Authors: Weiliang Zhao, Daniel Ben-Levi, Junfeng Yang, Chengzhi Mao
Abstract: We have uncovered a powerful jailbreak technique that leverages large language models' ability to diverge from prior context, enabling them to bypass safety constraints and generate harmful outputs. By simply instructing the LLM to deviate and obfuscate previous attacks, our method dramatically outperforms existing approaches, achieving up to a 62% higher success rate in compromising nine leading chatbots, including GPT-4, Gemini, and Llama, while using only 13% of the queries. This revelation exposes a critical flaw in current LLM safety training, suggesting that existing methods may merely mask vulnerabilities rather than eliminate them. Our findings sound an urgent alarm for the need to revolutionize testing methodologies to ensure robust and reliable LLM security.
Authors: Saketh Bachu, Erfan Shayegani, Trishna Chakraborty, Rohit Lal, Arindam Dutta, Chengyu Song, Yue Dong, Nael Abu-Ghazaleh, Amit K. Roy-Chowdhury
Abstract: Vision-language models (VLMs) have improved significantly in multi-modal tasks, but their more complex architecture makes their safety alignment more challenging than the alignment of large language models (LLMs). In this paper, we reveal an unfair distribution of safety across the layers of VLM's vision encoder, with earlier and middle layers being disproportionately vulnerable to malicious inputs compared to the more robust final layers. This 'cross-layer' vulnerability stems from the model's inability to generalize its safety training from the default architectural settings used during training to unseen or out-of-distribution scenarios, leaving certain layers exposed. We conduct a comprehensive analysis by projecting activations from various intermediate layers and demonstrate that these layers are more likely to generate harmful outputs when exposed to malicious inputs. Our experiments with LLaVA-1.5 and Llama 3.2 show discrepancies in attack success rates and toxicity scores across layers, indicating that current safety alignment strategies focused on a single default layer are insufficient.
Authors: Hellina Hailu Nigatu, Zeerak Talat
Abstract: Mainstream Natural Language Processing (NLP) research has ignored the majority of the world's languages. In moving from excluding the majority of the world's languages to blindly adopting what we make for English, we first risk importing the same harms we have at best mitigated and at least measured for English. However, in evaluating and mitigating harms arising from adopting new technologies into such contexts, we often disregard (1) the actual community needs of Language Technologies, and (2) biases and fairness issues within the context of the communities. In this extended abstract, we consider fairness, bias, and inclusion in Language Technologies through the lens of the Capabilities Approach. The Capabilities Approach centers on what people are capable of achieving, given their intersectional social, political, and economic contexts instead of what resources are (theoretically) available to them. We detail the Capabilities Approach, its relationship to multilingual and multicultural evaluation, and how the framework affords meaningful collaboration with community members in defining and measuring the harms of Language Technologies.
Authors: Anand Syamkumar, Nora Tseng, Kaycie Barron, Shanglin Yang, Shamya Karumbaiah, Rheeya Uppal, Junjie Hu
Abstract: Large language models (LLMs) offer promise in generating educational content, providing instructor feedback, and reducing teacher workload on assessments. While prior studies have focused on studying LLM-powered learning analytics, limited research has examined how effective LLMs are in a bilingual context. In this paper, we study the effectiveness of multilingual large language models (MLLMs) across monolingual (English-only, Spanish-only) and bilingual (Spanglish) student writing. We present a learning analytics use case that details LLM performance in assessing acceptable and unacceptable explanations of Science and Social Science concepts. Our findings reveal a significant bias in the grading performance of pre-trained models for bilingual writing compared to English-only and Spanish-only writing. Following this, we fine-tune open-source MLLMs including Llama 3.1 and Mistral NeMo using synthetic datasets generated in English, Spanish, and Spanglish. Our experiments indicate that the models perform significantly better for all three languages after fine-tuning with bilingual data. This study highlights the potential of enhancing MLLM effectiveness to support authentic language practices amongst bilingual learners. It also aims to illustrate the value of incorporating non-English languages into the design and implementation of language models in education.
Authors: Melusi Malinga, Isaac Lupanda, Mike Wa Nkongolo, Phil van Deventer
Abstract: South Africa and the Democratic Republic of Congo (DRC) present a complex linguistic landscape with languages such as Zulu, Sepedi, Afrikaans, French, English, and Tshiluba (Ciluba), which creates unique challenges for AI-driven translation and sentiment analysis systems due to a lack of accurately labeled data. This study seeks to address these challenges by developing a multilingual lexicon designed for French and Tshiluba, now expanded to include translations in English, Afrikaans, Sepedi, and Zulu. The lexicon enhances cultural relevance in sentiment classification by integrating language-specific sentiment scores. A comprehensive testing corpus is created to support translation and sentiment analysis tasks, with machine learning models such as Random Forest, Support Vector Machine (SVM), Decision Trees, and Gaussian Naive Bayes (GNB) trained to predict sentiment across low resource languages (LRLs). Among them, the Random Forest model performed particularly well, capturing sentiment polarity and handling language-specific nuances effectively. Furthermore, Bidirectional Encoder Representations from Transformers (BERT), a Large Language Model (LLM), is applied to predict context-based sentiment with high accuracy, achieving 99% accuracy and 98% precision, outperforming other models. The BERT predictions were clarified using Explainable AI (XAI), improving transparency and fostering confidence in sentiment classification. Overall, findings demonstrate that the proposed lexicon and machine learning models significantly enhance translation and sentiment analysis for LRLs in South Africa and the DRC, laying a foundation for future AI models that support underrepresented languages, with applications across education, governance, and business in multilingual contexts.
Authors: Manuel Nunez Martinez, Sonja Schmer-Galunder, Zoey Liu, Sangpil Youm, Chathuri Jayaweera, Bonnie J. Dorr
Abstract: The unchecked spread of digital information, combined with increasing political polarization and the tendency of individuals to isolate themselves from opposing political viewpoints, has driven researchers to develop systems for automatically detecting political bias in media. This trend has been further fueled by discussions on social media. We explore methods for categorizing bias in US news articles, comparing rule-based and deep learning approaches. The study highlights the sensitivity of modern self-learning systems to unconstrained data ingestion, while reconsidering the strengths of traditional rule-based systems. Applying both models to left-leaning (CNN) and right-leaning (FOX) news articles, we assess their effectiveness on data beyond the original training and test sets.This analysis highlights each model's accuracy, offers a framework for exploring deep-learning explainability, and sheds light on political bias in US news media. We contrast the opaque architecture of a deep learning model with the transparency of a linguistically informed rule-based model, showing that the rule-based model performs consistently across different data conditions and offers greater transparency, whereas the deep learning model is dependent on the training set and struggles with unseen data.
Authors: Jierui Li, Hung Le, Yinbo Zhou, Caiming Xiong, Silvio Savarese, Doyen Sahoo
Abstract: Pre-trained on massive amounts of code and text data, large language models (LLMs) have demonstrated remarkable achievements in performing code generation tasks. With additional execution-based feedback, these models can act as agents with capabilities to self-refine and improve generated code autonomously. However, on challenging coding tasks with extremely large search space, current agentic approaches still struggle with multi-stage planning, generating, and debugging. To address this problem, we propose CodeTree, a framework for LLM agents to efficiently explore the search space in different stages of the code generation process. Specifically, we adopted a unified tree structure to explicitly explore different coding strategies, generate corresponding coding solutions, and subsequently refine the solutions. In each stage, critical decision-making (ranking, termination, expanding) of the exploration process is guided by both the environmental execution-based feedback and LLM-agent-generated feedback. We comprehensively evaluated CodeTree on 7 code generation benchmarks and demonstrated the significant performance gains of CodeTree against strong baselines. Using GPT-4o as the base model, we consistently achieved top results of 95.1 on HumanEval, 98.7 on MBPP, and 43.0 on CodeContests. On the challenging SWEBench benchmark, our approach led to significant performance gains.
Authors: Jason Wei, Nguyen Karina, Hyung Won Chung, Yunxin Joy Jiao, Spencer Papay, Amelia Glaese, John Schulman, William Fedus
Abstract: We present SimpleQA, a benchmark that evaluates the ability of language models to answer short, fact-seeking questions. We prioritized two properties in designing this eval. First, SimpleQA is challenging, as it is adversarially collected against GPT-4 responses. Second, responses are easy to grade, because questions are created such that there exists only a single, indisputable answer. Each answer in SimpleQA is graded as either correct, incorrect, or not attempted. A model with ideal behavior would get as many questions correct as possible while not attempting the questions for which it is not confident it knows the correct answer. SimpleQA is a simple, targeted evaluation for whether models "know what they know," and our hope is that this benchmark will remain relevant for the next few generations of frontier models. SimpleQA can be found at https://github.com/openai/simple-evals.
Authors: Yicheng Gao, Gonghan Xu, Zhe Wang, Arman Cohan
Abstract: Recent advances in large language models (LLMs) show the potential of using LLMs as evaluators for assessing the quality of text generations from LLMs. However, applying LLM evaluators naively to compare or judge between different systems can lead to unreliable results due to the intrinsic win rate estimation bias of LLM evaluators. In order to mitigate this problem, we propose two calibration methods, Bayesian Win Rate Sampling (BWRS) and Bayesian Dawid-Skene, both of which leverage Bayesian inference to more accurately infer the true win rate of generative language models. We empirically validate our methods on six datasets covering story generation, summarization, and instruction following tasks. We show that both our methods are effective in improving the accuracy of win rate estimation using LLMs as evaluators, offering a promising direction for reliable automatic text quality evaluation.
Authors: Ishika Agarwal, Krishna Killamsetty, Lucian Popa, Marina Danilevksy
Abstract: Fine-tuning large language models (LLMs) is essential for enhancing their performance on specific tasks but is often resource-intensive due to redundant or uninformative data. To address this inefficiency, we introduce DELIFT (Data Efficient Language model Instruction Fine-Tuning), a novel algorithm that systematically optimizes data selection across the three key stages of fine-tuning: (1) instruction tuning, (2) task-specific fine-tuning (e.g., reasoning, question-answering), and (3) continual fine-tuning (e.g., incorporating new data versions). Unlike existing methods that focus on single-stage optimization or rely on computationally intensive gradient calculations, DELIFT operates efficiently across all stages. Central to our approach is a pairwise utility metric that quantifies how beneficial a data sample is for improving the model's responses to other samples, effectively measuring the informational value relative to the model's current capabilities. By leveraging different submodular functions applied to this metric, DELIFT selects diverse and optimal subsets that are useful across all stages of fine-tuning. Experiments across various tasks and model scales demonstrate that DELIFT can reduce the fine-tuning data size by up to 70% without compromising performance, offering significant computational savings and outperforming existing methods in both efficiency and efficacy.
Authors: Sonia K. Murthy, Tomer Ullman, Jennifer Hu
Abstract: Researchers in social science and psychology have recently proposed using large language models (LLMs) as replacements for humans in behavioral research. In addition to arguments about whether LLMs accurately capture population-level patterns, this has raised questions about whether LLMs capture human-like conceptual diversity. Separately, it is debated whether post-training alignment (RLHF or RLAIF) affects models' internal diversity. Inspired by human studies, we use a new way of measuring the conceptual diversity of synthetically-generated LLM "populations" by relating the internal variability of simulated individuals to the population-level variability. We use this approach to evaluate non-aligned and aligned LLMs on two domains with rich human behavioral data. While no model reaches human-like diversity, aligned models generally display less diversity than their instruction fine-tuned counterparts. Our findings highlight potential trade-offs between increasing models' value alignment and decreasing the diversity of their conceptual representations.
Authors: Whanhee Cho
Abstract: The attention to table understanding using recent natural language models has been growing. However, most related works tend to focus on learning the structure of the table directly. Just as humans improve their understanding of sentences by comparing them, they can also enhance their understanding by comparing tables. With this idea, in this paper, we introduce ACCIO, tAble understanding enhanCed via Contrastive learnIng with aggregatiOns, a novel approach to enhancing table understanding by contrasting original tables with their pivot summaries through contrastive learning. ACCIO trains an encoder to bring these table pairs closer together. Through validation via column type annotation, ACCIO achieves competitive performance with a macro F1 score of 91.1 compared to state-of-the-art methods. This work represents the first attempt to utilize pairs of tables for table embedding, promising significant advancements in table comprehension. Our code is available at https://github.com/whnhch/ACCIO/.
Authors: Jared Fernandez, Yonatan Bisk, Emma Strubell
Abstract: Large Language Models (LLMs) trained on web-scale text corpora have been shown to capture world knowledge in their parameters. However, the mechanism by which language models store different types of knowledge is poorly understood. In this work, we examine two types of knowledge relating to temporally sensitive entities and demonstrate that each type is localized to different sets of parameters within the LLMs. We hypothesize that the lack of consideration of the locality of knowledge in existing continual learning methods contributes to both: the failed uptake of new information, and catastrophic forgetting of previously learned information. We observe that sequences containing references to updated and newly mentioned entities exhibit larger gradient norms in a subset of layers. We demonstrate that targeting parameter updates to these relevant layers can improve the performance of continually pretraining on language containing temporal drift.
Authors: Yohei Seki, Hakusen Shu, Ana\"is Lhuissier, Hanwool Lee, Juyeon Kang, Min-Yuh Day, Chung-Chi Chen
Abstract: Promises made by politicians, corporate leaders, and public figures have a significant impact on public perception, trust, and institutional reputation. However, the complexity and volume of such commitments, coupled with difficulties in verifying their fulfillment, necessitate innovative methods for assessing their credibility. This paper introduces the concept of Promise Verification, a systematic approach involving steps such as promise identification, evidence assessment, and the evaluation of timing for verification. We propose the first multilingual dataset, ML-Promise, which includes English, French, Chinese, Japanese, and Korean, aimed at facilitating in-depth verification of promises, particularly in the context of Environmental, Social, and Governance (ESG) reports. Given the growing emphasis on corporate environmental contributions, this dataset addresses the challenge of evaluating corporate promises, especially in light of practices like greenwashing. Our findings also explore textual and image-based baselines, with promising results from retrieval-augmented generation (RAG) approaches. This work aims to foster further discourse on the accountability of public commitments across multiple languages and domains.
Authors: Young-Jun Lee, Dokyong Lee, Junyoung Youn, Kyeongjin Oh, Ho-Jin Choi
Abstract: To increase social bonding with interlocutors, humans naturally acquire the ability to respond appropriately in a given situation by considering which conversational skill is most suitable for the response - a process we call skill-of-mind. For large language model (LLM)-based conversational agents, planning appropriate conversational skills, as humans do, is challenging due to the complexity of social dialogue, especially in interactive scenarios. To address this, we propose a skill-of-mind-annotated conversation dataset, named Multifaceted Skill-of-Mind, which includes multi-turn and multifaceted conversational skills across various interactive scenarios (e.g., long-term, counseling, task-oriented), grounded in diverse social contexts (e.g., demographics, persona, rules of thumb). This dataset consists of roughly 100K conversations. Using this dataset, we introduce a new family of skill-of-mind-infused LLMs, named Thanos, with model sizes of 1B, 3B, and 8B parameters. With extensive experiments, these models successfully demonstrate the skill-of-mind process and exhibit strong generalizability in inferring multifaceted skills across a variety of domains. Moreover, we show that Thanos significantly enhances the quality of responses generated by LLM-based conversational agents and promotes prosocial behavior in human evaluations.
