on 5 zero-shot language modeling benchmarks (measured by perplexity) at the GPT-2 scale.
Authors: Yanming Liu, Xinyue Peng, Yuwei Zhang, Jiannan Cao, Xuhong Zhang, Sheng Cheng, Xun Wang, Jianwei Yin, Tianyu Du
Abstract: Large language models (LLMs) have demonstrated exceptional reasoning capabilities, enabling them to solve various complex problems. Recently, this ability has been applied to the paradigm of tool learning. Tool learning involves providing examples of tool usage and their corresponding functions, allowing LLMs to formulate plans and demonstrate the process of invoking and executing each tool. LLMs can address tasks that they cannot complete independently, thereby enhancing their potential across different tasks. However, this approach faces two key challenges. First, redundant error correction leads to unstable planning and long execution time. Additionally, designing a correct plan among multiple tools is also a challenge in tool learning. To address these issues, we propose Tool-Planner, a task-processing framework based on toolkits. Tool-Planner groups tools based on the API functions with the same function into a toolkit and allows LLMs to implement planning across the various toolkits. When a tool error occurs, the language model can reselect and adjust tools based on the toolkit. Experiments show that our approach demonstrates a high pass and win rate across different datasets and optimizes the planning scheme for tool learning in models such as GPT-4 and Claude 3, showcasing the potential of our method.
Authors: Stefan Gerd Fritsch, Cennet Oguz, Vitor Fortes Rey, Lala Ray, Maximilian Kiefer-Emmanouilidis, Paul Lukowicz
Abstract: Human Activity Recognition is a longstanding problem in AI with applications in a broad range of areas: from healthcare, sports and fitness, security, and human computer interaction to robotics. The performance of HAR in real-world settings is strongly dependent on the type and quality of the input signal that can be acquired. Given an unobstructed, high-quality camera view of a scene, computer vision systems, in particular in conjunction with foundational models (e.g., CLIP), can today fairly reliably distinguish complex activities. On the other hand, recognition using modalities such as wearable sensors (which are often more broadly available, e.g, in mobile phones and smartwatches) is a more difficult problem, as the signals often contain less information and labeled training data is more difficult to acquire. In this work, we show how we can improve HAR performance across different modalities using multimodal contrastive pretraining. Our approach MuJo (Multimodal Joint Feature Space Learning), learns a multimodal joint feature space with video, language, pose, and IMU sensor data. The proposed approach combines contrastive and multitask learning methods and analyzes different multitasking strategies for learning a compact shared representation. A large dataset with parallel video, language, pose, and sensor data points is also introduced to support the research, along with an analysis of the robustness of the multimodal joint space for modal-incomplete and low-resource data. On the MM-Fit dataset, our model achieves an impressive Macro F1-Score of up to 0.992 with only 2% of the train data and 0.999 when using all available training data for classification tasks. Moreover, in the scenario where the MM-Fit dataset is unseen, we demonstrate a generalization performance of up to 0.638.
Authors: Eleonora Mancini, Ana Tanevska, Andrea Galassi, Alessio Galatolo, Federico Ruggeri, Paolo Torroni
Abstract: Current research in machine learning and artificial intelligence is largely centered on modeling and performance evaluation, less so on data collection. However, recent research demonstrated that limitations and biases in data may negatively impact trustworthiness and reliability. These aspects are particularly impactful on sensitive domains such as mental health and neurological disorders, where speech data are used to develop AI applications aimed at improving the health of patients and supporting healthcare providers. In this paper, we chart the landscape of available speech datasets for this domain, to highlight possible pitfalls and opportunities for improvement and promote fairness and diversity. We present a comprehensive list of desiderata for building speech datasets for mental health and neurological disorders and distill it into a checklist focused on ethical concerns to foster more responsible research.
Authors: Zhiheng Xi, Yiwen Ding, Wenxiang Chen, Boyang Hong, Honglin Guo, Junzhe Wang, Dingwen Yang, Chenyang Liao, Xin Guo, Wei He, Songyang Gao, Lu Chen, Rui Zheng, Yicheng Zou, Tao Gui, Qi Zhang, Xipeng Qiu, Xuanjing Huang, Zuxuan Wu, Yu-Gang Jiang
Abstract: Building generalist agents that can handle diverse tasks and evolve themselves across different environments is a long-term goal in the AI community. Large language models (LLMs) are considered a promising foundation to build such agents due to their generalized capabilities. Current approaches either have LLM-based agents imitate expert-provided trajectories step-by-step, requiring human supervision, which is hard to scale and limits environmental exploration; or they let agents explore and learn in isolated environments, resulting in specialist agents with limited generalization. In this paper, we take the first step towards building generally-capable LLM-based agents with self-evolution ability. We identify a trinity of ingredients: 1) diverse environments for agent exploration and learning, 2) a trajectory set to equip agents with basic capabilities and prior knowledge, and 3) an effective and scalable evolution method. We propose AgentGym, a new framework featuring a variety of environments and tasks for broad, real-time, uni-format, and concurrent agent exploration. AgentGym also includes a database with expanded instructions, a benchmark suite, and high-quality trajectories across environments. Next, we propose a novel method, AgentEvol, to investigate the potential of agent self-evolution beyond previously seen data across tasks and environments. Experimental results show that the evolved agents can achieve results comparable to SOTA models. We release the AgentGym suite, including the platform, dataset, benchmark, checkpoints, and algorithm implementations. The AgentGym suite is available on https://github.com/WooooDyy/AgentGym.
Authors: Larisa Markeeva, Sean McLeish, Borja Ibarz, Wilfried Bounsi, Olga Kozlova, Alex Vitvitskyi, Charles Blundell, Tom Goldstein, Avi Schwarzschild, Petar Veli\v{c}kovi\'c
Abstract: Eliciting reasoning capabilities from language models (LMs) is a critical direction on the path towards building intelligent systems. Most recent studies dedicated to reasoning focus on out-of-distribution performance on procedurally-generated synthetic benchmarks, bespoke-built to evaluate specific skills only. This trend makes results hard to transfer across publications, slowing down progress. Three years ago, a similar issue was identified and rectified in the field of neural algorithmic reasoning, with the advent of the CLRS benchmark. CLRS is a dataset generator comprising graph execution traces of classical algorithms from the Introduction to Algorithms textbook. Inspired by this, we propose CLRS-Text -- a textual version of these algorithmic traces. Out of the box, CLRS-Text is capable of procedurally generating trace data for thirty diverse, challenging algorithmic tasks across any desirable input distribution, while offering a standard pipeline in which any additional algorithmic tasks may be created in the benchmark. We fine-tune and evaluate various LMs as generalist executors on this benchmark, validating prior work and revealing a novel, interesting challenge for the LM reasoning community. Our code is available at https://github.com/google-deepmind/clrs/tree/master/clrs/_src/clrs_text.
URLs: https://github.com/google-deepmind/clrs/tree/master/clrs/_src/clrs_text.
Authors: Max Mueller-Eberstein, Dianna Yee, Karren Yang, Gautam Varma Mantena, Colin Lea
Abstract: Parameter-efficient fine-tuning (PEFT) for personalizing automatic speech recognition (ASR) has recently shown promise for adapting general population models to atypical speech. However, these approaches assume a priori knowledge of the atypical speech disorder being adapted for -- the diagnosis of which requires expert knowledge that is not always available. Even given this knowledge, data scarcity and high inter/intra-speaker variability further limit the effectiveness of traditional fine-tuning. To circumvent these challenges, we first identify the minimal set of model parameters required for ASR adaptation. Our analysis of each individual parameter's effect on adaptation performance allows us to reduce Word Error Rate (WER) by half while adapting 0.03\% of all weights. Alleviating the need for cohort-specific models, we next propose the novel use of a meta-learned hypernetwork to generate highly individualized, utterance-level adaptations on-the-fly for a diverse set of atypical speech characteristics. Evaluating adaptation at the global, cohort and individual-level, we show that hypernetworks generalize better to out-of-distribution speakers, while maintaining an overall relative WER reduction of 75.2% using 0.1% of the full parameter budget.
Authors: Junjie Zhou, Yan Shu, Bo Zhao, Boya Wu, Shitao Xiao, Xi Yang, Yongping Xiong, Bo Zhang, Tiejun Huang, Zheng Liu
Abstract: The evaluation of Long Video Understanding (LVU) performance poses an important but challenging research problem. Despite previous efforts, the existing video understanding benchmarks are severely constrained by several issues, especially the insufficient lengths of videos, a lack of diversity in video types and evaluation tasks, and the inappropriateness for evaluating LVU performances. To address the above problems, we propose a new benchmark, called MLVU (Multi-task Long Video Understanding Benchmark), for the comprehensive and in-depth evaluation of LVU. MLVU presents the following critical values: 1) The substantial and flexible extension of video lengths, which enables the benchmark to evaluate LVU performance across a wide range of durations. 2) The inclusion of various video genres, e.g., movies, surveillance footage, egocentric videos, cartoons, game videos, etc., which reflects the models' LVU performances in different scenarios. 3) The development of diversified evaluation tasks, which enables a comprehensive examination of MLLMs' key abilities in long-video understanding. The empirical study with 20 latest MLLMs reveals significant room for improvement in today's technique, as all existing methods struggle with most of the evaluation tasks and exhibit severe performance degradation when handling longer videos. Additionally, it suggests that factors such as context length, image-understanding quality, and the choice of LLM backbone can play critical roles in future advancements. We anticipate that MLVU will advance the research of long video understanding by providing a comprehensive and in-depth analysis of MLLMs.
Authors: Xiang Ji, Sanjeev Kulkarni, Mengdi Wang, Tengyang Xie
Abstract: This work studies the challenge of aligning large language models (LLMs) with offline preference data. We focus on alignment by Reinforcement Learning from Human Feedback (RLHF) in particular. While popular preference optimization methods exhibit good empirical performance in practice, they are not theoretically guaranteed to converge to the optimal policy and can provably fail when the data coverage is sparse by classical offline reinforcement learning (RL) results. On the other hand, a recent line of work has focused on theoretically motivated preference optimization methods with provable guarantees, but these are not computationally efficient for large-scale applications like LLM alignment. To bridge this gap, we propose SPAC, a new offline preference optimization method with self-play, inspired by the on-average pessimism technique from the offline RL literature, to be the first provable and scalable approach to LLM alignment. We both provide theoretical analysis for its convergence under single-policy concentrability for the general function approximation setting and demonstrate its competitive empirical performance for LLM alignment on a 7B Mistral model with Open LLM Leaderboard evaluations.
Authors: Junjie Zhou, Zheng Liu, Shitao Xiao, Bo Zhao, Yongping Xiong
Abstract: Multi-modal retrieval becomes increasingly popular in practice. However, the existing retrievers are mostly text-oriented, which lack the capability to process visual information. Despite the presence of vision-language models like CLIP, the current methods are severely limited in representing the text-only and image-only data. In this work, we present a new embedding model VISTA for universal multi-modal retrieval. Our work brings forth threefold technical contributions. Firstly, we introduce a flexible architecture which extends a powerful text encoder with the image understanding capability by introducing visual token embeddings. Secondly, we develop two data generation strategies, which bring high-quality composed image-text to facilitate the training of the embedding model. Thirdly, we introduce a multi-stage training algorithm, which first aligns the visual token embedding with the text encoder using massive weakly labeled data, and then develops multi-modal representation capability using the generated composed image-text data. In our experiments, VISTA achieves superior performances across a variety of multi-modal retrieval tasks in both zero-shot and supervised settings. Our model, data, and source code are available at https://github.com/FlagOpen/FlagEmbedding.
Authors: Caleb Ziems, William Held, Jane Dwivedi-Yu, Diyi Yang
Abstract: Information Retrieval (IR) systems are designed to deliver relevant content, but traditional systems may not optimize rankings for fairness, neutrality, or the balance of ideas. Consequently, IR can often introduce indexical biases, or biases in the positional order of documents. Although indexical bias can demonstrably affect people's opinion, voting patterns, and other behaviors, these issues remain understudied as the field lacks reliable metrics and procedures for automatically measuring indexical bias. Towards this end, we introduce the PAIR framework, which supports automatic bias audits for ranked documents or entire IR systems. After introducing DUO, the first general-purpose automatic bias metric, we run an extensive evaluation of 8 IR systems on a new corpus of 32k synthetic and 4.7k natural documents, with 4k queries spanning 1.4k controversial issue topics. A human behavioral study validates our approach, showing that our bias metric can help predict when and how indexical bias will shift a reader's opinion.
Authors: Andy Zou, Long Phan, Justin Wang, Derek Duenas, Maxwell Lin, Maksym Andriushchenko, Rowan Wang, Zico Kolter, Matt Fredrikson, Dan Hendrycks
Abstract: AI systems can take harmful actions and are highly vulnerable to adversarial attacks. We present an approach, inspired by recent advances in representation engineering, that "short-circuits" models as they respond with harmful outputs. Existing techniques aimed at improving alignment, such as refusal training, are often bypassed. Techniques such as adversarial training try to plug these holes by countering specific attacks. As an alternative to refusal training and adversarial training, short-circuiting directly controls the representations that are responsible for harmful outputs in the first place. Our technique can be applied to both text-only and multimodal language models to prevent the generation of harmful outputs without sacrificing utility -- even in the presence of powerful unseen attacks. Notably, while adversarial robustness in standalone image recognition remains an open challenge, short-circuiting allows the larger multimodal system to reliably withstand image "hijacks" that aim to produce harmful content. Finally, we extend our approach to AI agents, demonstrating considerable reductions in the rate of harmful actions when they are under attack. Our approach represents a significant step forward in the development of reliable safeguards to harmful behavior and adversarial attacks.
Authors: Tim Z. Xiao, Robert Bamler, Bernhard Sch\"olkopf, Weiyang Liu
Abstract: Motivated by the large progress made by large language models (LLMs), we introduce the framework of verbalized machine learning (VML). In contrast to conventional machine learning models that are typically optimized over a continuous parameter space, VML constrains the parameter space to be human-interpretable natural language. Such a constraint leads to a new perspective of function approximation, where an LLM with a text prompt can be viewed as a function parameterized by the text prompt. Guided by this perspective, we revisit classical machine learning problems, such as regression and classification, and find that these problems can be solved by an LLM-parameterized learner and optimizer. The major advantages of VML include (1) easy encoding of inductive bias: prior knowledge about the problem and hypothesis class can be encoded in natural language and fed into the LLM-parameterized learner; (2) automatic model class selection: the optimizer can automatically select a concrete model class based on data and verbalized prior knowledge, and it can update the model class during training; and (3) interpretable learner updates: the LLM-parameterized optimizer can provide explanations for why each learner update is performed. We conduct several studies to empirically evaluate the effectiveness of VML, and hope that VML can serve as a stepping stone to stronger interpretability and trustworthiness in ML.
Authors: Ryan Cotterell, Georg Heigold
Abstract: Even for common NLP tasks, sufficient supervision is not available in many languages -- morphological tagging is no exception. In the work presented here, we explore a transfer learning scheme, whereby we train character-level recurrent neural taggers to predict morphological taggings for high-resource languages and low-resource languages together. Learning joint character representations among multiple related languages successfully enables knowledge transfer from the high-resource languages to the low-resource ones, improving accuracy by up to 30% over a monolingual model.
Authors: Ronald Cardenas, Matthias Galle, Shay B. Cohen
Abstract: Extractive summaries are usually presented as lists of sentences with no expected cohesion between them and with plenty of redundant information if not accounted for. In this paper, we investigate the trade-offs incurred when aiming to control for inter-sentential cohesion and redundancy in extracted summaries, and their impact on their informativeness. As case study, we focus on the summarization of long, highly redundant documents and consider two optimization scenarios, reward-guided and with no supervision. In the reward-guided scenario, we compare systems that control for redundancy and cohesion during sentence scoring. In the unsupervised scenario, we introduce two systems that aim to control all three properties -- informativeness, redundancy, and cohesion -- in a principled way. Both systems implement a psycholinguistic theory that simulates how humans keep track of relevant content units and how cohesion and non-redundancy constraints are applied in short-term memory during reading. Extensive automatic and human evaluations reveal that systems optimizing for -- among other properties -- cohesion are capable of better organizing content in summaries compared to systems that optimize only for redundancy, while maintaining comparable informativeness. We find that the proposed unsupervised systems manage to extract highly cohesive summaries across varying levels of document redundancy, although sacrificing informativeness in the process. Finally, we lay evidence as to how simulated cognitive processes impact the trade-off between the analyzed summary properties.
Authors: Chongyang Tao, Chang Liu, Tao Shen, Can Xu, Xiubo Geng, Binxing Jiao, Daxin Jiang
Abstract: To improve the performance of the dual-encoder retriever, one effective approach is knowledge distillation from the cross-encoder ranker. Existing works construct the candidate passages following the supervised learning setting where a query is paired with a positive passage and a batch of negatives. However, through empirical observation, we find that even the hard negatives from advanced methods are still too trivial for the teacher to distinguish, preventing the teacher from transferring abundant dark knowledge to the student through its soft label. To alleviate this issue, we propose ADAM, a knowledge distillation framework that can better transfer the dark knowledge held in the teacher with Adaptive Dark exAMples. Different from previous works that only rely on one positive and hard negatives as candidate passages, we create dark examples that all have moderate relevance to the query through mixing-up and masking in discrete space. Furthermore, as the quality of knowledge held in different training instances varies as measured by the teacher's confidence score, we propose a self-paced distillation strategy that adaptively concentrates on a subset of high-quality instances to conduct our dark-example-based knowledge distillation to help the student learn better. We conduct experiments on two widely-used benchmarks and verify the effectiveness of our method.
Authors: Chi Han, Jialiang Xu, Manling Li, Yi Fung, Chenkai Sun, Nan Jiang, Tarek Abdelzaher, Heng Ji
Abstract: Language models (LMs) automatically learn word embeddings during pre-training on language corpora. Although word embeddings are usually interpreted as feature vectors for individual words, their roles in language model generation remain underexplored. In this work, we theoretically and empirically revisit output word embeddings and find that their linear transformations are equivalent to steering language model generation styles. We name such steers LM-Steers and find them existing in LMs of all sizes. It requires learning parameters equal to 0.2% of the original LMs' size for steering each style. On tasks such as language model detoxification and sentiment control, LM-Steers can achieve comparable or superior performance compared with state-of-the-art controlled generation methods while maintaining a better balance with generation quality. The learned LM-Steer serves as a lens in text styles: it reveals that word embeddings are interpretable when associated with language model generations and can highlight text spans that most indicate the style differences. An LM-Steer is transferrable between different language models by an explicit form calculation. One can also continuously steer LMs simply by scaling the LM-Steer or compose multiple LM-Steers by adding their transformations. Our codes are publicly available at \url{https://github.com/Glaciohound/LM-Steer}.
Authors: Ruohao Guo, Wei Xu, Alan Ritter
Abstract: Language style is often used by writers to convey their intentions, identities, and mastery of language. In this paper, we show that current large language models struggle to capture some language styles without fine-tuning. To address this challenge, we investigate whether LLMs can be meta-trained based on representative lexicons to recognize new styles they have not been fine-tuned on. Experiments on 13 established style classification tasks, as well as 63 novel tasks generated using LLMs, demonstrate that meta-training with style lexicons consistently improves zero-shot transfer across styles. We release the code and data at http://github.com/octaviaguo/Style-LLM .
Authors: Afra Amini, Li Du, Ryan Cotterell
Abstract: Gradient-based sampling algorithms have demonstrated their effectiveness in text generation, especially in the context of controlled text generation. However, there exists a lack of theoretically grounded and principled approaches for this task. In this paper, we take an important step toward building a principled approach for sampling from language models with gradient-based methods. We use discrete distributions given by language models to define densities and develop an algorithm based on Hamiltonian Monte Carlo to sample from them. We name our gradient-based technique Structured Voronoi Sampling (SVS). In an experimental setup where the reference distribution is known, we show that the empirical distribution of SVS samples is closer to the reference distribution compared to alternative sampling schemes. Furthermore, in a controlled generation task, SVS is able to generate fluent and diverse samples while following the control targets significantly better than other methods.
