Authors: Sanjana Sankar, Martin Lenglet, Gerard Bailly, Denis Beautemps, Thomas Hueber
Abstract: This paper presents a novel approach for the automatic generation of Cued Speech (ACSG), a visual communication system used by people with hearing impairment to better elicit the spoken language. We explore transfer learning strategies by leveraging a pre-trained audiovisual autoregressive text-to-speech model (AVTacotron2). This model is reprogrammed to infer Cued Speech (CS) hand and lip movements from text input. Experiments are conducted on two publicly available datasets, including one recorded specifically for this study. Performance is assessed using an automatic CS recognition system. With a decoding accuracy at the phonetic level reaching approximately 77%, the results demonstrate the effectiveness of our approach.
Authors: Allison Lahnala, Vasudha Varadarajan, Lucie Flek, H. Andrew Schwartz, Ryan L. Boyd
Abstract: The proliferation of ideological movements into extremist factions via social media has become a global concern. While radicalization has been studied extensively within the context of specific ideologies, our ability to accurately characterize extremism in more generalizable terms remains underdeveloped. In this paper, we propose a novel method for extracting and analyzing extremist discourse across a range of online community forums. By focusing on verbal behavioral signatures of extremist traits, we develop a framework for quantifying extremism at both user and community levels. Our research identifies 11 distinct factors, which we term ``The Extremist Eleven,'' as a generalized psychosocial model of extremism. Applying our method to various online communities, we demonstrate an ability to characterize ideologically diverse communities across the 11 extremist traits. We demonstrate the power of this method by analyzing user histories from members of the incel community. We find that our framework accurately predicts which users join the incel community up to 10 months before their actual entry with an AUC of $>0.6$, steadily increasing to AUC ~0.9 three to four months before the event. Further, we find that upon entry into an extremist forum, the users tend to maintain their level of extremism within the community, while still remaining distinguishable from the general online discourse. Our findings contribute to the study of extremism by introducing a more holistic, cross-ideological approach that transcends traditional, trait-specific models.
Authors: \c{S}aziye Bet\"ul \"Ozate\c{s}, Tar{\i}k Emre T{\i}ra\c{s}, Ece Elif Adak, Berat Do\u{g}an, Fatih Burak Karag\"oz, Efe Eren Gen\c{c}, Esma F. Bilgin Ta\c{s}demir
Abstract: This paper introduces foundational resources and models for natural language processing (NLP) of historical Turkish, a domain that has remained underexplored in computational linguistics. We present the first named entity recognition (NER) dataset, HisTR and the first Universal Dependencies treebank, OTA-BOUN for a historical form of the Turkish language along with transformer-based models trained using these datasets for named entity recognition, dependency parsing, and part-of-speech tagging tasks. Additionally, we introduce Ottoman Text Corpus (OTC), a clean corpus of transliterated historical Turkish texts that spans a wide range of historical periods. Our experimental results show significant improvements in the computational analysis of historical Turkish, achieving promising results in tasks that require understanding of historical linguistic structures. They also highlight existing challenges, such as domain adaptation and language variations across time periods. All of the presented resources and models are made available at https://huggingface.co/bucolin to serve as a benchmark for future progress in historical Turkish NLP.
Authors: Sara Bourbour Hosseinbeigi, Sina Asghari, Mohammad Ali Seif Kashani, Mohammad Hossein Shalchian, Mohammad Amin Abbasi
Abstract: This paper examines the specific obstacles of constructing Retrieval-Augmented Generation(RAG) systems in low-resource languages, with a focus on Persian's complicated morphology and versatile syntax. The research aims to improve retrieval and generation accuracy by introducing Persian-specific models, namely MatinaRoberta(a masked language model) and MatinaSRoberta(a fine-tuned Sentence-BERT), along with a comprehensive benchmarking framework. Three datasets-general knowledge(PQuad), scientifically specialized texts, and organizational reports, were used to assess these models after they were trained on a varied corpus of 73.11 billion Persian tokens. The methodology involved extensive pretraining, fine-tuning with tailored loss functions, and systematic evaluations using both traditional metrics and the Retrieval-Augmented Generation Assessment framework. The results show that MatinaSRoberta outperformed previous embeddings, achieving superior contextual relevance and retrieval accuracy across datasets. Temperature tweaking, chunk size modifications, and document summary indexing were explored to enhance RAG setups. Larger models like Llama-3.1 (70B) consistently demonstrated the highest generation accuracy, while smaller models faced challenges with domain-specific and formal contexts. The findings underscore the potential for developing RAG systems in Persian through customized embeddings and retrieval-generation settings and highlight the enhancement of NLP applications such as search engines and legal document analysis in low-resource languages.
Authors: Long Mai, Julie Carson-Berndsen
Abstract: Recent advancements in large language models (LLMs) have led to significant progress in text-based dialogue systems. These systems can now generate high-quality responses that are accurate and coherent across a wide range of topics and tasks. However, spoken dialogue systems still lag behind in terms of naturalness. They tend to produce robotic interactions, with issues such as slow response times, overly generic or cautious replies, and a lack of natural rhythm and fluid turn-taking. This shortcoming is largely due to the over-reliance on the traditional cascaded design, which involve separate, sequential components, as well as the use of text as an intermediate representation. This paper propose a real-time, textless spoken dialogue generation model (RTTL-DG) that aims to overcome these challenges. Our system enables fluid turn-taking and generates responses with minimal delay by processing streaming spoken conversation directly. Additionally, our model incorporates backchannels, filters, laughter, and other paralinguistic signals, which are often absent in cascaded dialogue systems, to create more natural and human-like interactions. The implementations and generated samples are available in our repository: https://github.com/mailong25/rts2s-dg
Authors: Tommaso Soru, Jim Marshall
Abstract: In the constantly changing field of data-driven decision making, accurately predicting future events is crucial for strategic planning in various sectors. The emergence of Large Language Models (LLMs) marks a significant advancement in this area, offering advanced tools that utilise extensive text data for prediction. In this industry paper, we introduce a novel method for AI-driven foresight using LLMs. Building on top of previous research, we employ data on current trends and their trajectories for generating forecasts on 15 different topics. Subsequently, we estimate their probabilities via a multi-step approach based on log probabilities. We show we achieve a Brier score of 0.186, meaning a +26% improvement over random chance and a +19% improvement over widely-available AI systems.
Authors: Hanna Zubkova, Ji-Hoon Park, Seong-Whan Lee
Abstract: Bearing in mind the limited parametric knowledge of Large Language Models (LLMs), retrieval-augmented generation (RAG) which supplies them with the relevant external knowledge has served as an approach to mitigate the issue of hallucinations to a certain extent. However, uniformly retrieving supporting context makes response generation source-inefficient, as triggering the retriever is not always necessary, or even inaccurate, when a model gets distracted by noisy retrieved content and produces an unhelpful answer. Motivated by these issues, we introduce Semantic Uncertainty Guided Adaptive Retrieval (SUGAR), where we leverage context-based entropy to actively decide whether to retrieve and to further determine between single-step and multi-step retrieval. Our empirical results show that selective retrieval guided by semantic uncertainty estimation improves the performance across diverse question answering tasks, as well as achieves a more efficient inference.
Authors: Jun-Hyeok Cha, Seung-Bin Kim, Hyung-Seok Oh, Seong-Whan Lee
Abstract: Recently, there has been a growing demand for conversational speech synthesis (CSS) that generates more natural speech by considering the conversational context. To address this, we introduce JELLY, a novel CSS framework that integrates emotion recognition and context reasoning for generating appropriate speech in conversation by fine-tuning a large language model (LLM) with multiple partial LoRA modules. We propose an Emotion-aware Q-former encoder, which enables the LLM to perceive emotions in speech. The encoder is trained to align speech emotions with text, utilizing datasets of emotional speech. The entire model is then fine-tuned with conversational speech data to infer emotional context for generating emotionally appropriate speech in conversation. Our experimental results demonstrate that JELLY excels in emotional context modeling, synthesizing speech that naturally aligns with conversation, while mitigating the scarcity of emotional conversational speech datasets.
Authors: Wei Tang, Jiawei Yu, Yuang Li, Yanqing Zhao, Weidong Zhang, Wei Feng, Min Zhang, Hao Yang
Abstract: The inaccurate translation of numbers can lead to significant security issues, ranging from financial setbacks to medical inaccuracies. While large language models (LLMs) have made significant advancements in machine translation, their capacity for translating numbers has not been thoroughly explored. This study focuses on evaluating the reliability of LLM-based machine translation systems when handling numerical data. In order to systematically test the numerical translation capabilities of currently open source LLMs, we have constructed a numerical translation dataset between Chinese and English based on real business data, encompassing ten types of numerical translation. Experiments on the dataset indicate that errors in numerical translation are a common issue, with most open-source LLMs faltering when faced with our test scenarios. Especially when it comes to numerical types involving large units like ``million", ``billion", and "yi", even the latest llama3.1 8b model can have error rates as high as 20%. Finally, we introduce three potential strategies to mitigate the numerical mistranslations for large units.