Authors: Xinyu Zhang, Jing Lu, Vinh Q. Tran, Tal Schuster, Donald Metzler, Jimmy Lin
Abstract: Human understanding of language is robust to different word choices as far as they represent similar semantic concepts. To what extent does our human intuition transfer to language models, which represent all subwords as distinct embeddings? In this work, we take an initial step on measuring the role of shared semantics among subwords in the encoder-only multilingual language models (mLMs). To this end, we form "semantic tokens" by merging the semantically similar subwords and their embeddings, and evaluate the updated mLMs on 5 heterogeneous multilingual downstream tasks. Results show that the general shared semantics could get the models a long way in making the predictions on mLMs with different tokenizers and model sizes. Inspections on the grouped subwords show that they exhibit a wide range of semantic similarities, including synonyms and translations across many languages and scripts. Lastly, we found the zero-shot results with semantic tokens are on par or even better than the original models on certain classification tasks, suggesting that the shared subword-level semantics may serve as the anchors for cross-lingual transferring.
Authors: Lisa Alazraki, Marek Rei
Abstract: External tools help large language models (LLMs) succeed at tasks where they would otherwise typically fail. In existing frameworks, LLMs learn tool use either by in-context demonstrations or via full model fine-tuning on annotated data. As these approaches do not easily scale, a recent trend is to abandon them in favor of lightweight, parameter-efficient tuning paradigms. These methods allow quickly alternating between the frozen LLM and its specialised fine-tuned version, by switching on or off a handful of additional custom parameters. Hence, we postulate that the generalization ability of the frozen model can be leveraged to improve tool selection. We present Tool selECTion via meta-reasONing (TECTON), a two-phase system that first reasons over a task using a custom fine-tuned LM head and outputs candidate tools. Then, with the custom head disabled, it meta-reasons (i.e., it reasons over the previous reasoning process) to make a final choice. We show that TECTON results in substantial gains - both in-distribution and out-of-distribution - on a range of math reasoning datasets.
Authors: Rohan Kumar Yadav, Bimal Bhattarai, Abhik Jana, Lei Jiao, Seid Muhie Yimam
Abstract: Designing an explainable model becomes crucial now for Natural Language Processing(NLP) since most of the state-of-the-art machine learning models provide a limited explanation for the prediction. In the spectrum of an explainable model, Tsetlin Machine(TM) is promising because of its capability of providing word-level explanation using proposition logic. However, concern rises over the elaborated combination of literals (propositional logic) in the clause that makes the model difficult for humans to comprehend, despite having a transparent learning process. In this paper, we design a post-hoc pruning of clauses that eliminate the randomly placed literals in the clause thereby making the model more efficiently interpretable than the vanilla TM. Experiments on the publicly available YELP-HAT Dataset demonstrate that the proposed pruned TM's attention map aligns more with the human attention map than the vanilla TM's attention map. In addition, the pairwise similarity measure also surpasses the attention map-based neural network models. In terms of accuracy, the proposed pruning method does not degrade the accuracy significantly but rather enhances the performance up to 4% to 9% in some test data.
Authors: Leena G Pillai, Kavya Manohar, Basil K Raju, Elizabeth Sherly
Abstract: This paper presents a novel multistage fine-tuning strategy designed to enhance automatic speech recognition (ASR) performance in low-resource languages using OpenAI's Whisper model. In this approach we aim to build ASR model for languages with limited digital resources by sequentially adapting the model across linguistically similar languages. We experimented this on the Malasar language, a Dravidian language spoken by approximately ten thousand people in the Western Ghats of South India. Malasar language faces critical challenges for technological intervention due to its lack of a native script and absence of digital or spoken data resources. Working in collaboration with Wycliffe India and Malasar community members, we created a spoken Malasar corpus paired with transcription in Tamil script, a closely related major language. In our approach to build ASR model for Malasar, we first build an intermediate Tamil ASR, leveraging higher data availability for Tamil annotated speech. This intermediate model is subsequently fine-tuned on Malasar data, allowing for more effective ASR adaptation despite limited resources. The multistage fine-tuning strategy demonstrated significant improvements over direct fine-tuning on Malasar data alone, achieving a word error rate (WER) of 51.9%, which is 4.5% absolute reduction when compared to the direct fine-tuning method. Further a WER reduction to 47.3% was achieved through punctuation removal in post-processing, which addresses formatting inconsistencies that impact evaluation. Our results underscore the effectiveness of sequential multistage fine-tuning combined with targeted post-processing as a scalable strategy for ASR system development in low-resource languages, especially where linguistic similarities can be leveraged to bridge gaps in training data.
Authors: Noam Dahan, Gabriel Stanovsky
Abstract: Automatic summarization has consistently attracted attention, due to its versatility and wide application in various downstream tasks. Despite its popularity, we find that annotation efforts have largely been disjointed, and have lacked common terminology. Consequently, it is challenging to discover existing resources or identify coherent research directions. To address this, we survey a large body of work spanning 133 datasets in over 100 languages, creating a novel ontology covering sample properties, collection methods and distribution. With this ontology we make key observations, including the lack in accessible high-quality datasets for low-resource languages, and the field's over-reliance on the news domain and on automatically collected distant supervision. Finally, we make available a web interface that allows users to interact and explore our ontology and dataset collection, as well as a template for a summarization data card, which can be used to streamline future research into a more coherent body of work.
Authors: Ahlam Alrehili, Areej Alhothali
Abstract: Natural language processing (NLP) utilizes text data augmentation to overcome sample size constraints. Increasing the sample size is a natural and widely used strategy for alleviating these challenges. In this study, we chose Arabic to increase the sample size and correct grammatical errors. Arabic is considered one of the languages with limited resources for grammatical error correction (GEC). Furthermore, QALB-14 and QALB-15 are the only datasets used in most Arabic grammatical error correction research, with approximately 20,500 parallel examples, which is considered low compared with other languages. Therefore, this study aims to develop an Arabic corpus called "Tibyan" for grammatical error correction using ChatGPT. ChatGPT is used as a data augmenter tool based on a pair of Arabic sentences containing grammatical errors matched with a sentence free of errors extracted from Arabic books, called guide sentences. Multiple steps were involved in establishing our corpus, including the collection and pre-processing of a pair of Arabic texts from various sources, such as books and open-access corpora. We then used ChatGPT to generate a parallel corpus based on the text collected previously, as a guide for generating sentences with multiple types of errors. By engaging linguistic experts to review and validate the automatically generated sentences, we ensured that they were correct and error-free. The corpus was validated and refined iteratively based on feedback provided by linguistic experts to improve its accuracy. Finally, we used the Arabic Error Type Annotation tool (ARETA) to analyze the types of errors in the Tibyan corpus. Our corpus contained 49 of errors, including seven types: orthography, morphology, syntax, semantics, punctuation, merge, and split. The Tibyan corpus contains approximately 600 K tokens.
Authors: Amin Abdedaiem, Abdelhalim Hafedh Dahou, Mohamed Amine Cheragui, Brigitte Mathiak
Abstract: In the context of low-resource languages, the Algerian dialect (AD) faces challenges due to the absence of annotated corpora, hindering its effective processing, notably in Machine Learning (ML) applications reliant on corpora for training and assessment. This study outlines the development process of a specialized corpus for Fake News (FN) detection and sentiment analysis (SA) in AD called FASSILA. This corpus comprises 10,087 sentences, encompassing over 19,497 unique words in AD, and addresses the significant lack of linguistic resources in the language and covers seven distinct domains. We propose an annotation scheme for FN detection and SA, detailing the data collection, cleaning, and labelling process. Remarkable Inter-Annotator Agreement indicates that the annotation scheme produces consistent annotations of high quality. Subsequent classification experiments using BERT-based models and ML models are presented, demonstrate promising results and highlight avenues for further research. The dataset is made freely available on GitHub (https://github.com/amincoding/FASSILA) to facilitate future advancements in the field.
Authors: Ekaterina Artemova, Akim Tsvigun, Dominik Schlechtweg, Natalia Fedorova, Sergei Tilga, Boris Obmoroshev
Abstract: Training and deploying machine learning models relies on a large amount of human-annotated data. As human labeling becomes increasingly expensive and time-consuming, recent research has developed multiple strategies to speed up annotation and reduce costs and human workload: generating synthetic training data, active learning, and hybrid labeling. This tutorial is oriented toward practical applications: we will present the basics of each strategy, highlight their benefits and limitations, and discuss in detail real-life case studies. Additionally, we will walk through best practices for managing human annotators and controlling the quality of the final dataset. The tutorial includes a hands-on workshop, where attendees will be guided in implementing a hybrid annotation setup. This tutorial is designed for NLP practitioners from both research and industry backgrounds who are involved in or interested in optimizing data labeling projects.
Authors: Sparsh Jain, Ashwin Sankar, Devilal Choudhary, Dhairya Suman, Nikhil Narasimhan, Mohammed Safi Ur Rahman Khan, Anoop Kunchukuttan, Mitesh M Khapra, Raj Dabre
Abstract: Automatic Speech Translation (AST) datasets for Indian languages remain critically scarce, with public resources covering fewer than 10 of the 22 official languages. This scarcity has resulted in AST systems for Indian languages lagging far behind those available for high-resource languages like English. In this paper, we first evaluate the performance of widely-used AST systems on Indian languages, identifying notable performance gaps and challenges. Our findings show that while these systems perform adequately on read speech, they struggle significantly with spontaneous speech, including disfluencies like pauses and hesitations. Additionally, there is a striking absence of systems capable of accurately translating colloquial and informal language, a key aspect of everyday communication. To this end, we introduce BhasaAnuvaad, the largest publicly available dataset for AST involving 14 scheduled Indian languages spanning over 44,400 hours and 17M text segments. BhasaAnuvaad contains data for English speech to Indic text, as well as Indic speech to English text. This dataset comprises three key categories: (1) Curated datasets from existing resources, (2) Large-scale web mining, and (3) Synthetic data generation. By offering this diverse and expansive dataset, we aim to bridge the resource gap and promote advancements in AST for low-resource Indian languages, especially in handling spontaneous and informal speech patterns.
Authors: Aniket Deroy, Subhankar Maity
Abstract: Code-mixing, the integration of lexical and grammatical elements from multiple languages within a single sentence, is a widespread linguistic phenomenon, particularly prevalent in multilingual societies. In India, social media users frequently engage in code-mixed conversations using the Roman script, especially among migrant communities who form online groups to share relevant local information. This paper focuses on the challenges of extracting relevant information from code-mixed conversations, specifically within Roman transliterated Bengali mixed with English. This study presents a novel approach to address these challenges by developing a mechanism to automatically identify the most relevant answers from code-mixed conversations. We have experimented with a dataset comprising of queries and documents from Facebook, and Query Relevance files (QRels) to aid in this task. Our results demonstrate the effectiveness of our approach in extracting pertinent information from complex, code-mixed digital conversations, contributing to the broader field of natural language processing in multilingual and informal text environments. We use GPT-3.5 Turbo via prompting alongwith using the sequential nature of relevant documents to frame a mathematical model which helps to detect relevant documents corresponding to a query.
Authors: Hung Tuan Le, Long Truong To, Manh Trong Nguyen, Quyen Nguyen, Trong-Hop Do
Abstract: Determining the difficulty of a text involves assessing various textual features that may impact the reader's text comprehension, yet current research in Vietnamese has only focused on statistical features. This paper introduces a new approach that integrates statistical and semantic approaches to assessing text readability. Our research utilized three distinct datasets: the Vietnamese Text Readability Dataset (ViRead), OneStopEnglish, and RACE, with the latter two translated into Vietnamese. Advanced semantic analysis methods were employed for the semantic aspect using state-of-the-art language models such as PhoBERT, ViDeBERTa, and ViBERT. In addition, statistical methods were incorporated to extract syntactic and lexical features of the text. We conducted experiments using various machine learning models, including Support Vector Machine (SVM), Random Forest, and Extra Trees and evaluated their performance using accuracy and F1 score metrics. Our results indicate that a joint approach that combines semantic and statistical features significantly enhances the accuracy of readability classification compared to using each method in isolation. The current study emphasizes the importance of considering both statistical and semantic aspects for a more accurate assessment of text difficulty in Vietnamese. This contribution to the field provides insights into the adaptability of advanced language models in the context of Vietnamese text readability. It lays the groundwork for future research in this area.
Authors: Yuxin Zuo, Wenxuan Jiang, Wenxuan Liu, Zixuan Li, Long Bai, Hanbin Wang, Yutao Zeng, Xiaolong Jin, Jiafeng Guo, Xueqi Cheng
Abstract: Empirical evidence suggests that LLMs exhibit spontaneous cross-lingual alignment. Our findings suggest that although LLMs also demonstrate promising cross-lingual alignment in Information Extraction, there remains significant imbalance across languages, revealing an underlying deficiency in the IE alignment. To address this issue, we propose AlignXIE, a powerful code-based LLM that significantly enhances cross-lingual IE alignment through two strategies. Firstly, AlignXIE formulates IE across different languages, especially non-English ones, as code generation tasks, standardizing the representation of various schemas using Python classes to ensure consistency of the same ontology in different languages and align the schema. Secondly, it incorporates an IE cross-lingual alignment phase through a translated instance prediction task proposed in this paper to align the extraction process, utilizing ParallelNER, an IE bilingual parallel dataset with 257,190 samples, generated by our proposed LLM-based automatic pipeline for IE parallel data construction, with manual annotation to ensure quality. Ultimately, we obtain AlignXIE through multilingual IE instruction tuning. Although without training in 9 unseen languages, AlignXIE surpasses ChatGPT by $30.17\%$ and SoTA by $20.03\%$, thereby demonstrating superior cross-lingual IE capabilities. Comprehensive evaluations on 63 IE benchmarks in Chinese and English under various settings, demonstrate that AlignXIE significantly enhances cross-lingual and multilingual IE through boosting the IE alignment.
Authors: Xingyu Lu, Yuhang Hu, Changyi Liu, Tianke Zhang, Zhenyu Yang, Zhixiang Ding, Shengsheng Qian, Meng Du, Ruiwen Kang, Kaiyu Tang, Fan Yang, Tingting Gao, Di Zhang, Hai-Tao Zheng, Bin Wen
Abstract: Mathematical reasoning presents a significant challenge to the cognitive capabilities of LLMs. Various methods have been proposed to enhance the mathematical ability of LLMs. However, few recognize the value of state transition for LLM reasoning. In this work, we define mathematical problem-solving as a process of transiting from an initial unsolved state to the final resolved state, and propose Kwai-STaR framework, which transforms LLMs into State-Transition Reasoners to improve their intuitive reasoning capabilities. Our approach comprises three main steps: (1) Define the state space tailored to the mathematical reasoning. (2) Generate state-transition data based on the state space. (3) Convert original LLMs into State-Transition Reasoners via a curricular training strategy. Our experiments validate the effectiveness of Kwai-STaR in enhancing mathematical reasoning: After training on the small-scale Kwai-STaR dataset, general LLMs, including Mistral-7B and LLaMA-3, achieve considerable performance gain on the GSM8K and GSM-Hard dataset. Additionally, the state transition-based design endows Kwai-STaR with remarkable training and inference efficiency. Further experiments are underway to establish the generality of Kwai-STaR.