Authors: Kai Lv, Yuqing Yang, Tengxiao Liu, Qinghui Gao, Qipeng Guo, Xipeng Qiu
Abstract: Large Language Models (LLMs) have revolutionized Natural Language Processing (NLP) but demand massive GPU resources for training. Lowering the threshold for LLMs training would encourage greater participation from researchers, benefiting both academia and society. While existing approaches have focused on parameter-efficient fine-tuning, which tunes or adds a small number of parameters, few have addressed the challenge of tuning the full parameters of LLMs with limited resources. In this work, we propose a new optimizer, LOw-Memory Optimization (LOMO), which fuses the gradient computation and the parameter update in one step to reduce memory usage. By integrating LOMO with existing memory saving techniques, we reduce memory usage to 10.8% compared to the standard approach (DeepSpeed solution). Consequently, our approach enables the full parameter fine-tuning of a 65B model on a single machine with 8 RTX 3090, each with 24GB memory.Code and data are available at https://github.com/OpenLMLab/LOMO.
Authors: Yibo Hu, Erick Skorupa Parolin, Latifur Khan, Patrick T. Brandt, Javier Osorio, Vito J. D'Orazio
Abstract: Is it possible accurately classify political relations within evolving event ontologies without extensive annotations? This study investigates zero-shot learning methods that use expert knowledge from existing annotation codebook, and evaluates the performance of advanced ChatGPT (GPT-3.5/4) and a natural language inference (NLI)-based model called ZSP. ChatGPT uses codebook's labeled summaries as prompts, whereas ZSP breaks down the classification task into context, event mode, and class disambiguation to refine task-specific hypotheses. This decomposition enhances interpretability, efficiency, and adaptability to schema changes. The experiments reveal ChatGPT's strengths and limitations, and crucially show ZSP's outperformance of dictionary-based methods and its competitive edge over some supervised models. These findings affirm the value of ZSP for validating event records and advancing ontology development. Our study underscores the efficacy of leveraging transfer learning and existing domain expertise to enhance research efficiency and scalability.
Authors: Tongtao Ling, Yutao Lai, Lei Chen, Shilei Huang, Yi Liu
Abstract: In clinical dictation, utterances after automatic speech recognition (ASR) without explicit punctuation marks may lead to the misunderstanding of dictated reports. To give a precise and understandable clinical report with ASR, automatic punctuation restoration is required. Considering a practical scenario, we propose a fast and light pre-trained model for Chinese medical punctuation restoration based on 'pretraining and fine-tuning' paradigm. In this work, we distill pre-trained models by incorporating supervised contrastive learning and a novel auxiliary pre-training task (Punctuation Mark Prediction) to make it well-suited for punctuation restoration. Our experiments on various distilled models reveal that our model can achieve 95% performance while 10% model size relative to state-of-the-art Chinese RoBERTa.
Authors: Julius Steen, Katja Markert
Abstract: Summarization is an important application of large language models (LLMs). Most previous evaluation of summarization models has focused on their content selection, faithfulness, grammaticality and coherence. However, it is well known that LLMs can reproduce and reinforce harmful social biases. This raises the question: Do biases affect model outputs in a constrained setting like summarization? To help answer this question, we first motivate and introduce a number of definitions for biased behaviours in summarization models, along with practical operationalizations. Since we find that biases inherent to input documents can confound bias analysis in summaries, we propose a method to generate input documents with carefully controlled demographic attributes. This allows us to study summarizer behavior in a controlled setting, while still working with realistic input documents. We measure gender bias in English summaries generated by both purpose-built summarization models and general purpose chat models as a case study. We find content selection in single document summarization to be largely unaffected by gender bias, while hallucinations exhibit evidence of bias. To demonstrate the generality of our approach, we additionally investigate racial bias, including intersectional settings.
Authors: Junzhe Liu, Jianwei Yu, Xie Chen
Abstract: Adapting End-to-End ASR models to out-of-domain datasets with text data is challenging. Factorized neural Transducer (FNT) aims to address this issue by introducing a separate vocabulary decoder to predict the vocabulary. Nonetheless, this approach has limitations in fusing acoustic and language information seamlessly. Moreover, a degradation in word error rate (WER) on the general test sets was also observed, leading to doubts about its overall performance. In response to this challenge, we present the improved factorized neural Transducer (IFNT) model structure designed to comprehensively integrate acoustic and language information while enabling effective text adaptation. We assess the performance of our proposed method on English and Mandarin datasets. The results indicate that IFNT not only surpasses the neural Transducer and FNT in baseline performance in both scenarios but also exhibits superior adaptation ability compared to FNT. On source domains, IFNT demonstrated statistically significant accuracy improvements, achieving a relative enhancement of 1.2% to 2.8% in baseline accuracy compared to the neural Transducer. On out-of-domain datasets, IFNT shows relative WER(CER) improvements of up to 30.2% over the standard neural Transducer with shallow fusion, and relative WER(CER) reductions ranging from 1.1% to 2.8% on test sets compared to the FNT model.
Authors: Zheng Chu, Jingchang Chen, Qianglong Chen, Weijiang Yu, Tao He, Haotian Wang, Weihua Peng, Ming Liu, Bing Qin, Ting Liu
Abstract: Reasoning, a fundamental cognitive process integral to human intelligence, has garnered substantial interest within artificial intelligence. Notably, recent studies have revealed that chain-of-thought prompting significantly enhances LLM's reasoning capabilities, which attracts widespread attention from both academics and industry. In this paper, we systematically investigate relevant research, summarizing advanced methods through a meticulous taxonomy that offers novel perspectives. Moreover, we delve into the current frontiers and delineate the challenges and future directions, thereby shedding light on future research. Furthermore, we engage in a discussion about open questions. We hope this paper serves as an introduction for beginners and fosters future research. Resources have been made publicly available at https://github.com/zchuz/CoT-Reasoning-Survey
Authors: Junmo Kang, Hongyin Luo, Yada Zhu, Jacob Hansen, James Glass, David Cox, Alan Ritter, Rogerio Feris, Leonid Karlinsky
Abstract: Recent works have demonstrated the effectiveness of self-alignment in which a large language model is aligned to follow general instructions using instructional data generated from the model itself starting from a handful of human-written seeds. Instead of general alignment, in this work, we focus on self-alignment for expert domain specialization (e.g., biomedicine, finance). As a preliminary, we quantitively show the marginal effect that generic instruction-following training has on downstream expert domains' performance. To remedy this, we propose self-specialization - allowing for effective model specialization while achieving cross-task generalization by leveraging only a few labeled seeds. Self-specialization offers a data- and parameter-efficient way of "carving out" an expert model out of a generalist pre-trained LLM. Exploring a variety of popular open large models as a base for specialization, our experimental results in both biomedical and financial domains show that our self-specialized models outperform their base models by a large margin, and even larger models that are generally instruction-tuned or that have been adapted to the target domain by other means.
Authors: Junjie Liu, Shaotian Yan, Chen Shen, Liang Xie, Wenxiao Wang, Jieping Ye
Abstract: Exploiting large language models (LLMs) to tackle reasoning has garnered growing attention. It still remains highly challenging to achieve satisfactory results in complex logical problems, characterized by plenty of premises within the prompt and requiring multi-hop reasoning. In particular, the reasoning capabilities of LLMs are brittle to disorder and distractibility. In this work, we first examine the mechanism from the perspective of information flow and reveal that LLMs exhibit failure patterns akin to human-like cognitive biases when dealing with disordered and irrelevant content in reasoning tasks. However, in contrast to LLMs, disordered and irrelevant content does not significantly decrease human performance, as humans have a propensity to distill the most relevant information and systematically organize their thoughts, aiding them in responding to questions. Stem from that, we further propose a novel reasoning approach named Concise and Organized Perception (COP). COP carefully analyzes the given statements to identify the most pertinent information while eliminating redundancy efficiently. It then prompts the LLMs in a more organized form that adapts to the model's inference process. By perceiving concise and organized context, the reasoning abilities of LLMs can be better elicited. Extensive experimental results on several popular logical benchmarks (ProofWriter, PrOntoQA, PrOntoQA-OOD, and FOLIO) and math benchmark (DI-GSM) show that COP significantly outperforms previous state-of-the-art methods.
Authors: Rasul Tutunov, Antoine Grosnit, Juliusz Ziomek, Jun Wang, Haitham Bou-Ammar
Abstract: This paper delves into the capabilities of large language models (LLMs), specifically focusing on advancing the theoretical comprehension of chain-of-thought prompting. We investigate how LLMs can be effectively induced to generate a coherent chain of thoughts. To achieve this, we introduce a two-level hierarchical graphical model tailored for natural language generation. Within this framework, we establish a compelling geometrical convergence rate that gauges the likelihood of an LLM-generated chain of thoughts compared to those originating from the true language. Our findings provide a theoretical justification for the ability of LLMs to produce the correct sequence of thoughts (potentially) explaining performance gains in tasks demanding reasoning skills.
Authors: Ziyin Zhang, Yikang Liu, Weifang Huang, Junyu Mao, Rui Wang, Hai Hu
Abstract: In this work, we present the largest benchmark to date on linguistic acceptability: Multilingual Evaluation of Linguistic Acceptability -- MELA, with 46K samples covering 10 languages from a diverse set of language families. We establish LLM baselines on this benchmark, and investigate cross-lingual transfer in acceptability judgements with XLM-R. In pursuit of multilingual interpretability, we conduct probing experiments with fine-tuned XLM-R to explore the process of syntax capability acquisition. Our results show that GPT-4o exhibits a strong multilingual ability, outperforming fine-tuned XLM-R, while open-source multilingual models lag behind by a noticeable gap. Cross-lingual transfer experiments show that transfer in acceptability judgment is non-trivial: 500 Icelandic fine-tuning examples lead to 23 MCC performance in a completely unrelated language -- Chinese. Results of our probing experiments indicate that training on MELA improves the performance of XLM-R on syntax-related tasks. Our data is available at https://github.com/sjtu-compling/MELA.
Authors: Serwan Jassim, Mario Holubar, Annika Richter, Cornelius Wolff, Xenia Ohmer, Elia Bruni
Abstract: This paper presents GRASP, a novel benchmark to evaluate the language grounding and physical understanding capabilities of video-based multimodal large language models (LLMs). This evaluation is accomplished via a two-tier approach leveraging Unity simulations. The first level tests for language grounding by assessing a model's ability to relate simple textual descriptions with visual information. The second level evaluates the model's understanding of "Intuitive Physics" principles, such as object permanence and continuity. In addition to releasing the benchmark, we use it to evaluate several state-of-the-art multimodal LLMs. Our evaluation reveals significant shortcomings in the language grounding and intuitive physics capabilities of these models. Although they exhibit at least some grounding capabilities, particularly for colors and shapes, these capabilities depend heavily on the prompting strategy. At the same time, all models perform below or at the chance level of 50% in the Intuitive Physics tests, while human subjects are on average 80% correct. These identified limitations underline the importance of using benchmarks like GRASP to monitor the progress of future models in developing these competencies.
Authors: Yusuke Sakai, Hidetaka Kamigaito, Katsuhiko Hayashi, Taro Watanabe
Abstract: Knowledge graphs (KGs) consist of links that describe relationships between entities. Due to the difficulty of manually enumerating all relationships between entities, automatically completing them is essential for KGs. Knowledge Graph Completion (KGC) is a task that infers unseen relationships between entities in a KG. Traditional embedding-based KGC methods, such as RESCAL, TransE, DistMult, ComplEx, RotatE, HAKE, HousE, etc., infer missing links using only the knowledge from training data. In contrast, the recent Pre-trained Language Model (PLM)-based KGC utilizes knowledge obtained during pre-training. Therefore, PLM-based KGC can estimate missing links between entities by reusing memorized knowledge from pre-training without inference. This approach is problematic because building KGC models aims to infer unseen links between entities. However, conventional evaluations in KGC do not consider inference and memorization abilities separately. Thus, a PLM-based KGC method, which achieves high performance in current KGC evaluations, may be ineffective in practical applications. To address this issue, we analyze whether PLM-based KGC methods make inferences or merely access memorized knowledge. For this purpose, we propose a method for constructing synthetic datasets specified in this analysis and conclude that PLMs acquire the inference abilities required for KGC through pre-training, even though the performance improvements mostly come from textual information of entities and relations.
Authors: Jorge Leandro, Sudha Rao, Michael Xu, Weijia Xu, Nebosja Jojic, Chris Brockett, Bill Dolan
Abstract: Dialogue-based Role Playing Games (RPGs) require powerful storytelling. The narratives of these may take years to write and typically involve a large creative team. In this work, we demonstrate the potential of large generative text models to assist this process. \textbf{GENEVA}, a prototype tool, generates a rich narrative graph with branching and reconverging storylines that match a high-level narrative description and constraints provided by the designer. A large language model (LLM), GPT-4, is used to generate the branching narrative and to render it in a graph format in a two-step process. We illustrate the use of GENEVA in generating new branching narratives for four well-known stories under different contextual constraints. This tool has the potential to assist in game development, simulations, and other applications with game-like properties.
Authors: Kuan-Hao Huang, I-Hung Hsu, Tanmay Parekh, Zhiyu Xie, Zixuan Zhang, Premkumar Natarajan, Kai-Wei Chang, Nanyun Peng, Heng Ji
Abstract: Event extraction has gained considerable interest due to its wide-ranging applications. However, recent studies draw attention to evaluation issues, suggesting that reported scores may not accurately reflect the true performance. In this work, we identify and address evaluation challenges, including inconsistency due to varying data assumptions or preprocessing steps, the insufficiency of current evaluation frameworks that may introduce dataset or data split bias, and the low reproducibility of some previous approaches. To address these challenges, we present TextEE, a standardized, fair, and reproducible benchmark for event extraction. TextEE comprises standardized data preprocessing scripts and splits for 16 datasets spanning eight diverse domains and includes 14 recent methodologies, conducting a comprehensive benchmark reevaluation. We also evaluate five varied large language models on our TextEE benchmark and demonstrate how they struggle to achieve satisfactory performance. Inspired by our reevaluation results and findings, we discuss the role of event extraction in the current NLP era, as well as future challenges and insights derived from TextEE. We believe TextEE, the first standardized comprehensive benchmarking tool, will significantly facilitate future event extraction research.
Authors: Liang Chen, Yatao Bian, Yang Deng, Deng Cai, Shuaiyi Li, Peilin Zhao, Kam-fai Wong
Abstract: Text watermarking has emerged as a pivotal technique for identifying machine-generated text. However, existing methods often rely on arbitrary vocabulary partitioning during decoding to embed watermarks, which compromises the availability of suitable tokens and significantly degrades the quality of responses. This study assesses the impact of watermarking on different capabilities of large language models (LLMs) from a cognitive science lens. Our finding highlights a significant disparity; knowledge recall and logical reasoning are more adversely affected than language generation. These results suggest a more profound effect of watermarking on LLMs than previously understood. To address these challenges, we introduce Watermarking with Mutual Exclusion (WatME), a novel approach leveraging linguistic prior knowledge of inherent lexical redundancy in LLM vocabularies to seamlessly integrate watermarks. Specifically, WatME dynamically optimizes token usage during the decoding process by applying a mutually exclusive rule to the identified lexical redundancies. This strategy effectively prevents the unavailability of appropriate tokens and preserves the expressive power of LLMs. We provide both theoretical analysis and empirical evidence showing that WatME effectively preserves the diverse capabilities of LLMs while ensuring watermark detectability.
Authors: Navreet Kaur, Monojit Choudhury, Danish Pruthi
Abstract: As corporations rush to integrate large language models (LLMs) to their search offerings, it is critical that they provide factually accurate information that is robust to any presuppositions that a user may express. In this work, we introduce UPHILL, a dataset consisting of health-related queries with varying degrees of presuppositions. Using UPHILL, we evaluate the factual accuracy and consistency of InstructGPT, ChatGPT, and BingChat models. We find that while model responses rarely disagree with true health claims (posed as questions), they often fail to challenge false claims: responses from InstructGPT agree with 32% of the false claims, ChatGPT 26% and BingChat 23%. As we increase the extent of presupposition in input queries, the responses from InstructGPT and ChatGPT agree with the claim considerably more often, regardless of its veracity. Responses from BingChat, which rely on retrieved webpages, are not as susceptible. Given the moderate factual accuracy, and the inability of models to consistently correct false assumptions, our work calls for a careful assessment of current LLMs for use in high-stakes scenarios.
Authors: Weiwen Xu, Deng Cai, Zhisong Zhang, Wai Lam, Shuming Shi
Abstract: As humans, we consistently interact with our peers and receive feedback in the form of natural language. This language feedback allows us to maintain appropriate behavior, and rectify potential errors. The question arises naturally: can we use language feedback to align large language models (LLMs)? In contrast to previous research that aligns LLMs with scalar rewards, we present the first systematic exploration of alignment through the lens of language feedback (i.e., judgment). We start with an in-depth investigation of potential methods that can be adapted for aligning LLMs with judgments, revealing that these methods cannot fully capitalize on judgments. To facilitate more effective utilization of judgments, we propose a novel framework, Contrastive Unlikelihood Training (CUT), that allows for fine-grained inappropriate content detection and correction based on judgments. Our results show that, with merely 1317 off-the-shelf judgment data, CUT (LLaMA2-13b) can beat the 175B DaVinci003 and surpass the best baseline by 50.84 points on AlpacaEval. CUT (LLaMA2-chat-13b) can also align LLMs in an iterative fashion using up-to-date model-specific judgments, improving performance from 81.09 to 91.68 points on AlpacaEval. Further analysis suggests that judgments hold greater potential than rewards in LLM alignment.
Authors: Brian Thompson, Mehak Preet Dhaliwal, Peter Frisch, Tobias Domhan, Marcello Federico
Abstract: We show that content on the web is often translated into many languages, and the low quality of these multi-way translations indicates they were likely created using Machine Translation (MT). Multi-way parallel, machine generated content not only dominates the translations in lower resource languages; it also constitutes a large fraction of the total web content in those languages. We also find evidence of a selection bias in the type of content which is translated into many languages, consistent with low quality English content being translated en masse into many lower resource languages, via MT. Our work raises serious concerns about training models such as multilingual large language models on both monolingual and bilingual data scraped from the web.
Authors: Xu Huang, Zhirui Zhang, Xiang Geng, Yichao Du, Jiajun Chen, Shujian Huang
Abstract: This study investigates how Large Language Models (LLMs) leverage source and reference data in machine translation evaluation task, aiming to better understand the mechanisms behind their remarkable performance in this task. We design the controlled experiments across various input modes and model types, and employ both coarse-grained and fine-grained prompts to discern the utility of source versus reference information. We find that reference information significantly enhances the evaluation accuracy, while surprisingly, source information sometimes is counterproductive, indicating LLMs' inability to fully leverage the cross-lingual capability when evaluating translations. Further analysis of the fine-grained evaluation and fine-tuning experiments show similar results. These findings also suggest a potential research direction for LLMs that fully exploits the cross-lingual capability of LLMs to achieve better performance in machine translation evaluation tasks.
Authors: Giorgos Vernikos, Andrei Popescu-Belis
Abstract: Neural machine translation systems estimate probabilities of target sentences given source sentences, yet these estimates may not align with human preferences. This work introduces QE-fusion, a method that synthesizes translations using a quality estimation metric (QE), which correlates better with human judgments. QE-fusion leverages a pool of candidates sampled from a model, combining spans from different candidates using a QE metric such as CometKiwi. We compare QE-fusion against beam search and recent reranking techniques, such as Minimum Bayes Risk decoding or QE-reranking. Our method consistently improves translation quality in terms of COMET and BLEURT scores when applied to large language models (LLMs) used for translation (PolyLM, XGLM, Llama2, Mistral, ALMA, and Tower) and to multilingual translation models (NLLB), over five language pairs. Notably, QE-fusion exhibits larger improvements for LLMs due to their ability to generate diverse outputs. We demonstrate that our approach generates novel translations in over half of the cases and consistently outperforms other methods across varying numbers of candidates (5-200). Furthermore, we empirically establish that QE-fusion scales linearly with the number of candidates in the pool.
Authors: Weixiang Zhao, Shilong Wang, Yulin Hu, Yanyan Zhao, Bing Qin, Xuanyu Zhang, Qing Yang, Dongliang Xu, Wanxiang Che
Abstract: The continual learning (CL) ability is vital for deploying large language models (LLMs) in the dynamic world. Existing methods devise the learning module to acquire task-specific knowledge with parameter-efficient tuning (PET) block and the selection module to pick out the corresponding one for the testing input, aiming at handling the challenges of catastrophic forgetting and knowledge transfer in CL. However, these methods tend to address only one of the challenges, ignoring the potential of aligning the two modules to effectively address catastrophic forgetting and knowledge transfer simultaneously. To this end, we propose a novel Shared Attention Framework (SAPT), to align the PET learning and selection via the Shared Attentive Learning \& Selection module. Extensive Experiments on two CL benchmarks demonstrate the superiority of SAPT. Moreover, SAPT consistently demonstrates its superiority when we scale it to different model sizes (from 770M to 13B), different model architectures (T5 and LLaMA-2) and unseen tasks.