Authors: Qingyu Ren, Jie Zeng, Qianyu He, Jiaqing Liang, Yanghua Xiao, Weikang Zhou, Zeye Sun, Fei Yu
Abstract: It is crucial for large language models (LLMs) to follow instructions that involve multiple constraints. However, soft constraints are semantically related and difficult to verify through automated methods. These constraints remain a significant challenge for LLMs. To enhance the ability of LLMs to follow soft constraints, we initially design a pipeline to obtain high-quality outputs automatically. Additionally, to fully utilize the acquired data, we introduce a training paradigm based on curriculum learning. We experimentally evaluate the effectiveness of our methods in improving LLMs' soft constraint following ability and analyze the factors driving the improvements. The datasets and code are publicly available at https://github.com/Rainier-rq/FollowSoftConstraints.
Authors: Zixuan Ke, Yifei Ming, Xuan-Phi Nguyen, Caiming Xiong, Shafiq Joty
Abstract: Domain-adaptive post-training of large language models (LLMs) has emerged as a promising approach for specialized domains such as medicine and finance. However, significant challenges remain in identifying optimal adaptation criteria and training strategies across varying data and model configurations. To address these challenges, we introduce FINDAP, a systematic and fine-grained investigation into domain-adaptive post-training of LLMs for the finance domain. Our approach begins by identifying the core capabilities required for the target domain and designing a comprehensive evaluation suite aligned with these needs. We then analyze the effectiveness of key post-training stages, including continual pretraining, instruction tuning, and preference alignment. Building on these insights, we propose an effective training recipe centered on a novel preference data distillation method, which leverages process signals from a generative reward model. The resulting model, Llama-Fin, achieves state-of-the-art performance across a wide range of financial tasks. Our analysis also highlights how each post-training stage contributes to distinct capabilities, uncovering specific challenges and effective solutions, providing valuable insights for domain adaptation of LLMs. Project page: https://github.com/SalesforceAIResearch/FinDap
Authors: Wenqian Cui, Xiaoqi Jiao, Ziqiao Meng, Irwin King
Abstract: With the growing demand for developing speech-based interaction models, end-to-end Spoken Language Models (SLMs) have emerged as a promising solution. When engaging in conversations with humans, it is essential for these models to comprehend a wide range of world knowledge. In this paper, we introduce VoxEval, a novel speech question-answering benchmark specifically designed to assess SLMs' knowledge understanding through purely speech-based interactions. Unlike existing AudioQA benchmarks, VoxEval maintains speech format for both questions and answers, evaluates model robustness across diverse audio conditions (varying timbres, audio qualities, and speaking styles), and pioneers the assessment of challenging domains like mathematical problem-solving in spoken format. Our comprehensive evaluation of recent SLMs using VoxEval reveals significant performance limitations in current models, highlighting crucial areas for future improvements.
Authors: Benjamin Reichman, Xiaofan Yu, Lanxiang Hu, Jack Truxal, Atishay Jain, Rushil Chandrupatla, Tajana \v{S}imuni\'c Rosing, Larry Heck
Abstract: With the rapid growth in sensor data, effectively interpreting and interfacing with these data in a human-understandable way has become crucial. While existing research primarily focuses on learning classification models, fewer studies have explored how end users can actively extract useful insights from sensor data, often hindered by the lack of a proper dataset. To address this gap, we introduce \Dataset, the first human-created question-answering (QA) dataset for long-term time-series sensor data for daily life monitoring. \Dataset is created by human workers and includes 5.6K diverse and practical queries that reflect genuine human interests, paired with accurate answers derived from sensor data. We further establish benchmarks for state-of-the-art AI models on this dataset and evaluate their performance on typical edge devices. Our results reveal a gap between current models and optimal QA performance and efficiency, highlighting the need for new contributions. The dataset and code are available at: \url{https://github.com/benjamin-reichman/SensorQA}.
Authors: Ziwei He, Jian Yuan, Haoli Bai, Jingwen Leng, Bo Jiang
Abstract: Efficient key-value (KV) cache compression is critical for scaling transformer-based Large Language Models (LLMs) in long sequences and resource-limited settings. Existing methods evict tokens based on their positions or importance scores, but position-based strategies can miss crucial information outside predefined regions, while those relying on global importance scores resulting in strong regional biases, limiting the KV cache's overall context retention and potentially impairing the performance of LLMs on complex tasks. Our wavelet analysis reveals that as tokens approach the end of sequence, their contributions to generation gradually increase and tends to diverge more from neighboring tokens, indicating a smooth transition with increasing complexity and variability from distant to nearby context. Motivated by this observation, we propose TreeKV, an intuitive, training-free method that employs a tree structure for smooth cache compression. TreeKV maintains a fixed cache size, allowing LLMs to deliver high-quality output even in long text scenarios. Unlike most compression methods, TreeKV is applicable to both the generation and prefilling stages. It consistently surpasses all baseline models in language modeling tasks on PG19 and OpenWebText2, allowing LLMs trained with short context window to generalize to longer window with a 16x cache reduction. On the Longbench benchmark, TreeKV achieves the best performance with only 6\% of the budget at optimal efficiency.
Authors: Ethem Ya\u{g}{\i}z \c{C}al{\i}k, Talha R\"uzgar Akku\c{s}
Abstract: This paper explores the advancements in making large language models (LLMs) more human-like. We focus on techniques that enhance natural language understanding, conversational coherence, and emotional intelligence in AI systems. The study evaluates various approaches, including fine-tuning with diverse datasets, incorporating psychological principles, and designing models that better mimic human reasoning patterns. Our findings demonstrate that these enhancements not only improve user interactions but also open new possibilities for AI applications across different domains. Future work will address the ethical implications and potential biases introduced by these human-like attributes.
Authors: Chengxing Xie, Bowen Li, Chang Gao, He Du, Wai Lam, Difan Zou, Kai Chen
Abstract: Large Language Models (LLMs) have demonstrated remarkable proficiency across a variety of complex tasks. One significant application of LLMs is in tackling software engineering challenges, particularly in resolving real-world tasks on GitHub by fixing code based on the issues reported by the users. However, many current approaches rely on proprietary LLMs, which limits reproducibility, accessibility, and transparency. The critical components of LLMs for addressing software engineering issues and how their capabilities can be effectively enhanced remain unclear. To address these challenges, we introduce SWE-Fixer, a novel open-source LLM designed to effectively and efficiently resolve GitHub issues. SWE-Fixer comprises two essential modules: a code file retrieval module and a code editing module. The retrieval module employs BM25 along with a lightweight LLM model to achieve coarse-to-fine file retrieval. Subsequently, the code editing module utilizes the other LLM model to generate patches for the identified files. Then, to mitigate the lack of publicly available datasets, we compile an extensive dataset that includes 110K GitHub issues along with their corresponding patches, and train the two modules of SWE-Fixer separately. We assess our approach on the SWE-Bench Lite and Verified benchmarks, achieving state-of-the-art performance among open-source models with scores of 23.3% and 30.2%, respectively. These outcomes highlight the efficacy of our approach. We will make our model, dataset, and code publicly available at https://github.com/InternLM/SWE-Fixer.
Authors: Gregor Geigle, Florian Schneider, Carolin Holtermann, Chris Biemann, Radu Timofte, Anne Lauscher, Goran Glava\v{s}
Abstract: Most Large Vision-Language Models (LVLMs) to date are trained predominantly on English data, which makes them struggle to understand non-English input and fail to generate output in the desired target language. Existing efforts mitigate these issues by adding multilingual training data, but do so in a largely ad-hoc manner, lacking insight into how different training mixes tip the scale for different groups of languages. In this work, we present a comprehensive investigation into the training strategies for massively multilingual LVLMs. First, we conduct a series of multi-stage experiments spanning 13 downstream vision-language tasks and 43 languages, systematically examining: (1) the number of training languages that can be included without degrading English performance and (2) optimal language distributions of pre-training as well as (3) instruction-tuning data. Further, we (4) investigate how to improve multilingual text-in-image understanding, and introduce a new benchmark for the task. Surprisingly, our analysis reveals that one can (i) include as many as 100 training languages simultaneously (ii) with as little as 25-50\% of non-English data, to greatly improve multilingual performance while retaining strong English performance. We further find that (iii) including non-English OCR data in pre-training and instruction-tuning is paramount for improving multilingual text-in-image understanding. Finally, we put all our findings together and train Centurio, a 100-language LVLM, offering state-of-the-art performance in an evaluation covering 14 tasks and 56 languages.
Authors: Yufei Shang, Yanrong Guo, Shijie Hao, Richang Hong
Abstract: Document-Level Biomedical Relation Extraction (Bio-RE) aims to identify relations between biomedical entities within extensive texts, serving as a crucial subfield of biomedical text mining. Existing Bio-RE methods struggle with cross-sentence inference, which is essential for capturing relations spanning multiple sentences. Moreover, previous methods often overlook the incompleteness of documents and lack the integration of external knowledge, limiting contextual richness. Besides, the scarcity of annotated data further hampers model training. Recent advancements in large language models (LLMs) have inspired us to explore all the above issues for document-level Bio-RE. Specifically, we propose a document-level Bio-RE framework via LLM Adaptive Document-Relation Cross-Mapping (ADRCM) Fine-Tuning and Concept Unique Identifier (CUI) Retrieval-Augmented Generation (RAG). First, we introduce the Iteration-of-REsummary (IoRs) prompt for solving the data scarcity issue. In this way, Bio-RE task-specific synthetic data can be generated by guiding ChatGPT to focus on entity relations and iteratively refining synthetic data. Next, we propose ADRCM fine-tuning, a novel fine-tuning recipe that establishes mappings across different documents and relations, enhancing the model's contextual understanding and cross-sentence inference capabilities. Finally, during the inference, a biomedical-specific RAG approach, named CUI RAG, is designed to leverage CUIs as indexes for entities, narrowing the retrieval scope and enriching the relevant document contexts. Experiments conducted on three Bio-RE datasets (GDA, CDR, and BioRED) demonstrate the state-of-the-art performance of our proposed method by comparing it with other related works.