Authors: Alistair Plum, Caroline D\"ohmer, Emilia Milano, Anne-Marie Lutgen, Christoph Purschke
Abstract: The Universal Dependencies (UD) project has significantly expanded linguistic coverage across 161 languages, yet Luxembourgish, a West Germanic language spoken by approximately 400,000 people, has remained absent until now. In this paper, we introduce LuxBank, the first UD Treebank for Luxembourgish, addressing the gap in syntactic annotation and analysis for this `low-research' language. We establish formal guidelines for Luxembourgish language annotation, providing the foundation for the first large-scale quantitative analysis of its syntax. LuxBank serves not only as a resource for linguists and language learners but also as a tool for developing spell checkers and grammar checkers, organising existing text archives and even training large language models. By incorporating Luxembourgish into the UD framework, we aim to enhance the understanding of syntactic variation within West Germanic languages and offer a model for documenting smaller, semi-standardised languages. This work positions Luxembourgish as a valuable resource in the broader linguistic and NLP communities, contributing to the study of languages with limited research and resources.
Authors: Seyoung Song, Haneul Yoo, Jiho Jin, Kyunghyun Cho, Alice Oh
Abstract: Historical and linguistic connections within the Sinosphere have led researchers to use Classical Chinese resources for cross-lingual transfer when processing historical documents from Korea and Japan. In this paper, we question the assumption of cross-lingual transferability from Classical Chinese to Hanja and Kanbun, the ancient written languages of Korea and Japan, respectively. Our experiments across machine translation, named entity recognition, and punctuation restoration tasks show minimal impact of Classical Chinese datasets on language model performance for ancient Korean documents written in Hanja, with performance differences within $\pm{}0.0068$ F1-score for sequence labeling tasks and up to $+0.84$ BLEU score for translation. These limitations persist consistently across various model sizes, architectures, and domain-specific datasets. Our analysis reveals that the benefits of Classical Chinese resources diminish rapidly as local language data increases for Hanja, while showing substantial improvements only in extremely low-resource scenarios for both Korean and Japanese historical documents. These mixed results emphasize the need for careful empirical validation rather than assuming benefits from indiscriminate cross-lingual transfer.
Authors: Ming Cheng, Jiaying Gong, Chenhan Yuan, William A. Ingram, Edward Fox, Hoda Eldardiry
Abstract: Existing text simplification or paraphrase datasets mainly focus on sentence-level text generation in a general domain. These datasets are typically developed without using domain knowledge. In this paper, we release a novel dataset, VTechAGP, which is the first academic-to-general-audience text paraphrase dataset consisting of 4,938 document-level these and dissertation academic and general-audience abstract pairs from 8 colleges authored over 25 years. We also propose a novel dynamic soft prompt generative language model, DSPT5. For training, we leverage a contrastive-generative loss function to learn the keyword vectors in the dynamic prompt. For inference, we adopt a crowd-sampling decoding strategy at both semantic and structural levels to further select the best output candidate. We evaluate DSPT5 and various state-of-the-art large language models (LLMs) from multiple perspectives. Results demonstrate that the SOTA LLMs does not provide satisfactory outcomes, while the lightweight DSPT5 can achieve competitive results. To the best of our knowledge, we are the first to build a benchmark dataset and solutions for academic-to-general-audience text paraphrase dataset.
Authors: Fujie Zhang, Peiqi Yu, Biao Yi, Baolei Zhang, Tong Li, Zheli Liu
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities across a variety of tasks in different domains. However, they sometimes generate responses that are logically coherent but factually incorrect or misleading, which is known as LLM hallucinations. Data-driven supervised methods train hallucination detectors by leveraging the internal states of LLMs, but detectors trained on specific domains often struggle to generalize well to other domains. In this paper, we aim to enhance the cross-domain performance of supervised detectors with only in-domain data. We propose a novel framework, prompt-guided internal states for hallucination detection of LLMs, namely PRISM. By utilizing appropriate prompts to guide changes in the structure related to text truthfulness within the LLM's internal states, we make this structure more salient and consistent across texts from different domains. We integrated our framework with existing hallucination detection methods and conducted experiments on datasets from different domains. The experimental results indicate that our framework significantly enhances the cross-domain generalization of existing hallucination detection methods.
Authors: Chuqiao Song, Shunzhang Chen, Xinyi Cai, Hao Chen
Abstract: Title: Sentiment Analysis of Spanish Political Party Communications on Twitter Using Pre-trained Language Models Authors: Chuqiao Song, Shunzhang Chen, Xinyi Cai, Hao Chen Comments: 21 pages, 6 figures Abstract: This study investigates sentiment patterns within Spanish political party communications on Twitter by leveraging BETO and RoBERTuito, two pre-trained language models optimized for Spanish text. Using a dataset of tweets from major Spanish political parties: PSOE, PP, Vox, Podemos, and Ciudadanos, spanning 2019 to 2024, this research analyzes sentiment distributions and explores the relationship between sentiment expression and party ideology. The findings indicate that both models consistently identify a predominant Neutral sentiment across all parties, with significant variations in Negative and Positive sentiments that align with ideological distinctions. Specifically, Vox exhibits higher levels of Negative sentiment, while PSOE demonstrates relatively high Positive sentiment, supporting the hypothesis that emotional appeals in political messaging reflect ideological stances. This study underscores the potential of pre-trained language models for non-English sentiment analysis on social media, providing insights into sentiment dynamics that shape public discourse within Spain's multi-party political system. Keywords: Spanish politics, sentiment analysis, pre-trained language models, Twitter, BETO, RoBERTuito, political ideology, multi-party system
Authors: Siming Huang, Tianhao Cheng, Jason Klein Liu, Jiaran Hao, Liuyihan Song, Yang Xu, J. Yang, J. H. Liu, Chenchen Zhang, Linzheng Chai, Ruifeng Yuan, Zhaoxiang Zhang, Jie Fu, Qian Liu, Ge Zhang, Zili Wang, Yuan Qi, Yinghui Xu, Wei Chu
Abstract: Large language models (LLMs) for code have become indispensable in various domains, including code generation, reasoning tasks and agent systems.While open-access code LLMs are increasingly approaching the performance levels of proprietary models, high-quality code LLMs suitable for rigorous scientific investigation, particularly those with reproducible data processing pipelines and transparent training protocols, remain limited. The scarcity is due to various challenges, including resource constraints, ethical considerations, and the competitive advantages of keeping models advanced. To address the gap, we introduce OpenCoder, a top-tier code LLM that not only achieves performance comparable to leading models but also serves as an ``open cookbook'' for the research community. Unlike most prior efforts, we release not only model weights and inference code, but also the reproducible training data, complete data processing pipeline, rigorous experimental ablation results, and detailed training protocols for open scientific research. Through this comprehensive release, we identify the key ingredients for building a top-tier code LLM: (1) code optimized heuristic rules for data cleaning and methods for data deduplication, (2) recall of text corpus related to code and (3) high-quality synthetic data in both annealing and supervised fine-tuning stages. By offering this level of openness, we aim to broaden access to all aspects of a top-tier code LLM, with OpenCoder serving as both a powerful model and an open foundation to accelerate research, and enable reproducible advancements in code AI.
Authors: Manuel Frank, Haithem Afli
Abstract: We propose an approach to enhance sentence embeddings by applying generative text models for data augmentation at inference time. Unlike conventional data augmentation that utilises synthetic training data, our approach does not require access to model parameters or the computational resources typically required for fine-tuning state-of-the-art models. Generatively Augmented Sentence Encoding uses diverse linguistic synthetic variants of input texts generated by paraphrasing, summarising, or extracting keywords, followed by pooling the original and synthetic embeddings. Experimental results on the Massive Text Embedding Benchmark for Semantic Textual Similarity (STS) demonstrate performance improvements across a range of embedding models using different generative models for augmentation. We find that generative augmentation leads to larger performance improvements for embedding models with lower baseline performance. These findings suggest that integrating generative augmentation at inference time adds semantic diversity and can enhance the robustness and generalizability of sentence embeddings for embedding models. Our results show that the degree to which generative augmentation can improve STS performance depends not only on the embedding model but also on the dataset. From a broader perspective, the approach allows trading training for inference compute.
Authors: Yujia Hu, Shrestha Ghosh, Tuan-Phong Nugyen, Simon Razniewski
Abstract: General-domain knowledge bases (KB), in particular the "big three" -- Wikidata, Yago and DBpedia -- are the backbone of many intelligent applications. While these three have seen steady development, comprehensive KB construction at large has seen few fresh attempts. In this work, we propose to build a large general-domain KB entirely from a large language model (LLM). We demonstrate the feasibility of large-scale KB construction from LLMs, while highlighting specific challenges arising around entity recognition, entity and property canonicalization, and taxonomy construction. As a prototype, we use GPT-4o-mini to construct GPTKB, which contains 105 million triples for more than 2.9 million entities, at a cost 100x less than previous KBC projects. Our work is a landmark for two fields: For NLP, for the first time, it provides \textit{constructive} insights into the knowledge (or beliefs) of LLMs. For the Semantic Web, it shows novel ways forward for the long-standing challenge of general-domain KB construction. GPTKB is accessible at https://gptkb.org.
URLs: https://gptkb.org.
Authors: Gideon Yoffe, Nachum Dershowitz, Ariel Vishne, Barak Sober
Abstract: Stylometry aims to distinguish authors by analyzing literary traits assumed to reflect semi-conscious choices distinct from elements like genre or theme. However, these components often overlap, complicating text classification based solely on feature distributions. While some literary properties, such as thematic content, are likely to manifest as correlations between adjacent text units, others, like authorial style, may be independent thereof. We introduce a hypothesis-testing approach to evaluate the influence of sequentially correlated literary properties on text classification, aiming to determine when these correlations drive classification. Using a multivariate binary distribution, our method models sequential correlations between text units as a stochastic process, assessing the likelihood of clustering across varying adjacency scales. This enables us to examine whether classification is dominated by sequentially correlated properties or remains independent. In experiments on a diverse English prose corpus, our analysis integrates traditional and neural embeddings within supervised and unsupervised frameworks. Results demonstrate that our approach effectively identifies when textual classification is not primarily influenced by sequentially correlated literary properties, particularly in cases where texts differ in authorial style or genre rather than by a single author within a similar genre.
Authors: Hongyu Wang, Shuming Ma, Furu Wei
Abstract: Recent research on the 1-bit Large Language Models (LLMs), such as BitNet b1.58, presents a promising direction for reducing the inference cost of LLMs while maintaining their performance. In this work, we introduce BitNet a4.8, enabling 4-bit activations for 1-bit LLMs. BitNet a4.8 employs a hybrid quantization and sparsification strategy to mitigate the quantization errors introduced by the outlier channels. Specifically, we utilize 4-bit activations for inputs to the attention and feed-forward network layers, while sparsifying intermediate states followed with 8-bit quantization. Extensive experiments demonstrate that BitNet a4.8 achieves performance comparable to BitNet b1.58 with equivalent training costs, while being faster in inference with enabling 4-bit (INT4/FP4) kernels. Additionally, BitNet a4.8 activates only 55% of parameters and supports 3-bit KV cache, further enhancing the efficiency of large-scale LLM deployment and inference.
Authors: Gabriele Oliaro, Zhihao Jia, Daniel Campos, Aurick Qiao
Abstract: We present SuffixDecoding, a novel model-free approach to accelerating large language model (LLM) inference through speculative decoding. Unlike existing methods that rely on draft models or specialized decoding heads, SuffixDecoding leverages suffix trees built from previously generated outputs to efficiently predict candidate token sequences. Our approach enables flexible tree-structured speculation without the overhead of maintaining and orchestrating additional models. SuffixDecoding builds and dynamically updates suffix trees to capture patterns in the generated text, using them to construct speculation trees through a principled scoring mechanism based on empirical token frequencies. SuffixDecoding requires only CPU memory which is plentiful and underutilized on typical LLM serving nodes. We demonstrate that SuffixDecoding achieves competitive speedups compared to model-based approaches across diverse workloads including open-domain chat, code generation, and text-to-SQL tasks. For open-ended chat and code generation tasks, SuffixDecoding achieves up to $1.4\times$ higher output throughput than SpecInfer and up to $1.1\times$ lower time-per-token (TPOT) latency. For a proprietary multi-LLM text-to-SQL application, SuffixDecoding achieves up to $2.9\times$ higher output throughput and $3\times$ lower latency than speculative decoding. Our evaluation shows that SuffixDecoding maintains high acceptance rates even with small reference corpora of 256 examples, while continuing to improve performance as more historical outputs are incorporated.
Authors: Zhaofeng Wu, Xinyan Velocity Yu, Dani Yogatama, Jiasen Lu, Yoon Kim
Abstract: Modern language models can process inputs across diverse languages and modalities. We hypothesize that models acquire this capability through learning a shared representation space across heterogeneous data types (e.g., different languages and modalities), which places semantically similar inputs near one another, even if they are from different modalities/languages. We term this the semantic hub hypothesis, following the hub-and-spoke model from neuroscience (Patterson et al., 2007) which posits that semantic knowledge in the human brain is organized through a transmodal semantic "hub" which integrates information from various modality-specific "spokes" regions. We first show that model representations for semantically equivalent inputs in different languages are similar in the intermediate layers, and that this space can be interpreted using the model's dominant pretraining language via the logit lens. This tendency extends to other data types, including arithmetic expressions, code, and visual/audio inputs. Interventions in the shared representation space in one data type also predictably affect model outputs in other data types, suggesting that this shared representations space is not simply a vestigial byproduct of large-scale training on broad data, but something that is actively utilized by the model during input processing.
Authors: Weixin Liang, Lili Yu, Liang Luo, Srinivasan Iyer, Ning Dong, Chunting Zhou, Gargi Ghosh, Mike Lewis, Wen-tau Yih, Luke Zettlemoyer, Xi Victoria Lin
Abstract: The development of large language models (LLMs) has expanded to multi-modal systems capable of processing text, images, and speech within a unified framework. Training these models demands significantly larger datasets and computational resources compared to text-only LLMs. To address the scaling challenges, we introduce Mixture-of-Transformers (MoT), a sparse multi-modal transformer architecture that significantly reduces pretraining computational costs. MoT decouples non-embedding parameters of the model by modality -- including feed-forward networks, attention matrices, and layer normalization -- enabling modality-specific processing with global self-attention over the full input sequence. We evaluate MoT across multiple settings and model scales. In the Chameleon 7B setting (autoregressive text-and-image generation), MoT matches the dense baseline's performance using only 55.8\% of the FLOPs. When extended to include speech, MoT reaches speech performance comparable to the dense baseline with only 37.2\% of the FLOPs. In the Transfusion setting, where text and image are trained with different objectives, a 7B MoT model matches the image modality performance of the dense baseline with one third of the FLOPs, and a 760M MoT model outperforms a 1.4B dense baseline across key image generation metrics. System profiling further highlights MoT's practical benefits, achieving dense baseline image quality in 47.2\% of the wall-clock time and text quality in 75.6\% of the wall-clock time (measured on AWS p4de.24xlarge instances with NVIDIA A100 GPUs).