Authors: Zden\v{e}k Kasner, Ond\v{r}ej Du\v{s}ek
Abstract: We analyze the behaviors of open large language models (LLMs) on the task of data-to-text (D2T) generation, i.e., generating coherent and relevant text from structured data. To avoid the issue of LLM training data contamination with standard benchmarks, we design Quintd - a tool for collecting novel structured data records from public APIs. We find that open LLMs (Llama 2, Mistral, and Zephyr) can generate fluent and coherent texts in zero-shot settings from data in common formats collected with Quintd. However, we show that the semantic accuracy of the outputs is a major issue: both according to human annotators and our reference-free metric based on GPT-4, more than 80% of the outputs of open LLMs contain at least one semantic error. We publicly release the code, data, and model outputs.
Authors: Yuang Li, Jiawei Yu, Min Zhang, Mengxin Ren, Yanqing Zhao, Xiaofeng Zhao, Shimin Tao, Jinsong Su, Hao Yang
Abstract: Mapping speech tokens to the same feature space as text tokens has become the paradigm for the integration of speech modality into decoder-only large language models (LLMs). An alternative approach is to use an encoder-decoder architecture that incorporates speech features through cross-attention. This approach, however, has received less attention in the literature. In this work, we connect the Whisper encoder with ChatGLM3 and provide in-depth comparisons of these two approaches using Chinese automatic speech recognition (ASR) and name entity recognition (NER) tasks. We evaluate them not only by conventional metrics like the F1 score but also by a novel fine-grained taxonomy of ASR-NER errors. Our experiments reveal that encoder-decoder architecture outperforms decoder-only architecture with a short context, while decoder-only architecture benefits from a long context as it fully exploits all layers of the LLM. By using LLM, we significantly reduced the entity omission errors and improved the entity ASR accuracy compared to the Conformer baseline. Additionally, we obtained a state-of-the-art (SOTA) F1 score of 0.805 on the AISHELL-NER test set by using chain-of-thought (CoT) NER which first infers long-form ASR transcriptions and then predicts NER labels.
Authors: David Duki\'c, Jan \v{S}najder
Abstract: Pre-trained language models based on masked language modeling (MLM) excel in natural language understanding (NLU) tasks. While fine-tuned MLM-based encoders consistently outperform causal language modeling decoders of comparable size, recent decoder-only large language models (LLMs) perform on par with smaller MLM-based encoders. Although their performance improves with scale, LLMs fall short of achieving state-of-the-art results in information extraction (IE) tasks, many of which are formulated as sequence labeling (SL). We hypothesize that LLMs' poor SL performance stems from causal masking, which prevents the model from attending to tokens on the right of the current token. Yet, how exactly and to what extent LLMs' performance on SL can be improved remains unclear. We explore techniques for improving the SL performance of open LLMs on IE tasks by applying layer-wise removal of the causal mask (CM) during LLM fine-tuning. This approach yields performance gains competitive with state-of-the-art SL models, matching or outperforming the results of CM removal from all blocks. Our findings hold for diverse SL tasks, demonstrating that open LLMs with layer-dependent CM removal outperform strong MLM-based encoders and even instruction-tuned LLMs.
Authors: Berta Franzluebbers, Donald Dunagan, Milo\v{s} Stanojevi\'c, Jan Buys, John T. Hale
Abstract: Humans understand sentences word-by-word, in the order that they hear them. This incrementality entails resolving temporary ambiguities about syntactic relationships. We investigate how humans process these syntactic ambiguities by correlating predictions from incremental generative dependency parsers with timecourse data from people undergoing functional neuroimaging while listening to an audiobook. In particular, we compare competing hypotheses regarding the number of developing syntactic analyses in play during word-by-word comprehension: one vs more than one. This comparison involves evaluating syntactic surprisal from a state-of-the-art dependency parser with LLM-adapted encodings against an existing fMRI dataset. In both English and Chinese data, we find evidence for multipath parsing. Brain regions associated with this multipath effect include bilateral superior temporal gyrus.
Authors: Dongping Chen, Ruoxi Chen, Shilin Zhang, Yinuo Liu, Yaochen Wang, Huichi Zhou, Qihui Zhang, Pan Zhou, Yao Wan, Lichao Sun
Abstract: Multimodal Large Language Models (MLLMs) have gained significant attention recently, showing remarkable potential in artificial general intelligence. However, assessing the utility of MLLMs presents considerable challenges, primarily due to the absence of multimodal benchmarks that align with human preferences. Drawing inspiration from the concept of LLM-as-a-Judge within LLMs, this paper introduces a novel benchmark, termed MLLM-as-a-Judge, to assess the ability of MLLMs in assisting judges across diverse modalities, encompassing three distinct tasks: Scoring Evaluation, Pair Comparison, and Batch Ranking. Our study reveals that, while MLLMs demonstrate remarkable human-like discernment in Pair Comparison, there is a significant divergence from human preferences in Scoring Evaluation and Batch Ranking. Furthermore, a closer examination reveals persistent challenges in the judgment capacities of LLMs, including diverse biases, hallucinatory responses, and inconsistencies in judgment, even in advanced models such as GPT-4V. These findings emphasize the pressing need for enhancements and further research efforts to be undertaken before regarding MLLMs as fully reliable evaluators. In light of this, we advocate for additional efforts dedicated to supporting the continuous development within the domain of MLLM functioning as judges. The code and dataset are publicly available at our project homepage: \url{https://mllm-judge.github.io/}.
Authors: Pragya Srivastava, Satvik Golechha, Amit Deshpande, Amit Sharma
Abstract: Recent work shows that in-context learning and optimization of in-context examples (ICE) can significantly improve the accuracy of large language models (LLMs) on a wide range of tasks, leading to an apparent consensus that ICE optimization is crucial for better performance. However, most of these studies assume a fixed or no instruction provided in the prompt. We challenge this consensus by investigating the necessity of optimizing ICE when task-specific instructions are provided and find that there are many tasks for which it yields diminishing returns. In particular, using a diverse set of tasks and a systematically created instruction set with gradually added details, we find that as the prompt instruction becomes more detailed, the returns on ICE optimization diminish. To characterize this behavior, we introduce a task-specific metric called Normalized Invariability to Choice of Examples (NICE) that quantifies the learnability of tasks from a given instruction, and provides a heuristic to help decide whether to optimize instructions or ICE for a new task. Given a task, the proposed metric can reliably predict the utility of optimizing ICE compared to using random ICE. Our code is available at https://github.com/microsoft/nice-icl.
Authors: Shanshan Xu, T. Y. S. S Santosh, Oana Ichim, Barbara Plank, Matthias Grabmair
Abstract: In legal decisions, split votes (SV) occur when judges cannot reach a unanimous decision, posing a difficulty for lawyers who must navigate diverse legal arguments and opinions. In high-stakes domains, understanding the alignment of perceived difficulty between humans and AI systems is crucial to build trust. However, existing NLP calibration methods focus on a classifier's awareness of predictive performance, measured against the human majority class, overlooking inherent human label variation (HLV). This paper explores split votes as naturally observable human disagreement and value pluralism. We collect judges' vote distributions from the European Court of Human Rights (ECHR), and present SV-ECHR, a case outcome classification (COC) dataset with SV information. We build a taxonomy of disagreement with SV-specific subcategories. We further assess the alignment of perceived difficulty between models and humans, as well as confidence- and human-calibration of COC models. We observe limited alignment with the judge vote distribution. To our knowledge, this is the first systematic exploration of calibration to human judgements in legal NLP. Our study underscores the necessity for further research on measuring and enhancing model calibration considering HLV in legal decision tasks.
Authors: Shir Ashury-Tahan, Ariel Gera, Benjamin Sznajder, Leshem Choshen, Liat Ein-Dor, Eyal Shnarch
Abstract: Model selection for a given target task can be costly, as it may entail extensive annotation of the quality of outputs of different models. We introduce DiffUse, an efficient method to make an informed decision between candidate text generation models based on preference annotations. DiffUse reduces the required amount of annotations, thus saving valuable time and resources in performing evaluation. DiffUse intelligently selects instances by clustering embeddings that represent the semantic differences between model outputs. Thus, it is able to identify a subset of examples that are more informative for preference decisions. Our method is model-agnostic, and can be applied to any text generation model for selecting between models, prompts and configurations. Moreover, we propose a practical iterative approach for dynamically determining how many instances to annotate. In a series of experiments over hundreds of model pairs, we demonstrate that DiffUse can dramatically reduce the required number of annotations -- by up to 75% -- while maintaining high evaluation reliability.
Authors: Nur Lan, Emmanuel Chemla, Roni Katzir
Abstract: Neural networks offer good approximation to many tasks but consistently fail to reach perfect generalization, even when theoretical work shows that such perfect solutions can be expressed by certain architectures. Using the task of formal language learning, we focus on one simple formal language and show that the theoretically correct solution is in fact not an optimum of commonly used objectives -- even with regularization techniques that according to common wisdom should lead to simple weights and good generalization (L1, L2) or other meta-heuristics (early-stopping, dropout). On the other hand, replacing standard targets with the Minimum Description Length objective (MDL) results in the correct solution being an optimum.
Authors: Weixiang Zhao, Zhuojun Li, Shilong Wang, Yang Wang, Yulin Hu, Yanyan Zhao, Chen Wei, Bing Qin
Abstract: Emotional Intelligence (EI), consisting of emotion perception, emotion cognition and emotion expression, plays the critical roles in improving user interaction experience for the current large language model (LLM) based conversational general AI assistants. Previous works mainly focus on raising the emotion perception ability of them via naive fine-tuning on EI-related classification or regression tasks. However, this leads to the incomplete enhancement of EI and catastrophic forgetting of the general intelligence (GI). To this end, we first introduce \textsc{EiBench}, a large-scale collection of EI-related tasks in the text-to-text formation with task instructions that covers all three aspects of EI, which lays a solid foundation for the comprehensive EI enhancement of LLMs. Then a novel \underline{\textbf{Mo}}dular \underline{\textbf{E}}motional \underline{\textbf{I}}ntelligence enhancement method (\textbf{MoEI}), consisting of Modular Parameter Expansion and intra-inter modulation, is proposed to comprehensively enhance the EI of LLMs without compromise their GI. Extensive experiments on two representative LLM-based assistants, Flan-T5 and LLaMA-2-Chat, demonstrate the effectiveness of MoEI to improving EI while maintain GI.
Authors: Ioannis Tsiamas, Gerard I. G\'allego, Jos\'e A. R. Fonollosa, Marta R. Costa-juss\`a
Abstract: Data scarcity and the modality gap between the speech and text modalities are two major obstacles of end-to-end Speech Translation (ST) systems, thus hindering their performance. Prior work has attempted to mitigate these challenges by leveraging external MT data and optimizing distance metrics that bring closer the speech-text representations. However, achieving competitive results typically requires some ST data. For this reason, we introduce ZeroSwot, a method for zero-shot ST that bridges the modality gap without any paired ST data. Leveraging a novel CTC compression and Optimal Transport, we train a speech encoder using only ASR data, to align with the representation space of a massively multilingual MT model. The speech encoder seamlessly integrates with the MT model at inference, enabling direct translation from speech to text, across all languages supported by the MT model. Our experiments show that we can effectively close the modality gap without ST data, while our results on MuST-C and CoVoST demonstrate our method's superiority over not only previous zero-shot models, but also supervised ones, achieving state-of-the-art results.
Authors: Afra Amini, Tim Vieira, Ryan Cotterell
Abstract: Direct preference optimization (DPO) is a successful fine-tuning strategy for aligning large language models with human preferences without the need to train a reward model or employ reinforcement learning. DPO, as originally formulated, relies on binary preference data and fine-tunes a language model to increase the likelihood of a preferred response over a dispreferred response. However, not all preference pairs are equal. Sometimes, the preferred response is only slightly better than the dispreferred one. In other cases, the preference is much stronger. For instance, if a response contains harmful or toxic content, the annotator will have a strong preference for that response. In this paper, we propose a generalization of DPO, termed DPO with an offset (ODPO), that does not treat every preference pair equally during fine-tuning. Intuitively, ODPO requires the difference between the likelihood of the preferred and dispreferred response to be greater than an offset value. The offset is determined based on the extent to which one response is preferred over another. Our experiments on various tasks suggest that ODPO significantly outperforms DPO in aligning language models, especially when the number of preference pairs is limited.
Authors: Chris Wendler, Veniamin Veselovsky, Giovanni Monea, Robert West
Abstract: We ask whether multilingual language models trained on unbalanced, English-dominated corpora use English as an internal pivot language -- a question of key importance for understanding how language models function and the origins of linguistic bias. Focusing on the Llama-2 family of transformer models, our study uses carefully constructed non-English prompts with a unique correct single-token continuation. From layer to layer, transformers gradually map an input embedding of the final prompt token to an output embedding from which next-token probabilities are computed. Tracking intermediate embeddings through their high-dimensional space reveals three distinct phases, whereby intermediate embeddings (1) start far away from output token embeddings; (2) already allow for decoding a semantically correct next token in the middle layers, but give higher probability to its version in English than in the input language; (3) finally move into an input-language-specific region of the embedding space. We cast these results into a conceptual model where the three phases operate in "input space", "concept space", and "output space", respectively. Crucially, our evidence suggests that the abstract "concept space" lies closer to English than to other languages, which may have important consequences regarding the biases held by multilingual language models.
Authors: Tuc Nguyen, Thai Le
Abstract: Several parameter-efficient fine-tuning methods based on adapters have been proposed as a streamlined approach to incorporate not only a single specialized knowledge into existing Pre-Trained Language Models (PLMs) but also multiple of them at once. Recent works such as AdapterSoup propose to mix not all but only a selective sub-set of domain-specific adapters during inference via model weight averaging to optimize performance on novel, unseen domains with excellent computational efficiency. However, the essential generalizability of this emerging weight-space adapter mixing mechanism on \textit{unseen, in-domain examples} remains unexplored. Thus, in this study, we conduct a comprehensive analysis to elucidate the generalizability of domain-specific adapter mixtures in in-domain evaluation. We also provide investigations into the inner workings of the mixture of domain-specific adapters by analyzing their weight signs, yielding critical analysis on the negative correlation between their fraction of weight sign difference and their mixtures' generalizability.
Authors: Ziru Chen, Michael White, Raymond Mooney, Ali Payani, Yu Su, Huan Sun
Abstract: In this paper, we examine how large language models (LLMs) solve multi-step problems under a language agent framework with three components: a generator, a discriminator, and a planning method. We investigate the practical utility of two advanced planning methods, iterative correction and tree search. We present a comprehensive analysis of how discrimination accuracy affects the overall performance of agents when using these two methods or a simpler method, re-ranking. Experiments on two tasks, text-to-SQL parsing and mathematical reasoning, show that: (1) advanced planning methods demand discriminators with at least 90% accuracy to achieve significant improvements over re-ranking; (2) current LLMs' discrimination abilities have not met the needs of advanced planning methods to achieve such improvements; (3) with LLM-based discriminators, advanced planning methods may not adequately balance accuracy and efficiency. For example, compared to the other two methods, tree search is at least 10--20 times slower but leads to negligible performance gains, which hinders its real-world applications. Code and data are available at https://github.com/OSU-NLP-Group/llm-planning-eval.
Authors: Tianyi Lorena Yan, Fei Wang, James Y. Huang, Wenxuan Zhou, Fan Yin, Aram Galstyan, Wenpeng Yin, Muhao Chen
Abstract: Instruction tuning has been used as a promising approach to improve the performance of large language models (LLMs) on unseen tasks. However, current LLMs exhibit limited robustness to unseen instructions, generating inconsistent outputs when the same instruction is phrased with slightly varied forms or language styles. This behavior indicates LLMs' lack of robustness to textual variations and generalizability to unseen instructions, potentially leading to trustworthiness issues. Accordingly, we propose Contrastive Instruction Tuning, which maximizes the similarity between the hidden representations of semantically equivalent instruction-instance pairs while minimizing the similarity between semantically different ones. To facilitate this approach, we augment the existing FLAN collection by paraphrasing task instructions. Experiments on the PromptBench benchmark show that CoIN consistently improves LLMs' robustness to unseen instructions with variations across character, word, sentence, and semantic levels by an average of +2.5% in accuracy. Code is available at https://github.com/luka-group/CoIN.
Authors: Bruce W. Lee, JaeHyuk Lim
Abstract: We argue that language-only models don't learn the physical manifestation of language. We present an empirical investigation of visual-auditory properties of language through a series of tasks, termed H-Test. These tasks highlight a fundamental gap between human linguistic understanding and the sensory-deprived linguistic understanding of LLMs. In support of our hypothesis, 1. deliberate reasoning (Chain-of-Thought), 2. few-shot examples, or 3. stronger LLM from the same model family (LLaMA 2 13B -> LLaMA 2 70B) has no significant effect on H-Test performance.
We bring in the philosophical case of Mary, who learns about the world in a sensory-deprived environment as a useful conceptual framework to understand how language-only models learn about the world (Jackson, 1986). Our experiments show that some of the strongest proprietary LLMs stay near random chance baseline accuracy of 50%, highlighting the limitations of linguistic knowledge acquired in the absence of sensory experience. Our code and data are available at
Authors: Ikuya Yamada, Ryokan Ri
Abstract: Adapting English-based large language models (LLMs) to other languages has become increasingly popular due to the efficiency and potential of cross-lingual transfer. However, existing language adaptation methods often overlook the benefits of cross-lingual supervision. In this study, we introduce LEIA, a language adaptation tuning method that utilizes Wikipedia entity names aligned across languages. This method involves augmenting the target language corpus with English entity names and training the model using left-to-right language modeling. We assess LEIA on diverse question answering datasets using 7B-parameter LLMs, demonstrating significant performance gains across various non-English languages. The source code is available at https://github.com/studio-ousia/leia.
Authors: Zijin Hong, Zheng Yuan, Hao Chen, Qinggang Zhang, Feiran Huang, Xiao Huang
Abstract: Generating accurate SQL queries for user questions (text-to-SQL) has been a long-standing challenge since it requires a deep understanding of both the user's question and the corresponding database schema in order to retrieve the desired content accurately. Existing methods rely on the comprehensive capability of large language models (LLMs) to generate the SQL. However, some necessary knowledge is not explicitly included in the database schema and user question or has been learned by LLMs. Thus, the generated SQL of the knowledge-insufficient questions may be inaccurate, negatively influencing the text-to-SQL models' performance and robustness. To address this challenge, we propose the Knowledge-to-SQL framework, which employs tailored Data Expert LLM (DELLM) to provide helpful knowledge for all text-to-SQL models. Specifically, we introduce the detailed implementation of DELLM regarding table reading and the basic fine-tuning process. We further propose a Preference Learning via Database Feedback (PLDBF) strategy, refining the DELLM to generate more helpful knowledge for LLMs. Extensive experiments verify that DELLM can enhance the state-of-the-art approaches for text-to-SQL tasks. The corresponding code of DELLM is released for further research.
Authors: Guijin Son, Hanwool Lee, Sungdong Kim, Seungone Kim, Niklas Muennighoff, Taekyoon Choi, Cheonbok Park, Kang Min Yoo, Stella Biderman
Abstract: We propose KMMLU, a new Korean benchmark with 35,030 expert-level multiple-choice questions across 45 subjects ranging from humanities to STEM. While prior Korean benchmarks are translated from existing English benchmarks, KMMLU is collected from original Korean exams, capturing linguistic and cultural aspects of the Korean language. We test 27 public and proprietary LLMs and observe the best public model to score 50.5%, leaving significant room for improvement. This model was primarily trained for English and Chinese, not Korean. Current LLMs tailored to Korean, such as Polyglot-Ko, perform far worse. Surprisingly, even the most capable proprietary LLMs, e.g., GPT-4 and HyperCLOVA X do not exceed 60%. This suggests that further work is needed to improve LLMs for Korean, and we believe KMMLU offers the appropriate tool to track this progress. We make our dataset publicly available on the Hugging Face Hub and integrate the benchmark into EleutherAI's Language Model Evaluation Harness.