Authors: Dimitris Kouremenos, Klimis Ntalianis
Abstract: The Greek Language Multimodal Lip Reading with Integrated Sign Language Accessibility (GLaM-Sign) [1] is a groundbreaking resource in accessibility and multimodal AI, designed to support Deaf and Hard-of-Hearing (DHH) individuals. Developed from the FEELIT project [2], it integrates high-resolution audio, video, textual transcriptions, and Greek Sign Language translations for applications like real-time sign language translation and enhanced subtitle synchronization. While its primary focus is on promoting inclusivity in the Greek tourism sector, its adaptability extends to education, healthcare, and public services. Future advancements will enhance word-level precision and scalability to additional languages, supported by advanced AI methodologies and collaborations with diverse stakeholders. This dataset underscores the transformative potential of multimodal resources in bridging communication gaps, fostering innovation, and setting a benchmark for ethical AI and inclusive technologies.
Authors: L\'eane Jourdan, Nicolas Hernandez, Richard Dufour, Florian Boudin, Akiko Aizawa
Abstract: Revision is a crucial step in scientific writing, where authors refine their work to improve clarity, structure, and academic quality. Existing approaches to automated writing assistance often focus on sentence-level revisions, which fail to capture the broader context needed for effective modification. In this paper, we explore the impact of shifting from sentence-level to paragraph-level scope for the task of scientific text revision. The paragraph level definition of the task allows for more meaningful changes, and is guided by detailed revision instructions rather than general ones. To support this task, we introduce ParaRev, the first dataset of revised scientific paragraphs with an evaluation subset manually annotated with revision instructions. Our experiments demonstrate that using detailed instructions significantly improves the quality of automated revisions compared to general approaches, no matter the model or the metric considered.
Authors: Tomas Goldsack, Carolina Scarton, Chenghua Lin
Abstract: In this work, we explore the application of Large Language Models to zero-shot Lay Summarisation. We propose a novel two-stage framework for Lay Summarisation based on real-life processes, and find that summaries generated with this method are increasingly preferred by human judges for larger models. To help establish best practices for employing LLMs in zero-shot settings, we also assess the ability of LLMs as judges, finding that they are able to replicate the preferences of human judges. Finally, we take the initial steps towards Lay Summarisation for Natural Language Processing (NLP) articles, finding that LLMs are able to generalise to this new domain, and further highlighting the greater utility of summaries generated by our proposed approach via an in-depth human evaluation.
Authors: Artem Fedorchenko, Tanel Alum\"ae
Abstract: This paper presents an approach for generating high-quality, same-language subtitles for Estonian TV content. We fine-tune the Whisper model on human-generated Estonian subtitles and enhance it with iterative pseudo-labeling and large language model (LLM) based post-editing. Our experiments demonstrate notable subtitle quality improvement through pseudo-labeling with an unlabeled dataset. We find that applying LLM-based editing at test time enhances subtitle accuracy, while its use during training does not yield further gains. This approach holds promise for creating subtitle quality close to human standard and could be extended to real-time applications.
Authors: Atharva Mutsaddi, Aditya Choudhary
Abstract: Plagiarism involves using another person's work or concepts without proper attribution, presenting them as original creations. With the growing amount of data communicated in regional languages such as Marathi -- one of India's regional languages -- it is crucial to design robust plagiarism detection systems tailored for low-resource languages. Language models like Bidirectional Encoder Representations from Transformers (BERT) have demonstrated exceptional capability in text representation and feature extraction, making them essential tools for semantic analysis and plagiarism detection. However, the application of BERT for low-resource languages remains under-explored, particularly in the context of plagiarism detection. This paper presents a method to enhance the accuracy of plagiarism detection for Marathi texts using BERT sentence embeddings in conjunction with Term Frequency-Inverse Document Frequency (TF-IDF) feature representation. This approach effectively captures statistical, semantic, and syntactic aspects of text features through a weighted voting ensemble of machine learning models.
Authors: Hantao Lou, Jiaming Ji, Kaile Wang, Yaodong Yang
Abstract: The rapid advancement of large language models (LLMs) has led to significant improvements in their capabilities, but also to increased concerns about their alignment with human values and intentions. Current alignment strategies, including adaptive training and inference-time methods, have demonstrated potential in this area. However, these approaches still struggle to balance deployment complexity and capability across various tasks and difficulties. In this work, we introduce the Streaming Distribution Induce Aligner (Stream Aligner), a novel alignment paradigm that combines efficiency with enhanced performance in various tasks throughout the generation process. Stream Aligner achieves dynamic sentence-level correction by using a small model to learn the preferences of the suffix sentence, iteratively correcting the suffix sentence output by the upstream model, and then using the corrected sentence to replace the suffix sentence in subsequent generations. Compared to Aligner, our experiments demonstrate that Stream Aligner reduces reliance on the capabilities of additional models, enhances the reasoning abilities of LLMs, and decreases latency during user interaction. Specifically, Stream Aligner-2B model has achieved an improvement of 76.1% in helpfulness, 36.0% in harmlessness on the tested Llama2-70B-chat model, and Stream Aligner-8B has achieved an improvement of 3.5% on the math ability of the tested Llama3-70B-Instruct model.
Authors: Yongkang Du, Jen-tse Huang, Jieyu Zhao, Lu Lin
Abstract: Large language models (LLMs) have demonstrated significant capability in code generation, drawing increasing attention to the evaluation of the quality and safety of their outputs. However, research on bias in code generation remains limited. Existing studies typically assess bias by applying malicious prompts or reapply tasks and dataset for discriminative models. Given that LLMs are often aligned with human values and that prior datasets are not fully optimized for code-related tasks, there is a pressing need for benchmarks specifically designed for evaluating code models. In this study, we introduce FairCode, a novel benchmark for evaluating bias in code generation. FairCode comprises two tasks: function implementation and test case generation, each evaluating social bias through diverse scenarios. Additionally, we propose a new metric, FairScore, to assess model performance on this benchmark. We conduct experiments on widely used LLMs and provide a comprehensive analysis of the results. The findings reveal that all tested LLMs exhibit bias. The code is available at https://github.com/YongkDu/FairCode.
Authors: Xi Ye, Fangcong Yin, Yinghui He, Joie Zhang, Howard Yen, Tianyu Gao, Greg Durrett, Danqi Chen
Abstract: Existing benchmarks for evaluating long-context language models (LCLMs) primarily focus on long-context recall, requiring models to produce short responses based on a few critical snippets while processing thousands of irrelevant tokens. We introduce LongProc (Long Procedural Generation), a new benchmark that requires both the integration of highly dispersed information and long-form generation. LongProc consists of six diverse procedural generation tasks, such as extracting structured information from HTML pages into a TSV format and executing complex search procedures to create travel plans. These tasks challenge LCLMs by testing their ability to follow detailed procedural instructions, synthesize and reason over dispersed information, and generate structured, long-form outputs (up to 8K tokens). Furthermore, as these tasks adhere to deterministic procedures and yield structured outputs, they enable reliable rule-based evaluation. We evaluate 17 LCLMs on LongProc across three difficulty levels, with maximum numbers of output tokens set at 500, 2K, and 8K. Notably, while all tested models claim a context window size above 32K tokens, open-weight models typically falter on 2K-token tasks, and closed-source models like GPT-4o show significant degradation on 8K-token tasks. Further analysis reveals that LCLMs struggle to maintain long-range coherence in long-form generations. These findings highlight critical limitations in current LCLMs and suggest substantial room for improvement. Data and code available at: https://princeton-pli.github.io/LongProc
Authors: Jose A. Diaz-Garcia, Joao Paulo Carvalho
Abstract: The success of social media platforms has facilitated the emergence of various forms of online abuse within digital communities. This abuse manifests in multiple ways, including hate speech, cyberbullying, emotional abuse, grooming, and sexting. In this paper, we present a comprehensive analysis of the different forms of abuse prevalent in social media, with a particular focus on how emerging technologies, such as Language Models (LMs) and Large Language Models (LLMs), are reshaping both the detection and generation of abusive content within these networks. We delve into the mechanisms through which social media abuse is perpetuated, exploring the psychological and social impact. Additionally, we examine the dual role of advanced language models-highlighting their potential to enhance automated detection systems for abusive behavior while also acknowledging their capacity to generate harmful content. This paper aims to contribute to the ongoing discourse on online safety and ethics, offering insights into the evolving landscape of cyberabuse and the technological innovations that both mitigate and exacerbate it.