Authors: Jonathan Roberts, Kai Han, Samuel Albanie
Abstract: As the context limits of Large Language Models (LLMs) increase, the range of possible applications and downstream functions broadens. In many real-world tasks, decisions depend on details scattered across collections of often disparate documents containing mostly irrelevant information. Long-context LLMs appear well-suited to this form of complex information retrieval and reasoning, which has traditionally proven costly and time-consuming. However, although the development of longer context models has seen rapid gains in recent years, our understanding of how effectively LLMs use their context has not kept pace. To address this, we conduct a set of retrieval experiments designed to evaluate the capabilities of 17 leading LLMs, such as their ability to follow threads of information through the context window. Strikingly, we find that many models are remarkably threadsafe: capable of simultaneously following multiple threads without significant loss in performance. Still, for many models, we find the effective context limit is significantly shorter than the supported context length, with accuracy decreasing as the context window grows. Our study also highlights the important point that token counts from different tokenizers should not be directly compared -- they often correspond to substantially different numbers of written characters. We release our code and long-context experimental data.
Authors: Dingjie Song, Sicheng Lai, Shunian Chen, Lichao Sun, Benyou Wang
Abstract: The rapid progression of multimodal large language models (MLLMs) has demonstrated superior performance on various multimodal benchmarks. However, the issue of data contamination during training creates challenges in performance evaluation and comparison. While numerous methods exist for detecting dataset contamination in large language models (LLMs), they are less effective for MLLMs due to their various modalities and multiple training phases. In this study, we introduce a multimodal data contamination detection framework, MM-Detect, designed for MLLMs. Our experimental results indicate that MM-Detect is sensitive to varying degrees of contamination and can highlight significant performance improvements due to leakage of the training set of multimodal benchmarks. Furthermore, We also explore the possibility of contamination originating from the pre-training phase of LLMs used by MLLMs and the fine-tuning phase of MLLMs, offering new insights into the stages at which contamination may be introduced.
Authors: Jun Cui
Abstract: This paper compares the impact of Test-Driven Development (TDD) and Behavior-Driven Development (BDD) on software delivery effectiveness within enterprise environments. Using a qualitative research design, data were collected through in-depth interviews with developers and project managers from enterprises adopting TDD or BDD. Moreover, the findings reveal distinct effects of each model on delivery speed, software quality, and team collaboration. Specifically, TDD emphasizes early testing and iterative development, leading to enhanced code quality and fewer defects, while BDD improves cross-functional communication by focusing on behavior specifications that involve stakeholders directly. However, TDD may create a higher initial time investment, and BDD might encounter challenges in requirement clarity. These differences highlight gaps in understanding how each model aligns with varying project types and stakeholder needs, which can guide enterprises in selecting the most suitable model for their unique requirements. The study contributes to the literature by providing insights into the practical application and challenges of TDD and BDD, suggesting future research on their long-term impacts in diverse settings.
Authors: Fei Yang, Xuenan Xu, Mengyue Wu, Kai Yu
Abstract: Pathological speech analysis has been of interest in the detection of certain diseases like depression and Alzheimer's disease and attracts much interest from researchers. However, previous pathological speech analysis models are commonly designed for a specific disease while overlooking the connection between diseases, which may constrain performance and lower training efficiency. Instead of fine-tuning deep models for different tasks, prompt tuning is a much more efficient training paradigm. We thus propose a unified pathological speech analysis system for as many as three diseases with the prompt tuning technique. This system uses prompt tuning to adjust only a small part of the parameters to detect different diseases from speeches of possible patients. Our system leverages a pre-trained spoken language model and demonstrates strong performance across multiple disorders while only fine-tuning a fraction of the parameters. This efficient training approach leads to faster convergence and improved F1 scores by allowing knowledge to be shared across tasks. Our experiments on Alzheimer's disease, Depression, and Parkinson's disease show competitive results, highlighting the effectiveness of our method in pathological speech analysis.
Authors: Jun Cui
Abstract: This study investigates the impact of software design model capabilities and data structure algorithm abilities on microservices architecture design within enterprises. Utilizing a qualitative methodology, the research involved in-depth interviews with software architects and developers who possess extensive experience in microservices implementation. The findings reveal that organizations emphasizing robust design models and efficient algorithms achieve superior scalability, performance, and flexibility in their microservices architecture. Notably, participants highlighted that a strong foundation in these areas facilitates better service decomposition, optimizes data processing, and enhances system responsiveness. Despite these insights, gaps remain regarding the integration of emerging technologies and the evolving nature of software design practices. This paper contributes to the existing literature by underscoring the critical role of these competencies in fostering effective microservices architectures and suggests avenues for future research to address identified gaps
Authors: Tianhua Tao, Junbo Li, Bowen Tan, Hongyi Wang, William Marshall, Bhargav M Kanakiya, Joel Hestness, Natalia Vassilieva, Zhiqiang Shen, Eric P. Xing, Zhengzhong Liu
Abstract: Large Language Models (LLMs) specializing in code generation (which are also often referred to as code LLMs), e.g., StarCoder and Code Llama, play increasingly critical roles in various software development scenarios. It is also crucial for code LLMs to possess both code generation and natural language abilities for many specific applications, such as code snippet retrieval using natural language or code explanations. The intricate interaction between acquiring language and coding skills complicates the development of strong code LLMs. Furthermore, there is a lack of thorough prior studies on the LLM pretraining strategy that mixes code and natural language. In this work, we propose a pretraining strategy to enhance the integration of natural language and coding capabilities within a single LLM. Specifically, it includes two phases of training with appropriately adjusted code/language ratios. The resulting model, Crystal, demonstrates remarkable capabilities in both domains. Specifically, it has natural language and coding performance comparable to that of Llama 2 and Code Llama, respectively. Crystal exhibits better data efficiency, using 1.4 trillion tokens compared to the more than 2 trillion tokens used by Llama 2 and Code Llama. We verify our pretraining strategy by analyzing the training process and observe consistent improvements in most benchmarks. We also adopted a typical application adaptation phase with a code-centric data mixture, only to find that it did not lead to enhanced performance or training efficiency, underlining the importance of a carefully designed data recipe. To foster research within the community, we commit to open-sourcing every detail of the pretraining, including our training datasets, code, loggings and 136 checkpoints throughout the training.
Authors: Nana Lin, Youxiang Zhu, Xiaohui Liang, John A. Batsis, Caroline Summerour
Abstract: Mild cognitive impairment (MCI) is a major public health concern due to its high risk of progressing to dementia. This study investigates the potential of detecting MCI with spontaneous voice assistant (VA) commands from 35 older adults in a controlled setting. Specifically, a command-generation task is designed with pre-defined intents for participants to freely generate commands that are more associated with cognitive ability than read commands. We develop MCI classification and regression models with audio, textual, intent, and multimodal fusion features. We find the command-generation task outperforms the command-reading task with an average classification accuracy of 82%, achieved by leveraging multimodal fusion features. In addition, generated commands correlate more strongly with memory and attention subdomains than read commands. Our results confirm the effectiveness of the command-generation task and imply the promise of using longitudinal in-home commands for MCI detection.
Authors: Haolin Chen, Yihao Feng, Zuxin Liu, Weiran Yao, Akshara Prabhakar, Shelby Heinecke, Ricky Ho, Phil Mui, Silvio Savarese, Caiming Xiong, Huan Wang
Abstract: Large language models (LLMs) have shown impressive capabilities, but still struggle with complex reasoning tasks requiring multiple steps. While prompt-based methods like Chain-of-Thought (CoT) can improve LLM reasoning at inference time, optimizing reasoning capabilities during training remains challenging. We introduce LaTent Reasoning Optimization (LaTRO), a principled framework that formulates reasoning as sampling from a latent distribution and optimizes it via variational approaches. LaTRO enables LLMs to concurrently improve both their reasoning process and ability to evaluate reasoning quality, without requiring external feedback or reward models. We validate LaTRO through experiments on GSM8K and ARC-Challenge datasets using multiple model architectures. On GSM8K, LaTRO improves zero-shot accuracy by an average of 12.5% over base models and 9.6% over supervised fine-tuning across Phi-3.5-mini, Mistral-7B, and Llama-3.1-8B. Our findings suggest that pre-trained LLMs possess latent reasoning capabilities that can be unlocked and enhanced through our proposed optimization approach in a self-improvement manner. The code of LaTRO is available at \url{https://github.com/SalesforceAIResearch/LaTRO}.
Authors: Tanishq Kumar, Zachary Ankner, Benjamin F. Spector, Blake Bordelon, Niklas Muennighoff, Mansheej Paul, Cengiz Pehlevan, Christopher R\'e, Aditi Raghunathan
Abstract: Low precision training and inference affect both the quality and cost of language models, but current scaling laws do not account for this. In this work, we devise "precision-aware" scaling laws for both training and inference. We propose that training in lower precision reduces the model's "effective parameter count," allowing us to predict the additional loss incurred from training in low precision and post-train quantization. For inference, we find that the degradation introduced by post-training quantization increases as models are trained on more data, eventually making additional pretraining data actively harmful. For training, our scaling laws allow us to predict the loss of a model with different parts in different precisions, and suggest that training larger models in lower precision may be compute optimal. We unify the scaling laws for post and pretraining quantization to arrive at a single functional form that predicts degradation from training and inference in varied precisions. We fit on over 465 pretraining runs and validate our predictions on model sizes up to 1.7B parameters trained on up to 26B tokens.
Authors: Vaibhav Seth, Arinjay Pathak, Ayan Sengupta, Natraj Raman, Sriram Gopalakrishnan, Tanmoy Chakraborty
Abstract: Large Language Models (LLMs) are highly resource-intensive to fine-tune due to their enormous size. While low-rank adaptation is a prominent parameter-efficient fine-tuning approach, it suffers from sensitivity to hyperparameter choices, leading to instability in model performance on fine-tuning downstream tasks. This paper highlights the importance of effective parameterization in low-rank fine-tuning to reduce estimator variance and enhance the stability of final model outputs. We propose MonteCLoRA, an efficient fine-tuning technique, employing Monte Carlo estimation to learn an unbiased posterior estimation of low-rank parameters with low expected variance, which stabilizes fine-tuned LLMs with only O(1) additional parameters. MonteCLoRA shows significant improvements in accuracy and robustness, achieving up to 3.8% higher accuracy and 8.6% greater robustness than existing efficient fine-tuning methods on natural language understanding tasks with pre-trained RoBERTa-base. Furthermore, in generative tasks with pre-trained LLaMA-1-7B, MonteCLoRA demonstrates robust zero-shot performance with 50% lower variance than the contemporary efficient fine-tuning methods. The theoretical and empirical results presented in the paper underscore how parameterization and hyperpriors balance exploration-exploitation in the low-rank parametric space, therefore leading to more optimal and robust parameter estimation during efficient fine-tuning.
Authors: Bai Cong, Nico Daheim, Yuesong Shen, Daniel Cremers, Rio Yokota, Mohammad Emtiyaz Khan, Thomas M\"ollenhoff
Abstract: We show that variational learning can significantly improve the accuracy and calibration of Low-Rank Adaptation (LoRA) without a substantial increase in the cost. We replace AdamW by the Improved Variational Online Newton (IVON) algorithm to finetune large language models. For Llama-2 with 7 billion parameters, IVON improves the accuracy over AdamW by 2.8% and expected calibration error by 4.6%. The accuracy is also better than the other Bayesian alternatives, yet the cost is lower and the implementation is easier. Our work provides additional evidence for the effectiveness of IVON for large language models. The code is available at https://github.com/team-approx-bayes/ivon-lora.
Authors: Dezhi Ye, Junwei Hu, Jiabin Fan, Bowen Tian, Jie Liu, Haijin Liang, Jin Ma
Abstract: Recent studies have highlighted the significant potential of Large Language Models (LLMs) as zero-shot relevance rankers. These methods predominantly utilize prompt learning to assess the relevance between queries and documents by generating a ranked list of potential documents. Despite their promise, the substantial costs associated with LLMs pose a significant challenge for their direct implementation in commercial search systems. To overcome this barrier and fully exploit the capabilities of LLMs for text ranking, we explore techniques to transfer the ranking expertise of LLMs to a more compact model similar to BERT, using a ranking loss to enable the deployment of less resource-intensive models. Specifically, we enhance the training of LLMs through Continued Pre-Training, taking the query as input and the clicked title and summary as output. We then proceed with supervised fine-tuning of the LLM using a rank loss, assigning the final token as a representative of the entire sentence. Given the inherent characteristics of autoregressive language models, only the final token can encapsulate all preceding tokens. Additionally, we introduce a hybrid point-wise and margin MSE loss to transfer the ranking knowledge from LLMs to smaller models like BERT. This method creates a viable solution for environments with strict resource constraints. Both offline and online evaluations have confirmed the efficacy of our approach, and our model has been successfully integrated into a commercial web search engine as of February 2024.
Authors: Ruiyang Ren, Yuhao Wang, Kun Zhou, Wayne Xin Zhao, Wenjie Wang, Jing Liu, Ji-Rong Wen, Tat-Seng Chua
Abstract: Large language models (LLMs), with advanced linguistic capabilities, have been employed in reranking tasks through a sequence-to-sequence approach. In this paradigm, multiple passages are reranked in a listwise manner and a textual reranked permutation is generated. However, due to the limited context window of LLMs, this reranking paradigm requires a sliding window strategy to iteratively handle larger candidate sets. This not only increases computational costs but also restricts the LLM from fully capturing all the comparison information for all candidates. To address these challenges, we propose a novel self-calibrated listwise reranking method, which aims to leverage LLMs to produce global relevance scores for ranking. To achieve it, we first propose the relevance-aware listwise reranking framework, which incorporates explicit list-view relevance scores to improve reranking efficiency and enable global comparison across the entire candidate set. Second, to ensure the comparability of the computed scores, we propose self-calibrated training that uses point-view relevance assessments generated internally by the LLM itself to calibrate the list-view relevance assessments. Extensive experiments and comprehensive analysis on the BEIR benchmark and TREC Deep Learning Tracks demonstrate the effectiveness and efficiency of our proposed method.
Authors: Zijian Zhang, Vinay Setty, Yumeng Wang, Avishek Anand
Abstract: With the rapid advancement of neural language models, the deployment of over-parameterized models has surged, increasing the need for interpretable explanations comprehensible to human inspectors. Existing post-hoc interpretability methods, which often focus on unigram features of single input textual instances, fail to capture the models' decision-making process fully. Additionally, many methods do not differentiate between decisions based on spurious correlations and those based on a holistic understanding of the input. Our paper introduces DISCO, a novel method for discovering global, rule-based explanations by identifying causal n-gram associations with model predictions. This method employs a scalable sequence mining technique to extract relevant text spans from training data, associate them with model predictions, and conduct causality checks to distill robust rules that elucidate model behavior. These rules expose potential overfitting and provide insights into misleading feature combinations. We validate DISCO through extensive testing, demonstrating its superiority over existing methods in offering comprehensive insights into complex model behaviors. Our approach successfully identifies all shortcuts manually introduced into the training data (100% detection rate on the MultiRC dataset), resulting in an 18.8% regression in model performance -- a capability unmatched by any other method. Furthermore, DISCO supports interactive explanations, enabling human inspectors to distinguish spurious causes in the rule-based output. This alleviates the burden of abundant instance-wise explanations and helps assess the model's risk when encountering out-of-distribution (OOD) data.