Authors: Guijin Son, Sangwon Baek, Sangdae Nam, Ilgyun Jeong, Seungone Kim
Abstract: Large language models (LLMs) are typically prompted to follow a single instruction per inference call. In this work, we analyze whether LLMs also hold the capability to handle multiple instructions simultaneously, denoted as Multi-Task Inference. For this purpose, we introduce the MTI Bench(Multi-Task Inference Benchmark), a comprehensive evaluation benchmark encompassing 5,000 instances across 25 tasks. Each task in the MTI Bench involves 2 to 3 sub-tasks. As expected, we first demonstrate that Multi-Task Inference reduces the total inference time by 1.46 times in average since it does not require multiple inference calls. Interestingly, contrary to the expectation that LLMs would perform better when tasks are divided, we find that state-of-the-art LLMs, such as Llama-2-Chat-70B and GPT-4, show up to 7.3% and 12.4% improved performance with Multi-Task Inference compared to Single-Task Inference on the MTI Bench. We release the MTI Bench dataset and our code at this link https://github.com/guijinSON/MTI-Bench.
Authors: Jiahao Ying, Yixin Cao, Yushi Bai, Qianru Sun, Bo Wang, Wei Tang, Zhaojun Ding, Yizhe Yang, Xuanjing Huang, Shuicheng Yan
Abstract: Large language models (LLMs) have achieved impressive performance across various natural language benchmarks, prompting a continual need to curate more difficult datasets for larger LLMs, which is costly and time-consuming. In this paper, we propose to automate dataset updating and provide systematic analysis regarding its effectiveness in dealing with benchmark leakage issue, difficulty control, and stability. Thus, once the current benchmark has been mastered or leaked, we can update it for timely and reliable evaluation. There are two updating strategies: 1) mimicking strategy to generate similar samples based on original data, preserving stylistic and contextual essence, and 2) extending strategy that further expands existing samples at varying cognitive levels by adapting Bloom's taxonomy of educational objectives. Extensive experiments on updated MMLU and BIG-Bench demonstrate the stability of the proposed strategies and find that the mimicking strategy can effectively alleviate issues of overestimation from benchmark leakage. In cases where the efficient mimicking strategy fails, our extending strategy still shows promising results. Additionally, by controlling the difficulty, we can better discern the models' performance and enable fine-grained analysis neither too difficult nor too easy an exam can fairly judge students' learning status. To the best of our knowledge, we are the first to automate updating benchmarks for reliable and timely evaluation. Our demo leaderboard can be found at https://yingjiahao14.github.io/Automating-DatasetUpdates/.
URLs: https://yingjiahao14.github.io/Automating-DatasetUpdates/.
Authors: Zhanhui Zhou, Jie Liu, Zhichen Dong, Jiaheng Liu, Chao Yang, Wanli Ouyang, Yu Qiao
Abstract: Large language models (LLMs) undergo safety alignment to ensure safe conversations with humans. However, this paper introduces a training-free attack method capable of reversing safety alignment, converting the outcomes of stronger alignment into greater potential for harm by accessing only LLM output token distributions. Specifically, our method achieves this reversal by contrasting the output token distribution of a safety-aligned language model (e.g., Llama-2-chat) against its pre-trained version (e.g., Llama-2), so that the token predictions are shifted towards the opposite direction of safety alignment. We name this method emulated disalignment (ED) because sampling from this contrastive distribution provably emulates the result of fine-tuning to minimize a safety reward. Our experiments with ED across three evaluation datasets and four model families (Llama-1, Llama-2, Mistral, and Alpaca) show that ED doubles the harmfulness of pre-trained models and outperforms strong baselines, achieving the highest harmful rates in 43 out of 48 evaluation subsets by a large margin. Eventually, given ED's reliance on language model output token distributions, which particularly compromises open-source models, our findings highlight the need to reassess the open accessibility of language models, even if they have been safety-aligned. Code is available at https://github.com/ZHZisZZ/emulated-disalignment.
Authors: Ryo Yoshida, Taiga Someya, Yohei Oseki
Abstract: Syntactic Language Models (SLMs) can be trained efficiently to reach relatively high performance; however, they have trouble with inference efficiency due to the explicit generation of syntactic structures. In this paper, we propose a new method dubbed tree-planting: instead of explicitly generating syntactic structures, we "plant" trees into attention weights of unidirectional Transformer LMs to implicitly reflect syntactic structures of natural language. Specifically, unidirectional Transformer LMs trained with tree-planting will be called Tree-Planted Transformers (TPT), which inherit the training efficiency from SLMs without changing the inference efficiency of their underlying Transformer LMs. Targeted syntactic evaluations on the SyntaxGym benchmark demonstrated that TPTs, despite the lack of explicit generation of syntactic structures, significantly outperformed not only vanilla Transformer LMs but also various SLMs that generate hundreds of syntactic structures in parallel. This result suggests that TPTs can learn human-like syntactic knowledge as data-efficiently as SLMs while maintaining the modeling space of Transformer LMs unchanged.
Authors: Han Wang, Archiki Prasad, Elias Stengel-Eskin, Mohit Bansal
Abstract: Generations from large language models (LLMs) can be improved by sampling and scoring multiple solutions to select a final answer. Current "sample and select" methods such as self-consistency (SC) rely on majority voting to score answers. However, when tasks have many distinct and valid answers, selection by voting requires a large number of samples. This makes SC prohibitively expensive for interactive tasks that involve generating multiple actions (answers) sequentially. After establishing that majority voting fails to provide consistent gains on such tasks, we demonstrate how to increase success rates by softening the scoring criterion. We introduce Soft Self-Consistency (SOFT-SC), which replaces SC's discontinuous scoring with a continuous score computed from model likelihoods, allowing for selection even when actions are sparsely distributed. SOFT-SC improves both performance and efficiency on long-horizon interactive tasks, requiring half as many samples as SC for comparable or better performance. For a fixed number of samples, SOFT-SC leads to a 1.3% increase over SC in absolute success rate on writing bash programs, a 6.6% increase on online shopping (WebShop), and a 4.7% increase for an interactive household game (ALFWorld). Finally, we show that SOFT-SC can be applied to both open-source and black-box models.
Authors: Haoyu Liu, Jianfeng Liu, Shaohan Huang, Yuefeng Zhan, Hao Sun, Weiwei Deng, Furu Wei, Qi Zhang
Abstract: The remarkable capability of large language models (LLMs) for in-context learning (ICL) needs to be activated by demonstration examples. Prior work has extensively explored the selection of examples for ICL, predominantly following the "select then organize" paradigm, such approaches often neglect the internal relationships between examples and exist an inconsistency between the training and inference. In this paper, we formulate the problem as a $Se$quential $Se$lection problem and introduce $Se^2$, a sequential-aware method that leverages the LLM's feedback on varying context, aiding in capturing inter-relationships and sequential information among examples, significantly enriching the contextuality and relevance of ICL prompts. Meanwhile, we utilize beam search to seek and construct example sequences, enhancing both quality and diversity. Extensive experiments across 23 NLP tasks from 8 distinct categories illustrate that $Se^2$ markedly surpasses competitive baselines and achieves 42\% relative improvement over random selection. Further in-depth analysis shows the effectiveness of proposed strategies, highlighting $Se^2$'s exceptional stability and adaptability across various scenarios. Code available at https://github.com/microsoft/LMOps.
Authors: Chaoqun He, Renjie Luo, Yuzhuo Bai, Shengding Hu, Zhen Leng Thai, Junhao Shen, Jinyi Hu, Xu Han, Yujie Huang, Yuxiang Zhang, Jie Liu, Lei Qi, Zhiyuan Liu, Maosong Sun
Abstract: Recent advancements have seen Large Language Models (LLMs) and Large Multimodal Models (LMMs) surpassing general human capabilities in various tasks, approaching the proficiency level of human experts across multiple domains. With traditional benchmarks becoming less challenging for these models, new rigorous challenges are essential to gauge their advanced abilities. In this work, we present OlympiadBench, an Olympiad-level bilingual multimodal scientific benchmark, featuring 8,476 problems from Olympiad-level mathematics and physics competitions, including the Chinese college entrance exam. Each problem is detailed with expert-level annotations for step-by-step reasoning. Evaluating top-tier models on OlympiadBench, we implement a comprehensive assessment methodology to accurately evaluate model responses. Notably, the best-performing model, GPT-4V, attains an average score of 17.97% on OlympiadBench, with a mere 10.74% in physics, highlighting the benchmark rigor and the intricacy of physical reasoning. Our analysis orienting GPT-4V points out prevalent issues with hallucinations, knowledge omissions, and logical fallacies. We hope that our challenging benchmark can serve as a valuable resource for helping future AGI research endeavors. The data and evaluation code are available at \url{https://github.com/OpenBMB/OlympiadBench}
Authors: Andrew Zhu, Alyssa Hwang, Liam Dugan, Chris Callison-Burch
Abstract: One type of question that is commonly found in day-to-day scenarios is ``fan-out'' questions, complex multi-hop, multi-document reasoning questions that require finding information about a large number of entities. However, there exist few resources to evaluate this type of question-answering capability among large language models. To evaluate complex reasoning in LLMs more fully, we present FanOutQA, a high-quality dataset of fan-out question-answer pairs and human-annotated decompositions with English Wikipedia as the knowledge base. We formulate three benchmark settings across our dataset and benchmark 7 LLMs, including GPT-4, LLaMA 2, Claude-2.1, and Mixtral-8x7B, finding that contemporary models still have room to improve reasoning over inter-document dependencies in a long context. We provide our dataset and open-source tools to run models to encourage evaluation at https://fanoutqa.com
URLs: https://fanoutqa.com
Authors: Bin Liang, Ang Li, Jingqian Zhao, Lin Gui, Min Yang, Yue Yu, Kam-Fai Wong, Ruifeng Xu
Abstract: Stance detection is a challenging task that aims to identify public opinion from social media platforms with respect to specific targets. Previous work on stance detection largely focused on pure texts. In this paper, we study multi-modal stance detection for tweets consisting of texts and images, which are prevalent in today's fast-growing social media platforms where people often post multi-modal messages. To this end, we create five new multi-modal stance detection datasets of different domains based on Twitter, in which each example consists of a text and an image. In addition, we propose a simple yet effective Targeted Multi-modal Prompt Tuning framework (TMPT), where target information is leveraged to learn multi-modal stance features from textual and visual modalities. Experimental results on our five benchmark datasets show that the proposed TMPT achieves state-of-the-art performance in multi-modal stance detection.
Authors: Zhaoyi Li, Gangwei Jiang, Hong Xie, Linqi Song, Defu Lian, Ying Wei
Abstract: LLMs have marked a revolutonary shift, yet they falter when faced with compositional reasoning tasks. Our research embarks on a quest to uncover the root causes of compositional reasoning failures of LLMs, uncovering that most of them stem from the improperly generated or leveraged implicit reasoning results. Inspired by our empirical findings, we resort to Logit Lens and an intervention experiment to dissect the inner hidden states of LLMs. This deep dive reveals that implicit reasoning results indeed surface within middle layers and play a causative role in shaping the final explicit reasoning results. Our exploration further locates multi-head self-attention (MHSA) modules within these layers, which emerge as the linchpins in accurate generation and leveraing of implicit reasoning results. Grounded on the above findings, we develop CREME, a lightweight method to patch errors in compositional reasoning via editing the located MHSA modules. Our empirical evidence stands testament to CREME's effectiveness, paving the way for autonomously and continuously enhancing compositional reasoning capabilities in language models.
Authors: Zhisheng Lin, Han Fu, Chenghao Liu, Zhuo Li, Jianling Sun
Abstract: Parameter-efficient fine-tuning (PEFT) has emerged as an effective method for adapting pre-trained language models to various tasks efficiently. Recently, there has been a growing interest in transferring knowledge from one or multiple tasks to the downstream target task to achieve performance improvements. However, current approaches typically either train adapters on individual tasks or distill shared knowledge from source tasks, failing to fully exploit task-specific knowledge and the correlation between source and target tasks. To overcome these limitations, we propose PEMT, a novel parameter-efficient fine-tuning framework based on multi-task transfer learning. PEMT extends the mixture-of-experts (MoE) framework to capture the transferable knowledge as a weighted combination of adapters trained on source tasks. These weights are determined by a gated unit, measuring the correlation between the target and each source task using task description prompt vectors. To fully exploit the task-specific knowledge, we also propose the Task Sparsity Loss to improve the sparsity of the gated unit. We conduct experiments on a broad range of tasks over 17 datasets. The experimental results demonstrate our PEMT yields stable improvements over full fine-tuning, and state-of-the-art PEFT and knowledge transferring methods on various tasks. The results highlight the effectiveness of our method which is capable of sufficiently exploiting the knowledge and correlation features across multiple tasks.
Authors: Yanzheng Xiang, Hanqi Yan, Lin Gui, Yulan He
Abstract: In-context learning has become a popular paradigm in natural language processing. However, its performance can be significantly influenced by the order of in-context demonstration examples. In this paper, we found that causal language models (CausalLMs) are more sensitive to this order compared to prefix language models (PrefixLMs). We attribute this phenomenon to the auto-regressive attention masks within CausalLMs, which restrict each token from accessing information from subsequent tokens. This results in different receptive fields for samples at different positions, thereby leading to representation disparities across positions. To tackle this challenge, we introduce an unsupervised fine-tuning method, termed the Information-Augmented and Consistency-Enhanced approach. This approach utilizes contrastive learning to align representations of in-context examples across different positions and introduces a consistency loss to ensure similar representations for inputs with different permutations. This enhances the model's predictive consistency across permutations. Experimental results on five benchmarks suggest that our proposed method can reduce the sensitivity of CausalLMs to the order of in-context examples and exhibit robust generalizability, particularly when demonstrations are sourced from a candidate pool different from that used in the training phase, or when the number of in-context examples differs from what is used during training.
Authors: Tianyi Tang, Wenyang Luo, Haoyang Huang, Dongdong Zhang, Xiaolei Wang, Xin Zhao, Furu Wei, Ji-Rong Wen
Abstract: Large language models (LLMs) demonstrate remarkable multilingual capabilities without being pre-trained on specially curated multilingual parallel corpora. It remains a challenging problem to explain the underlying mechanisms by which LLMs process multilingual texts. In this paper, we delve into the composition of Transformer architectures in LLMs to pinpoint language-specific regions. Specially, we propose a novel detection method, language activation probability entropy (LAPE), to identify language-specific neurons within LLMs. Based on LAPE, we conduct comprehensive experiments on several representative LLMs, such as LLaMA-2, BLOOM, and Mistral. Our findings indicate that LLMs' proficiency in processing a particular language is predominantly due to a small subset of neurons, primarily situated in the models' top and bottom layers. Furthermore, we showcase the feasibility to "steer" the output language of LLMs by selectively activating or deactivating language-specific neurons. Our research provides important evidence to the understanding and exploration of the multilingual capabilities of LLMs.
Authors: Renren Jin, Jiangcun Du, Wuwei Huang, Wei Liu, Jian Luan, Bin Wang, Deyi Xiong
Abstract: Increasing the number of parameters in large language models (LLMs) usually improves performance in downstream tasks but raises compute and memory costs, making deployment difficult in resource-limited settings. Quantization techniques, which reduce the bits needed for model weights or activations with minimal performance loss, have become popular due to the rise of LLMs. However, most quantization studies use pre-trained LLMs, and the impact of quantization on instruction-tuned LLMs and the relationship between perplexity and benchmark performance of quantized LLMs are not well understood. Evaluation of quantized LLMs is often limited to language modeling and a few classification tasks, leaving their performance on other benchmarks unclear. To address these gaps, we propose a structured evaluation framework consisting of three critical dimensions: (1) knowledge \& capacity, (2) alignment, and (3) efficiency, and conduct extensive experiments across ten diverse benchmarks. Our experimental results indicate that LLMs with 4-bit quantization can retain performance comparable to their non-quantized counterparts, and perplexity can serve as a proxy metric for quantized LLMs on most benchmarks. Furthermore, quantized LLMs with larger parameter scales can outperform smaller LLMs. Despite the memory savings achieved through quantization, it can also slow down the inference speed of LLMs. Consequently, substantial engineering efforts and hardware support are imperative to achieve a balanced optimization of decoding speed and memory consumption in the context of quantized LLMs.
Authors: Mansi Goel, Ayush Agarwal, Shubham Agrawal, Janak Kapuriya, Akhil Vamshi Konam, Rishabh Gupta, Shrey Rastogi, Niharika, Ganesh Bagler
Abstract: Food touches our lives through various endeavors, including flavor, nourishment, health, and sustainability. Recipes are cultural capsules transmitted across generations via unstructured text. Automated protocols for recognizing named entities, the building blocks of recipe text, are of immense value for various applications ranging from information extraction to novel recipe generation. Named entity recognition is a technique for extracting information from unstructured or semi-structured data with known labels. Starting with manually-annotated data of 6,611 ingredient phrases, we created an augmented dataset of 26,445 phrases cumulatively. Simultaneously, we systematically cleaned and analyzed ingredient phrases from RecipeDB, the gold-standard recipe data repository, and annotated them using the Stanford NER. Based on the analysis, we sampled a subset of 88,526 phrases using a clustering-based approach while preserving the diversity to create the machine-annotated dataset. A thorough investigation of NER approaches on these three datasets involving statistical, fine-tuning of deep learning-based language models and few-shot prompting on large language models (LLMs) provides deep insights. We conclude that few-shot prompting on LLMs has abysmal performance, whereas the fine-tuned spaCy-transformer emerges as the best model with macro-F1 scores of 95.9%, 96.04%, and 95.71% for the manually-annotated, augmented, and machine-annotated datasets, respectively.
Authors: Shiyao Li, Xuefei Ning, Luning Wang, Tengxuan Liu, Xiangsheng Shi, Shengen Yan, Guohao Dai, Huazhong Yang, Yu Wang
Abstract: Post-training quantization (PTQ) has emerged as a promising technique to reduce the cost of large language models (LLMs). Specifically, PTQ can effectively mitigate memory consumption and reduce computational overhead in LLMs. To meet the requirements of both high efficiency and performance across diverse scenarios, a comprehensive evaluation of quantized LLMs is essential to guide the selection of quantization methods. This paper presents a thorough evaluation of these factors by evaluating the effect of PTQ on Weight, Activation, and KV Cache on 11 model families, including OPT, LLaMA2, Falcon, Bloomz, Mistral, ChatGLM, Vicuna, LongChat, StableLM, Gemma, and Mamba, with parameters ranging from 125M to 180B. The evaluation encompasses five types of tasks: basic NLP, emergent ability, trustworthiness, dialogue, and long-context tasks. Moreover, we also evaluate the state-of-the-art (SOTA) quantization methods to demonstrate their applicability. Based on the extensive experiments, we systematically summarize the effect of quantization, provide recommendations to apply quantization techniques, and point out future directions. The code can be found in https://github.com/thu-nics/qllm-eval.
Authors: Nihal V. Nayak, Yiyang Nan, Avi Trost, Stephen H. Bach
Abstract: We introduce Bonito, an open-source model for conditional task generation that converts unannotated text into task-specific training datasets for instruction tuning. We aim to enable zero-shot task adaptation of large language models on users' specialized, private data. We train Bonito by fine-tuning a pretrained large language model on a new large-scale dataset with 1.65M examples created by remixing existing instruction tuning datasets into meta-templates. The meta-templates for a dataset produce training examples where the input is the unannotated text and the task attribute and the output consists of the instruction and the response. We use Bonito to generate synthetic tasks for seven datasets from specialized domains with unannotated text across three task types -- yes-no question answering, extractive question answering, and natural language inference -- and adapt language models. We show that Bonito significantly improves the average performance of pretrained and instruction tuned models over the de facto self supervised baseline. For example, adapting Mistral-Instruct-v2 and instruction tuned variants of Mistral and Llama2 with Bonito improves the strong zero-shot performance by 22.1 F1 points whereas the next word prediction objective undoes some of the benefits of instruction tuning and reduces the average performance by 0.8 F1 points. We conduct additional experiments with Bonito to understand the effects of the domain, the size of the training set, and the choice of alternative synthetic task generators. Overall, we show that learning with synthetic instruction tuning datasets is an effective way to adapt language models to new domains. The model, dataset, and code are available at https://github.com/BatsResearch/bonito.