Authors: Yongkang Li, Panagiotis Eustratiadis, Evangelos Kanoulas
Abstract: HotFlip is a topical gradient-based word substitution method for attacking language models. Recently, this method has been further applied to attack retrieval systems by generating malicious passages that are injected into a corpus, i.e., corpus poisoning. However, HotFlip is known to be computationally inefficient, with the majority of time being spent on gradient accumulation for each query-passage pair during the adversarial token generation phase, making it impossible to generate an adequate number of adversarial passages in a reasonable amount of time. Moreover, the attack method itself assumes access to a set of user queries, a strong assumption that does not correspond to how real-world adversarial attacks are usually performed. In this paper, we first significantly boost the efficiency of HotFlip, reducing the adversarial generation process from 4 hours per document to only 15 minutes, using the same hardware. We further contribute experiments and analysis on two additional tasks: (1) transfer-based black-box attacks, and (2) query-agnostic attacks. Whenever possible, we provide comparisons between the original method and our improved version. Our experiments demonstrate that HotFlip can effectively attack a variety of dense retrievers, with an observed trend that its attack performance diminishes against more advanced and recent methods. Interestingly, we observe that while HotFlip performs poorly in a black-box setting, indicating limited capacity for generalization, in query-agnostic scenarios its performance is correlated to the volume of injected adversarial passages.
Authors: Jihwan Lee, Tiantian Feng, Aditya Kommineni, Sudarsana Reddy Kadiri, Shrikanth Narayanan
Abstract: Brain-computer interfaces (BCI) offer numerous human-centered application possibilities, particularly affecting people with neurological disorders. Text or speech decoding from brain activities is a relevant domain that could augment the quality of life for people with impaired speech perception. We propose a novel approach to enhance listened speech decoding from electroencephalography (EEG) signals by utilizing an auxiliary phoneme predictor that simultaneously decodes textual phoneme sequences. The proposed model architecture consists of three main parts: EEG module, speech module, and phoneme predictor. The EEG module learns to properly represent EEG signals into EEG embeddings. The speech module generates speech waveforms from the EEG embeddings. The phoneme predictor outputs the decoded phoneme sequences in text modality. Our proposed approach allows users to obtain decoded listened speech from EEG signals in both modalities (speech waveforms and textual phoneme sequences) simultaneously, eliminating the need for a concatenated sequential pipeline for each modality. The proposed approach also outperforms previous methods in both modalities. The source code and speech samples are publicly available.
Authors: Jun-Hak Yun, Seung-Bin Kim, Seong-Whan Lee
Abstract: Audio super-resolution is challenging owing to its ill-posed nature. Recently, the application of diffusion models in audio super-resolution has shown promising results in alleviating this challenge. However, diffusion-based models have limitations, primarily the necessity for numerous sampling steps, which causes significantly increased latency when synthesizing high-quality audio samples. In this paper, we propose FLowHigh, a novel approach that integrates flow matching, a highly efficient generative model, into audio super-resolution. We also explore probability paths specially tailored for audio super-resolution, which effectively capture high-resolution audio distributions, thereby enhancing reconstruction quality. The proposed method generates high-fidelity, high-resolution audio through a single-step sampling process across various input sampling rates. The experimental results on the VCTK benchmark dataset demonstrate that FLowHigh achieves state-of-the-art performance in audio super-resolution, as evaluated by log-spectral distance and ViSQOL while maintaining computational efficiency with only a single-step sampling process.
Authors: Shiji Zhao, Ranjie Duan, Fengxiang Wang, Chi Chen, Caixin Kang, Jialing Tao, YueFeng Chen, Hui Xue, Xingxing Wei
Abstract: Multimodal Large Language Models (MLLMs) have achieved impressive performance and have been put into practical use in commercial applications, but they still have potential safety mechanism vulnerabilities. Jailbreak attacks are red teaming methods that aim to bypass safety mechanisms and discover MLLMs' potential risks. Existing MLLMs' jailbreak methods often bypass the model's safety mechanism through complex optimization methods or carefully designed image and text prompts. Despite achieving some progress, they have a low attack success rate on commercial closed-source MLLMs. Unlike previous research, we empirically find that there exists a Shuffle Inconsistency between MLLMs' comprehension ability and safety ability for the shuffled harmful instruction. That is, from the perspective of comprehension ability, MLLMs can understand the shuffled harmful text-image instructions well. However, they can be easily bypassed by the shuffled harmful instructions from the perspective of safety ability, leading to harmful responses. Then we innovatively propose a text-image jailbreak attack named SI-Attack. Specifically, to fully utilize the Shuffle Inconsistency and overcome the shuffle randomness, we apply a query-based black-box optimization method to select the most harmful shuffled inputs based on the feedback of the toxic judge model. A series of experiments show that SI-Attack can improve the attack's performance on three benchmarks. In particular, SI-Attack can obviously improve the attack success rate for commercial MLLMs such as GPT-4o or Claude-3.5-Sonnet.
Authors: Zeyd Boukhers, Cong Yang
Abstract: The availability of metadata for scientific documents is pivotal in propelling scientific knowledge forward and for adhering to the FAIR principles (i.e. Findability, Accessibility, Interoperability, and Reusability) of research findings. However, the lack of sufficient metadata in published documents, particularly those from smaller and mid-sized publishers, hinders their accessibility. This issue is widespread in some disciplines, such as the German Social Sciences, where publications often employ diverse templates. To address this challenge, our study evaluates various feature learning and prediction methods, including natural language processing (NLP), computer vision (CV), and multimodal approaches, for extracting metadata from documents with high template variance. We aim to improve the accessibility of scientific documents and facilitate their wider use. To support our comparison of these methods, we provide comprehensive experimental results, analyzing their accuracy and efficiency in extracting metadata. Additionally, we provide valuable insights into the strengths and weaknesses of various feature learning and prediction methods, which can guide future research in this field.
Authors: Matteo Esposito
Abstract: Context. Developing secure and reliable software remains a key challenge in software engineering (SE). The ever-evolving technological landscape offers both opportunities and threats, creating a dynamic space where chaos and order compete. Secure software engineering (SSE) must continuously address vulnerabilities that endanger software systems and carry broader socio-economic risks, such as compromising critical national infrastructure and causing significant financial losses. Researchers and practitioners have explored methodologies like Static Application Security Testing Tools (SASTTs) and artificial intelligence (AI) approaches, including machine learning (ML) and large language models (LLMs), to detect and mitigate these vulnerabilities. Each method has unique strengths and limitations. Aim. This thesis seeks to bring order to the chaos in SSE by addressing domain-specific differences that impact AI accuracy. Methodology. The research employs a mix of empirical strategies, such as evaluating effort-aware metrics, analyzing SASTTs, conducting method-level analysis, and leveraging evidence-based techniques like systematic dataset reviews. These approaches help characterize vulnerability prediction datasets. Results. Key findings include limitations in static analysis tools for identifying vulnerabilities, gaps in SASTT coverage of vulnerability types, weak relationships among vulnerability severity scores, improved defect prediction accuracy using just-in-time modeling, and threats posed by untouched methods. Conclusions. This thesis highlights the complexity of SSE and the importance of contextual knowledge in improving AI-driven vulnerability and defect prediction. The comprehensive analysis advances effective prediction models, benefiting both researchers and practitioners.
Authors: Ziqing Li, Mutlu Cukurova, Sahan Bulathwela
Abstract: The development of Automatic Question Generation (QG) models has the potential to significantly improve educational practices by reducing the teacher workload associated with creating educational content. This paper introduces a novel approach to educational question generation that controls the topical focus of questions. The proposed Topic-Controlled Question Generation (T-CQG) method enhances the relevance and effectiveness of the generated content for educational purposes. Our approach uses fine-tuning on a pre-trained T5-small model, employing specially created datasets tailored to educational needs. The research further explores the impacts of pre-training strategies, quantisation, and data augmentation on the model's performance. We specifically address the challenge of generating semantically aligned questions with paragraph-level contexts, thereby improving the topic specificity of the generated questions. In addition, we introduce and explore novel evaluation methods to assess the topical relatedness of the generated questions. Our results, validated through rigorous offline and human-backed evaluations, demonstrate that the proposed models effectively generate high-quality, topic-focused questions. These models have the potential to reduce teacher workload and support personalised tutoring systems by serving as bespoke question generators. With its relatively small number of parameters, the proposals not only advance the capabilities of question generation models for handling specific educational topics but also offer a scalable solution that reduces infrastructure costs. This scalability makes them feasible for widespread use in education without reliance on proprietary large language models like ChatGPT.
Authors: Yewei Song, Cedric Lothritz, Xunzhu Tang, Saad Ezzini, Jacques Klein, Tegawend\'e F. Bissyand\'e, Andrey Boytsov, Ulrick Ble, Anne Goujon
Abstract: Interacting with a software system via a chatbot can be challenging, especially when the chatbot needs to generate API calls, in the right order and with the right parameters, to communicate with the system. API calling in chatbot systems poses significant challenges, particularly in complex, multi-step tasks requiring accurate API selection and execution. We contribute to this domain in three ways: first, by introducing a novel dataset designed to assess models on API function selection, parameter generation, and nested API calls; second, by benchmarking state-of-the-art language models across varying levels of complexity to evaluate their performance in API function generation and parameter accuracy; and third, by proposing an enhanced API routing method that combines general-purpose large language models for API selection with fine-tuned models for parameter generation and some prompt engineering approach. These approaches lead to substantial improvements in handling complex API tasks, offering practical advancements for real-world API-driven chatbot systems.
