Large language models (LLMs) have excelled in various NLP tasks, including machine translation (MT), yet most studies focus on sentence-level translation. This work investigates the inherent capability of instruction-tuned LLMs for document-level translation (docMT). Unlike prior approaches that require specialized techniques, we evaluate LLMs by directly prompting them to translate entire documents in a single pass. Our results show that this method improves translation quality compared to translating sentences separately, even without document-level fine-tuning. However, this advantage is not reflected in BLEU scores, which often favor sentence-based translations. We propose using the LLM-as-a-judge paradigm for evaluation, where GPT-4 is used to assess document coherence, accuracy, and fluency in a more nuanced way than n-gram-based metrics. Overall, our work demonstrates that instruction-tuned LLMs can effectively leverage document context for translation. However, we caution against using BLEU scores for evaluating docMT, as they often provide misleading outcomes, failing to capture the quality of document-level translation.

This paper presents our system, InsightBuddy-AI, designed for extracting medication mentions and their associated attributes, and for linking these entities to established clinical terminology resources, including SNOMED-CT, the British National Formulary (BNF), ICD, and the Dictionary of Medicines and Devices (dm+d).To perform medication extraction, we investigated various ensemble learning approaches, including stacked and voting ensembles (using first, average, and max voting methods) built upon eight pre-trained language models (PLMs). These models include general-domain PLMs—BERT, RoBERTa, and RoBERTa-Large—as well as domain-specific models such as BioBERT, BioClinicalBERT, BioMedRoBERTa, ClinicalBERT, and PubMedBERT.The system targets the extraction of drug-related attributes such as adverse drug effects (ADEs), dosage, duration, form, frequency, reason, route, and strength.Experiments conducted on the n2c2-2018 shared task dataset demonstrate that ensemble learning methods outperformed individually fine-tuned models, with notable improvements of 2.43% in Precision and 1.35% in F1-score.We have also developed cross-platform desktop applications for both entity recognition and entity linking, available for Windows and macOS.The InsightBuddy-AI application is freely accessible for research use at https://github.com/HECTA-UoM/InsightBuddy-AI.

Most studies on the linguistic information encoded by BERT primarily focus on English. Our study examines a monolingual German BERT model using a semantic classification task on newspaper articles, analysing the linguistic features influencing classification decisions through SHAP values. We use the TüBa-D/Z corpus, a resource with gold-standard annotations for a set of linguistic features, including POS, inflectional morphology, phrasal, clausal, and dependency structures. Semantic features of nouns are evaluated via the GermaNet ontology using shared hypernyms. Our results indicate that the features identified in English also affect classification in German but suggests important language- and task-specific features as well.

Knowledge Graph Embedding (KGE) methods are widely used to map entities and relations from knowledge graphs (KGs) into continuous vector spaces, enabling non-classical reasoning over knowledge structures. Despite their effectiveness, the uncertainty of KGE methods has not been extensively studied in the literature. This gap poses significant challenges, particularly when deploying KGE models in high-stakes domains like medicine, where reliability and risk assessment are critical. This dissertation seeks to investigate various types of uncertainty in KGE methods and explore strategies to quantify, mitigate, and reason under uncertainty effectively. The outcomes of this research will contribute to enhancing the reliability of KGE methods, providing greater confidence in their use beyond benchmark datasets, and supporting their application in real-world, high-stakes domains.

In the digital age, social media platforms like Twitter serve as an extensive repository of public discourse, including instances of sexism. It is important to identify such behavior since radicalized ideologies can lead to real-world violent acts. This project aims to develop a deep learning-based tool that leverages a combination of BERT (both English and multilingual versions) and GraphSAGE, a Graph Neural Network (GNN) model, alongside sentiment analysis and natural language processing (NLP) techniques. The tool is designed to analyze tweets for sexism detection and classify them into five categories.

Neural codec language models (or codec LMs) are emerging as a powerful framework for audio generation tasks like text-to-speech (TTS). These models leverage advancements in language modeling and residual vector quantization (RVQ)-based audio codecs, which compress audios into discrete codes for LMs to process. Despite the close interdependence of codecs and LMs in these systems, research on codecs and LMs has largely remained siloed. In this work, we propose three techniques for better codec-LM co-design: (i) a frame-wise codec encoder that improves both LM log-likelihood and end-to-end TTS metrics, (ii) LM codebook level dropout, a method to efficiently navigate a portion of the codec-LM design space by training a single LM, and (iii) increased codec frame duration, which we show can accelerate inference while maintaining end-to-end performance. Our experiments demonstrate that combining all three co-design techniques results in doubled inference speed, and improvements in intelligibility, audio quality, and speaker control in TTS relative to a siloed baseline.