Authors: Linhai Zhang, Jialong Wu, Deyu Zhou, Guoqiang Xu
Abstract: Though Large Language Models (LLMs) have demonstrated the powerful capabilities of few-shot learning through prompting methods, supervised training is still necessary for complex reasoning tasks. Because of their extensive parameters and memory consumption, both Parameter-Efficient Fine-Tuning (PEFT) methods and Memory-Efficient Fine-Tuning methods have been proposed for LLMs. Nevertheless, the issue of large annotated data consumption, the aim of Data-Efficient Fine-Tuning, remains unexplored. One obvious way is to combine the PEFT method with active learning. However, the experimental results show that such a combination is not trivial and yields inferior results. Through probe experiments, such observation might be explained by two main reasons: uncertainty gap and poor model calibration. Therefore, in this paper, we propose a novel approach to effectively integrate uncertainty-based active learning and LoRA. Specifically, for the uncertainty gap, we introduce a dynamic uncertainty measurement that combines the uncertainty of the base model and the uncertainty of the full model during the iteration of active learning. For poor model calibration, we incorporate the regularization method during LoRA training to keep the model from being over-confident, and the Monte-Carlo dropout mechanism is employed to enhance the uncertainty estimation. Experimental results show that the proposed approach outperforms existing baseline models on three complex reasoning tasks.
Authors: Jialong Wu, Linhai Zhang, Deyu Zhou, Guoqiang Xu
Abstract: Though notable progress has been made, neural-based aspect-based sentiment analysis (ABSA) models are prone to learn spurious correlations from annotation biases, resulting in poor robustness on adversarial data transformations. Among the debiasing solutions, causal inference-based methods have attracted much research attention, which can be mainly categorized into causal intervention methods and counterfactual reasoning methods. However, most of the present debiasing methods focus on single-variable causal inference, which is not suitable for ABSA with two input variables (the target aspect and the review). In this paper, we propose a novel framework based on multi-variable causal inference for debiasing ABSA. In this framework, different types of biases are tackled based on different causal intervention methods. For the review branch, the bias is modeled as indirect confounding from context, where backdoor adjustment intervention is employed for debiasing. For the aspect branch, the bias is described as a direct correlation with labels, where counterfactual reasoning is adopted for debiasing. Extensive experiments demonstrate the effectiveness of the proposed method compared to various baselines on the two widely used real-world aspect robustness test set datasets.
Authors: Leon Weber-Genzel, Siyao Peng, Marie-Catherine de Marneffe, Barbara Plank
Abstract: Human label variation arises when annotators assign different labels to the same item for valid reasons, while annotation errors occur when labels are assigned for invalid reasons. These two issues are prevalent in NLP benchmarks, yet existing research has studied them in isolation. To the best of our knowledge, there exists no prior work that focuses on teasing apart error from signal, especially in cases where signal is beyond black-and-white. To fill this gap, we introduce a systematic methodology and a new dataset, VariErr (variation versus error), focusing on the NLI task in English. We propose a 2-round annotation procedure with annotators explaining each label and subsequently judging the validity of label-explanation pairs. VariErr contains 7,732 validity judgments on 1,933 explanations for 500 re-annotated MNLI items. We assess the effectiveness of various automatic error detection (AED) methods and GPTs in uncovering errors versus human label variation. We find that state-of-the-art AED methods significantly underperform GPTs and humans. While GPT-4 is the best system, it still falls short of human performance. Our methodology is applicable beyond NLI, offering fertile ground for future research on error versus plausible variation, which in turn can yield better and more trustworthy NLP systems.
Authors: Saeed Najafi, Alona Fyshe
Abstract: Pre-trained Language Models (PLMs) can be accurately fine-tuned for downstream text processing tasks. Recently, researchers have introduced several parameter-efficient fine-tuning methods that optimize input prompts or adjust a small number of model parameters (e.g LoRA). In this study, we explore the impact of altering the input text of the original task in conjunction with parameter-efficient fine-tuning methods. To most effectively rewrite the input text, we train a few-shot paraphrase model with a Maximum-Marginal Likelihood objective. Using six few-shot text classification datasets, we show that enriching data with paraphrases at train and test time enhances the performance beyond what can be achieved with parameter-efficient fine-tuning alone. The code used for our experiments can be found at https://github.com/SaeedNajafi/RIFF.
Authors: Adil Soubki, John Murzaku, Arash Yousefi Jordehi, Peter Zeng, Magdalena Markowska, Seyed Abolghasem Mirroshandel, Owen Rambow
Abstract: Evaluating the theory of mind (ToM) capabilities of language models (LMs) has recently received a great deal of attention. However, many existing benchmarks rely on synthetic data, which risks misaligning the resulting experiments with human behavior. We introduce the first ToM dataset based on naturally occurring spoken dialogs, Common-ToM, and show that LMs struggle to demonstrate ToM. We then show that integrating a simple, explicit representation of beliefs improves LM performance on Common-ToM.
Authors: Kaiyan Zhang, Jianyu Wang, Ermo Hua, Biqing Qi, Ning Ding, Bowen Zhou
Abstract: With the advancement of language models (LMs), their exposure to private data is increasingly inevitable, and their deployment (especially for smaller ones) on personal devices, such as PCs and smartphones, has become a prevailing trend. In contexts laden with user information, enabling models to both safeguard user privacy and execute commands efficiently emerges as an essential research imperative. In this paper, we propose CoGenesis, a collaborative generation framework integrating large (hosted on cloud infrastructure) and small models (deployed on local devices) to address privacy concerns logically. Initially, we design a pipeline to create personalized writing instruction datasets enriched with extensive context details as the testbed of this research issue. Subsequently, we introduce two variants of CoGenesis based on sketch and logits respectively. Our experimental findings, based on our synthesized dataset and two additional open-source datasets, indicate that: 1) Large-scale models perform well when provided with user context but struggle in the absence of such context. 2) While specialized smaller models fine-tuned on the synthetic dataset show promise, they still lag behind their larger counterparts. 3) Our CoGenesis framework, utilizing mixed-scale models, showcases competitive performance, providing a feasible solution to privacy issues.
Authors: Arda Uzunoglu, Abdalfatah Rashid Safa, G\"ozde G\"ul \c{S}ahin
Abstract: Recently, there has been growing interest within the community regarding whether large language models are capable of planning or executing plans. However, most prior studies use LLMs to generate high-level plans for simplified scenarios lacking linguistic complexity and domain diversity, limiting analysis of their planning abilities. These setups constrain evaluation methods (e.g., predefined action space), architectural choices (e.g., only generative models), and overlook the linguistic nuances essential for realistic analysis. To tackle this, we present PARADISE, an abductive reasoning task using Q\&A format on practical procedural text sourced from wikiHow. It involves warning and tip inference tasks directly associated with goals, excluding intermediary steps, with the aim of testing the ability of the models to infer implicit knowledge of the plan solely from the given goal. Our experiments, utilizing fine-tuned language models and zero-shot prompting, reveal the effectiveness of task-specific small models over large language models in most scenarios. Despite advancements, all models fall short of human performance. Notably, our analysis uncovers intriguing insights, such as variations in model behavior with dropped keywords, struggles of BERT-family and GPT-4 with physical and abstract goals, and the proposed tasks offering valuable prior knowledge for other unseen procedural tasks. The PARADISE dataset and associated resources are publicly available for further research exploration with https://github.com/GGLAB-KU/paradise.
Authors: Yanming Liu, Xinyue Peng, Xuhong Zhang, Weihao Liu, Jianwei Yin, Jiannan Cao, Tianyu Du
Abstract: Large language models (LLMs) demonstrate exceptional performance in numerous tasks but still heavily rely on knowledge stored in their parameters. Moreover, updating this knowledge incurs high training costs. Retrieval-augmented generation (RAG) methods address this issue by integrating external knowledge. The model can answer questions it couldn't previously by retrieving knowledge relevant to the query. This approach improves performance in certain scenarios for specific tasks. However, if irrelevant texts are retrieved, it may impair model performance. In this paper, we propose Retrieval Augmented Iterative Self-Feedback (RA-ISF), a framework that iteratively decomposes tasks and processes them in three submodules to enhance the model's problem-solving capabilities. Experiments show that our method outperforms existing benchmarks, performing well on models like GPT3.5, Llama2, significantly enhancing factual reasoning capabilities and reducing hallucinations.
Authors: Yanming Liu, Xinyue Peng, Tianyu Du, Jianwei Yin, Weihao Liu, Xuhong Zhang
Abstract: Large language models (LLMs) have achieved commendable accomplishments in various natural language processing tasks. However, LLMs still encounter significant challenges when dealing with complex scenarios involving multiple entities. These challenges arise from the presence of implicit relationships that demand multi-step reasoning. In this paper, we propose a novel approach ERA-CoT, which aids LLMs in understanding context by capturing relationships between entities and supports the reasoning of diverse tasks through Chain-of-Thoughts (CoT). Experimental results show that ERA-CoT demonstrates the superior performance of our proposed method compared to current CoT prompting methods, achieving a significant improvement of an average of 5.1\% on GPT3.5 compared to previous SOTA baselines. Our analysis indicates that ERA-CoT increases the LLM's understanding of entity relationships, significantly improves the accuracy of question answering, and enhances the reasoning ability of LLMs.
Authors: Weihang Su, Yichen Tang, Qingyao Ai, Zhijing Wu, Yiqun Liu
Abstract: Dynamic retrieval augmented generation (RAG) paradigm actively decides when and what to retrieve during the text generation process of Large Language Models (LLMs). There are two key elements of this paradigm: identifying the optimal moment to activate the retrieval module (deciding when to retrieve) and crafting the appropriate query once retrieval is triggered (determining what to retrieve). However, current dynamic RAG methods fall short in both aspects. Firstly, the strategies for deciding when to retrieve often rely on static rules. Moreover, the strategies for deciding what to retrieve typically limit themselves to the LLM's most recent sentence or the last few tokens, while the LLM's real-time information needs may span across the entire context. To overcome these limitations, we introduce a new framework, DRAGIN, i.e., Dynamic Retrieval Augmented Generation based on the real-time Information Needs of LLMs. Our framework is specifically designed to make decisions on when and what to retrieve based on the LLM's real-time information needs during the text generation process. We evaluate DRAGIN along with existing methods comprehensively over 4 knowledge-intensive generation datasets. Experimental results show that DRAGIN achieves superior performance on all tasks, demonstrating the effectiveness of our method. We have open-sourced all the code, data, and models in GitHub: https://github.com/oneal2000/DRAGIN/tree/main
Authors: Taiqi He, Kwanghee Choi, Lindia Tjuatja, Nathaniel R. Robinson, Jiatong Shi, Shinji Watanabe, Graham Neubig, David R. Mortensen, Lori Levin
Abstract: Thousands of the world's languages are in danger of extinction--a tremendous threat to cultural identities and human language diversity. Interlinear Glossed Text (IGT) is a form of linguistic annotation that can support documentation and resource creation for these languages' communities. IGT typically consists of (1) transcriptions, (2) morphological segmentation, (3) glosses, and (4) free translations to a majority language. We propose Wav2Gloss: a task in which these four annotation components are extracted automatically from speech, and introduce the first dataset to this end, Fieldwork: a corpus of speech with all these annotations, derived from the work of field linguists, covering 37 languages, with standard formatting, and train/dev/test splits. We provide various baselines to lay the groundwork for future research on IGT generation from speech, such as end-to-end versus cascaded, monolingual versus multilingual, and single-task versus multi-task approaches.
Authors: I-Hung Hsu, Zihan Xue, Nilay Pochh, Sahil Bansal, Premkumar Natarajan, Jayanth Srinivasa, Nanyun Peng
Abstract: Event linking connects event mentions in text with relevant nodes in a knowledge base (KB). Prior research in event linking has mainly borrowed methods from entity linking, overlooking the distinct features of events. Compared to the extensively explored entity linking task, events have more complex structures and can be more effectively distinguished by examining their associated arguments. Moreover, the information-rich nature of events leads to the scarcity of event KBs. This emphasizes the need for event linking models to identify and classify event mentions not in the KB as ``out-of-KB,'' an area that has received limited attention. In this work, we tackle these challenges by introducing an argument-aware approach. First, we improve event linking models by augmenting input text with tagged event argument information, facilitating the recognition of key information about event mentions. Subsequently, to help the model handle ``out-of-KB'' scenarios, we synthesize out-of-KB training examples from in-KB instances through controlled manipulation of event arguments. Our experiment across two test datasets showed significant enhancements in both in-KB and out-of-KB scenarios, with a notable 22% improvement in out-of-KB evaluations.
Authors: Jakub Hoscilowicz, Adam Wiacek, Jan Chojnacki, Adam Cieslak, Leszek Michon, Vitalii Urbanevych, Artur Janicki
Abstract: In this work, we explore LLM's internal representation space to identify attention heads that contain the most truthful and accurate information. We further developed the Inference Time Intervention (ITI) framework, which lets bias LLM without the need for fine-tuning. The improvement manifests in introducing a non-linear multi-token probing and multi-token intervention: Non-Linear ITI (NL-ITI), which significantly enhances performance on evaluation benchmarks. NL-ITI is tested on diverse multiple-choice datasets, including TruthfulQA, on which we report over 16% relative MC1 (accuracy of model pointing to the correct answer) improvement with respect to the baseline ITI results. Moreover, we achieved a 10% relative improvement over the recently released Truth Forest (TrFf) method that also focused on ITI improvement.
Authors: Manuel Tonneau, Pedro Vitor Quinta de Castro, Karim Lasri, Ibrahim Farouq, Lakshminarayanan Subramanian, Victor Orozco-Olvera, Samuel P. Fraiberger
Abstract: To address the global issue of online hate, hate speech detection (HSD) systems are typically developed on datasets from the United States, thereby failing to generalize to English dialects from the Majority World. Furthermore, HSD models are often evaluated on non-representative samples, raising concerns about overestimating model performance in real-world settings. In this work, we introduce NaijaHate, the first dataset annotated for HSD which contains a representative sample of Nigerian tweets. We demonstrate that HSD evaluated on biased datasets traditionally used in the literature consistently overestimates real-world performance by at least two-fold. We then propose NaijaXLM-T, a pretrained model tailored to the Nigerian Twitter context, and establish the key role played by domain-adaptive pretraining and finetuning in maximizing HSD performance. Finally, owing to the modest performance of HSD systems in real-world conditions, we find that content moderators would need to review about ten thousand Nigerian tweets flagged as hateful daily to moderate 60% of all hateful content, highlighting the challenges of moderating hate speech at scale as social media usage continues to grow globally. Taken together, these results pave the way towards robust HSD systems and a better protection of social media users from hateful content in low-resource settings.
Authors: Yuemei Xu, Ling Hu, Jiayi Zhao, Zihan Qiu, Yuqi Ye, Hanwen Gu
Abstract: Based on the foundation of Large Language Models (LLMs), Multilingual Large Language Models (MLLMs) have been developed to address the challenges of multilingual natural language processing tasks, hoping to achieve knowledge transfer from high-resource to low-resource languages. However, significant limitations and challenges still exist, such as language imbalance, multilingual alignment, and inherent bias. In this paper, we aim to provide a comprehensive analysis of MLLMs, delving deeply into discussions surrounding these critical issues. First of all, we start by presenting an overview of MLLMs, covering their evolution, key techniques, and multilingual capacities. Secondly, we explore widely utilized multilingual corpora for MLLMs' training and multilingual datasets oriented for downstream tasks that are crucial for enhancing the cross-lingual capability of MLLMs. Thirdly, we survey the existing studies on multilingual representations and investigate whether the current MLLMs can learn a universal language representation. Fourthly, we discuss bias on MLLMs including its category and evaluation metrics, and summarize the existing debiasing techniques. Finally, we discuss existing challenges and point out promising research directions. By demonstrating these aspects, this paper aims to facilitate a deeper understanding of MLLMs and their potentiality in various domains.
Authors: Abhinav Rao, Akhila Yerukola, Vishwa Shah, Katharina Reinecke, Maarten Sap
Abstract: The integration of Large Language Models (LLMs) into various global cultures fundamentally presents a cultural challenge: LLMs must navigate interactions, respect social norms, and avoid transgressing cultural boundaries. However, it is still unclear if LLMs can adapt their outputs to diverse cultural norms. Our study focuses on this aspect. We introduce NormAd, a novel dataset, which includes 2.6k stories that represent social and cultural norms from 75 countries, to assess the ability of LLMs to adapt to different granular levels of socio-cultural contexts such as the country of origin, its associated cultural values, and prevalent social norms. Our study reveals that LLMs struggle with cultural reasoning across all contextual granularities, showing stronger adaptability to English-centric cultures over those from the Global South. Even with explicit social norms, the top-performing model, Mistral-7b-Instruct, achieves only 81.8\% accuracy, lagging behind the 95.6\% achieved by humans. Evaluation on NormAd further reveals that LLMs struggle to adapt to stories involving gift-giving across cultures. Due to inherent agreement or sycophancy biases, LLMs find it considerably easier to assess the social acceptability of stories that adhere to cultural norms than those that deviate from them. Our benchmark measures the cultural adaptability (or lack thereof) of LLMs, emphasizing the potential to make these technologies more equitable and useful for global audiences. We release the NormAd dataset and its associated code on GitHub.
Authors: Haoyi Qiu, Wenbo Hu, Zi-Yi Dou, Nanyun Peng
Abstract: Large Vision-Language Models (LVLMs) suffer from hallucination issues, wherein the models generate plausible-sounding but factually incorrect outputs, undermining their reliability. A comprehensive quantitative evaluation is necessary to identify and understand the extent of hallucinations in these models. However, existing benchmarks are often limited in scope, focusing mainly on object hallucinations. Furthermore, current evaluation methods struggle to effectively address the subtle semantic distinctions between model outputs and reference data, as well as the balance between hallucination and informativeness. To address these issues, we introduce a multi-dimensional benchmark covering objects, attributes, and relations, with challenging images selected based on associative biases. Moreover, we propose a large language model (LLM)-based two-stage evaluation framework that generalizes the popular CHAIR metric and incorporates both faithfulness and coverage into the evaluation. Experiments on 10 established LVLMs demonstrate that our evaluation metric is more comprehensive and better correlated with humans than existing work when evaluating on our challenging human-annotated benchmark dataset. Our work also highlights the critical balance between faithfulness and coverage of model outputs, and encourages future works to address hallucinations in LVLMs while keeping their outputs informative.
Authors: Javier Rando, Francesco Croce, Kry\v{s}tof Mitka, Stepan Shabalin, Maksym Andriushchenko, Nicolas Flammarion, Florian Tram\`er
Abstract: Large language models are aligned to be safe, preventing users from generating harmful content like misinformation or instructions for illegal activities. However, previous work has shown that the alignment process is vulnerable to poisoning attacks. Adversaries can manipulate the safety training data to inject backdoors that act like a universal sudo command: adding the backdoor string to any prompt enables harmful responses from models that, otherwise, behave safely. Our competition, co-located at IEEE SaTML 2024, challenged participants to find universal backdoors in several large language models. This report summarizes the key findings and promising ideas for future research.
Authors: Derek Powell, Walter Gerych, Thomas Hartvigsen
Abstract: Humans rarely learn one fact in isolation. Instead, learning a new fact induces knowledge of other facts about the world. For example, in learning a korat is a type of cat, you also infer it is a mammal and has claws, ensuring your model of the world is consistent. Knowledge editing aims to inject new facts into language models to improve their factuality, but current benchmarks fail to evaluate consistency, which is critical to ensure efficient, accurate, and generalizable edits. We manually create TAXI, a new benchmark dataset specifically created to evaluate consistency in categorical knowledge edits. TAXI contains 11,120 multiple-choice queries for 976 edits spanning 41 categories (e.g., Dogs), 164 subjects (e.g., Labrador), and 183 properties (e.g., is a mammal). We then use TAXI to evaluate popular editors' categorical consistency, measuring how often editing a subject's category appropriately edits its properties. We find that 1) the editors achieve marginal, yet non-random consistency, 2) their consistency far underperforms human baselines, and 3) consistency is more achievable when editing atypical subjects Our code and data are available at https://github.com/derekpowell/taxi.
Authors: Mihir Parmar, Nisarg Patel, Neeraj Varshney, Mutsumi Nakamura, Man Luo, Santosh Mashetty, Arindam Mitra, Chitta Baral
Abstract: Recently developed large language models (LLMs) have been shown to perform remarkably well on a wide range of language understanding tasks. But, can they really "reason" over the natural language? This question has been receiving significant research attention and many reasoning skills such as commonsense, numerical, and qualitative have been studied. However, the crucial skill pertaining to 'logical reasoning' has remained underexplored. Existing work investigating this reasoning ability of LLMs has focused only on a couple of inference rules (such as modus ponens and modus tollens) of propositional and first-order logic. Addressing the above limitation, we comprehensively evaluate the logical reasoning ability of LLMs on 25 different reasoning patterns spanning over propositional, first-order, and non-monotonic logics. To enable systematic evaluation, we introduce LogicBench, a natural language question-answering dataset focusing on the use of a single inference rule. We conduct detailed analysis with a range of LLMs such as GPT-4, ChatGPT, Gemini, Llama-2, and Mistral using chain-of-thought prompting. Experimental results show that existing LLMs do not fare well on LogicBench; especially, they struggle with instances involving complex reasoning and negations. Furthermore, they sometimes overlook contextual information necessary for reasoning to arrive at the correct conclusion. We believe that our work and findings facilitate future research for evaluating and enhancing the logical reasoning ability of LLMs. Data and code are available at https://github.com/Mihir3009/LogicBench.