Authors: Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, Peitian Zhang, Zhicheng Dou
Abstract: Large reasoning models (LRMs) like OpenAI-o1 have demonstrated impressive long stepwise reasoning capabilities through large-scale reinforcement learning. However, their extended reasoning processes often suffer from knowledge insufficiency, leading to frequent uncertainties and potential errors. To address this limitation, we introduce \textbf{Search-o1}, a framework that enhances LRMs with an agentic retrieval-augmented generation (RAG) mechanism and a Reason-in-Documents module for refining retrieved documents. Search-o1 integrates an agentic search workflow into the reasoning process, enabling dynamic retrieval of external knowledge when LRMs encounter uncertain knowledge points. Additionally, due to the verbose nature of retrieved documents, we design a separate Reason-in-Documents module to deeply analyze the retrieved information before injecting it into the reasoning chain, minimizing noise and preserving coherent reasoning flow. Extensive experiments on complex reasoning tasks in science, mathematics, and coding, as well as six open-domain QA benchmarks, demonstrate the strong performance of Search-o1. This approach enhances the trustworthiness and applicability of LRMs in complex reasoning tasks, paving the way for more reliable and versatile intelligent systems. The code is available at \url{https://github.com/sunnynexus/Search-o1}.
Authors: Xingyu Fu, Minqian Liu, Zhengyuan Yang, John Corring, Yijuan Lu, Jianwei Yang, Dan Roth, Dinei Florencio, Cha Zhang
Abstract: Structured image understanding, such as interpreting tables and charts, requires strategically refocusing across various structures and texts within an image, forming a reasoning sequence to arrive at the final answer. However, current multimodal large language models (LLMs) lack this multihop selective attention capability. In this work, we introduce ReFocus, a simple yet effective framework that equips multimodal LLMs with the ability to generate "visual thoughts" by performing visual editing on the input image through code, shifting and refining their visual focuses. Specifically, ReFocus enables multimodal LLMs to generate Python codes to call tools and modify the input image, sequentially drawing boxes, highlighting sections, and masking out areas, thereby enhancing the visual reasoning process. We experiment upon a wide range of structured image understanding tasks involving tables and charts. ReFocus largely improves performance on all tasks over GPT-4o without visual editing, yielding an average gain of 11.0% on table tasks and 6.8% on chart tasks. We present an in-depth analysis of the effects of different visual edits, and reasons why ReFocus can improve the performance without introducing additional information. Further, we collect a 14k training set using ReFocus, and prove that such visual chain-of-thought with intermediate information offers a better supervision than standard VQA data, reaching a 8.0% average gain over the same model trained with QA pairs and 2.6% over CoT.
Authors: Tongshuang Wu, Haiyi Zhu, Maya Albayrak, Alexis Axon, Amanda Bertsch, Wenxing Deng, Ziqi Ding, Bill Guo, Sireesh Gururaja, Tzu-Sheng Kuo, Jenny T. Liang, Ryan Liu, Ihita Mandal, Jeremiah Milbauer, Xiaolin Ni, Namrata Padmanabhan, Subhashini Ramkumar, Alexis Sudjianto, Jordan Taylor, Ying-Jui Tseng, Patricia Vaidos, Zhijin Wu, Wei Wu, Chenyang Yang
Abstract: LLMs have shown promise in replicating human-like behavior in crowdsourcing tasks that were previously thought to be exclusive to human abilities. However, current efforts focus mainly on simple atomic tasks. We explore whether LLMs can replicate more complex crowdsourcing pipelines. We find that modern LLMs can simulate some of crowdworkers' abilities in these ``human computation algorithms,'' but the level of success is variable and influenced by requesters' understanding of LLM capabilities, the specific skills required for sub-tasks, and the optimal interaction modality for performing these sub-tasks. We reflect on human and LLMs' different sensitivities to instructions, stress the importance of enabling human-facing safeguards for LLMs, and discuss the potential of training humans and LLMs with complementary skill sets. Crucially, we show that replicating crowdsourcing pipelines offers a valuable platform to investigate 1) the relative LLM strengths on different tasks (by cross-comparing their performances on sub-tasks) and 2) LLMs' potential in complex tasks, where they can complete part of the tasks while leaving others to humans.
Authors: Ali Modarressi, Abdullatif K\"oksal, Ayyoob Imani, Mohsen Fayyaz, Hinrich Sch\"utze
Abstract: While current large language models (LLMs) perform well on many knowledge-related tasks, they are limited by relying on their parameters as an implicit storage mechanism. As a result, they struggle with memorizing rare events and with updating their memory as facts change over time. In addition, the uninterpretable nature of parametric memory makes it challenging to prevent hallucination. Model editing and augmenting LLMs with parameters specialized for memory are only partial solutions. In this paper, we introduce MemLLM, a novel method of enhancing LLMs by integrating a structured and explicit read-and-write memory module. MemLLM tackles the aforementioned challenges by enabling dynamic interaction with the memory and improving the LLM's capabilities in using stored knowledge. Our experiments indicate that MemLLM enhances the LLM's performance and interpretability, in language modeling in general and knowledge-intensive tasks in particular. We see MemLLM as an important step towards making LLMs more grounded and factual through memory augmentation.
Authors: Junrui Ni, Liming Wang, Yang Zhang, Kaizhi Qian, Heting Gao, Mark Hasegawa-Johnson, Chang D. Yoo
Abstract: Recent advancements in supervised automatic speech recognition (ASR) have achieved remarkable performance, largely due to the growing availability of large transcribed speech corpora. However, most languages lack sufficient paired speech and text data to effectively train these systems. In this article, we tackle the challenge of developing ASR systems without paired speech and text corpora by proposing the removal of reliance on a phoneme lexicon. We explore a new research direction: word-level unsupervised ASR, and experimentally demonstrate that an unsupervised speech recognizer can emerge from joint speech-to-speech and text-to-text masked token-infilling. Using a curated speech corpus containing a fixed number of English words, our system iteratively refines the word segmentation structure and achieves a word error rate of between 20-23%, depending on the vocabulary size, without parallel transcripts, oracle word boundaries, or a pronunciation lexicon. This innovative model surpasses the performance of previous unsupervised ASR models under the lexicon-free setting.
Authors: Andrew Halterman, Katherine A. Keith
Abstract: Codebooks -- documents that operationalize concepts and outline annotation procedures -- are used almost universally by social scientists when coding political texts. To code these texts automatically, researchers are increasing turning to generative large language models (LLMs). However, there is limited empirical evidence on whether "off-the-shelf" LLMs faithfully follow real-world codebook operationalizations and measure complex political constructs with sufficient accuracy. To address this, we gather and curate three real-world political science codebooks -- covering protest events, political violence and manifestos -- along with their unstructured texts and human labels. We also propose a five-stage framework for codebook-LLM measurement: preparing a codebook for both humans and LLMs, testing LLMs' basic capabilities on a codebook, evaluating zero-shot measurement accuracy (i.e. off-the-shelf performance), analyzing errors, and further (parameter-efficient) supervised training of LLMs. We provide an empirical demonstration of this framework using our three codebook datasets and several pretrained 7-12 billion open-weight LLMs. We find current open-weight LLMs have limitations in following codebooks zero-shot, but that supervised instruction tuning can substantially improve performance. Rather than suggesting the "best" LLM, our contribution lies in our codebook datasets, evaluation framework, and guidance for applied researchers who wish to implement their own codebook-LLM measurement projects.
Authors: Jiale Wang, Junhui Yu, Huanyong Liu, Chenanran Kong
Abstract: Hierarchical and complex Mathematical Expression Recognition (MER) is challenging due to multiple possible interpretations of a formula, complicating both parsing and evaluation. In this paper, we introduce the Hierarchical Detail-Focused Recognition dataset (HDR), the first dataset specifically designed to address these issues. It consists of a large-scale training set, HDR-100M, offering an unprecedented scale and diversity with one hundred million training instances. And the test set, HDR-Test, includes multiple interpretations of complex hierarchical formulas for comprehensive model performance evaluation. Additionally, the parsing of complex formulas often suffers from errors in fine-grained details. To address this, we propose the Hierarchical Detail-Focused Recognition Network (HDNet), an innovative framework that incorporates a hierarchical sub-formula module, focusing on the precise handling of formula details, thereby significantly enhancing MER performance. Experimental results demonstrate that HDNet outperforms existing MER models across various datasets.