Authors: Xuewen Han, Neng Wang, Shangkun Che, Hongyang Yang, Kunpeng Zhang, Sean Xin Xu
Abstract: In recent years, the application of generative artificial intelligence (GenAI) in financial analysis and investment decision-making has gained significant attention. However, most existing approaches rely on single-agent systems, which fail to fully utilize the collaborative potential of multiple AI agents. In this paper, we propose a novel multi-agent collaboration system designed to enhance decision-making in financial investment research. The system incorporates agent groups with both configurable group sizes and collaboration structures to leverage the strengths of each agent group type. By utilizing a sub-optimal combination strategy, the system dynamically adapts to varying market conditions and investment scenarios, optimizing performance across different tasks. We focus on three sub-tasks: fundamentals, market sentiment, and risk analysis, by analyzing the 2023 SEC 10-K forms of 30 companies listed on the Dow Jones Index. Our findings reveal significant performance variations based on the configurations of AI agents for different tasks. The results demonstrate that our multi-agent collaboration system outperforms traditional single-agent models, offering improved accuracy, efficiency, and adaptability in complex financial environments. This study highlights the potential of multi-agent systems in transforming financial analysis and investment decision-making by integrating diverse analytical perspectives.
Authors: Jaemin Cho, Debanjan Mahata, Ozan Irsoy, Yujie He, Mohit Bansal
Abstract: Document visual question answering (DocVQA) pipelines that answer questions from documents have broad applications. Existing methods focus on handling single-page documents with multi-modal language models (MLMs), or rely on text-based retrieval-augmented generation (RAG) that uses text extraction tools such as optical character recognition (OCR). However, there are difficulties in applying these methods in real-world scenarios: (a) questions often require information across different pages or documents, where MLMs cannot handle many long documents; (b) documents often have important information in visual elements such as figures, but text extraction tools ignore them. We introduce M3DocRAG, a novel multi-modal RAG framework that flexibly accommodates various document contexts (closed-domain and open-domain), question hops (single-hop and multi-hop), and evidence modalities (text, chart, figure, etc.). M3DocRAG finds relevant documents and answers questions using a multi-modal retriever and an MLM, so that it can efficiently handle single or many documents while preserving visual information. Since previous DocVQA datasets ask questions in the context of a specific document, we also present M3DocVQA, a new benchmark for evaluating open-domain DocVQA over 3,000+ PDF documents with 40,000+ pages. In three benchmarks (M3DocVQA/MMLongBench-Doc/MP-DocVQA), empirical results show that M3DocRAG with ColPali and Qwen2-VL 7B achieves superior performance than many strong baselines, including state-of-the-art performance in MP-DocVQA. We provide comprehensive analyses of different indexing, MLMs, and retrieval models. Lastly, we qualitatively show that M3DocRAG can successfully handle various scenarios, such as when relevant information exists across multiple pages and when answer evidence only exists in images.
Authors: Yanjun Gao, Skatje Myers, Shan Chen, Dmitriy Dligach, Timothy A Miller, Danielle Bitterman, Guanhua Chen, Anoop Mayampurath, Matthew Churpek, Majid Afshar
Abstract: Large language models (LLMs) are being explored for diagnostic decision support, yet their ability to estimate pre-test probabilities, vital for clinical decision-making, remains limited. This study evaluates two LLMs, Mistral-7B and Llama3-70B, using structured electronic health record data on three diagnosis tasks. We examined three current methods of extracting LLM probability estimations and revealed their limitations. We aim to highlight the need for improved techniques in LLM confidence estimation.
Authors: Weiquan Huang, Aoqi Wu, Yifan Yang, Xufang Luo, Yuqing Yang, Liang Hu, Qi Dai, Xiyang Dai, Dongdong Chen, Chong Luo, Lili Qiu
Abstract: CLIP is one of the most important multimodal foundational models today. What powers CLIP's capabilities? The rich supervision signals provided by natural language, the carrier of human knowledge, shape a powerful cross-modal representation space. However, with the rapid advancements in large language models LLMs like GPT-4 and LLaMA, the boundaries of language comprehension and generation are continually being pushed. This raises an intriguing question: can the capabilities of LLMs be harnessed to further improve multimodal representation learning? The potential benefits of incorporating LLMs into CLIP are clear. LLMs' strong textual understanding can fundamentally improve CLIP's ability to handle image captions, drastically enhancing its ability to process long and complex texts, a well-known limitation of vanilla CLIP. Moreover, LLMs are trained on a vast corpus of text, possessing open-world knowledge. This allows them to expand on caption information during training, increasing the efficiency of the learning process. In this paper, we propose LLM2CLIP, a novel approach that embraces the power of LLMs to unlock CLIP's potential. By fine-tuning the LLM in the caption space with contrastive learning, we extract its textual capabilities into the output embeddings, significantly improving the output layer's textual discriminability. We then design an efficient training process where the fine-tuned LLM acts as a powerful teacher for CLIP's visual encoder. Thanks to the LLM's presence, we can now incorporate longer and more complex captions without being restricted by vanilla CLIP's text encoder's context window and ability limitations. Our experiments demonstrate that this approach brings substantial improvements in cross-modal tasks.
Authors: David M. Chan, Rodolfo Corona, Joonyong Park, Cheol Jun Cho, Yutong Bai, Trevor Darrell
Abstract: With the introduction of transformer-based models for vision and language tasks, such as LLaVA and Chameleon, there has been renewed interest in the discrete tokenized representation of images. These models often treat image patches as discrete tokens, analogous to words in natural language, learning joint alignments between visual and human languages. However, little is known about the statistical behavior of these visual languages - whether they follow similar frequency distributions, grammatical structures, or topologies as natural languages. In this paper, we take a natural-language-centric approach to analyzing discrete visual languages and uncover striking similarities and fundamental differences. We demonstrate that, although visual languages adhere to Zipfian distributions, higher token innovation drives greater entropy and lower compression, with tokens predominantly representing object parts, indicating intermediate granularity. We also show that visual languages lack cohesive grammatical structures, leading to higher perplexity and weaker hierarchical organization compared to natural languages. Finally, we demonstrate that, while vision models align more closely with natural languages than other models, this alignment remains significantly weaker than the cohesion found within natural languages. Through these experiments, we demonstrate how understanding the statistical properties of discrete visual languages can inform the design of more effective computer vision models.
Authors: Xiaoyang Wang, Chen Li, Jianqiao Zhao, Dong Yu
Abstract: In this paper, we propose a Chinese multi-turn topic-driven conversation dataset, NaturalConv, which allows the participants to chat anything they want as long as any element from the topic is mentioned and the topic shift is smooth. Our corpus contains 19.9K conversations from six domains, and 400K utterances with an average turn number of 20.1. These conversations contain in-depth discussions on related topics or widely natural transition between multiple topics. We believe either way is normal for human conversation. To facilitate the research on this corpus, we provide results of several benchmark models. Comparative results show that for this dataset, our current models are not able to provide significant improvement by introducing background knowledge/topic. Therefore, the proposed dataset should be a good benchmark for further research to evaluate the validity and naturalness of multi-turn conversation systems. Our dataset is available at https://ailab.tencent.com/ailab/nlp/dialogue/#datasets.
Authors: Maximillian Chen, Zhou Yu
Abstract: Speech models have long been known to overfit individual speakers for many classification tasks. This leads to poor generalization in settings where the speakers are out-of-domain or out-of-distribution, as is common in production environments. We view speaker adaptation as a few-shot learning problem and propose investigating transfer learning approaches inspired by recent success with pre-trained models in natural language tasks. We propose pre-finetuning speech models on difficult tasks to distill knowledge into few-shot downstream classification objectives. We pre-finetune Wav2Vec2.0 on every permutation of four multiclass emotional speech recognition corpora and evaluate our pre-finetuned models through 33,600 few-shot fine-tuning trials on the Emotional Speech Dataset.
Authors: Anuj Diwan, Anirudh Srinivasan, David Harwath, Eunsol Choi
Abstract: Existing speech-to-speech translation (S2ST) models fall into two camps: they either leverage text as an intermediate step or require hundreds of hours of parallel speech data. Both approaches are incompatible with textless languages or language pairs with limited parallel data. We present PFB, a framework for training textless S2ST models that require just dozens of hours of parallel speech data. We first pretrain a model on large-scale monolingual speech data, finetune it with a small amount of parallel speech data (20-60 hours), and lastly train with an unsupervised backtranslation objective. We train and evaluate our models for English-to-German, German-to-English and Marathi-to-English translation on three different domains (European Parliament, Common Voice, and All India Radio) with single-speaker synthesized speech. Evaluated using the ASR-BLEU metric, our models achieve reasonable performance on all three domains, with some being within 1-2 points of our higher-resourced topline.
Authors: Liangtai Sun, Yang Han, Zihan Zhao, Da Ma, Zhennan Shen, Baocai Chen, Lu Chen, Kai Yu
Abstract: Recently, there has been growing interest in using Large Language Models (LLMs) for scientific research. Numerous benchmarks have been proposed to evaluate the ability of LLMs for scientific research. However, current benchmarks are mostly based on pre-collected objective questions. This design suffers from data leakage problem and lacks the evaluation of subjective Q/A ability. In this paper, we propose SciEval, a comprehensive and multi-disciplinary evaluation benchmark to address these issues. Based on Bloom's taxonomy, SciEval covers four dimensions to systematically evaluate scientific research ability. In particular, we design a "dynamic" subset based on scientific principles to prevent evaluation from potential data leakage. Both objective and subjective questions are included in SciEval. These characteristics make SciEval a more effective benchmark for scientific research ability evaluation of LLMs. Comprehensive experiments on most advanced LLMs show that, although GPT-4 achieves SOTA performance compared to other LLMs, there is still substantial room for improvement, especially for dynamic questions. The codes and data are publicly available on https://github.com/OpenDFM/SciEval.
Authors: Sayan Mahapatra, Debtanu Datta, Shubham Soni, Adrijit Goswami, Saptarshi Ghosh
Abstract: Most legal text in the Indian judiciary is written in complex English due to historical reasons. However, only a small fraction of the Indian population is comfortable in reading English. Hence legal text needs to be made available in various Indian languages, possibly by translating the available legal text from English. Though there has been a lot of research on translation to and between Indian languages, to our knowledge, there has not been much prior work on such translation in the legal domain. In this work, we construct the first high-quality legal parallel corpus containing aligned text units in English and nine Indian languages, that includes several low-resource languages. We also benchmark the performance of a wide variety of Machine Translation (MT) systems over this corpus, including commercial MT systems, open-source MT systems and Large Language Models. Through a comprehensive survey by Law practitioners, we check how satisfied they are with the translations by some of these MT systems, and how well automatic MT evaluation metrics agree with the opinions of Law practitioners.
Authors: Stanis{\l}aw Wo\'zniak, Bart{\l}omiej Koptyra, Arkadiusz Janz, Przemys{\l}aw Kazienko, Jan Koco\'n
Abstract: Large language models (LLMs) have significantly advanced Natural Language Processing (NLP) tasks in recent years. However, their universal nature poses limitations in scenarios requiring personalized responses, such as recommendation systems and chatbots. This paper investigates methods to personalize LLMs, comparing fine-tuning and zero-shot reasoning approaches on subjective tasks. Results demonstrate that personalized fine-tuning improves model reasoning compared to non-personalized models. Experiments on datasets for emotion recognition and hate speech detection show consistent performance gains with personalized methods across different LLM architectures. These findings underscore the importance of personalization for enhancing LLM capabilities in subjective text perception tasks.
Authors: Jerry Wei, Chengrun Yang, Xinying Song, Yifeng Lu, Nathan Hu, Jie Huang, Dustin Tran, Daiyi Peng, Ruibo Liu, Da Huang, Cosmo Du, Quoc V. Le
Abstract: Large language models (LLMs) often generate content that contains factual errors when responding to fact-seeking prompts on open-ended topics. To benchmark a model's long-form factuality in open domains, we first use GPT-4 to generate LongFact, a prompt set comprising thousands of questions spanning 38 topics. We then propose that LLM agents can be used as automated evaluators for long-form factuality through a method which we call Search-Augmented Factuality Evaluator (SAFE). SAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results. Furthermore, we propose extending F1 score as an aggregated metric for long-form factuality. To do so, we balance the percentage of supported facts in a response (precision) with the percentage of provided facts relative to a hyperparameter representing a user's preferred response length (recall). Empirically, we demonstrate that LLM agents can outperform crowdsourced human annotators - on a set of ~16k individual facts, SAFE agrees with crowdsourced human annotators 72% of the time, and on a random subset of 100 disagreement cases, SAFE wins 76% of the time. At the same time, SAFE is more than 20 times cheaper than human annotators. We also benchmark thirteen language models on LongFact across four model families (Gemini, GPT, Claude, and PaLM-2), finding that larger language models generally achieve better long-form factuality. LongFact, SAFE, and all experimental code are available at https://github.com/google-deepmind/long-form-factuality.
URLs: https://github.com/google-deepmind/long-form-factuality.
Authors: Dawei Zhu, Liang Wang, Nan Yang, Yifan Song, Wenhao Wu, Furu Wei, Sujian Li
Abstract: Embedding models play a pivot role in modern NLP applications such as IR and RAG. While the context limit of LLMs has been pushed beyond 1 million tokens, embedding models are still confined to a narrow context window not exceeding 8k tokens, refrained from application scenarios requiring long inputs such as legal contracts. This paper explores context window extension of existing embedding models, pushing the limit to 32k without requiring additional training. First, we examine the performance of current embedding models for long context retrieval on our newly constructed LongEmbed benchmark. LongEmbed comprises two synthetic tasks and four carefully chosen real-world tasks, featuring documents of varying length and dispersed target information. Benchmarking results underscore huge room for improvement in these models. Based on this, comprehensive experiments show that training-free context window extension strategies like position interpolation can effectively extend the context window of existing embedding models by several folds, regardless of their original context being 512 or beyond 4k. Furthermore, for models employing absolute position encoding (APE), we show the possibility of further fine-tuning to harvest notable performance gains while strictly preserving original behavior for short inputs. For models using rotary position embedding (RoPE), significant enhancements are observed when employing RoPE-specific methods, such as NTK and SelfExtend, indicating RoPE's superiority over APE for context window extension. To facilitate future research, we release E5-Base-4k and E5-RoPE-Base, along with the LongEmbed benchmark.
Authors: Jonathan Mamou, Oren Pereg, Daniel Korat, Moshe Berchansky, Nadav Timor, Moshe Wasserblat, Roy Schwartz
Abstract: Speculative decoding is commonly used for reducing the inference latency of large language models. Its effectiveness depends highly on the speculation lookahead (SL)-the number of tokens generated by the draft model at each iteration. In this work we show that the common practice of using the same SL for all iterations (static SL) is suboptimal. We introduce DISCO (DynamIc SpeCulation lookahead Optimization), a novel method for dynamically selecting the SL. Our experiments with four datasets show that DISCO reaches an average speedup of 10% compared to the best static SL baseline, while generating the exact same text.