Authors: Melissa Ailem, Katerina Marazopoulou, Charlotte Siska, James Bono
Abstract: Benchmarks have emerged as the central approach for evaluating Large Language Models (LLMs). The research community often relies on a model's average performance across the test prompts of a benchmark to evaluate the model's performance. This is consistent with the assumption that the test prompts within a benchmark represent a random sample from a real-world distribution of interest. We note that this is generally not the case; instead, we hold that the distribution of interest varies according to the specific use case. We find that (1) the correlation in model performance across test prompts is non-random, (2) accounting for correlations across test prompts can change model rankings on major benchmarks, (3) explanatory factors for these correlations include semantic similarity and common LLM failure points.
Authors: Yunxiang Zhang, Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim, Moontae Lee, Honglak Lee, Lu Wang
Abstract: Self-correction has emerged as a promising solution to boost the reasoning performance of large language models (LLMs), where LLMs refine their solutions using self-generated critiques that pinpoint the errors. This work explores whether small (<= 13B) language models (LMs) have the ability of self-correction on reasoning tasks with minimal inputs from stronger LMs. We propose a novel pipeline that prompts smaller LMs to collect self-correction data that supports the training of self-refinement abilities. First, we leverage correct solutions to guide the model in critiquing their incorrect responses. Second, the generated critiques, after filtering, are used for supervised fine-tuning of the self-correcting reasoner through solution refinement. Our experimental results show improved self-correction abilities of two models on five datasets spanning math and commonsense reasoning, with notable performance gains when paired with a strong GPT-4-based verifier, though limitations are identified when using a weak self-verifier for determining when to correct.
Authors: Farima Fatahi Bayat, Xin Liu, H. V. Jagadish, Lu Wang
Abstract: Large language models (LLMs) can generate long-form and coherent text, yet they often hallucinate facts, which undermines their reliability. To mitigate this issue, inference-time methods steer LLM representations toward the "truthful directions" previously learned for truth elicitation. However, applying these truthful directions with the same intensity fails to generalize across different query contexts. We propose LITO, a Learnable Intervention method for Truthfulness Optimization that automatically identifies the optimal intervention intensity tailored to each specific context. LITO explores a sequence of model generations based on increasing levels of intervention intensities. It selects the most accurate response or refuses to answer when the predictions are highly uncertain. Experiments on multiple LLMs and question-answering datasets demonstrate that LITO improves truthfulness while preserving task accuracy. The adaptive nature of LITO counters the limitations of one-size-fits-all intervention methods, maximizing truthfulness by reflecting the model's internal knowledge only when it is confident. Our code is available at https://github.com/launchnlp/LITO.
Authors: Donya Rooein, Paul Rottger, Anastassia Shaitarova, Dirk Hovy
Abstract: Using large language models (LLMs) for educational applications like dialogue-based teaching is a hot topic. Effective teaching, however, requires teachers to adapt the difficulty of content and explanations to the education level of their students. Even the best LLMs today struggle to do this well. If we want to improve LLMs on this adaptation task, we need to be able to measure adaptation success reliably. However, current Static metrics for text difficulty, like the Flesch-Kincaid Reading Ease score, are known to be crude and brittle. We, therefore, introduce and evaluate a new set of Prompt-based metrics for text difficulty. Based on a user study, we create Prompt-based metrics as inputs for LLMs. They leverage LLM's general language understanding capabilities to capture more abstract and complex features than Static metrics. Regression experiments show that adding our Prompt-based metrics significantly improves text difficulty classification over Static metrics alone. Our results demonstrate the promise of using LLMs to evaluate text adaptation to different education levels.
Authors: Yizhe Yang, Palakorn Achananuparp, Heyan Huang, Jing Jiang, Ee-Peng Lim
Abstract: The recent success of large language models (LLMs) has attracted widespread interest to develop role-playing conversational agents personalized to the characteristics and styles of different speakers to enhance their abilities to perform both general and special purpose dialogue tasks. However, the ability to personalize the generated utterances to speakers, whether conducted by human or LLM, has not been well studied. To bridge this gap, our study introduces a novel evaluation challenge: speaker verification in agent-generated conversations, which aimed to verify whether two sets of utterances originate from the same speaker. To this end, we assemble a large dataset collection encompassing thousands of speakers and their utterances. We also develop and evaluate speaker verification models under experiment setups. We further utilize the speaker verification models to evaluate the personalization abilities of LLM-based role-playing models. Comprehensive experiments suggest that the current role-playing models fail in accurately mimicking speakers, primarily due to their inherent linguistic characteristics.
Authors: Zijin Hong, Jian Liu
Abstract: Event Extraction (EE) is an essential information extraction task that aims to extract event-related information from unstructured texts. The paradigm of this task has shifted from conventional classification-based methods to more contemporary question-answering-based (QA-based) approaches. However, in QA-based EE, the quality of the questions dramatically affects the extraction accuracy, and how to generate high-quality questions for QA-based EE remains a challenge. In this work, to tackle this challenge, we suggest four criteria to evaluate the quality of a question and propose a reinforcement learning method, RLQG, for QA-based EE that can generate generalizable, high-quality, and context-dependent questions and provides clear guidance to QA models. The extensive experiments conducted on ACE and RAMS datasets have strongly validated our approach's effectiveness, which also demonstrates its robustness in scenarios with limited training data. The corresponding code of RLQG is released for further research.
Authors: Jiahuan Pei, Irene Viola, Haochen Huang, Junxiao Wang, Moonisa Ahsan, Fanghua Ye, Jiang Yiming, Yao Sai, Di Wang, Zhumin Chen, Pengjie Ren, Pablo Cesar
Abstract: Autonomous artificial intelligence (AI) agents have emerged as promising protocols for automatically understanding the language-based environment, particularly with the exponential development of large language models (LLMs). However, a fine-grained, comprehensive understanding of multimodal environments remains under-explored. This work designs an autonomous workflow tailored for integrating AI agents seamlessly into extended reality (XR) applications for fine-grained training. We present a demonstration of a multimodal fine-grained training assistant for LEGO brick assembly in a pilot XR environment. Specifically, we design a cerebral language agent that integrates LLM with memory, planning, and interaction with XR tools and a vision-language agent, enabling agents to decide their actions based on past experiences. Furthermore, we introduce LEGO-MRTA, a multimodal fine-grained assembly dialogue dataset synthesized automatically in the workflow served by a commercial LLM. This dataset comprises multimodal instruction manuals, conversations, XR responses, and vision question answering. Last, we present several prevailing open-resource LLMs as benchmarks, assessing their performance with and without fine-tuning on the proposed dataset. We anticipate that the broader impact of this workflow will advance the development of smarter assistants for seamless user interaction in XR environments, fostering research in both AI and HCI communities.
Authors: Ruochen Zhao, Wenxuan Zhang, Yew Ken Chia, Deli Zhao, Lidong Bing
Abstract: As LLMs evolve on a daily basis, there is an urgent need for a trustworthy evaluation method that can provide robust evaluation results in a timely fashion. Currently, as static benchmarks are prone to contamination concerns, users tend to trust human voting platforms, such as Chatbot Arena. However, human annotations require extensive manual efforts. To provide an automatic, robust, and trustworthy evaluation framework, we innovatively propose the Auto-Arena of LLMs, which automates the entire evaluation process with LLM agents. Firstly, an examiner LLM devises queries. Then, a pair of candidate LLMs engage in a multi-round peer-battle around the query, during which the LLM's true performance gaps become visible. Finally, a committee of LLM judges collectively discuss and determine the winner, which alleviates bias and promotes fairness. In our extensive experiment on the 17 newest LLMs, Auto-Arena shows the highest correlation with human preferences, providing a promising alternative to human evaluation platforms.
Authors: Laura Cabello, Uchenna Akujuobi
Abstract: Aspect-Based Sentiment Analysis (ABSA) involves extracting opinions from textual data about specific entities and their corresponding aspects through various complementary subtasks. Several prior research has focused on developing ad hoc designs of varying complexities for these subtasks. In this paper, we present a generative framework extensible to any ABSA subtask. We build upon the instruction tuned model proposed by Scaria et al. (2023), who present an instruction-based model with task descriptions followed by in-context examples on ABSA subtasks. We propose PFInstruct, an extension to this instruction learning paradigm by appending an NLP-related task prefix to the task description. This simple approach leads to improved performance across all tested SemEval subtasks, surpassing previous state-of-the-art (SOTA) on the ATE subtask (Rest14) by +3.28 F1-score, and on the AOOE subtask by an average of +5.43 F1-score across SemEval datasets. Furthermore, we explore the impact of the prefix-enhanced prompt quality on the ABSA subtasks and find that even a noisy prefix enhances model performance compared to the baseline. Our method also achieves competitive results on a biomedical domain dataset (ERSA).
Authors: Mengge Xue, Zhenyu Hu, Liqun Liu, Kuo Liao, Shuang Li, Honglin Han, Meng Zhao, Chengguo Yin
Abstract: Multiple-Choice Questions (MCQs) constitute a critical area of research in the study of Large Language Models (LLMs). Previous works have investigated the selection bias problem in MCQs within few-shot scenarios, in which the LLM's performance may be influenced by the presentation of answer choices, leaving the selection bias during Supervised Fine-Tuning (SFT) unexplored. In this paper, we reveal that selection bias persists in the SFT phase , primarily due to the LLM's inadequate Multiple Choice Symbol Binding (MCSB) ability. This limitation implies that the model struggles to associate the answer options with their corresponding symbols (e.g., A/B/C/D) effectively. To enhance the model's MCSB capability, we first incorporate option contents into the loss function and subsequently adjust the weights of the option symbols and contents, guiding the model to understand the option content of the current symbol. Based on this, we introduce an efficient SFT algorithm for MCQs, termed Point-wise Intelligent Feedback (PIF). PIF constructs negative instances by randomly combining the incorrect option contents with all candidate symbols, and proposes a point-wise loss to provide feedback on these negative samples into LLMs. Our experimental results demonstrate that PIF significantly reduces the model's selection bias by improving its MCSB capability. Remarkably, PIF exhibits a substantial enhancement in the accuracy for MCQs.
Authors: Da Ma, Lu Chen, Pengyu Wang, Hongshen Xu, Hanqi Li, Liangtai Sun, Su Zhu, Shuai Fan, Kai Yu
Abstract: Large language models (LLMs) have demonstrated proficiency across various natural language processing (NLP) tasks but often require additional training, such as continual pre-training and supervised fine-tuning. However, the costs associated with this, primarily due to their large parameter count, remain high. This paper proposes leveraging \emph{sparsity} in pre-trained LLMs to expedite this training process. By observing sparsity in activated neurons during forward iterations, we identify the potential for computational speed-ups by excluding inactive neurons. We address associated challenges by extending existing neuron importance evaluation metrics and introducing a ladder omission rate scheduler. Our experiments on Llama-2 demonstrate that Sparsity-Accelerated Training (SAT) achieves comparable or superior performance to standard training while significantly accelerating the process. Specifically, SAT achieves a $45\%$ throughput improvement in continual pre-training and saves $38\%$ training time in supervised fine-tuning in practice. It offers a simple, hardware-agnostic, and easily deployable framework for additional LLM training. Our code is available at https://github.com/OpenDFM/SAT.
Authors: Haozheng Luo, Jiahao Yu, Wenxin Zhang, Jialong Li, Jerry Yao-Chieh Hu, Xinyu Xing, Han Liu
Abstract: We introduce a low-resource safety enhancement method for aligning large language models (LLMs) without the need for supervised fine-tuning (SFT) or reinforcement learning from human feedback (RLHF). Our main idea is to exploit knowledge distillation to extract the alignment information from existing well-aligned LLMs and integrate it into unaligned LLMs in a plug-and-play fashion. Methodology, we employ delta debugging to identify the critical components of knowledge necessary for effective distillation. On the harmful question dataset, our method significantly enhances the average defense success rate by approximately 14.41%, reaching as high as 51.39%, in 17 unaligned pre-trained LLMs, without compromising performance.
Authors: Saminu Mohammad Aliyu, Gregory Maksha Wajiga, Muhammad Murtala
Abstract: The proliferation of online offensive language necessitates the development of effective detection mechanisms, especially in multilingual contexts. This study addresses the challenge by developing and introducing novel datasets for offensive language detection in three major Nigerian languages: Hausa, Yoruba, and Igbo. We collected data from Twitter and manually annotated it to create datasets for each of the three languages, using native speakers. We used pre-trained language models to evaluate their efficacy in detecting offensive language in our datasets. The best-performing model achieved an accuracy of 90\%. To further support research in offensive language detection, we plan to make the dataset and our models publicly available.
Authors: Zengkui Sun, Yijin Liu, Fandong Meng, Jinan Xu, Yufeng Chen, Jie Zhou
Abstract: Multilingual neural machine translation models generally distinguish translation directions by the language tag (LT) in front of the source or target sentences. However, current LT strategies cannot indicate the desired target language as expected on zero-shot translation, i.e., the off-target issue. Our analysis reveals that the indication of the target language is sensitive to the placement of the target LT. For example, when placing the target LT on the decoder side, the indication would rapidly degrade along with decoding steps, while placing the target LT on the encoder side would lead to copying or paraphrasing the source input. To address the above issues, we propose a simple yet effective strategy named Language Converter Strategy (LCS). By introducing the target language embedding into the top encoder layers, LCS mitigates confusion in the encoder and ensures stable language indication for the decoder. Experimental results on MultiUN, TED, and OPUS-100 datasets demonstrate that LCS could significantly mitigate the off-target issue, with language accuracy up to 95.28%, 96.21%, and 85.35% meanwhile outperforming the vanilla LT strategy by 3.07, 3,3, and 7.93 BLEU scores on zero-shot translation, respectively.
Authors: Zengkui Sun, Yijin Liu, Jiaan Wang, Fandong Meng, Jinan Xu, Yufeng Chen, Jie Zhou
Abstract: Recently, Knowledge Editing has received increasing attention, since it could update the specific knowledge from outdated ones in pretrained models without re-training. However, as pointed out by recent studies, existing related methods tend to merely memorize the superficial word composition of the edited knowledge, rather than truly learning and absorbing it. Consequently, on the reasoning questions, we discover that existing methods struggle to utilize the edited knowledge to reason the new answer, and tend to retain outdated responses, which are generated by the original models utilizing original knowledge. Nevertheless, the outdated responses are unexpected for the correct answers to reasoning questions, which we named as the outdated issue. To alleviate this issue, in this paper, we propose a simple yet effective decoding strategy, i.e., outDated ISsue aware deCOding (DISCO), to enhance the performance of edited models on reasoning questions. Specifically, we capture the difference in the probability distribution between the original and edited models. Further, we amplify the difference of the token prediction in the edited model to alleviate the outdated issue, and thus enhance the model performance w.r.t the edited knowledge. Experimental results suggest that applying DISCO could enhance edited models to reason, e.g., on reasoning questions, DISCO outperforms the prior SOTA method by 12.99 F1 scores, and reduces the ratio of the outdated issue to 5.78% on the zsRE dataset.
Authors: Rongzhi Zhang, Jiaming Shen, Tianqi Liu, Haorui Wang, Zhen Qin, Feng Han, Jialu Liu, Simon Baumgartner, Michael Bendersky, Chao Zhang
Abstract: Large Language Models (LLMs) have exhibited impressive capabilities in various tasks, yet their vast parameter sizes restrict their applicability in resource-constrained settings. Knowledge distillation (KD) offers a viable solution by transferring expertise from large teacher models to compact student models. However, traditional KD techniques face specific challenges when applied to LLMs, including restricted access to LLM outputs, significant teacher-student capacity gaps, and the inherited mis-calibration issue. In this work, we present PLaD, a novel preference-based LLM distillation framework. PLaD exploits the teacher-student capacity discrepancy to generate pseudo-preference pairs where teacher outputs are preferred over student outputs. Then, PLaD leverages a ranking loss to re-calibrate student's estimation of sequence likelihood, which steers the student's focus towards understanding the relative quality of outputs instead of simply imitating the teacher. PLaD bypasses the need for access to teacher LLM's internal states, tackles the student's expressivity limitations, and mitigates the student mis-calibration issue. Through extensive experiments on two sequence generation tasks and with various LLMs, we demonstrate the effectiveness of our proposed PLaD framework.
Authors: Hao Li, Yuping Wu, Viktor Schlegel, Riza Batista-Navarro, Tharindu Madusanka, Iqra Zahid, Jiayan Zeng, Xiaochi Wang, Xinran He, Yizhi Li, Goran Nenadic
Abstract: With the recent advances of large language models (LLMs), it is no longer infeasible to build an automated debate system that helps people to synthesise persuasive arguments. Previous work attempted this task by integrating multiple components. In our work, we introduce an argument mining dataset that captures the end-to-end process of preparing an argumentative essay for a debate, which covers the tasks of claim and evidence identification (Task 1 ED), evidence convincingness ranking (Task 2 ECR), argumentative essay summarisation and human preference ranking (Task 3 ASR) and metric learning for automated evaluation of resulting essays, based on human feedback along argument quality dimensions (Task 4 SQE). Our dataset contains 14k examples of claims that are fully annotated with the various properties supporting the aforementioned tasks. We evaluate multiple generative baselines for each of these tasks, including representative LLMs. We find, that while they show promising results on individual tasks in our benchmark, their end-to-end performance on all four tasks in succession deteriorates significantly, both in automated measures as well as in human-centred evaluation. This challenge presented by our proposed dataset motivates future research on end-to-end argument mining and summarisation. The repository of this project is available at https://github.com/HarrywillDr/ArgSum-Datatset
Authors: Farhad Nooralahzadeh, Yi Zhang, Ellery Smith, Sabine Maennel, Cyril Matthey-Doret, Rapha\"el de Fondville, Kurt Stockinger
Abstract: The potential for improvements brought by Large Language Models (LLMs) in Text-to-SQL systems is mostly assessed on monolingual English datasets. However, LLMs' performance for other languages remains vastly unexplored. In this work, we release the StatBot.Swiss dataset, the first bilingual benchmark for evaluating Text-to-SQL systems based on real-world applications. The StatBot.Swiss dataset contains 455 natural language/SQL-pairs over 35 big databases with varying level of complexity for both English and German. We evaluate the performance of state-of-the-art LLMs such as GPT-3.5-Turbo and mixtral-8x7b-instruct for the Text-to-SQL translation task using an in-context learning approach. Our experimental analysis illustrates that current LLMs struggle to generalize well in generating SQL queries on our novel bilingual dataset.
Authors: Pierluigi Cassotti, Stefano De Pascale, Nina Tahmasebi
Abstract: There is abundant evidence of the fact that the way words change their meaning can be classified in different types of change, highlighting the relationship between the old and new meanings (among which generalization, specialization and co-hyponymy transfer). In this paper, we present a way of detecting these types of change by constructing a model that leverages information both from synchronic lexical relations and definitions of word meanings. Specifically, we use synset definitions and hierarchy information from WordNet and test it on a digitized version of Blank's (1997) dataset of semantic change types. Finally, we show how the sense relationships can improve models for both approximation of human judgments of semantic relatedness as well as binary Lexical Semantic Change Detection.
Authors: Deven Santosh Shah, Shiying He, Gosuddin Kamaruddin Siddiqi, Radhika Bansal
Abstract: Local news articles are a subset of news that impact users in a geographical area, such as a city, county, or state. Detecting local news (Step 1) and subsequently deciding its geographical location as well as radius of impact (Step 2) are two important steps towards accurate local news recommendation. Naive rule-based methods, such as detecting city names from the news title, tend to give erroneous results due to lack of understanding of the news content. Empowered by the latest development in natural language processing, we develop an integrated pipeline that enables automatic local news detection and content-based local news recommendations. In this paper, we focus on Step 1 of the pipeline, which highlights: (1) a weakly supervised framework incorporated with domain knowledge and auto data processing, and (2) scalability to multi-lingual settings. Compared with Stanford CoreNLP NER model, our pipeline has higher precision and recall evaluated on a real-world and human-labeled dataset. This pipeline has potential to more precise local news to users, helps local businesses get more exposure, and gives people more information about their neighborhood safety.