Authors: Khai Le-Duc, Phuc Phan, Tan-Hanh Pham, Bach Phan Tat, Minh-Huong Ngo, Truong-Son Hy
Abstract: Multilingual automatic speech recognition (ASR) in the medical domain serves as a foundational task for various downstream applications such as speech translation, spoken language understanding, and voice-activated assistants. This technology enhances patient care by enabling efficient communication across language barriers, alleviating specialized workforce shortages, and facilitating improved diagnosis and treatment, particularly during pandemics. In this work, we introduce MultiMed, the first multilingual medical ASR dataset, along with the first collection of small-to-large end-to-end medical ASR models, spanning five languages: Vietnamese, English, German, French, and Mandarin Chinese. To our best knowledge, MultiMed stands as the world's largest medical ASR dataset across all major benchmarks: total duration, number of recording conditions, number of accents, and number of speaking roles. Furthermore, we present the first multilinguality study for medical ASR, which includes reproducible empirical baselines, a monolinguality-multilinguality analysis, Attention Encoder Decoder (AED) vs Hybrid comparative study, a layer-wise ablation study for the AED, and a linguistic analysis for multilingual medical ASR. All code, data, and models are available online: https://github.com/leduckhai/MultiMed/tree/master/MultiMed
URLs: https://github.com/leduckhai/MultiMed/tree/master/MultiMed
Authors: German Gritsai, Anastasia Voznyuk, Andrey Grabovoy, Yury Chekhovich
Abstract: The rapid development of autoregressive Large Language Models (LLMs) has significantly improved the quality of generated texts, necessitating reliable machine-generated text detectors. A huge number of detectors and collections with AI fragments have emerged, and several detection methods even showed recognition quality up to 99.9% according to the target metrics in such collections. However, the quality of such detectors tends to drop dramatically in the wild, posing a question: Are detectors actually highly trustworthy or do their high benchmark scores come from the poor quality of evaluation datasets? In this paper, we emphasise the need for robust and qualitative methods for evaluating generated data to be secure against bias and low generalising ability of future model. We present a systematic review of datasets from competitions dedicated to AI-generated content detection and propose methods for evaluating the quality of datasets containing AI-generated fragments. In addition, we discuss the possibility of using high-quality generated data to achieve two goals: improving the training of detection models and improving the training datasets themselves. Our contribution aims to facilitate a better understanding of the dynamics between human and machine text, which will ultimately support the integrity of information in an increasingly automated world.
Authors: Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, Yuanzhuo Wang, Jian Guo
Abstract: Accurate and consistent evaluation is crucial for decision-making across numerous fields, yet it remains a challenging task due to inherent subjectivity, variability, and scale. Large Language Models (LLMs) have achieved remarkable success across diverse domains, leading to the emergence of "LLM-as-a-Judge," where LLMs are employed as evaluators for complex tasks. With their ability to process diverse data types and provide scalable, cost-effective, and consistent assessments, LLMs present a compelling alternative to traditional expert-driven evaluations. However, ensuring the reliability of LLM-as-a-Judge systems remains a significant challenge that requires careful design and standardization. This paper provides a comprehensive survey of LLM-as-a-Judge, addressing the core question: How can reliable LLM-as-a-Judge systems be built? We explore strategies to enhance reliability, including improving consistency, mitigating biases, and adapting to diverse assessment scenarios. Additionally, we propose methodologies for evaluating the reliability of LLM-as-a-Judge systems, supported by a novel benchmark designed for this purpose. To advance the development and real-world deployment of LLM-as-a-Judge systems, we also discussed practical applications, challenges, and future directions. This survey serves as a foundational reference for researchers and practitioners in this rapidly evolving field.
Authors: Ben Liu, Jihai Zhang, Fangquan Lin, Cheng Yang, Min Peng
Abstract: Large Language Models (LLMs) present massive inherent knowledge and superior semantic comprehension capability, which have revolutionized various tasks in natural language processing. Despite their success, a critical gap remains in enabling LLMs to perform knowledge graph completion (KGC). Empirical evidence suggests that LLMs consistently perform worse than conventional KGC approaches, even through sophisticated prompt design or tailored instruction-tuning. Fundamentally, applying LLMs on KGC introduces several critical challenges, including a vast set of entity candidates, hallucination issue of LLMs, and under-exploitation of the graph structure. To address these challenges, we propose a novel instruction-tuning-based method, namely FtG. Specifically, we present a \textit{filter-then-generate} paradigm and formulate the KGC task into a multiple-choice question format. In this way, we can harness the capability of LLMs while mitigating the issue casused by hallucinations. Moreover, we devise a flexible ego-graph serialization prompt and employ a structure-text adapter to couple structure and text information in a contextualized manner. Experimental results demonstrate that FtG achieves substantial performance gain compared to existing state-of-the-art methods. The instruction dataset and code are available at \url{https://github.com/LB0828/FtG}.
Authors: Jaione Bengoetxea, Mikel Zubillaga, Ekhi Azurmendi, Maite Heredia, Julen Etxaniz, Markel Ferro, Jeremy Barnes
Abstract: In this paper we present our submission for the NorSID Shared Task as part of the 2025 VarDial Workshop (Scherrer et al., 2025), consisting of three tasks: Intent Detection, Slot Filling and Dialect Identification, evaluated using data in different dialects of the Norwegian language. For Intent Detection and Slot Filling, we have fine-tuned a multitask model in a cross-lingual setting, to leverage the xSID dataset available in 17 languages. In the case of Dialect Identification, our final submission consists of a model fine-tuned on the provided development set, which has obtained the highest scores within our experiments. Our final results on the test set show that our models do not drop in performance compared to the development set, likely due to the domain-specificity of the dataset and the similar distribution of both subsets. Finally, we also report an in-depth analysis of the provided datasets and their artifacts, as well as other sets of experiments that have been carried out but did not yield the best results. Additionally, we present an analysis on the reasons why some methods have been more successful than others; mainly the impact of the combination of languages and domain-specificity of the training data on the results.
Authors: Melody Y. Guan, Manas Joglekar, Eric Wallace, Saachi Jain, Boaz Barak, Alec Helyar, Rachel Dias, Andrea Vallone, Hongyu Ren, Jason Wei, Hyung Won Chung, Sam Toyer, Johannes Heidecke, Alex Beutel, Amelia Glaese
Abstract: As large-scale language models increasingly impact safety-critical domains, ensuring their reliable adherence to well-defined principles remains a fundamental challenge. We introduce Deliberative Alignment, a new paradigm that directly teaches the model safety specifications and trains it to explicitly recall and accurately reason over the specifications before answering. We used this approach to align OpenAI's o-series models, and achieved highly precise adherence to OpenAI's safety policies, without requiring human-written chain-of-thoughts or answers. Deliberative Alignment pushes the Pareto frontier by simultaneously increasing robustness to jailbreaks while decreasing overrefusal rates, and also improves out-of-distribution generalization. We demonstrate that reasoning over explicitly specified policies enables more scalable, trustworthy, and interpretable alignment.
Authors: Melkamu Mersha, Mingiziem Bitewa, Tsion Abay, Jugal Kalita
Abstract: Neural networks are widely regarded as black-box models, creating significant challenges in understanding their inner workings, especially in natural language processing (NLP) applications. To address this opacity, model explanation techniques like Local Interpretable Model-Agnostic Explanations (LIME) have emerged as essential tools for providing insights into the behavior of these complex systems. This study leverages LIME to interpret a multi-layer perceptron (MLP) neural network trained on a text classification task. By analyzing the contribution of individual features to model predictions, the LIME approach enhances interpretability and supports informed decision-making. Despite its effectiveness in offering localized explanations, LIME has limitations in capturing global patterns and feature interactions. This research highlights the strengths and shortcomings of LIME and proposes directions for future work to achieve more comprehensive interpretability in neural NLP models.
Authors: Rachneet Sachdeva, Rima Hazra, Iryna Gurevych
Abstract: Large language models, despite extensive alignment with human values and ethical principles, remain vulnerable to sophisticated jailbreak attacks that exploit their reasoning abilities. Existing safety measures often detect overt malicious intent but fail to address subtle, reasoning-driven vulnerabilities. In this work, we introduce POATE (Polar Opposite query generation, Adversarial Template construction, and Elaboration), a novel jailbreak technique that harnesses contrastive reasoning to provoke unethical responses. POATE crafts semantically opposing intents and integrates them with adversarial templates, steering models toward harmful outputs with remarkable subtlety. We conduct extensive evaluation across six diverse language model families of varying parameter sizes to demonstrate the robustness of the attack, achieving significantly higher attack success rates (~44%) compared to existing methods. To counter this, we propose Intent-Aware CoT and Reverse Thinking CoT, which decompose queries to detect malicious intent and reason in reverse to evaluate and reject harmful responses. These methods enhance reasoning robustness and strengthen the model's defense against adversarial exploits.
Authors: Zhaoyi Yan, Zhijie Sang, Yiming Zhang, Yuhao Fu, Baoyi He, Qi Zhou, Yining Di, Chunlin Ji, Shengyu Zhang, Fei Wu, Hongxia Yang
Abstract: Large Language Models (LLMs) have demonstrated strong performance across various reasoning tasks, yet building a single model that consistently excels across all domains remains challenging. This paper addresses this problem by exploring strategies to integrate multiple domain-specialized models into an efficient pivot model.We propose two fusion strategies to combine the strengths of multiple LLMs: (1) a pairwise, multi-step fusion approach that sequentially distills each source model into the pivot model, followed by a weight merging step to integrate the distilled models into the final model. This method achieves strong performance but requires substantial training effort; and (2) a unified fusion approach that aggregates all source models' outputs simultaneously.To improve the fusion process, we introduce a novel Rate-Skewness Adaptive Fusion (RSAF) technique, which dynamically adjusts top-K ratios during parameter merging for enhanced flexibility and stability.Furthermore, we propose an uncertainty-based weighting method for the unified approach, which dynamically balances the contributions of source models and outperforms other logits/distribution ensemble methods.We achieved accuracy improvements of 9.27%, 8.80%, and 8.89% on the GSM8K, MATH, and HumanEval tasks, respectively.