Machine translation for low-resource language pairs is a challenging task. This task could become extremely difficult once a speaker uses code switching. We present the first code-switching Kazakh-Russian parallel corpus.Additionally, we propose a method to build a machine translation model for code-switched Kazakh-Russian language pair with no labeled data. Our method is basing on generation of synthetic data. This method results in a model beating an existing commercial system by human evaluation.

In recommender systems, users often seek the best products through indirect, vague, or under-specified queries such as “best shoes for trail running.” These queries, referred to as implicit superlative queries, pose a challenge for standard retrieval and ranking systems due to their lack of explicit attribute mentions and the need for identifying and reasoning over complex attributes. We investigate how Large Language Models (LLMs) can generate implicit attributes for ranking and reason over them to improve product recommendations for such queries. As a first step, we propose a novel four-point schema, called SUPERB, for annotating the best product candidates for superlative queries, paired with LLM-based product annotations. We then empirically evaluate several existing retrieval and ranking approaches on our newly created dataset, providing insights and discussing how to integrate these findings into real-world e-commerce production systems.

Quizzes play a crucial role in education by reinforcing students’ understanding of key concepts and encouraging self-directed exploration. However, compiling high-quality quizzes can be challenging and require deep expertise and insight into specific subject matter. Although LLMs have greatly enhanced the efficiency of quiz generation, concerns remain regarding the quality of these AI-generated quizzes and their educational impact on students. To address these issues, we introduce ConQuer, a concept-based quiz generation framework that leverages external knowledge sources. We employ comprehensive evaluation dimensions to assess the quality of the generated quizzes, using LLMs as judges. Our experiment results demonstrate a 4.8% improvement in evaluation scores and a 77.52% win rate in pairwise comparisons against baseline quiz sets. Ablation studies further underscore the effectiveness of each component in our framework.

This paper presents a research thesis proposal to develop a generalizable Native American language identification system. Despite their cultural and historical significance, Native American languages remain entirely unsupported by major commercial language identification systems. This omission not only underscores the systemic neglect of endangered languages in technological development, but also highlights the urgent need for dedicated, community-driven solutions. We propose a two-pronged approach: (1) systematically curating linguistic resources across all Native American languages for robust training, and (2) tailored data augmentation to generate synthetic yet linguistically coherent training samples. As proof of concept, we extend an existing rudimentary Athabaskan language classifier by integrating Plains Apache, an extinct Southern Athabaskan language, as an additional language class. We also adapt a data generation framework for low-resource languages to create synthetic Plains Apache data, highlighting the potential of data augmentation. This proposal advocates for a community-driven, technological approach to supporting Native American languages.

The emergence of large language models (LLMs) has significantly influenced numerous fields, including healthcare, by enhancing the capabilities of automated systems to process and generate human-like text. However, despite their advancements, the reliability and accuracy of LLMs in medical contexts remain critical concerns. Current evaluation methods often lack robustness and fail to provide a comprehensive assessment of LLM performance, leading to potential risks in clinical settings. In this work, we propose Med-CoDE, a specifically designed evaluation framework for medical LLMs to address these challenges. The framework leverages a critique-based approach to quantitatively measure the degree of disagreement between model-generated responses and established medical ground truths. This framework captures both accuracy and reliability in medical settings. The proposed evaluation framework aims to fill the existing gap in LLM assessment by offering a systematic method to evaluate the quality and trustworthiness of medical LLMs. Through extensive experiments and case studies, we illustrate the practicality of our framework in providing a comprehensive and reliable evaluation of medical LLMs.

Supervised Fine-tuning (SFT) and preference optimization (PO) are key methods for enhancing language models and aligning them with human preferences. However, scaling preference datasets for PO training is challenging, leading AI customer support systems to rely on SFT. To address this, we propose the Sentiment-guided Automatic Generation of Preference Datasets (Sentimatic) methodology to automatically generate customer preference datasets without human intervention using a publicly available dataset constructed for SFT. Our approach classifies responses by sentiment, fine-tunes models on them, and applies advanced sampling and evaluation techniques to ensure diversity and quality. Ultimately, we generated 1,174 customer preference datasets based on 357 test datasets, and through experiments, we confirmed that the AI customer support system trained on these datasets is capable of carefully considering customer emotions and generating professional and appropriate responses.