Authors: Shuyue Stella Li, Vidhisha Balachandran, Shangbin Feng, Jonathan S. Ilgen, Emma Pierson, Pang Wei Koh, Yulia Tsvetkov
Abstract: Users typically engage with LLMs interactively, yet most existing benchmarks evaluate them in a static, single-turn format, posing reliability concerns in interactive scenarios. We identify a key obstacle towards reliability: LLMs are trained to answer any question, even with incomplete context or insufficient knowledge. In this paper, we propose to change the static paradigm to an interactive one, develop systems that proactively ask questions to gather more information and respond reliably, and introduce an benchmark - MediQ - to evaluate question-asking ability in LLMs. MediQ simulates clinical interactions consisting of a Patient System and an adaptive Expert System; with potentially incomplete initial information, the Expert refrains from making diagnostic decisions when unconfident, and instead elicits missing details via follow-up questions. We provide a pipeline to convert single-turn medical benchmarks into an interactive format. Our results show that directly prompting state-of-the-art LLMs to ask questions degrades performance, indicating that adapting LLMs to proactive information-seeking settings is nontrivial. We experiment with abstention strategies to better estimate model confidence and decide when to ask questions, improving diagnostic accuracy by 22.3%; however, performance still lags compared to an (unrealistic in practice) upper bound with complete information upfront. Further analyses show improved interactive performance with filtering irrelevant contexts and reformatting conversations. Overall, we introduce a novel problem towards LLM reliability, an interactive MediQ benchmark and a novel question-asking system, and highlight directions to extend LLMs' information-seeking abilities in critical domains.
Authors: Priyanka Kargupta, Ishika Agarwal, Dilek Hakkani-Tur, Jiawei Han
Abstract: Socratic questioning is an effective teaching strategy, encouraging critical thinking and problem-solving. The conversational capabilities of large language models (LLMs) show great potential for providing scalable, real-time student guidance. However, current LLMs often give away solutions directly, making them ineffective instructors. We tackle this issue in the code debugging domain with TreeInstruct, an Instructor agent guided by a novel state space-based planning algorithm. TreeInstruct asks probing questions to help students independently identify and resolve errors. It estimates a student's conceptual and syntactical knowledge to dynamically construct a question tree based on their responses and current knowledge state, effectively addressing both independent and dependent mistakes concurrently in a multi-turn interaction setting. In addition to using an existing single-bug debugging benchmark, we construct a more challenging multi-bug dataset of 150 coding problems, incorrect solutions, and bug fixes -- all carefully constructed and annotated by experts. Extensive evaluation shows TreeInstruct's state-of-the-art performance on both datasets, proving it to be a more effective instructor than baselines. Furthermore, a real-world case study with five students of varying skill levels further demonstrates TreeInstruct's ability to guide students to debug their code efficiently with minimal turns and highly Socratic questioning.
Authors: Candida M. Greco, Lucio La Cava, Andrea Tagarelli
Abstract: Verbs form the backbone of language, providing the structure and meaning to sentences. Yet, their intricate semantic nuances pose a longstanding challenge. Understanding verb relations through the concept of lexical entailment is crucial for comprehending sentence meanings and grasping verb dynamics. This work investigates the capabilities of eight Large Language Models in recognizing lexical entailment relations among verbs through differently devised prompting strategies and zero-/few-shot settings over verb pairs from two lexical databases, namely WordNet and HyperLex. Our findings unveil that the models can tackle the lexical entailment recognition task with moderately good performance, although at varying degree of effectiveness and under different conditions. Also, utilizing few-shot prompting can enhance the models' performance. However, perfectly solving the task arises as an unmet challenge for all examined LLMs, which raises an emergence for further research developments on this topic.
Authors: Zachary Horvitz, Ajay Patel, Kanishk Singh, Chris Callison-Burch, Kathleen McKeown, Zhou Yu
Abstract: The goal of text style transfer is to transform the style of texts while preserving their original meaning, often with only a few examples of the target style. Existing style transfer methods generally rely on the few-shot capabilities of large language models or on complex controllable text generation approaches that are inefficient and underperform on fluency metrics. We introduce TinyStyler, a lightweight but effective approach, which leverages a small language model (800M params) and pre-trained authorship embeddings to perform efficient, few-shot text style transfer. We evaluate on the challenging task of authorship style transfer and find TinyStyler outperforms strong approaches such as GPT-4. We also evaluate TinyStyler's ability to perform text attribute style transfer (formal $\leftrightarrow$ informal) with automatic and human evaluations and find that the approach outperforms recent controllable text generation methods. Our model has been made publicly available at https://huggingface.co/tinystyler/tinystyler .
Authors: Xingchen Wan, Ruoxi Sun, Hootan Nakhost, Sercan O. Arik
Abstract: Large language models have demonstrated remarkable capabilities, but their performance is heavily reliant on effective prompt engineering. Automatic prompt optimization (APO) methods are designed to automate this and can be broadly categorized into those targeting instructions (instruction optimization, IO) vs. those targeting exemplars (exemplar optimization, EO). Despite their shared objective, these have evolved rather independently, with IO receiving more research attention recently. This paper seeks to bridge this gap by comprehensively comparing the performance of representative IO and EO techniques both isolation and combination on a diverse set of challenging tasks. Our findings reveal that intelligently reusing model-generated input-output pairs obtained from evaluating prompts on the validation set as exemplars, consistently improves performance on top of IO methods but is currently under-investigated. We also find that despite the recent focus on IO, how we select exemplars can outweigh how we optimize instructions, with EO strategies as simple as random search outperforming state-of-the-art IO methods with seed instructions without any optimization. Moreover, we observe a synergy between EO and IO, with optimal combinations surpassing the individual contributions. We conclude that studying exemplar optimization both as a standalone method and its optimal combination with instruction optimization remain a crucial aspect of APO and deserve greater consideration in future research, even in the era of highly capable instruction-following models.
Authors: Yang Wang, Alberto Garcia Hernandez, Roman Kyslyi, Nicholas Kersting
Abstract: We present a comprehensive study of answer quality evaluation in Retrieval-Augmented Generation (RAG) applications using vRAG-Eval, a novel grading system that is designed to assess correctness, completeness, and honesty. We further map the grading of quality aspects aforementioned into a binary score, indicating an accept or reject decision, mirroring the intuitive "thumbs-up" or "thumbs-down" gesture commonly used in chat applications. This approach suits factual business contexts where a clear decision opinion is essential. Our assessment applies vRAG-Eval to two Large Language Models (LLMs), evaluating the quality of answers generated by a vanilla RAG application. We compare these evaluations with human expert judgments and find a substantial alignment between GPT-4's assessments and those of human experts, reaching 83% agreement on accept or reject decisions. This study highlights the potential of LLMs as reliable evaluators in closed-domain, closed-ended settings, particularly when human evaluations require significant resources.
Authors: Chani Jung, Dongkwan Kim, Jiho Jin, Jiseon Kim, Yeon Seonwoo, Yejin Choi, Alice Oh, Hyunwoo Kim
Abstract: While humans naturally develop theory of mind (ToM), the capability to understand other people's mental states and beliefs, state-of-the-art large language models (LLMs) underperform on simple ToM benchmarks. We posit that we can extend our understanding of LLMs' ToM abilities by evaluating key human ToM precursors$-$perception inference and perception-to-belief inference$-$in LLMs. We introduce two datasets, Percept-ToMi and Percept-FANToM, to evaluate these precursory inferences for ToM in LLMs by annotating characters' perceptions on ToMi and FANToM, respectively. Our evaluation of eight state-of-the-art LLMs reveals that the models generally perform well in perception inference while exhibiting limited capability in perception-to-belief inference (e.g., lack of inhibitory control). Based on these results, we present PercepToM, a novel ToM method leveraging LLMs' strong perception inference capability while supplementing their limited perception-to-belief inference. Experimental results demonstrate that PercepToM significantly enhances LLM's performance, especially in false belief scenarios.
Authors: Yangyang Yu, Zhiyuan Yao, Haohang Li, Zhiyang Deng, Yupeng Cao, Zhi Chen, Jordan W. Suchow, Rong Liu, Zhenyu Cui, Zhaozhuo Xu, Denghui Zhang, Koduvayur Subbalakshmi, Guojun Xiong, Yueru He, Jimin Huang, Dong Li, Qianqian Xie
Abstract: Large language models (LLMs) have demonstrated notable potential in conducting complex tasks and are increasingly utilized in various financial applications. However, high-quality sequential financial investment decision-making remains challenging. These tasks require multiple interactions with a volatile environment for every decision, demanding sufficient intelligence to maximize returns and manage risks. Although LLMs have been used to develop agent systems that surpass human teams and yield impressive investment returns, opportunities to enhance multi-sourced information synthesis and optimize decision-making outcomes through timely experience refinement remain unexplored. Here, we introduce the FinCon, an LLM-based multi-agent framework with CONceptual verbal reinforcement tailored for diverse FINancial tasks. Inspired by effective real-world investment firm organizational structures, FinCon utilizes a manager-analyst communication hierarchy. This structure allows for synchronized cross-functional agent collaboration towards unified goals through natural language interactions and equips each agent with greater memory capacity than humans. Additionally, a risk-control component in FinCon enhances decision quality by episodically initiating a self-critiquing mechanism to update systematic investment beliefs. The conceptualized beliefs serve as verbal reinforcement for the future agent's behavior and can be selectively propagated to the appropriate node that requires knowledge updates. This feature significantly improves performance while reducing unnecessary peer-to-peer communication costs. Moreover, FinCon demonstrates strong generalization capabilities in various financial tasks, including single stock trading and portfolio management.
Authors: Shuang Zhou, Mingquan Lin, Sirui Ding, Jiashuo Wang, Genevieve B. Melton, James Zou, Rui Zhang
Abstract: Automatic differential diagnosis (DDx) is an essential medical task that generates a list of potential diseases as differentials based on patient symptom descriptions. In practice, interpreting these differential diagnoses yields significant value but remains under-explored. Given the powerful capabilities of large language models (LLMs), we investigated using LLMs for interpretable DDx. Specifically, we curated the first DDx dataset with expert-derived interpretation on 570 clinical notes. Besides, we proposed Dual-Inf, a novel framework that enabled LLMs to conduct bidirectional inference (i.e., from symptoms to diagnoses and vice versa) for DDx interpretation. Both human and automated evaluation validated its efficacy in predicting and elucidating differentials across four base LLMs. In addition, Dual-Inf could reduce interpretation errors and hold promise for rare disease explanations. To the best of our knowledge, it is the first work that customizes LLMs for DDx explanation and comprehensively evaluates their interpretation performance. Overall, our study bridges a critical gap in DDx interpretation and enhances clinical decision-making.
Authors: Liang Shi, Zhengju Tang, Nan Zhang, Xiaotong Zhang, Zhi Yang
Abstract: The increasing volume of data in relational databases and the expertise needed for writing SQL queries pose challenges for users to access and analyze data. Text-to-SQL (Text2SQL) solves the issues by utilizing natural language processing (NLP) techniques to convert natural language into SQL queries. With the development of Large Language Models (LLMs), a range of LLM-based Text2SQL methods have emerged. This survey provides a comprehensive review of LLMs in Text2SQL tasks. We review benchmark datasets, prompt engineering methods, fine-tuning methods, and base models in LLM-based Text2SQL methods. We provide insights in each part and discuss future directions in this field.
Authors: Catarina G Belem, Markelle Kelly, Mark Steyvers, Sameer Singh, Padhraic Smyth
Abstract: _Uncertainty expressions_ such as "probably" or "highly unlikely" are pervasive in human language. While prior work has established that there is population-level agreement in terms of how humans quantitatively interpret these expressions, there has been little inquiry into the abilities of language models in the same context. In this paper, we investigate how language models map linguistic expressions of uncertainty to numerical responses. Our approach assesses whether language models can employ theory of mind in this setting: understanding the uncertainty of another agent about a particular statement, independently of the model's own certainty about that statement. We find that 7 out of 10 models are able to map uncertainty expressions to probabilistic responses in a human-like manner. However, we observe systematically different behavior depending on whether a statement is actually true or false. This sensitivity indicates that language models are substantially more susceptible to bias based on their prior knowledge (as compared to humans). These findings raise important questions and have broad implications for human-AI and AI-AI communication.
Authors: Wenxuan Xie, Gaochen Wu, Bowen Zhou
Abstract: Recent In-Context Learning based methods have achieved remarkable success in Text-to-SQL task. However, there is still a large gap between the performance of these models and human performance on datasets with complex database schema and difficult questions, such as BIRD. Besides, existing work has neglected to supervise intermediate steps when solving questions iteratively with question decomposition methods, and the schema linking methods used in these works are very rudimentary. To address these issues, we propose MAG-SQL, a multi-agent generative approach with soft schema linking and iterative Sub-SQL refinement. In our framework, an entity-based method with tables' summary is used to select the columns in database, and a novel targets-conditions decomposition method is introduced to decompose those complex questions. Additionally, we build a iterative generating module which includes a Sub-SQL Generator and Sub-SQL Refiner, introducing external oversight for each step of generation. Through a series of ablation studies, the effectiveness of each agent in our framework has been demonstrated. When evaluated on the BIRD benchmark with GPT-4, MAG-SQL achieves an execution accuracy of 61.08%, compared to the baseline accuracy of 46.35% for vanilla GPT-4 and the baseline accuracy of 57.56% for MAC-SQL. Besides, our approach makes similar progress on Spider. The codes are available at https://github.com/LancelotXWX/MAG-SQL.
Authors: Aman Priyanshu, Supriti Vijay
Abstract: This paper introduces FRACTURED-SORRY-Bench, a framework for evaluating the safety of Large Language Models (LLMs) against multi-turn conversational attacks. Building upon the SORRY-Bench dataset, we propose a simple yet effective method for generating adversarial prompts by breaking down harmful queries into seemingly innocuous sub-questions. Our approach achieves a maximum increase of +46.22\% in Attack Success Rates (ASRs) across GPT-4, GPT-4o, GPT-4o-mini, and GPT-3.5-Turbo models compared to baseline methods. We demonstrate that this technique poses a challenge to current LLM safety measures and highlights the need for more robust defenses against subtle, multi-turn attacks.
Authors: Raoyuan Zhao, Abdullatif K\"oksal, Yihong Liu, Leonie Weissweiler, Anna Korhonen, Hinrich Sch\"utze
Abstract: Traditional benchmarking in NLP typically involves using static held-out test sets. However, this approach often results in an overestimation of performance and lacks the ability to offer comprehensive, interpretable, and dynamic assessments of NLP models. Recently, works like DynaBench (Kiela et al., 2021) and CheckList (Ribeiro et al., 2020) have addressed these limitations through behavioral testing of NLP models with test types generated by a multistep human-annotated pipeline. Unfortunately, manually creating a variety of test types requires much human labor, often at prohibitive cost. In this work, we propose SYNTHEVAL, a hybrid behavioral testing framework that leverages large language models (LLMs) to generate a wide range of test types for a comprehensive evaluation of NLP models. SYNTHEVAL first generates sentences via LLMs using controlled generation, and then identifies challenging examples by comparing the predictions made by LLMs with task-specific NLP models. In the last stage, human experts investigate the challenging examples, manually design templates, and identify the types of failures the taskspecific models consistently exhibit. We apply SYNTHEVAL to two classification tasks, sentiment analysis and toxic language detection, and show that our framework is effective in identifying weaknesses of strong models on these tasks. We share our code in https://github.com/Loreley99/SynthEval_CheckList.