Authors: Run Luo, Ting-En Lin, Haonan Zhang, Yuchuan Wu, Xiong Liu, Min Yang, Yongbin Li, Longze Chen, Jiaming Li, Lei Zhang, Yangyi Chen, Hamid Alinejad-Rokny, Fei Huang
Abstract: Recent advancements in omnimodal learning have been achieved in understanding and generation across images, text, and speech, though mainly within proprietary models. Limited omnimodal datasets and the inherent challenges associated with real-time emotional speech generation have hindered open-source progress. To address these issues, we propose openomni, a two-stage training method combining omnimodal alignment and speech generation to develop a state-of-the-art omnimodal large language model. In the alignment phase, a pre-trained speech model is further trained on text-image tasks to generalize from vision to speech in a (near) zero-shot manner, outperforming models trained on tri-modal datasets. In the speech generation phase, a lightweight decoder facilitates real-time emotional speech through training on speech tasks and preference learning. Experiments demonstrate that openomni consistently improves across omnimodal, vision-language, and speech-language evaluations, enabling natural, emotion-rich dialogues and real-time emotional speech generation.
Authors: Sheng Zhang, Yanbo Xu, Naoto Usuyama, Hanwen Xu, Jaspreet Bagga, Robert Tinn, Sam Preston, Rajesh Rao, Mu Wei, Naveen Valluri, Cliff Wong, Andrea Tupini, Yu Wang, Matt Mazzola, Swadheen Shukla, Lars Liden, Jianfeng Gao, Angela Crabtree, Brian Piening, Carlo Bifulco, Matthew P. Lungren, Tristan Naumann, Sheng Wang, Hoifung Poon
Abstract: Biomedical data is inherently multimodal, comprising physical measurements and natural language narratives. A generalist biomedical AI model needs to simultaneously process different modalities of data, including text and images. Therefore, training an effective generalist biomedical model requires high-quality multimodal data, such as parallel image-text pairs. Here, we present PMC-15M, a novel dataset that is two orders of magnitude larger than existing biomedical multimodal datasets such as MIMIC-CXR, and spans a diverse range of biomedical image types. PMC-15M contains 15 million biomedical image-text pairs collected from 4.4 million scientific articles. Based on PMC-15M, we have pretrained BiomedCLIP, a multimodal foundation model, with domain-specific adaptations tailored to biomedical vision-language processing. We conducted extensive experiments and ablation studies on standard biomedical imaging tasks from retrieval to classification to visual question-answering (VQA). BiomedCLIP achieved new state-of-the-art results in a wide range of standard datasets, substantially outperforming prior approaches. Intriguingly, by large-scale pretraining on diverse biomedical image types, BiomedCLIP even outperforms state-of-the-art radiology-specific models such as BioViL in radiology-specific tasks such as RSNA pneumonia detection. In summary, BiomedCLIP is a fully open-access foundation model that achieves state-of-the-art performance on various biomedical tasks, paving the way for transformative multimodal biomedical discovery and applications. We release our models at https://aka.ms/biomedclip to facilitate future research in multimodal biomedical AI.
Authors: Jiaming Han, Kaixiong Gong, Yiyuan Zhang, Jiaqi Wang, Kaipeng Zhang, Dahua Lin, Yu Qiao, Peng Gao, Xiangyu Yue
Abstract: Multimodal large language models (MLLMs) have gained significant attention due to their strong multimodal understanding capability. However, existing works rely heavily on modality-specific encoders, which usually differ in architecture and are limited to common modalities. In this paper, we present OneLLM, an MLLM that aligns eight modalities to language using a unified framework. We achieve this through a unified multimodal encoder and a progressive multimodal alignment pipeline. In detail, we first train an image projection module to connect a vision encoder with LLM. Then, we build a universal projection module (UPM) by mixing multiple image projection modules and dynamic routing. Finally, we progressively align more modalities to LLM with the UPM. To fully leverage the potential of OneLLM in following instructions, we also curated a comprehensive multimodal instruction dataset, including 2M items from image, audio, video, point cloud, depth/normal map, IMU and fMRI brain activity. OneLLM is evaluated on 25 diverse benchmarks, encompassing tasks such as multimodal captioning, question answering and reasoning, where it delivers excellent performance. Code, data, model and online demo are available at https://github.com/csuhan/OneLLM
Authors: Yihong Tang, Zhaokai Wang, Ao Qu, Yihao Yan, Zhaofeng Wu, Dingyi Zhuang, Jushi Kai, Kebing Hou, Xiaotong Guo, Han Zheng, Tiange Luo, Jinhua Zhao, Zhan Zhao, Wei Ma
Abstract: Citywalk, a recently popular form of urban travel, requires genuine personalization and understanding of fine-grained requests compared to traditional itinerary planning. In this paper, we introduce the novel task of Open-domain Urban Itinerary Planning (OUIP), which generates personalized urban itineraries from user requests in natural language. We then present ITINERA, an OUIP system that integrates spatial optimization with large language models to provide customized urban itineraries based on user needs. This involves decomposing user requests, selecting candidate points of interest (POIs), ordering the POIs based on cluster-aware spatial optimization, and generating the itinerary. Experiments on real-world datasets and the performance of the deployed system demonstrate our system's capacity to deliver personalized and spatially coherent itineraries compared to current solutions. Source codes of ITINERA are available at https://github.com/YihongT/ITINERA.
Authors: Tobias Leemann, Alina Fastowski, Felix Pfeiffer, Gjergji Kasneci
Abstract: We address the critical challenge of applying feature attribution methods to the transformer architecture, which dominates current applications in natural language processing and beyond. Traditional attribution methods to explainable AI (XAI) explicitly or implicitly rely on linear or additive surrogate models to quantify the impact of input features on a model's output. In this work, we formally prove an alarming incompatibility: transformers are structurally incapable of representing linear or additive surrogate models used for feature attribution, undermining the grounding of these conventional explanation methodologies. To address this discrepancy, we introduce the Softmax-Linked Additive Log Odds Model (SLALOM), a novel surrogate model specifically designed to align with the transformer framework. SLALOM demonstrates the capacity to deliver a range of insightful explanations with both synthetic and real-world datasets. We highlight SLALOM's unique efficiency-quality curve by showing that SLALOM can produce explanations with substantially higher fidelity than competing surrogate models or provide explanations of comparable quality at a fraction of their computational costs. We release code for SLALOM as an open-source project online at https://github.com/tleemann/slalom_explanations.
Authors: Ranran Haoran Zhang, Bensu U\c{c}ar, Soumik Dey, Hansi Wu, Binbin Li, Rui Zhang
Abstract: Open-vocabulary Extreme Multi-label Classification (OXMC) extends traditional XMC by allowing prediction beyond an extremely large, predefined label set (typically $10^3$ to $10^{12}$ labels), addressing the dynamic nature of real-world labeling tasks. However, self-selection bias in data annotation leads to significant missing labels in both training and test data, particularly for less popular inputs. This creates two critical challenges: generation models learn to be "lazy'" by under-generating labels, and evaluation becomes unreliable due to insufficient annotation in the test set. In this work, we introduce Positive-Unlabeled Sequence Learning (PUSL), which reframes OXMC as an infinite keyphrase generation task, addressing the generation model's laziness. Additionally, we propose to adopt a suite of evaluation metrics, F1@$\mathcal{O}$ and newly proposed B@$k$, to reliably assess OXMC models with incomplete ground truths. In a highly imbalanced e-commerce dataset with substantial missing labels, PUSL generates 30% more unique labels, and 72% of its predictions align with actual user queries. On the less skewed EURLex-4.3k dataset, PUSL demonstrates superior F1 scores, especially as label counts increase from 15 to 30. Our approach effectively tackles both the modeling and evaluation challenges in OXMC with missing labels.
Authors: Jinzuomu Zhong, Korin Richmond, Zhiba Su, Siqi Sun
Abstract: While recent Zero-Shot Text-to-Speech (ZS-TTS) models have achieved high naturalness and speaker similarity, they fall short in accent fidelity and control. To address this issue, we propose zero-shot accent generation that unifies Foreign Accent Conversion (FAC), accented TTS, and ZS-TTS, with a novel two-stage pipeline. In the first stage, we achieve state-of-the-art (SOTA) on Accent Identification (AID) with 0.56 f1 score on unseen speakers. In the second stage, we condition a ZS-TTS system on the pretrained speaker-agnostic accent embeddings extracted by the AID model. The proposed system achieves higher accent fidelity on inherent/cross accent generation, and enables unseen accent generation.
Authors: Georg Ahnert, Max Pellert, David Garcia, Markus Strohmaier
Abstract: This paper proposes temporally aligned Large Language Models (LLMs) as a tool for longitudinal analysis of social media data. We fine-tune Temporal Adapters for Llama 3 8B on full timelines from a panel of British Twitter users, and extract longitudinal aggregates of emotions and attitudes with established questionnaires. We focus our analysis on the beginning of the COVID-19 pandemic that had a strong impact on public opinion and collective emotions. We validate our estimates against representative British survey data and find strong positive, significant correlations for several collective emotions. The obtained estimates are robust across multiple training seeds and prompt formulations, and in line with collective emotions extracted using a traditional classification model trained on labeled data. We demonstrate the flexibility of our method on questions of public opinion for which no pre-trained classifier is available. Our work extends the analysis of affect in LLMs to a longitudinal setting through Temporal Adapters. It enables flexible, new approaches towards the longitudinal analysis of social media data.