This paper presents a federated learning system with differential privacy for hate speech detection, tailored to low-resource languages. By fine-tuning pre-trained language models, ALBERT emerged as the most effective option for balancing performance and privacy. Experiments demonstrated that federated learning with differential privacy performs adequately in low-resource settings, though datasets with fewer than 20 sentences per client struggled due to excessive noise. Balanced datasets and augmenting hateful data with non-hateful examples proved critical for improving model utility. These findings offer a scalable and privacy-conscious framework for integrating hate speech detection into social media platforms and browsers, safeguarding user privacy while addressing online harm.

This study examines the performance of Large Language Models (LLMs) in Aspect-Based Sentiment Analysis (ABSA), with a focus on implicit aspect extraction in a novel domain. Using a synthetic sports feedback dataset, we evaluate open-weight LLMs’ ability to extract aspect-polarity pairs and propose a metric to facilitate the evaluation of aspect extraction with generative models. Our findings highlight both the potential and limitations of LLMs in the ABSA task.

CLIP is a foundational model that bridges images and text, widely adopted as a key component in numerous vision-language models.However, the lack of large-scale open Japanese image-text pairs poses a significant barrier to the development of Japanese vision-language models.In this study, we constructed a Japanese image-text pair dataset with 1.5 billion examples using machine translation with open-weight LLMs and pre-trained Japanese CLIP models on the dataset.The performance of the pre-trained models was evaluated across seven benchmark datasets, achieving competitive average scores compared to models of similar size without the need for extensive data curation. However, the results also revealed relatively low performance on tasks specific to Japanese culture, highlighting the limitations of translation-based approaches in capturing cultural nuances. Our dataset, models, and code are publicly available.

Past vocabulary learning techniques identify relevant vocabulary before training, relying on statistical and entropy-based assumptions that largely neglect the role of model training.Empirically, we observe that trained translation models are induced to use a byte-pair encoding (BPE) vocabulary subset distinct from the original BPE vocabulary, leading to performance improvements when retrained with the induced vocabulary.In this paper, we analyze this discrepancy in neural machine translation by examining vocabulary and entropy shifts during self-training—where each iteration generates a labeled dataset by pairing source sentences with the model’s predictions to define a new vocabulary.Building on these insights, we propose *self-vocabularizing training*, an iterative method that self-selects a smaller, more optimal vocabulary, yielding up to a 1.49 BLEU improvement.Moreover, we find that deeper model architectures lead to both an increase in unique token usage and a 6–8% reduction in vocabulary size.

We consider the well-known and important tasks of clone detection and information retrieval for source code. The most standard setup is to search clones inside the same language code snippets. But it is also useful to find code snippets with identical behaviour in different programming languages. Nevertheless multi- and cross-lingual clone detection has been little studied in literature. We present a novel training procedure, cross-consistency training (CCT) leveraging cross-lingual similarity, that we apply to train language models on source code in various programming languages. We show that this training is effective both for encoder- and decoder-based models.The trained encoder-based CCT-LM model%and fine-tuned with CCT,achieves a new state of the art on POJ-104 (monolingual C++ clone detection benchmark) with 96.73% MAP and AdvTest (monolingual Python code search benchmark) with 47.18% MRR. The decoder-based CCT-LM model shows comparable performance in these tasks. In addition, we formulate the multi- and cross-lingual clone detection problem and present XCD, a new benchmark dataset produced from CodeForces submissions.

We challenge the prevailing assumption that LLMs must rely fully on sub-word tokens for high-quality text generation. To this end, we propose the “Generative Pretrained Thoughtformer” (GPTHF), a hierarchical transformer language model capable of text generation by compressing text into sentence embeddings and employing a sentence attention mechanism. GPTHF retains GPT’s architecture, modifying only token interactions via dynamic sparse attention masks. Our experiments show that GPTHF achieves an up to an order of magnitude improvement in FLOPs efficiency and a threefold increase in runtime speed compared to equally-sized GPT models in the low-size regime. This is achieved through a unique generation method that caches and reuses sentence embeddings, allowing significant portions of the input to bypass large parts of the network.