Authors: Kavya Manohar, Leena G Pillai, Elizabeth Sherly
Abstract: This paper explores the pitfalls in evaluating multilingual automatic speech recognition (ASR) models, with a particular focus on Indic language scripts. We investigate the text normalization routine employed by leading ASR models, including OpenAI Whisper, Meta's MMS, Seamless, and Assembly AI's Conformer, and their unintended consequences on performance metrics. Our research reveals that current text normalization practices, while aiming to standardize ASR outputs for fair comparison, by removing inconsistencies such as variations in spelling, punctuation, and special characters, are fundamentally flawed when applied to Indic scripts. Through empirical analysis using text similarity scores and in-depth linguistic examination, we demonstrate that these flaws lead to artificially improved performance metrics for Indic languages. We conclude by proposing a shift towards developing text normalization routines that leverage native linguistic expertise, ensuring more robust and accurate evaluations of multilingual ASR models.
Authors: Liangtai Sun, Danyu Luo, Da Ma, Zihan Zhao, Baocai Chen, Zhennan Shen, Su Zhu, Lu Chen, Xin Chen, Kai Yu
Abstract: Recently, there has been a significant upsurge of interest in leveraging large language models (LLMs) to assist scientific discovery. However, most LLMs only focus on general science, while they lack domain-specific knowledge, such as chemical molecules and amino acid sequences. To bridge these gaps, we introduce SciDFM, a mixture-of-experts LLM, which is trained from scratch and is able to conduct college-level scientific reasoning and understand molecules and amino acid sequences. We collect a large-scale training corpus containing numerous scientific papers and books from different disciplines as well as data from domain-specific databases. We further fine-tune the pre-trained model on lots of instruction data to improve performances on downstream benchmarks. From experiment results, we show that SciDFM achieves strong performance on general scientific benchmarks such as SciEval and SciQ, and it reaches a SOTA performance on domain-specific benchmarks among models of similar size. We further analyze the expert layers and show that the results of expert selection vary with data from different disciplines. To benefit the broader research community, we open-source SciDFM at https://huggingface.co/OpenDFM/SciDFM-MoE-A5.6B-v1.0.
Authors: Ziyu Wang, Hao Li, Di Huang, Amir M. Rahmani
Abstract: In digital healthcare, large language models (LLMs) have primarily been utilized to enhance question-answering capabilities and improve patient interactions. However, effective patient care necessitates LLM chains that can actively gather information by posing relevant questions. This paper presents HealthQ, a novel framework designed to evaluate the questioning capabilities of LLM healthcare chains. We implemented several LLM chains, including Retrieval-Augmented Generation (RAG), Chain of Thought (CoT), and reflective chains, and introduced an LLM judge to assess the relevance and informativeness of the generated questions. To validate HealthQ, we employed traditional Natural Language Processing (NLP) metrics such as Recall-Oriented Understudy for Gisting Evaluation (ROUGE) and Named Entity Recognition (NER)-based set comparison, and constructed two custom datasets from public medical note datasets, ChatDoctor and MTS-Dialog. Our contributions are threefold: we provide the first comprehensive study on the questioning capabilities of LLMs in healthcare conversations, develop a novel dataset generation pipeline, and propose a detailed evaluation methodology.
Authors: Ruizhe Chen, Xiaotian Zhang, Meng Luo, Wenhao Chai, Zuozhu Liu
Abstract: Aligning with personalized preferences, which vary significantly across cultural, educational, and political differences, poses a significant challenge due to the computational costs and data demands of traditional alignment methods. In response, this paper presents Personalized Alignment at Decoding-time (PAD), a novel framework designed to align LLM outputs with diverse personalized preferences during the inference phase, eliminating the need for additional training. By introducing a unique personalized reward modeling strategy, this framework decouples the text generation process from personalized preferences, facilitating the generation of generalizable token-level personalized rewards. The PAD algorithm leverages these rewards to guide the decoding process, dynamically tailoring the base model's predictions to personalized preferences. Extensive experimental results demonstrate that PAD not only outperforms existing training-based alignment methods in terms of aligning with diverse preferences but also shows significant generalizability to preferences unseen during training and scalability across different base models. This work advances the capability of LLMs to meet user needs in real-time applications, presenting a substantial step forward in personalized LLM alignment.
Authors: Joseph James, Chenghao Xiao, Yucheng Li, Chenghua Lin
Abstract: Rigour is crucial for scientific research as it ensures the reproducibility and validity of results and findings. Despite its importance, little work exists on modelling rigour computationally, and there is a lack of analysis on whether these criteria can effectively signal or measure the rigour of scientific papers in practice. In this paper, we introduce a bottom-up, data-driven framework to automatically identify and define rigour criteria and assess their relevance in scientific writing. Our framework includes rigour keyword extraction, detailed rigour definition generation, and salient criteria identification. Furthermore, our framework is domain-agnostic and can be tailored to the evaluation of scientific rigour for different areas, accommodating the distinct salient criteria across fields. We conducted comprehensive experiments based on datasets collected from two high impact venues for Machine Learning and NLP (i.e., ICLR and ACL) to demonstrate the effectiveness of our framework in modelling rigour. In addition, we analyse linguistic patterns of rigour, revealing that framing certainty is crucial for enhancing the perception of scientific rigour, while suggestion certainty and probability uncertainty diminish it.
Authors: Anish Pahilajani, Devasha Trivedi, Jincen Shuai, Khin S. Yone, Samyak Rajesh Jain, Namyong Park, Ryan A. Rossi, Nesreen K. Ahmed, Franck Dernoncourt, Yu Wang
Abstract: Large Language Models (LLMs) have excelled in multi-hop question-answering (M-QA) due to their advanced reasoning abilities. However, the impact of the inherent reasoning structures on LLM M-QA performance remains unclear, largely due to the absence of QA datasets that provide fine-grained reasoning structures. To address this gap, we introduce the Graph Reasoning-Structured Question Answering Dataset (GRS-QA), which includes both semantic contexts and reasoning structures for QA pairs. Unlike existing M-QA datasets, where different reasoning structures are entangled together, GRS-QA explicitly captures intricate reasoning pathways by constructing reasoning graphs, where nodes represent textual contexts and edges denote logical flows. These reasoning graphs of different structures enable a fine-grained evaluation of LLM reasoning capabilities across various reasoning structures. Our empirical analysis reveals that LLMs perform differently when handling questions with varying reasoning structures. This finding facilitates the exploration of textual structures as compared with semantics.
Authors: Sam Blouir, Jimmy T. H. Smith, Antonios Anastasopoulos, Amarda Shehu
Abstract: Efficient state space models (SSMs), such as linear recurrent neural networks and linear attention variants, offer computational advantages over Transformers but struggle with tasks requiring long-range in-context retrieval-like text copying, associative recall, and question answering over long contexts. Previous efforts to address these challenges have focused on architectural modifications, often reintroducing computational inefficiencies. In this paper, we propose a novel training procedure, Birdie, that significantly enhances the in-context retrieval capabilities of SSMs without altering their architecture. Our approach combines bidirectional input processing with dynamic mixtures of specialized pre-training objectives, optimized via reinforcement learning. We introduce a new bidirectional SSM architecture that seamlessly transitions from bidirectional context processing to causal generation. Experimental evaluations demonstrate that Birdie markedly improves performance on retrieval-intensive tasks such as multi-number phone book lookup, long paragraph question-answering, and infilling. This narrows the performance gap with Transformers, while retaining computational efficiency. Our findings highlight the importance of training procedures in leveraging the fixed-state capacity of SSMs, offering a new direction to advance their capabilities. All code and pre-trained models are available at https://www.github.com/samblouir/birdie, with support for JAX and PyTorch.
Authors: Baizhou Huang, Xiao Pu, Xiaojun Wan
Abstract: Watermarking has emerged as a prominent technique for LLM-generated content detection by embedding imperceptible patterns. Despite supreme performance, its robustness against adversarial attacks remains underexplored. Previous work typically considers a grey-box attack setting, where the specific type of watermark is already known. Some even necessitates knowledge about hyperparameters of the watermarking method. Such prerequisites are unattainable in real-world scenarios. Targeting at a more realistic black-box threat model with fewer assumptions, we here propose $B^4$, a black-box scrubbing attack on watermarks. Specifically, we formulate the watermark scrubbing attack as a constrained optimization problem by capturing its objectives with two distributions, a Watermark Distribution and a Fidelity Distribution. This optimization problem can be approximately solved using two proxy distributions. Experimental results across 12 different settings demonstrate the superior performance of $B^4$ compared with other baselines.
Authors: Mete Ismayilzada, Claire Stevenson, Lonneke van der Plas
Abstract: Storytelling is a fundamental aspect of human communication, relying heavily on creativity to produce narratives that are novel, appropriate, and surprising. While large language models (LLMs) have recently demonstrated the ability to generate high-quality stories, their creative capabilities remain underexplored. Previous research has either focused on creativity tests requiring short responses or primarily compared model performance in story generation to that of professional writers. However, the question of whether LLMs exhibit creativity in writing short stories on par with the average human remains unanswered. In this work, we conduct a systematic analysis of creativity in short story generation across LLMs and everyday people. Using a five-sentence creative story task, commonly employed in psychology to assess human creativity, we automatically evaluate model- and human-generated stories across several dimensions of creativity, including novelty, surprise, and diversity. Our findings reveal that while LLMs can generate stylistically complex stories, they tend to fall short in terms of creativity when compared to average human writers.
Authors: Fan Nie, Xiaotian Hou, Shuhang Lin, James Zou, Huaxiu Yao, Linjun Zhang
Abstract: The propensity of Large Language Models (LLMs) to generate hallucinations and non-factual content undermines their reliability in high-stakes domains, where rigorous control over Type I errors (the conditional probability of incorrectly classifying hallucinations as truthful content) is essential. Despite its importance, formal verification of LLM factuality with such guarantees remains largely unexplored. In this paper, we introduce FactTest, a novel framework that statistically assesses whether a LLM can confidently provide correct answers to given questions with high-probability correctness guarantees. We formulate factuality testing as hypothesis testing problem to enforce an upper bound of Type I errors at user-specified significance levels. Notably, we prove that our framework also ensures strong Type II error control under mild conditions and can be extended to maintain its effectiveness when covariate shifts exist. Our approach is distribution-free and works for any number of human-annotated samples. It is model-agnostic and applies to any black-box or white-box LM. Extensive experiments on question-answering (QA) and multiple-choice benchmarks demonstrate that FactTest effectively detects hallucinations and improves the model's ability to abstain from answering unknown questions, leading to an over 40% accuracy improvement.
Authors: Xin Zhang, Victor S. Sheng
Abstract: We propose an innovative token representation and update method in a new ultra-small language model: the Wave network. Specifically, we use a complex vector to represent each token, encoding both global and local semantics of the input text. A complex vector consists of two components: a magnitude vector representing the global semantics of the input text, and a phase vector capturing the relationships between individual tokens and global semantics. Experiments on the AG News text classification task demonstrate that, when generating complex vectors from randomly initialized token embeddings, our single-layer Wave Network achieves 90.91% accuracy with wave interference and 91.66% with wave modulation - outperforming a single Transformer layer using BERT pre-trained embeddings by 19.23% and 19.98%, respectively, and approaching the accuracy of the pre-trained and fine-tuned BERT base model (94.64%). Additionally, compared to BERT base, the Wave Network reduces video memory usage and training time by 77.34% and 85.62% during wave modulation. In summary, we used a 2.4-million-parameter small language model to achieve accuracy comparable to a 100-million-parameter BERT model in text classification.
Authors: Dalal Waadallah Shehab
Abstract: This study investigates the translation of circumlocution from Arabic to English in a corpus of short stories by renowned Arabic authors. By analyzing the source and target texts, the study aims to identify and categorize circumlocution instances in Arabic and their corresponding renditions in English. The study employs Nida's (1964) translation theory as a framework to assess the appropriateness of the translation strategies employed. It examines the extent to which translators successfully rendered Arabic circumlocution into English, identifying potential challenges and limitations in the translation process. The findings reveal significant similarities between Arabic circumlocution categories and English metadiscourse categories, particularly in terms of textual and interpersonal functions. However, the study also highlights instances where translators encountered difficulties in accurately conveying the nuances of circumlocution, often resorting to strategies like addition, subtraction, and alteration.https://ntu.edu.iq/
URLs: https://ntu.edu.iq/
Authors: Yincen Qu, Chao Ma, Xiangying Dai, Hui Zhou, Yiting Wu, Hengyue Liu
Abstract: In the industry, numerous tasks are deployed online. Traditional approaches often tackle each task separately by its own network, which leads to excessive costs for developing and scaling models, especially in the context of large language models. Although multi-task methods can save costs through parameter sharing, they often struggle to outperform single-task methods in real-world applications. To tackle these challenges, we present a three-stage multi-task learning framework for large language models. It involves task filtering, followed by fine-tuning on high-resource tasks, and finally fine-tuning on all tasks. We conducted comprehensive experiments in single-task and multi-task settings. Our approach, exemplified on different benchmarks, demonstrates that it is able to achieve performance comparable to the single-task method while reducing up to 90.9\% of its overhead.
Authors: Laura Cabello, Carmen Martin-Turrero, Uchenna Akujuobi, Anders S{\o}gaard, Carlos Bobed
Abstract: Question answering is a natural language understanding task that involves reasoning over both explicit context and unstated, relevant domain knowledge. Large language models (LLMs), which underpin most contemporary question answering systems, struggle to induce how concepts relate in specialized domains such as medicine. Existing medical LLMs are also costly to train. In this work, we present MEG, a parameter-efficient approach for medical knowledge-augmented LLMs. MEG uses a lightweight mapping network to integrate graph embeddings into the LLM, enabling it to leverage external knowledge in a cost-effective way. We evaluate our method on four popular medical multiple-choice datasets and show that LLMs greatly benefit from the factual grounding provided by knowledge graph embeddings. MEG attains an average of +10.2% accuracy over the Mistral-Instruct baseline, and +6.7% over specialized models like BioMistral. We also show results based on Llama-3. Finally, we show that MEG's performance remains robust to the choice of graph encoder.
Authors: Guillermo Villate-Castillo, Javier Del Ser, Borja Sanz
Abstract: Content moderation typically combines the efforts of human moderators and machine learning models. However, these systems often rely on data where significant disagreement occurs during moderation, reflecting the subjective nature of toxicity perception. Rather than dismissing this disagreement as noise, we interpret it as a valuable signal that highlights the inherent ambiguity of the content,an insight missed when only the majority label is considered. In this work, we introduce a novel content moderation framework that emphasizes the importance of capturing annotation disagreement. Our approach uses multitask learning, where toxicity classification serves as the primary task and annotation disagreement is addressed as an auxiliary task. Additionally, we leverage uncertainty estimation techniques, specifically Conformal Prediction, to account for both the ambiguity in comment annotations and the model's inherent uncertainty in predicting toxicity and disagreement.The framework also allows moderators to adjust thresholds for annotation disagreement, offering flexibility in determining when ambiguity should trigger a review. We demonstrate that our joint approach enhances model performance, calibration, and uncertainty estimation, while offering greater parameter efficiency and improving the review process in comparison to single-task methods.