Authors: Yilun Liu, Yuhe Ji, Shimin Tao, Minggui He, Weibin Meng, Shenglin Zhang, Yongqian Sun, Yuming Xie, Boxing Chen, Hao Yang
Abstract: Automatic log analysis is essential for the efficient Operation and Maintenance (O&M) of software systems, providing critical insights into system behaviors. However, existing approaches mostly treat log analysis as training a model to perform an isolated task ( e.g., anomaly detection, log parsing, etc.) using task-specific log-label pairs. These task-based approaches are inflexible in generalizing to complex scenarios, depend on task-specific training data, and cost significantly when deploying multiple models. In this paper, we propose an instruction-based training approach that transforms log-label pairs from multiple tasks and domains into a unified format of instruction-response pairs. Our trained model, LogLM, can follow complex user instructions and generalize better across different tasks, thereby increasing flexibility and reducing the dependence on task-specific training data. By integrating major log analysis tasks into a single model, our approach also relieves model deployment burden. Experimentally, LogLM outperforms existing approaches across five log analysis capabilities, and exhibits strong generalization abilities on complex instructions and unseen tasks.
Authors: Haokun Liu, Yangqiaoyu Zhou, Mingxuan Li, Chenfei Yuan, Chenhao Tan
Abstract: AI holds promise for transforming scientific processes, including hypothesis generation. Prior work on hypothesis generation can be broadly categorized into theory-driven and data-driven approaches. While both have proven effective in generating novel and plausible hypotheses, it remains an open question whether they can complement each other. To address this, we develop the first method that combines literature-based insights with data to perform LLM-powered hypothesis generation. We apply our method on five different datasets and demonstrate that integrating literature and data outperforms other baselines (8.97\% over few-shot, 15.75\% over literature-based alone, and 3.37\% over data-driven alone). Additionally, we conduct the first human evaluation to assess the utility of LLM-generated hypotheses in assisting human decision-making on two challenging tasks: deception detection and AI generated content detection. Our results show that human accuracy improves significantly by 7.44\% and 14.19\% on these tasks, respectively. These findings suggest that integrating literature-based and data-driven approaches provides a comprehensive and nuanced framework for hypothesis generation and could open new avenues for scientific inquiry.
Authors: Yuanming Zhang, Jing Lu, Fei Chen, Haoliang Du, Xia Gao, Zhibin Lin
Abstract: Decoding the directional focus of an attended speaker from listeners' electroencephalogram (EEG) signals is essential for developing brain-computer interfaces to improve the quality of life for individuals with hearing impairment. Previous works have concentrated on binary directional focus decoding, i.e., determining whether the attended speaker is on the left or right side of the listener. However, a more precise decoding of the exact direction of the attended speaker is necessary for effective speech processing. Additionally, audio spatial information has not been effectively leveraged, resulting in suboptimal decoding results. In this paper, it is found that on the recently presented dataset with 14-class directional focus, models relying exclusively on EEG inputs exhibit significantly lower accuracy when decoding the directional focus in both leave-one-subject-out and leave-one-trial-out scenarios. By integrating audio spatial spectra with EEG features, the decoding accuracy can be effectively improved. The CNN, LSM-CNN, and Deformer models are employed to decode the directional focus from listeners' EEG signals and audio spatial spectra. The proposed Sp-EEG-Deformer model achieves notable 14-class decoding accuracies of 55.35% and 57.19% in leave-one-subject-out and leave-one-trial-out scenarios with a decision window of 1 second, respectively. Experiment results indicate increased decoding accuracy as the number of alternative directions reduces. These findings suggest the efficacy of our proposed dual modal directional focus decoding strategy.
Authors: Elia Cunegatti, Leonardo Lucio Custode, Giovanni Iacca
Abstract: Network pruning focuses on computational techniques that aim to reduce a given model's computational cost by removing a subset of its parameters while having minimal impact on performance. Throughout the last decade, the most widely used pruning paradigm has been pruning and re-training, which nowadays is inconvenient due to the vast amount of pre-trained models, which are in any case too expensive to re-train. In this paper, we exploit functional information from dense pre-trained models, i.e., their activations, to obtain sparse models that maximize the activations' alignment w.r.t. their corresponding dense models. Hence, we propose \textsc{NeuroAL}, a \emph{top-up} algorithm that can be used on top of any given pruning algorithm for LLMs, which modifies the block-wise and row-wise sparsity exploiting information from both the dense model and its sparse version to maximize the \emph{neuron alignment} among activations. Differently from existing methods, our approach adaptively selects the best hyperparameters for the block-wise and row-wise sparsity ratios w.r.t. the model and the desired sparsity, and requires \emph{no re-training}. We test our method over 276 cases combining four LLM families, three sparsity ratios, and ten language tasks (three language modeling and seven zero-shot datasets), showing how it consistently outperforms the latest state-of-the-art methods in terms of performance-runtime trade-off. The code is available at \href{https://github.com/eliacunegatti/NeuroAL}{https://github.com/eliacunegatti/NeuroAL}.
URLs: https://github.com/eliacunegatti/NeuroAL, https://github.com/eliacunegatti/NeuroAL
Authors: Feng Xiong, Runxi Cheng, Wang Chen, Zhanqiu Zhang, Yiwen Guo, Chun Yuan, Ruifeng Xu
Abstract: Model merging has recently gained attention as an economical and scalable approach to incorporate task-specific weights from various tasks into a unified multi-task model. For example, in Task Arithmetic (TA), adding the fine-tuned weights of different tasks can enhance the model's performance on those tasks, while subtracting them leads to task forgetting. Although TA is highly effective, interference among task still hampers the performance of the merged model. Existing methods for handling conflicts between task generally rely on empirical selection, resulting in suboptimal performance. In this paper, we introduce an Adaptive Weight Disentanglement method. We begin by theoretically proving that task vectors employed in model merging should be orthogonal to minimize interference among tasks. Guided by this insight, we initialize redundant vectors such that, when subtracted from the original task vectors, the resulting vectors exhibit increased orthogonality. Additionally, we impose an norm constraint on the redundant vectors to preserve the performance of the task-specific models. Experimental results demonstrate the effectiveness of our proposed technique: it successfully extracts redundant vectors, and after their subtraction, the task vectors not only retain robust performance but also achieve superior fusion outcomes. Our code is available at \href{https://github.com/FarisXiong/AWD.git}{https://github.com/FarisXiong/AWD.git}.
URLs: https://github.com/FarisXiong/AWD.git, https://github.com/FarisXiong/AWD.git
Authors: Taneesh Gupta, Rahul Madhavan, Xuchao Zhang, Chetan Bansal, Saravan Rajmohan
Abstract: We introduce REFA, a family of reference-free alignment methods that optimize over multiple user preferences while enforcing fine-grained length control. Our approach integrates deviation-based weighting to emphasize high-quality responses more strongly, length normalization to prevent trivial short-response solutions, and an EOS-probability regularizer to mitigate dataset-induced brevity biases. Theoretically, we show that under the Uncertainty Reduction with Sequence Length Assertion (URSLA), naive length normalization can still incentivize length-based shortcuts. By contrast, REFA corrects these subtle incentives, guiding models toward genuinely more informative and higher-quality outputs. Empirically, REFA sets a new state-of-the-art among reference-free alignment methods, producing richer responses aligned more closely with human preferences. Compared to a base supervised fine-tuned (SFT) mistral-7b model that achieves 8.4% length-controlled win rate (LC-WR) and 6.2% win rate (WR), our best REFA configuration attains 21.62% LC-WR and 19.87% WR on the AlpacaEval v2 benchmark. This represents a substantial improvement over both the strongest multi-preference baseline, InfoNCA (16.82% LC-WR, 10.44% WR), and the strongest reference-free baseline, SimPO (20.01% LC-WR, 17.65% WR)
Authors: Xiaoqing Zhang, Ang Lv, Yuhan Liu, Flood Sung, Wei Liu, Shuo Shang, Xiuying Chen, Rui Yan
Abstract: Large language models (LLMs) excel at few-shot in-context learning (ICL) without requiring parameter updates. However, as the number of ICL demonstrations increases from a few to many, performance tends to plateau and eventually decline. We identify two primary causes for this trend: the suboptimal negative log-likelihood (NLL) optimization objective and the incremental data noise. To address these issues, we introduce DrICL, a novel optimization method that enhances model performance through Differentiated Learning and advantage-based Reweighting objectives. Globally, DrICL utilizes differentiated learning to optimize the NLL objective, ensuring that many-shot performance surpasses zero-shot levels. Locally, it dynamically adjusts the weighting of many-shot demonstrations by leveraging cumulative advantages inspired by reinforcement learning, thereby improving generalization. This approach allows the model to handle varying numbers of shots effectively, mitigating the impact of noisy data. Recognizing the lack of multi-task datasets with diverse many-shot distributions, we develop the Many-Shot ICL Benchmark (ICL-50)-a large-scale benchmark of 50 tasks that cover shot numbers from 1 to 350 within sequences of up to 8,000 tokens-for fine-tuning purposes. ICL-50 facilitates the evaluation of many-shot ICL strategies across seven prominent NLP tasks and 50 distinct datasets. Experimental results demonstrate that LLMs enhanced with DrICL achieve significant improvements in many-shot setups across various tasks, including both in-domain and out-of-domain scenarios. We release the code and benchmark dataset hoping to facilitate further research in many-shot ICL.