Native Language Identification (NLI) is the task of automatically identifying the native language (L1) of individuals based on their second language (L2) production. The introduction of Large Language Models (LLMs) with billions of parameters has renewed interest in text-based NLI, with new studies exploring LLM-based approaches to NLI on English L2. The capabilities of state-of-the-art LLMs on non-English NLI corpora, however, have not yet been fully evaluated. To fill this important gap, we present the first evaluation of LLMs for multilingual NLI. We evaluated the performance of several LLMs compared to traditional statistical machine learning models and language-specific BERT-based models on NLI corpora in English, Italian, Norwegian, and Portuguese. Our results show that fine-tuned GPT-4 models achieve state-of-the-art NLI performance.

The abundance of medical records holds great promise for enhancing healthcare and advancing biomedical research. However, due to privacy constraints, access to such data is typically limited to internal use.Recent studies have attempted to overcome this challenge by generating synthetic data through Causal Language Modelling. Yet, this approach often fails to ensure patient anonymity and offers limited control over output diversity—unless additional computational cost is introduced.In response, we propose a method for generating synthetic free-text medical records based on Masked Language Modelling. Our approach retains key medical details while introducing variability in the generated texts and reducing the risk of patient re-identification. With a relatively lightweight architecture of approximately 120 million parameters, the system ensures low inference costs.Experimental results show that our method produces high-quality synthetic data, achieving a HIPAA-compliant PHI recall of 96% and a re-identification risk of only 3.5%. Furthermore, downstream evaluations reveal that models trained on the synthetic data perform comparably to those trained on real-world data. Our trained models are publicly available on Github as SynDeidMLM (at https://github.com/SamySam0/SynDeidMLM) (meaning synthetic and de-identified data generation using MLM).

When developing language technology, researchers have routinely turned to transfer learning to resolve the data scarcity conundrum presented in low-resource languages. As far as we know, this study is the first to evaluate the amount of documentation needed for transfer learning, specifically the smallest vocabulary size needed to create a sentence embedding space. In adopting widely spoken languages as a proxy for low-resource languages, our experiments show that the relationship between a sentence embedding’s vocabulary size and performance is logarithmic with performance leveling at a vocabulary size of 25,000. It should be noted that this relationship cannot be replicated across all languages and this level of documentation does not exist for many low-resource languages. We do observe, however, that performance accelerates at a vocabulary size of ≤ 1000, a quantity that is present in most low-resource language documentation. These results can aid researchers in understanding whether a low-resource language has enough documentation necessary to support the creation of a sentence embedding and language model.

A two-part affine approximation has been found to be a good approximation for transformer computations over certain subject-object relations. Adapting the Bigger Analogy Test Set, we show that the linear transformation W s , where s is a middle-layer representation of a subject token and W is derived from model derivatives, can accurately reproduce final object states for many relations. This linear technique achieves 90% faithfulness on morphological relations, with similar findings across languages and models. Our results suggest that some conceptual relationships in language models, such as morphology, are readily interpretable from latent space and are sparsely encoded by cross-layer linear transformations.

We introduce **SPY Dataset**: a novel synthetic dataset for the task of **Personal Identifiable Information (PII) detection**, underscoring the significance of protecting PII in modern data processing. Our research innovates by leveraging Large Language Models (LLMs) to generate a dataset that emulates real-world PII scenarios. Through evaluation, we validate the dataset’s quality, providing a benchmark for PII detection. Comparative analyses reveal that while PII and Named Entity Recognition (NER) share similarities, **dedicated NER models exhibit limitations** when applied to PII-specific contexts. This work contributes to the field by making the generation methodology and the generated dataset publicly, thereby enabling further research and development in this field.

Distinguishing vulnerable code from non-vulnerable code is challenging due to high inter-class similarity. Supervised contrastive learning (SCL) improves embedding separation but struggles with intra-class clustering, especially when variations within the same class are subtle. We propose Cluster-Enhanced Supervised Contrastive Loss (CESCL), an extension of SCL with a distance-based regularization term that tightens intra-class clustering while maintaining inter-class separation. Evaluating on CodeBERT and GraphCodeBERT with Binary Cross Entropy (BCE), BCE + SCL, and BCE + CESCL, our method improves F1 score by 1.76% on CodeBERT and 4.1% on GraphCodeBERT, demonstrating its effectiveness in code vulnerability detection and broader applicability to high-similarity classification tasks.