Authors: Zirui Liu, Guanchu Wang, Shaochen Zhong, Zhaozhuo Xu, Daochen Zha, Ruixiang Tang, Zhimeng Jiang, Kaixiong Zhou, Vipin Chaudhary, Shuai Xu, Xia Hu
Abstract: With the rapid growth in model size, fine-tuning the large pre-trained language model has become increasingly difficult due to its extensive memory usage. Previous works usually focus on reducing the number of trainable parameters in the network. While the model parameters do contribute to memory usage, the primary memory bottleneck during training arises from storing feature maps, also known as activations, as they are crucial for gradient calculation. Notably, neural networks are usually trained using stochastic gradient descent. We argue that in stochastic optimization, models can handle noisy gradients as long as the gradient estimator is unbiased with reasonable variance. Following this motivation, we propose a new family of unbiased estimators called WTA-CRS, for matrix production with reduced variance, which only requires storing the sub-sampled activations for calculating the gradient. Our work provides both theoretical and experimental evidence that, in the context of tuning transformers, our proposed estimators exhibit lower variance compared to existing ones. By replacing the linear operation with our approximated one in transformers, we can achieve up to 2.7$\times$ peak memory reduction with almost no accuracy drop and enables up to $6.4\times$ larger batch size. Under the same hardware, WTA-CRS enables better down-streaming task performance by applying larger models and/or faster training speed with larger batch sizes.
Authors: Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho
Abstract: Large Language Models (LLMs) are becoming a prominent generative AI tool, where the user enters a query and the LLM generates an answer. To reduce harm and misuse, efforts have been made to align these LLMs to human values using advanced training techniques such as Reinforcement Learning from Human Feedback (RLHF). However, recent studies have highlighted the vulnerability of LLMs to adversarial jailbreak attempts aiming at subverting the embedded safety guardrails. To address this challenge, this paper defines and investigates the Refusal Loss of LLMs and then proposes a method called Gradient Cuff to detect jailbreak attempts. Gradient Cuff exploits the unique properties observed in the refusal loss landscape, including functional values and its smoothness, to design an effective two-step detection strategy. Experimental results on two aligned LLMs (LLaMA-2-7B-Chat and Vicuna-7B-V1.5) and six types of jailbreak attacks (GCG, AutoDAN, PAIR, TAP, Base64, and LRL) show that Gradient Cuff can significantly improve the LLM's rejection capability for malicious jailbreak queries, while maintaining the model's performance for benign user queries by adjusting the detection threshold.
Authors: Jingnan Zheng, Han Wang, An Zhang, Tai D. Nguyen, Jun Sun, Tat-Seng Chua
Abstract: Large Language Models (LLMs) can elicit unintended and even harmful content when misaligned with human values, posing severe risks to users and society. To mitigate these risks, current evaluation benchmarks predominantly employ expert-designed contextual scenarios to assess how well LLMs align with human values. However, the labor-intensive nature of these benchmarks limits their test scope, hindering their ability to generalize to the extensive variety of open-world use cases and identify rare but crucial long-tail risks. Additionally, these static tests fail to adapt to the rapid evolution of LLMs, making it hard to evaluate timely alignment issues. To address these challenges, we propose ALI-Agent, an evaluation framework that leverages the autonomous abilities of LLM-powered agents to conduct in-depth and adaptive alignment assessments. ALI-Agent operates through two principal stages: Emulation and Refinement. During the Emulation stage, ALI-Agent automates the generation of realistic test scenarios. In the Refinement stage, it iteratively refines the scenarios to probe long-tail risks. Specifically, ALI-Agent incorporates a memory module to guide test scenario generation, a tool-using module to reduce human labor in tasks such as evaluating feedback from target LLMs, and an action module to refine tests. Extensive experiments across three aspects of human values--stereotypes, morality, and legality--demonstrate that ALI-Agent, as a general evaluation framework, effectively identifies model misalignment. Systematic analysis also validates that the generated test scenarios represent meaningful use cases, as well as integrate enhanced measures to probe long-tail risks. Our code is available at https://github.com/SophieZheng998/ALI-Agent.git
Authors: Henry Hengyuan Zhao, Pan Zhou, Difei Gao, Zechen Bai, Mike Zheng Shou
Abstract: Question answering, asking, and assessment are three innate human traits crucial for understanding the world and acquiring knowledge. By enhancing these capabilities, humans can more effectively utilize data, leading to better comprehension and learning outcomes. Current Multimodal Large Language Models (MLLMs) primarily focus on question answering, often neglecting the full potential of questioning and assessment skills. Inspired by the human learning mechanism, we introduce LOVA3, an innovative framework named "Learning tO Visual question Answering, Asking and Assessment," designed to equip MLLMs with these additional capabilities. Our approach involves the creation of two supplementary training tasks GenQA and EvalQA, aiming at fostering the skills of asking and assessing questions in the context of images. To develop the questioning ability, we compile a comprehensive set of multimodal foundational tasks. For assessment, we introduce a new benchmark called EvalQABench, comprising 64,000 training samples (split evenly between positive and negative samples) and 5,000 validation and testing samples. We posit that enhancing MLLMs with the capabilities to answer, ask, and assess questions will enhance their multimodal comprehension, ultimately improving overall performance. To validate this hypothesis, we train MLLMs using the LOVA3 framework and evaluate them on a range of multimodal datasets and benchmarks. Our results demonstrate consistent performance gains, underscoring the critical role of these additional tasks in fostering comprehensive intelligence in MLLMs. The code is available at https://github.com/showlab/LOVA3.
Authors: Hyunseok Lee, Jihoon Tack, Jinwoo Shin
Abstract: The remarkable capabilities and easy accessibility of large language models (LLMs) have significantly increased societal risks (e.g., fake news generation), necessitating the development of LLM-generated text (LGT) detection methods for safe usage. However, detecting LGTs is challenging due to the vast number of LLMs, making it impractical to account for each LLM individually; hence, it is crucial to identify the common characteristics shared by these models. In this paper, we draw attention to a common feature of recent powerful LLMs, namely the alignment training, i.e., training LLMs to generate human-preferable texts. Our key finding is that as these aligned LLMs are trained to maximize the human preferences, they generate texts with higher estimated preferences even than human-written texts; thus, such texts are easily detected by using the reward model (i.e., an LLM trained to model human preference distribution). Based on this finding, we propose two training schemes to further improve the detection ability of the reward model, namely (i) continual preference fine-tuning to make the reward model prefer aligned LGTs even further and (ii) reward modeling of Human/LLM mixed texts (a rephrased texts from human-written texts using aligned LLMs), which serves as a median preference text corpus between LGTs and human-written texts to learn the decision boundary better. We provide an extensive evaluation by considering six text domains across twelve aligned LLMs, where our method demonstrates state-of-the-art results. Code is available at https://github.com/hyunseoklee-ai/ReMoDetect.
Authors: Robert Wu, Vardan Papyan
Abstract: Neural collapse ($\mathcal{NC}$) is a phenomenon observed in classification tasks where top-layer representations collapse into their class means, which become equinorm, equiangular and aligned with the classifiers. These behaviors -- associated with generalization and robustness -- would manifest under specific conditions: models are trained towards zero loss, with noise-free labels belonging to balanced classes, which do not outnumber the model's hidden dimension. Recent studies have explored $\mathcal{NC}$ in the absence of one or more of these conditions to extend and capitalize on the associated benefits of ideal geometries. Language modeling presents a curious frontier, as \textit{training by token prediction} constitutes a classification task where none of the conditions exist: the vocabulary is imbalanced and exceeds the embedding dimension; different tokens might correspond to similar contextual embeddings; and large language models (LLMs) in particular are typically only trained for a few epochs. This paper empirically investigates the impact of scaling the architectures and training of causal language models (CLMs) on their progression towards $\mathcal{NC}$. We find that $\mathcal{NC}$ properties that develop with scale (and regularization) are linked to generalization. Moreover, there is evidence of some relationship between $\mathcal{NC}$ and generalization independent of scale. Our work thereby underscores the generality of $\mathcal{NC}$ as it extends to the novel and more challenging setting of language modeling. Downstream, we seek to inspire further research on the phenomenon to deepen our understanding of LLMs -- and neural networks at large -- and improve existing architectures based on $\mathcal{NC}$-related properties. Our code is hosted on GitHub at https://github.com/rhubarbwu/linguistic-collapse .
Authors: Jacob Dunefsky, Philippe Chlenski, Neel Nanda
Abstract: A key goal in mechanistic interpretability is circuit analysis: finding sparse subgraphs of models corresponding to specific behaviors or capabilities. However, MLP sublayers make fine-grained circuit analysis on transformer-based language models difficult. In particular, interpretable features -- such as those found by sparse autoencoders (SAEs) -- are typically linear combinations of extremely many neurons, each with its own nonlinearity to account for. Circuit analysis in this setting thus either yields intractably large circuits or fails to disentangle local and global behavior. To address this we explore transcoders, which seek to faithfully approximate a densely activating MLP layer with a wider, sparsely-activating MLP layer. We introduce a novel method for using transcoders to perform weights-based circuit analysis through MLP sublayers. The resulting circuits neatly factorize into input-dependent and input-invariant terms. We then successfully train transcoders on language models with 120M, 410M, and 1.4B parameters, and find them to perform at least on par with SAEs in terms of sparsity, faithfulness, and human-interpretability. Finally, we apply transcoders to reverse-engineer unknown circuits in the model, and we obtain novel insights regarding the "greater-than circuit" in GPT2-small. Our results suggest that transcoders can prove effective in decomposing model computations involving MLPs into interpretable circuits. Code is available at https://github.com/jacobdunefsky/transcoder_circuits/.
URLs: https://github.com/jacobdunefsky/transcoder_circuits/.
Authors: Yuyang Jiang, Chacha Chen, Dang Nguyen, Benjamin M. Mervak, Chenhao Tan
Abstract: GPT-4V's purported strong multimodal abilities raise interests in using it to automate radiology report writing, but there lacks thorough evaluations. In this work, we perform a systematic evaluation of GPT-4V in generating radiology reports on two chest X-ray report datasets: MIMIC-CXR and IU X-Ray. We attempt to directly generate reports using GPT-4V through different prompting strategies and find that it fails terribly in both lexical metrics and clinical efficacy metrics. To understand the low performance, we decompose the task into two steps: 1) the medical image reasoning step of predicting medical condition labels from images; and 2) the report synthesis step of generating reports from (groundtruth) conditions. We show that GPT-4V's performance in image reasoning is consistently low across different prompts. In fact, the distributions of model-predicted labels remain constant regardless of which groundtruth conditions are present on the image, suggesting that the model is not interpreting chest X-rays meaningfully. Even when given groundtruth conditions in report synthesis, its generated reports are less correct and less natural-sounding than a finetuned LLaMA-2. Altogether, our findings cast doubt on the viability of using GPT-4V in a radiology workflow.
Authors: Wei Xie, Shuoyoucheng Ma, Zhenhua Wang, Enze Wang, Kai Chen, Xiaobing Sun, Baosheng Wang
Abstract: Despite their proficiency in math tasks, the mechanisms underlying LLMs' mathematical reasoning abilities remain a subject of debate. Recent studies suggest that chain-of-thought (CoT) prompts can bolster mathematical reasoning by encouraging LLMs to employ human-like logical reasoning (System 2), enabling them to excel on the Cognitive Reflection Test (CRT). To assess whether LLMs genuinely possess System 2-like logical reasoning, we introduced targeted modifications to CRT problems. Our findings reveal that, despite the use of CoT prompts, mainstream LLMs, including the latest o1-preview model, continue to exhibit a significant error rate. Further analysis indicates that they predominantly rely on System 1-like intuitive reasoning and pattern matching derived from training data, rather than demonstrating mastery of mathematical thinking. This discovery challenges the prevailing notion that LLMs possess genuine logical reasoning abilities and that CoT can enhance them. Consequently, this work may temper overly optimistic projections regarding LLMs' advancement toward artificial general intelligence.
Authors: Xin Li, Qizhi Chu, Yubin Chen, Yang Liu, Yaoqi Liu, Zekai Yu, Weize Chen, Chen Qian, Chuan Shi, Cheng Yang
Abstract: Graphs are widely used for modeling relational data in real-world scenarios, such as social networks and urban computing. Existing LLM-based graph analysis approaches either integrate graph neural networks (GNNs) for specific machine learning tasks, limiting their transferability, or rely solely on LLMs' internal reasoning ability, resulting in suboptimal performance. To address these limitations, we take advantage of recent advances in LLM-based agents, which have shown capabilities of utilizing external knowledge or tools for problem solving. By simulating human problem-solving strategies such as analogy and collaboration, we propose a multi-agent system based on LLMs named GraphTeam, for graph analysis. GraphTeam consists of five LLM-based agents from three modules, and the agents with different specialities can collaborate with each other to address complex problems. Specifically, (1) input-output normalization module: the question agent extracts and refines four key arguments from the original question, facilitating the problem understanding, and the answer agent organizes the results to meet the output requirement; (2) external knowledge retrieval module: we first build a knowledge base consisting of relevant documentation and experience information, and then the search agent retrieves the most relevant entries for each question. (3) problem-solving module: given the retrieved information from search agent, the coding agent uses established algorithms via programming to generate solutions, and in case the coding agent does not work, the reasoning agent will directly compute the results without programming. Extensive experiments on six graph analysis benchmarks demonstrate that GraphTeam achieves state-of-the-art performance with an average 25.85% improvement over the best baseline in terms of accuracy. The code and data are available at https://github.com/BUPT-GAMMA/GraphTeam.
Authors: Edward Vendrow, Omiros Pantazis, Alexander Shepard, Gabriel Brostow, Kate E. Jones, Oisin Mac Aodha, Sara Beery, Grant Van Horn
Abstract: We introduce INQUIRE, a text-to-image retrieval benchmark designed to challenge multimodal vision-language models on expert-level queries. INQUIRE includes iNaturalist 2024 (iNat24), a new dataset of five million natural world images, along with 250 expert-level retrieval queries. These queries are paired with all relevant images comprehensively labeled within iNat24, comprising 33,000 total matches. Queries span categories such as species identification, context, behavior, and appearance, emphasizing tasks that require nuanced image understanding and domain expertise. Our benchmark evaluates two core retrieval tasks: (1) INQUIRE-Fullrank, a full dataset ranking task, and (2) INQUIRE-Rerank, a reranking task for refining top-100 retrievals. Detailed evaluation of a range of recent multimodal models demonstrates that INQUIRE poses a significant challenge, with the best models failing to achieve an mAP@50 above 50%. In addition, we show that reranking with more powerful multimodal models can enhance retrieval performance, yet there remains a significant margin for improvement. By focusing on scientifically-motivated ecological challenges, INQUIRE aims to bridge the gap between AI capabilities and the needs of real-world scientific inquiry, encouraging the development of retrieval systems that can assist with accelerating ecological and biodiversity research. Our dataset and code are available at https://inquire-benchmark.github.io
Authors: Guan Zhe Hong, Nishanth Dikkala, Enming Luo, Cyrus Rashtchian, Xin Wang, Rina Panigrahy
Abstract: Large language models (LLMs) have shown amazing performance on tasks that require planning and reasoning. Motivated by this, we investigate the internal mechanisms that underpin a network's ability to perform complex logical reasoning. We first construct a synthetic propositional logic problem that serves as a concrete test-bed for network training and evaluation. Crucially, this problem demands nontrivial planning to solve, but we can train a small transformer to achieve perfect accuracy. Building on our set-up, we then pursue an understanding of precisely how a three-layer transformer, trained from scratch, solves this problem. We are able to identify certain "planning" and "reasoning" circuits in the network that necessitate cooperation between the attention blocks to implement the desired logic. To expand our findings, we then study a larger model, Mistral 7B. Using activation patching, we characterize internal components that are critical in solving our logic problem. Overall, our work systemically uncovers novel aspects of small and large transformers, and continues the study of how they plan and reason.