Authors: Rohith Kuditipudi, John Thickstun, Tatsunori Hashimoto, Percy Liang
Abstract: We propose a methodology for planting watermarks in text from an autoregressive language model that are robust to perturbations without changing the distribution over text up to a certain maximum generation budget. We generate watermarked text by mapping a sequence of random numbers -- which we compute using a randomized watermark key -- to a sample from the language model. To detect watermarked text, any party who knows the key can align the text to the random number sequence. We instantiate our watermark methodology with two sampling schemes: inverse transform sampling and exponential minimum sampling. We apply these watermarks to three language models -- OPT-1.3B, LLaMA-7B and Alpaca-7B -- to experimentally validate their statistical power and robustness to various paraphrasing attacks. Notably, for both the OPT-1.3B and LLaMA-7B models, we find we can reliably detect watermarked text ($p \leq 0.01$) from $35$ tokens even after corrupting between $40$-$50\%$ of the tokens via random edits (i.e., substitutions, insertions or deletions). For the Alpaca-7B model, we conduct a case study on the feasibility of watermarking responses to typical user instructions. Due to the lower entropy of the responses, detection is more difficult: around $25\%$ of the responses -- whose median length is around $100$ tokens -- are detectable with $p \leq 0.01$, and the watermark is also less robust to certain automated paraphrasing attacks we implement.
Authors: Jingwen Fu, Tao Yang, Yuwang Wang, Yan Lu, Nanning Zheng
Abstract: In-context learning, i.e., learning from context examples, is an impressive ability of Transformer. Training Transformers to possess this in-context learning skill is computationally intensive due to the occurrence of learning plateaus, which are periods within the training process where there is minimal or no enhancement in the model's in-context learning capability. To study the mechanism behind the learning plateaus, we conceptually seperate a component within the model's internal representation that is exclusively affected by the model's weights. We call this the "weights component", and the remainder is identified as the "context component". By conducting meticulous and controlled experiments on synthetic tasks, we note that the persistence of learning plateaus correlates with compromised functionality of the weights component. Recognizing the impaired performance of the weights component as a fundamental behavior drives learning plateaus, we have developed three strategies to expedite the learning of Transformers. The effectiveness of these strategies is further confirmed in natural language processing tasks. In conclusion, our research demonstrates the feasibility of cultivating a powerful in-context learning ability within AI systems in an eco-friendly manner.
Authors: Yuang Li, Min Zhang, Chang Su, Yinglu Li, Xiaosong Qiao, Mengxin Ren, Miaomiao Ma, Daimeng Wei, Shimin Tao, Hao Yang
Abstract: The recognition of rare named entities, such as personal names and terminologies, is challenging for automatic speech recognition (ASR) systems, especially when they are not frequently observed in the training data. In this paper, we introduce keyword spotting enhanced Whisper (KWS-Whisper), a novel ASR system that leverages the Whisper model and performs open-vocabulary keyword spotting (OV-KWS) on the hidden states of the Whisper encoder to recognize user-defined named entities. These entities serve as prompts for the Whisper decoder. To optimize the model, we propose a multitask training approach that learns OV-KWS and contextual-ASR tasks. We evaluate our approach on Chinese Aishell hot word subsets and two internal code-switching test sets and show that it significantly improves the entity recall compared to the original Whisper model. Moreover, we demonstrate that the OV-KWS can be a plug-and-play module to enhance the ASR error correction methods and frozen Whisper models.
Authors: Sreyan Ghosh, Sonal Kumar, Chandra Kiran Reddy Evuru, Ramani Duraiswami, Dinesh Manocha
Abstract: We present RECAP (REtrieval-Augmented Audio CAPtioning), a novel and effective audio captioning system that generates captions conditioned on an input audio and other captions similar to the audio retrieved from a datastore. Additionally, our proposed method can transfer to any domain without the need for any additional fine-tuning. To generate a caption for an audio sample, we leverage an audio-text model CLAP to retrieve captions similar to it from a replaceable datastore, which are then used to construct a prompt. Next, we feed this prompt to a GPT-2 decoder and introduce cross-attention layers between the CLAP encoder and GPT-2 to condition the audio for caption generation. Experiments on two benchmark datasets, Clotho and AudioCaps, show that RECAP achieves competitive performance in in-domain settings and significant improvements in out-of-domain settings. Additionally, due to its capability to exploit a large text-captions-only datastore in a training-free fashion, RECAP shows unique capabilities of captioning novel audio events never seen during training and compositional audios with multiple events. To promote research in this space, we also release 150,000+ new weakly labeled captions for AudioSet, AudioCaps, and Clotho.
Authors: Andy Zhou, Kai Yan, Michal Shlapentokh-Rothman, Haohan Wang, Yu-Xiong Wang
Abstract: While language models (LMs) have shown potential across a range of decision-making tasks, their reliance on simple acting processes limits their broad deployment as autonomous agents. In this paper, we introduce Language Agent Tree Search (LATS) -- the first general framework that synergizes the capabilities of LMs in reasoning, acting, and planning. By leveraging the in-context learning ability of LMs, we integrate Monte Carlo Tree Search into LATS to enable LMs as agents, along with LM-powered value functions and self-reflections for proficient exploration and enhanced decision-making. A key feature of our approach is the incorporation of an environment for external feedback, which offers a more deliberate and adaptive problem-solving mechanism that surpasses the constraints of existing techniques. Our experimental evaluation across diverse domains, including programming, interactive question-answering (QA), web navigation, and math, validates the effectiveness and generality of LATS in decision-making while maintaining competitive or improved reasoning performance. Notably, LATS achieves state-of-the-art pass@1 accuracy (92.7%) for programming on HumanEval with GPT-4 and demonstrates gradient-free performance (average score of 75.9) comparable to gradient-based fine-tuning for web navigation on WebShop with GPT-3.5. Code can be found at https://github.com/lapisrocks/LanguageAgentTreeSearch
Authors: Kai Lv, Hang Yan, Qipeng Guo, Haijun Lv, Xipeng Qiu
Abstract: Large language models have achieved remarkable success, but their extensive parameter size necessitates substantial memory for training, thereby setting a high threshold. While the recently proposed low-memory optimization (LOMO) reduces memory footprint, its optimization technique, akin to stochastic gradient descent, is sensitive to hyper-parameters and exhibits suboptimal convergence, failing to match the performance of the prevailing optimizer for large language models, AdamW. Through empirical analysis of the Adam optimizer, we found that, compared to momentum, the adaptive learning rate is more critical for bridging the gap. Building on this insight, we introduce the low-memory optimization with adaptive learning rate (AdaLomo), which offers an adaptive learning rate for each parameter. To maintain memory efficiency, we employ non-negative matrix factorization for the second-order moment estimation in the optimizer state. Additionally, we suggest the use of a grouped update normalization to stabilize convergence. Our experiments with instruction-tuning and further pre-training demonstrate that AdaLomo achieves results on par with AdamW, while significantly reducing memory requirements, thereby lowering the hardware barrier to training large language models. The code is accessible at https://github.com/OpenLMLab/LOMO.
Authors: Yukiya Hono, Koh Mitsuda, Tianyu Zhao, Kentaro Mitsui, Toshiaki Wakatsuki, Kei Sawada
Abstract: Advances in machine learning have made it possible to perform various text and speech processing tasks, such as automatic speech recognition (ASR), in an end-to-end (E2E) manner. E2E approaches utilizing pre-trained models are gaining attention for conserving training data and resources. However, most of their applications in ASR involve only one of either a pre-trained speech or a language model. This paper proposes integrating a pre-trained speech representation model and a large language model (LLM) for E2E ASR. The proposed model enables the optimization of the entire ASR process, including acoustic feature extraction and acoustic and language modeling, by combining pre-trained models with a bridge network and also enables the application of remarkable developments in LLM utilization, such as parameter-efficient domain adaptation and inference optimization. Experimental results demonstrate that the proposed model achieves a performance comparable to that of modern E2E ASR models by utilizing powerful pre-training models with the proposed integrated approach.
Authors: Xu Yang, Yingzhe Peng, Haoxuan Ma, Shuo Xu, Chi Zhang, Yucheng Han, Hanwang Zhang
Abstract: As Archimedes famously said, ``Give me a lever long enough and a fulcrum on which to place it, and I shall move the world'', in this study, we propose to use a tiny Language Model (LM), \eg, a Transformer with 67M parameters, to lever much larger Vision-Language Models (LVLMs) with 9B parameters. Specifically, we use this tiny \textbf{Lever-LM} to configure effective in-context demonstration (ICD) sequences to improve the In-Context Learinng (ICL) performance of LVLMs. Previous studies show that diverse ICD configurations like the selection and ordering of the demonstrations heavily affect the ICL performance, highlighting the significance of configuring effective ICD sequences. Motivated by this and by re-considering the the process of configuring ICD sequence, we find this is a mirror process of human sentence composition and further assume that effective ICD configurations may contain internal statistical patterns that can be captured by Lever-LM. Then a dataset with effective ICD sequences is constructed to train Lever-LM. After training, given novel queries, new ICD sequences are configured by the trained Lever-LM to solve vision-language tasks through ICL. Experiments show that these ICD sequences can improve the ICL performance of two LVLMs compared with some strong baselines in Visual Question Answering and Image Captioning, validating that Lever-LM can really capture the statistical patterns for levering LVLMs.
Authors: Jinglong Luo, Yehong Zhang, Zhuo Zhang, Jiaqi Zhang, Xin Mu, Hui Wang, Yue Yu, Zenglin Xu
Abstract: With the growing use of large language models hosted on cloud platforms to offer inference services, privacy concerns are escalating, especially concerning sensitive data like investment plans and bank account details. Secure Multi-Party Computing (SMPC) emerges as a promising solution to protect the privacy of inference data and model parameters. However, the application of SMPC in Privacy-Preserving Inference (PPI) for large language models, particularly those based on the Transformer architecture, often leads to considerable slowdowns or declines in performance. This is largely due to the multitude of nonlinear operations in the Transformer architecture, which are not well-suited to SMPC and difficult to circumvent or optimize effectively. To address this concern, we introduce an advanced optimization framework called SecFormer, to achieve fast and accurate PPI for Transformer models. By implementing model design optimization, we successfully eliminate the high-cost exponential and maximum operations in PPI without sacrificing model performance. Additionally, we have developed a suite of efficient SMPC protocols that utilize segmented polynomials, Fourier series and Goldschmidt's method to handle other complex nonlinear functions within PPI, such as GeLU, LayerNorm, and Softmax. Our extensive experiments reveal that SecFormer outperforms MPCFormer in performance, showing improvements of $5.6\%$ and $24.2\%$ for BERT$_{\text{BASE}}$ and BERT$_{\text{LARGE}}$, respectively. In terms of efficiency, SecFormer is 3.56 and 3.58 times faster than Puma for BERT$_{\text{BASE}}$ and BERT$_{\text{LARGE}}$, demonstrating its effectiveness and speed.
Authors: Runchu Tian, Yining Ye, Yujia Qin, Xin Cong, Yankai Lin, Yinxu Pan, Yesai Wu, Haotian Hui, Weichuan Liu, Zhiyuan Liu, Maosong Sun
Abstract: Large Language Models (LLMs) have demonstrated exceptional coding capability. However, as another critical component of programming proficiency, the debugging capability of LLMs remains relatively unexplored. Previous evaluations of LLMs' debugging ability are significantly limited by the risk of data leakage, the scale of the dataset, and the variety of tested bugs. To overcome these deficiencies, we introduce `DebugBench', an LLM debugging benchmark consisting of 4,253 instances. It covers four major bug categories and 18 minor types in C++, Java, and Python. To construct DebugBench, we collect code snippets from the LeetCode community, implant bugs into source data with GPT-4, and assure rigorous quality checks. We evaluate two commercial and four open-source models in a zero-shot scenario. We find that (1) while closed-source models exhibit inferior debugging performance compared to humans, open-source models relatively lower pass rate scores; (2) the complexity of debugging notably fluctuates depending on the bug category; (3) incorporating runtime feedback has a clear impact on debugging performance which is not always helpful. As an extension, we also compare LLM debugging and code generation, revealing a strong correlation between them for closed-source models. These findings will benefit the development of LLMs in debugging.
Authors: Tianle Cai, Yuhong Li, Zhengyang Geng, Hongwu Peng, Jason D. Lee, Deming Chen, Tri Dao
Abstract: Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one's output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator's cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa substantially reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the backbone model's capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2x speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-3.6x.
Authors: Jing Yu Koh, Robert Lo, Lawrence Jang, Vikram Duvvur, Ming Chong Lim, Po-Yu Huang, Graham Neubig, Shuyan Zhou, Ruslan Salakhutdinov, Daniel Fried
Abstract: Autonomous agents capable of planning, reasoning, and executing actions on the web offer a promising avenue for automating computer tasks. However, the majority of existing benchmarks primarily focus on text-based agents, neglecting many natural tasks that require visual information to effectively solve. Given that most computer interfaces cater to human perception, visual information often augments textual data in ways that text-only models struggle to harness effectively. To bridge this gap, we introduce VisualWebArena, a benchmark designed to assess the performance of multimodal web agents on realistic \textit{visually grounded tasks}. VisualWebArena comprises of a set of diverse and complex web-based tasks that evaluate various capabilities of autonomous multimodal agents. To perform on this benchmark, agents need to accurately process image-text inputs, interpret natural language instructions, and execute actions on websites to accomplish user-defined objectives. We conduct an extensive evaluation of state-of-the-art LLM-based autonomous agents, including several multimodal models. Through extensive quantitative and qualitative analysis, we identify several limitations of text-only LLM agents, and reveal gaps in the capabilities of state-of-the-art multimodal language agents. VisualWebArena provides a framework for evaluating multimodal autonomous language agents, and offers insights towards building stronger autonomous agents for the web. Our code, baseline models, and data is publicly available at https://jykoh.com/vwa.
URLs: https://jykoh.com/vwa.
Authors: Elias Stengel-Eskin, Archiki Prasad, Mohit Bansal
Abstract: While large language models (LLMs) are increasingly being used for program synthesis, they lack the global view needed to develop useful abstractions; they generally predict programs one at a time, often repeating the same functionality. Generating redundant code from scratch is both inefficient and error-prone. To address this, we propose Refactoring for Generalizable Abstraction Learning (ReGAL), a gradient-free method for learning a library of reusable functions via code refactorization, i.e., restructuring code without changing its execution output. ReGAL learns from a small set of existing programs, iteratively verifying and refining its abstractions via execution. We find that the shared function libraries discovered by ReGAL make programs easier to predict across diverse domains. On five datasets -- LOGO graphics generation, Date reasoning, TextCraft (a Minecraft-based text-game) MATH, and TabMWP -- both open-source and proprietary LLMs improve in accuracy when predicting programs with ReGAL functions. For CodeLlama-13B, ReGAL results in absolute accuracy increases of 11.5% on LOGO, 26.1% on date understanding, and 8.1% on TextCraft, outperforming GPT-3.5 in two of three domains. Our analysis reveals ReGAL's abstractions encapsulate frequently-used subroutines as well as environment dynamics.
Authors: Andy Zhou, Bo Li, Haohan Wang
Abstract: Despite advances in AI alignment, large language models (LLMs) remain vulnerable to adversarial attacks or jailbreaking, in which adversaries can modify prompts to induce unwanted behavior. While some defenses have been proposed, they have not been adapted to newly proposed attacks and more challenging threat models. To address this, we propose an optimization-based objective for defending LLMs against jailbreaking attacks and an algorithm, Robust Prompt Optimization (RPO) to create robust system-level defenses. Our approach directly incorporates the adversary into the defensive objective and optimizes a lightweight and transferable suffix, enabling RPO to adapt to worst-case adaptive attacks. Our theoretical and experimental results show improved robustness to both jailbreaks seen during optimization and unknown jailbreaks, reducing the attack success rate (ASR) on GPT-4 to 6% and Llama-2 to 0% on JailbreakBench, setting the state-of-the-art. Code can be found at https://github.com/lapisrocks/rpo
Authors: Masoomali Fatehkia, Ji Kim Lucas, Sanjay Chawla
Abstract: Large Language Models (LLM) have shown remarkable language capabilities fueling attempts to integrate them into applications across a wide range of domains. An important application area is question answering over private enterprise documents where the main considerations are data security, which necessitates applications that can be deployed on-prem, limited computational resources and the need for a robust application that correctly responds to queries. Retrieval-Augmented Generation (RAG) has emerged as the most prominent framework for building LLM-based applications. While building a RAG is relatively straightforward, making it robust and a reliable application requires extensive customization and relatively deep knowledge of the application domain. We share our experiences building and deploying an LLM application for question answering over private organizational documents. Our application combines the use of RAG with a finetuned open-source LLM. Additionally, our system, which we call Tree-RAG (T-RAG), uses a tree structure to represent entity hierarchies within the organization. This is used to generate a textual description to augment the context when responding to user queries pertaining to entities within the organization's hierarchy. Our evaluations, including a Needle in a Haystack test, show that this combination performs better than a simple RAG or finetuning implementation. Finally, we share some lessons learned based on our experiences building an LLM application for real-world use.
Authors: Mingzhe Du, Anh Tuan Luu, Bin Ji, Qian Liu, See-Kiong Ng
Abstract: Amidst the recent strides in evaluating Large Language Models for Code (Code LLMs), existing benchmarks have mainly focused on the functional correctness of generated code, neglecting the importance of their computational efficiency. To fill the gap, we present Mercury, the first code efficiency benchmark for Code LLMs. It comprises 1,889 Python tasks, each accompanied by adequate solutions that serve as real-world efficiency baselines, enabling a comprehensive analysis of the runtime distribution. Based on the distribution, we introduce a new metric Beyond, which computes a runtime-percentile-weighted Pass score to reflect functional correctness and code efficiency simultaneously. On Mercury, leading Code LLMs can achieve 65% on Pass, while less than 50% on Beyond. Given that an ideal Beyond score would be aligned with the Pass score, it indicates that while Code LLMs exhibit impressive capabilities in generating functionally correct code, there remains a notable gap in their efficiency. Finally, our empirical experiments reveal that Direct Preference Optimization (DPO) serves as a robust baseline for enhancing code efficiency compared with Supervised Fine Tuning (SFT), which paves a promising avenue for future exploration of efficient code generation. Our code and data are available on GitHub: https://github.com/Elfsong/Mercury.
Authors: Naihao Deng, Zhenjie Sun, Ruiqi He, Aman Sikka, Yulong Chen, Lin Ma, Yue Zhang, Rada Mihalcea
Abstract: In this paper, we investigate the effectiveness of various LLMs in interpreting tabular data through different prompting strategies and data formats. Our analyses extend across six benchmarks for table-related tasks such as question-answering and fact-checking. We introduce for the first time the assessment of LLMs' performance on image-based table representations. Specifically, we compare five text-based and three image-based table representations, demonstrating the role of representation and prompting on LLM performance. Our study provides insights into the effective use of LLMs on table-related tasks.
Authors: Davide Caffagni, Federico Cocchi, Luca Barsellotti, Nicholas Moratelli, Sara Sarto, Lorenzo Baraldi, Lorenzo Baraldi, Marcella Cornia, Rita Cucchiara
Abstract: Connecting text and visual modalities plays an essential role in generative intelligence. For this reason, inspired by the success of large language models, significant research efforts are being devoted to the development of Multimodal Large Language Models (MLLMs). These models can seamlessly integrate visual and textual modalities, while providing a dialogue-based interface and instruction-following capabilities. In this paper, we provide a comprehensive review of recent visual-based MLLMs, analyzing their architectural choices, multimodal alignment strategies, and training techniques. We also conduct a detailed analysis of these models across a wide range of tasks, including visual grounding, image generation and editing, visual understanding, and domain-specific applications. Additionally, we compile and describe training datasets and evaluation benchmarks, conducting comparisons among existing models in terms of performance and computational requirements. Overall, this survey offers a comprehensive overview of the current state of the art, laying the groundwork for future MLLMs.
Authors: Runlong Zhou, Simon S. Du, Beibin Li
Abstract: As language models (LMs) demonstrate their capabilities in various fields, their application to tasks requiring multi-round interactions has become increasingly popular. These tasks usually have complex dynamics, so supervised fine-tuning (SFT) on a limited offline dataset does not yield good performance. However, only a few works attempted to directly train the LMs within interactive decision-making environments. We aim to create an effective approach to fine-tune LMs with online reinforcement learning (RL) in these environments. We propose Reflect-RL, a two-player system to fine-tune an LM using SFT and online RL, where a frozen reflection model (player) assists the policy model (player). To generate data for the warm-up SFT stage, we use negative example generation to enhance the error-correction ability of the reflection model. Furthermore, we designed single-prompt action enumeration and applied curriculum learning to allow the policy model to learn more efficiently. Empirically, we verify that Reflect-RL outperforms SFT and online RL without reflection. Testing results indicate GPT-2 XL 1.56B fine-tuned with Reflect-RL outperforms larger open-source LMs, such as Mistral 7B. The benchmarks, dataset, and code involved in this work are publicly available: https://github.com/zhourunlong/Reflect-RL.
Authors: Martin Gubri, Dennis Ulmer, Hwaran Lee, Sangdoo Yun, Seong Joon Oh
Abstract: Large Language Model (LLM) services and models often come with legal rules on who can use them and how they must use them. Assessing the compliance of the released LLMs is crucial, as these rules protect the interests of the LLM contributor and prevent misuse. In this context, we describe the novel fingerprinting problem of Black-box Identity Verification (BBIV). The goal is to determine whether a third-party application uses a certain LLM through its chat function. We propose a method called Targeted Random Adversarial Prompt (TRAP) that identifies the specific LLM in use. We repurpose adversarial suffixes, originally proposed for jailbreaking, to get a pre-defined answer from the target LLM, while other models give random answers. TRAP detects the target LLMs with over 95% true positive rate at under 0.2% false positive rate even after a single interaction. TRAP remains effective even if the LLM has minor changes that do not significantly alter the original function.
Authors: Victoria Lin, Eli Ben-Michael, Louis-Philippe Morency
Abstract: As large language models (LLMs) see greater use in academic and commercial settings, there is increasing interest in methods that allow language models to generate texts aligned with human preferences. In this paper, we present an initial exploration of language model optimization for human preferences from direct outcome datasets, where each sample consists of a text and an associated numerical outcome measuring the reader's response. We first propose that language model optimization should be viewed as a causal problem to ensure that the model correctly learns the relationship between the text and the outcome. We formalize this causal language optimization problem, and we develop a method--causal preference optimization (CPO)--that solves an unbiased surrogate objective for the problem. We further extend CPO with doubly robust CPO (DR-CPO), which reduces the variance of the surrogate objective while retaining provably strong guarantees on bias. Finally, we empirically demonstrate the effectiveness of (DR-)CPO in optimizing state-of-the-art LLMs for human preferences on direct outcome data, and we validate the robustness of DR-CPO under difficult confounding conditions.
Authors: Yuesong Shen, Nico Daheim, Bai Cong, Peter Nickl, Gian Maria Marconi, Clement Bazan, Rio Yokota, Iryna Gurevych, Daniel Cremers, Mohammad Emtiyaz Khan, Thomas M\"ollenhoff
Abstract: We give extensive empirical evidence against the common belief that variational learning is ineffective for large neural networks. We show that an optimizer called Improved Variational Online Newton (IVON) consistently matches or outperforms Adam for training large networks such as GPT-2 and ResNets from scratch. IVON's computational costs are nearly identical to Adam but its predictive uncertainty is better. We show several new use cases of IVON where we improve finetuning and model merging in Large Language Models, accurately predict generalization error, and faithfully estimate sensitivity to data. We find overwhelming evidence that variational learning is effective.
Authors: Mingjia Huo, Sai Ashish Somayajula, Youwei Liang, Ruisi Zhang, Farinaz Koushanfar, Pengtao Xie
Abstract: Large language models generate high-quality responses with potential misinformation, underscoring the need for regulation by distinguishing AI-generated and human-written texts. Watermarking is pivotal in this context, which involves embedding hidden markers in texts during the LLM inference phase, which is imperceptible to humans. Achieving both the detectability of inserted watermarks and the semantic quality of generated texts is challenging. While current watermarking algorithms have made promising progress in this direction, there remains significant scope for improvement. To address these challenges, we introduce a novel multi-objective optimization (MOO) approach for watermarking that utilizes lightweight networks to generate token-specific watermarking logits and splitting ratios. By leveraging MOO to optimize for both detection and semantic objective functions, our method simultaneously achieves detectability and semantic integrity. Experimental results show that our method outperforms current watermarking techniques in enhancing the detectability of texts generated by LLMs while maintaining their semantic coherence. Our code is available at https://github.com/mignonjia/TS_watermark.
Authors: Tarun Kalluri, Bodhisattwa Prasad Majumder, Manmohan Chandraker
Abstract: We introduce LaGTran, a novel framework that utilizes text supervision to guide robust transfer of discriminative knowledge from labeled source to unlabeled target data with domain gaps. While unsupervised adaptation methods have been established to address this problem, they show limitations in handling challenging domain shifts due to their exclusive operation within the pixel-space. Motivated by our observation that semantically richer text modality has more favorable transfer properties, we devise a transfer mechanism to use a source-trained text-classifier to generate predictions on the target text descriptions, and utilize these predictions as supervision for the corresponding images. Our approach driven by language guidance is surprisingly easy and simple, yet significantly outperforms all prior approaches on challenging datasets like GeoNet and DomainNet, validating its extreme effectiveness. To further extend the scope of our study beyond images, we introduce a new benchmark called Ego2Exo to study ego-exo transfer in videos and find that our language-aided approach LaGTran yields significant gains in this highly challenging and non-trivial transfer setting. Code, models, and proposed datasets are publicly available at https://tarun005.github.io/lagtran/.
Authors: Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, Ion Stoica
Abstract: Large Language Models (LLMs) applied to code-related applications have emerged as a prominent field, attracting significant interest from both academia and industry. However, as new and improved LLMs are developed, existing evaluation benchmarks (e.g., HumanEval, MBPP) are no longer sufficient for assessing their capabilities. In this work, we propose LiveCodeBench, a comprehensive and contamination-free evaluation of LLMs for code, which continuously collects new problems over time from contests across three competition platforms, namely LeetCode, AtCoder, and CodeForces. Notably, our benchmark also focuses on a broader range of code related capabilities, such as self-repair, code execution, and test output prediction, beyond just code generation. Currently, LiveCodeBench hosts four hundred high-quality coding problems that were published between May 2023 and May 2024. We have evaluated 18 base LLMs and 34 instruction-tuned LLMs on LiveCodeBench. We present empirical findings on contamination, holistic performance comparisons, potential overfitting in existing benchmarks as well as individual model comparisons. We will release all prompts and model completions for further community analysis, along with a general toolkit for adding new scenarios and model
Authors: Peter Baile Chen, Yi Zhang, Dan Roth
Abstract: Retrieving relevant tables containing the necessary information to accurately answer a given question over tables is critical to open-domain question-answering (QA) systems. Previous methods assume the answer to such a question can be found either in a single table or multiple tables identified through question decomposition or rewriting. However, neither of these approaches is sufficient, as many questions require retrieving multiple tables and joining them through a join plan that cannot be discerned from the user query itself. If the join plan is not considered in the retrieval stage, the subsequent steps of reasoning and answering based on those retrieved tables are likely to be incorrect. To address this problem, we introduce a method that uncovers useful join relations for any query and database during table retrieval. We use a novel re-ranking method formulated as a mixed-integer program that considers not only table-query relevance but also table-table relevance that requires inferring join relationships. Our method outperforms the state-of-the-art approaches for table retrieval by up to 9.3% in F1 score and for end-to-end QA by up to 5.4% in accuracy.
Authors: Surabhi S. Nath, Peter Dayan, Claire Stevenson
Abstract: Large language models appear quite creative, often performing on par with the average human on creative tasks. However, research on LLM creativity has focused solely on \textit{products}, with little attention on the creative \textit{process}. Process analyses of human creativity often require hand-coded categories or exploit response times, which do not apply to LLMs. We provide an automated method to characterise how humans and LLMs explore semantic spaces on the Alternate Uses Task, and contrast with behaviour in a Verbal Fluency Task. We use sentence embeddings to identify response categories and compute semantic similarities, which we use to generate jump profiles. Our results corroborate earlier work in humans reporting both persistent (deep search in few semantic spaces) and flexible (broad search across multiple semantic spaces) pathways to creativity, where both pathways lead to similar creativity scores. LLMs were found to be biased towards either persistent or flexible paths, that varied across tasks. Though LLMs as a population match human profiles, their relationship with creativity is different, where the more flexible models score higher on creativity. Our dataset and scripts are available on \href{https://github.com/surabhisnath/Creative_Process}{GitHub}.
Authors: Alejandro Tlaie
Abstract: Large Language Models (LLMs) have become central to advancing automation and decision-making across various sectors, raising significant ethical questions. This study proposes a comprehensive comparative analysis of the most advanced LLMs to assess their moral profiles. We subjected several state-of-the-art models to a selection of ethical dilemmas and found that all the proprietary ones are mostly utilitarian and all of the open-weights ones align mostly with values-based ethics. Furthermore, when using the Moral Foundations Questionnaire, all models we probed - except for Llama 2-7B - displayed a strong liberal bias. Lastly, in order to causally intervene in one of the studied models, we propose a novel similarity-specific activation steering technique. Using this method, we were able to reliably steer the model's moral compass to different ethical schools. All of these results showcase that there is an ethical dimension in already deployed LLMs, an aspect that is generally overlooked.
Authors: Zixian Guo, Ming Liu, Zhilong Ji, Jinfeng Bai, Yiwen Guo, Wangmeng Zuo
Abstract: Learning a skill generally relies on both practical experience by doer and insightful high-level guidance by instructor. Will this strategy also work well for solving complex non-convex optimization problems? Here, a common gradient-based optimizer acts like a disciplined doer, making locally optimal update at each step. Recent methods utilize large language models (LLMs) to optimize solutions for concrete problems by inferring from natural language instructions, akin to a high-level instructor. In this paper, we show that these two optimizers are complementary to each other, suggesting a collaborative optimization approach. The gradient-based optimizer and LLM-based optimizer are combined in an interleaved manner. We instruct LLMs using task descriptions and timely optimization trajectories recorded during gradient-based optimization. Inferred results from LLMs are used as restarting points for the next stage of gradient optimization. By leveraging both the locally rigorous gradient-based optimizer and the high-level deductive LLM-based optimizer, our combined optimization method consistently yields improvements over competitive baseline prompt tuning methods. Our results demonstrate the synergistic effect of conventional gradient-based optimization and the inference ability of LLMs. The code is released at https://github.com/guozix/LLM-catalyst.
Authors: Alaa Nfissi, Wassim Bouachir, Nizar Bouguila, Brian Mishara
Abstract: In speech emotion recognition (SER), using predefined features without considering their practical importance may lead to high dimensional datasets, including redundant and irrelevant information. Consequently, high-dimensional learning often results in decreasing model accuracy while increasing computational complexity. Our work underlines the importance of carefully considering and analyzing features in order to build efficient SER systems. We present a new supervised SER method based on an efficient feature engineering approach. We pay particular attention to the explainability of results to evaluate feature relevance and refine feature sets. This is performed iteratively through feature evaluation loop, using Shapley values to boost feature selection and improve overall framework performance. Our approach allows thus to balance the benefits between model performance and transparency. The proposed method outperforms human-level performance (HLP) and state-of-the-art machine learning methods in emotion recognition on the TESS dataset. The source code of this paper is publicly available at https://github.com/alaaNfissi/Iterative-Feature-Boosting-for-Explainable-Speech-Emotion-Recognition.
URLs: https://github.com/alaaNfissi/Iterative-Feature-Boosting-for-Explainable-Speech-Emotion-Recognition.
Authors: Jiaqi Xue, Mengxin Zheng, Yebowen Hu, Fei Liu, Xun Chen, Qian Lou
Abstract: Large Language Models (LLMs) are constrained by outdated information and a tendency to generate incorrect data, commonly referred to as "hallucinations." Retrieval-Augmented Generation (RAG) addresses these limitations by combining the strengths of retrieval-based methods and generative models. This approach involves retrieving relevant information from a large, up-to-date dataset and using it to enhance the generation process, leading to more accurate and contextually appropriate responses. Despite its benefits, RAG introduces a new attack surface for LLMs, particularly because RAG databases are often sourced from public data, such as the web. In this paper, we propose \TrojRAG{} to identify the vulnerabilities and attacks on retrieval parts (RAG database) and their indirect attacks on generative parts (LLMs). Specifically, we identify that poisoning several customized content passages could achieve a retrieval backdoor, where the retrieval works well for clean queries but always returns customized poisoned adversarial queries. Triggers and poisoned passages can be highly customized to implement various attacks. For example, a trigger could be a semantic group like "The Republican Party, Donald Trump, etc." Adversarial passages can be tailored to different contents, not only linked to the triggers but also used to indirectly attack generative LLMs without modifying them. These attacks can include denial-of-service attacks on RAG and semantic steering attacks on LLM generations conditioned by the triggers. Our experiments demonstrate that by just poisoning 10 adversarial passages can induce 98.2\% success rate to retrieve the adversarial passages. Then, these passages can increase the reject ratio of RAG-based GPT-4 from 0.01\% to 74.6\% or increase the rate of negative responses from 0.22\% to 72\% for targeted queries.
Authors: Gregory Schwartzman
Abstract: We report that ChatGPT 4 and 4o are susceptible to a prompt injection attack that allows an attacker to exfiltrate users' personal data. It is applicable without the use of any 3rd party tools and all users are currently affected. This vulnerability is exacerbated by the recent introduction of ChatGPT's memory feature, which allows an attacker to command ChatGPT to monitor the user for the desired personal data.
Authors: Bowen Jiang, Yangxinyu Xie, Xiaomeng Wang, Weijie J. Su, Camillo J. Taylor, Tanwi Mallick
Abstract: Rationality is the quality of being guided by reason, characterized by logical thinking and decision-making that align with evidence and logical rules. This quality is essential for effective problem-solving, as it ensures that solutions are well-founded and systematically derived. Despite the advancements of large language models (LLMs) in generating human-like text with remarkable accuracy, they present biases inherited from the training data, inconsistency across different contexts, and difficulty understanding complex scenarios involving multiple layers of context. Therefore, recent research attempts to leverage the strength of multiple agents working collaboratively with various types of data and tools for enhanced consistency and reliability. To that end, this paper aims to understand whether multi-modal and multi-agent systems are advancing toward rationality by surveying the state-of-the-art works, identifying advancements over single-agent and single-modal systems in terms of rationality, and discussing open problems and future directions. We maintain an open repository at https://github.com/bowen-upenn/MMMA_Rationality.
Authors: Alaa Nfissi, Wassim Bouachir, Nizar Bouguila, Brian Mishara
Abstract: Speech emotion recognition (SER) has gained significant attention due to its several application fields, such as mental health, education, and human-computer interaction. However, the accuracy of SER systems is hindered by high-dimensional feature sets that may contain irrelevant and redundant information. To overcome this challenge, this study proposes an iterative feature boosting approach for SER that emphasizes feature relevance and explainability to enhance machine learning model performance. Our approach involves meticulous feature selection and analysis to build efficient SER systems. In addressing our main problem through model explainability, we employ a feature evaluation loop with Shapley values to iteratively refine feature sets. This process strikes a balance between model performance and transparency, which enables a comprehensive understanding of the model's predictions. The proposed approach offers several advantages, including the identification and removal of irrelevant and redundant features, leading to a more effective model. Additionally, it promotes explainability, facilitating comprehension of the model's predictions and the identification of crucial features for emotion determination. The effectiveness of the proposed method is validated on the SER benchmarks of the Toronto emotional speech set (TESS), Berlin Database of Emotional Speech (EMO-DB), Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), and Surrey Audio-Visual Expressed Emotion (SAVEE) datasets, outperforming state-of-the-art methods. To the best of our knowledge, this is the first work to incorporate model explainability into an SER framework. The source code of this paper is publicly available via this https://github.com/alaaNfissi/Unveiling-Hidden-Factors-Explainable-AI-for-Feature-Boosting-in-Speech-Emotion-Recognition.
Authors: Lun Wang, Om Thakkar, Zhong Meng, Nicole Rafidi, Rohit Prabhavalkar, Arun Narayanan
Abstract: Gradient clipping plays a vital role in training large-scale automatic speech recognition (ASR) models. It is typically applied to minibatch gradients to prevent gradient explosion, and to the individual sample gradients to mitigate unintended memorization. This work systematically investigates the impact of a specific granularity of gradient clipping, namely per-core clip-ping (PCC), across training a wide range of ASR models. We empirically demonstrate that PCC can effectively mitigate unintended memorization in ASR models. Surprisingly, we find that PCC positively influences ASR performance metrics, leading to improved convergence rates and reduced word error rates. To avoid tuning the additional hyperparameter introduced by PCC, we further propose a novel variant, adaptive per-core clipping (APCC), for streamlined optimization. Our findings highlight the multifaceted benefits of PCC as a strategy for robust, privacy-forward ASR model training.
Authors: Marianna Nezhurina, Lucia Cipolina-Kun, Mehdi Cherti, Jenia Jitsev
Abstract: Large Language Models (LLMs) are often described as being instances of foundation models - that is, models that transfer strongly across various tasks and conditions in few-show or zero-shot manner, while exhibiting scaling laws that predict function improvement when increasing the pre-training scale. These claims of excelling in different functions and tasks rely on measurements taken across various sets of standardized benchmarks showing high scores for such models. We demonstrate here a dramatic breakdown of function and reasoning capabilities of state-of-the-art models trained at the largest available scales which claim strong function, using a simple, short, conventional common sense problem formulated in concise natural language, easily solvable by humans. The breakdown is dramatic, as models also express strong overconfidence in their wrong solutions, while providing often non-sensical "reasoning"-like explanations akin to confabulations to justify and backup the validity of their clearly failed responses, making them sound plausible. Various standard interventions in an attempt to get the right solution, like various type of enhanced prompting, or urging the models to reconsider the wrong solutions again by multi step re-evaluation, fail. We take these initial observations to the scientific and technological community to stimulate urgent re-assessment of the claimed capabilities of current generation of LLMs, Such re-assessment also requires common action to create standardized benchmarks that would allow proper detection of such basic reasoning deficits that obviously manage to remain undiscovered by current state-of-the-art evaluation procedures and benchmarks. Code for reproducing experiments in the paper and raw experiments data can be found at https://github.com/LAION-AI/AIW
Authors: Wenyan Li, Jiaang Li, Rita Ramos, Raphael Tang, Desmond Elliott
Abstract: Recent advances in retrieval-augmented models for image captioning highlight the benefit of retrieving related captions for efficient, lightweight models with strong domain-transfer capabilities. While these models demonstrate the success of retrieval augmentation, retrieval models are still far from perfect in practice: the retrieved information can sometimes mislead the model, resulting in incorrect generation and worse performance. In this paper, we analyze the robustness of a retrieval-augmented captioning model SmallCap. Our analysis shows that the model is sensitive to tokens that appear in the majority of the retrieved captions, and the input attribution shows that those tokens are likely copied into the generated output. Given these findings, we propose to train the model by sampling retrieved captions from more diverse sets. This decreases the chance that the model learns to copy majority tokens, and improves both in-domain and cross-domain performance.
Authors: Xiaoyu Zhang, Yishan Li, Jiayin Wang, Bowen Sun, Weizhi Ma, Peijie Sun, Min Zhang
Abstract: The explainability of recommender systems has attracted significant attention in academia and industry. Many efforts have been made for explainable recommendations, yet evaluating the quality of the explanations remains a challenging and unresolved issue. In recent years, leveraging LLMs as evaluators presents a promising avenue in Natural Language Processing tasks (e.g., sentiment classification, information extraction), as they perform strong capabilities in instruction following and common-sense reasoning. However, evaluating recommendation explanatory texts is different from these NLG tasks, as its criteria are related to human perceptions and are usually subjective. In this paper, we investigate whether LLMs can serve as evaluators of recommendation explanations. To answer the question, we utilize real user feedback on explanations given from previous work and additionally collect third-party annotations and LLM evaluations. We design and apply a 3-level meta evaluation strategy to measure the correlation between evaluator labels and the ground truth provided by users. Our experiments reveal that LLMs, such as GPT4, can provide comparable evaluations with appropriate prompts and settings. We also provide further insights into combining human labels with the LLM evaluation process and utilizing ensembles of multiple heterogeneous LLM evaluators to enhance the accuracy and stability of evaluations. Our study verifies that utilizing LLMs as evaluators can be an accurate, reproducible and cost-effective solution for evaluating recommendation explanation texts. Our code is available at https://github.com/Xiaoyu-SZ/LLMasEvaluator.