Arabic calligraphy carries rich historical information and meaning. However, the complexity of its artistic elements and the absence of a consistent baseline make text extraction from such works highly challenging. In this paper, we provide an in-depth analysis of the unique obstacles in processing and interpreting these images, including the variability in calligraphic styles, the influence of artistic distortions, and the challenges posed by missing or damaged text elements. We explore potential solutions by leveraging state-of-the-art architectures and deep learning models, including visual language models, to improve text extraction and script completion.

Accurate detection of offensive language is essential for a number of applications related to social media safety. There is a sharp contrast in performance in this task between low and high-resource languages. In this paper, we adapt fine-tuning strategies that have not been previously explored for Sinhala in the downstream task of offensive language detection. Using this approach, we introduce four models: “Subasa-XLM-R”, which incorporates an intermediate Pre-Finetuning step using Masked Rationale Prediction. Two variants of “Subasa-Llama” and “Subasa-Mistral”, are fine-tuned versions of Llama (3.2) and Mistral (v0.3), respectively, with a task-specific strategy. We evaluate our models on the SOLD benchmark dataset for Sinhala offensive language detection. All our models outperform existing baselines. Subasa-XLM-R achieves the highest Macro F1 score (0.84) surpassing state-of-the-art large language models like GPT-4o when evaluated on the same SOLD benchmark dataset under zero-shot settings. The models and code are publicly available.

Code-generating Large Language Models (LLMs) have become essential tools in modern software development, enhancing productivity and accelerating development. This paper aims to investigate the fine-tuning of code-generating LLMs using Reinforcement Learning and Direct Preference Optimization, further improving their performance. To achieve this, we enhance the training data for the reward model with the help of symbolic execution techniques, ensuring more comprehensive and objective data. With symbolic execution, we create a custom dataset that better captures the nuances in code evaluation. Our reward models, fine-tuned on this dataset, demonstrate significant improvements over the baseline, CodeRL, in estimating the quality of generated code. Our code-generating LLMs, trained with the help of reward model feedback, achieve similar results compared to the CodeRL benchmark.

Measuring how real images look is a complex task in artificial intelligence research. For example, an image of Albert Einstein holding a smartphone violates common-sense because modern smartphone were invented after Einstein’s death. We introduce a novel method, which we called Through the Looking Glass (TLG), to assess image common sense consistency using Large Vision-Language Models (LVLMs) and Transformer-based encoder. By leveraging LVLM to extract atomic facts from these images, we obtain a mix of accurate facts. We proceed by fine-tuning a compact attention-pooling classifier over encoded atomic facts. Our TLG has achieved a new state-of-the-art performance on the WHOOPS! and WEIRD datasets while leveraging a compact fine-tuning component.

With the utilization of Transformer architecture, large Vision and Language (V&L) models have shown promising performance in even zero-shot settings. Several studies, however, indicate a lack of robustness of the models when dealing with complex linguistics and visual attributes. In this work, we introduce a novel V&L benchmark - ColorFoil, by creating color-related foils to assess the models’ perception ability to detect colors like red, white, green, etc. We evaluate seven state-of-the-art V&L models including CLIP, ViLT, GroupViT, and BridgeTower, etc. in a zero-shot setting and present intriguing findings from the V&L models. The experimental evaluation indicates that ViLT and BridgeTower demonstrate much better color perception capabilities compared to CLIP and its variants and GroupViT. Moreover, CLIP-based models and GroupViT struggle to distinguish colors that are visually distinct to humans with normal color perception ability.

The frontier of large language model (LLM) development has largely been substantiated by knowledge-intensive tasks specified in English. In this proposed thesis, I argue for the key role that multilinguality occupies in the development of practical and knowledgeable LLMs.First, I consider practical methods to improve LLM’s performance on standard natural language processing (NLP) tasks by leveraging their existing multilingual knowledge.Then, I investigate the underlying multilingual knowledge of LLMs with two benchmarks: on complex reasoning, and on territorial disputes. These benchmarks reveal LLMs’ inconsistent performance across languages. I then design efficient techniques, both at inference-time and training-time, to address these discrepancies. Finally, I extend the territorial disputes benchmark to retrieval-augmented generation (RAG) setting, comparing the effects of different retrieval settings on cross-lingual robustness. My proposal shows that informed use of multilinguality enhances LLMs’ capabilities, and our understanding thereof.

Identifying commercial posts in resource-constrained languages among diverse and unstructured content remains a significant challenge for automatic text classification tasks. To address this, this work introduces a novel dataset named MDC³ (Multimodal Dataset for Commercial Content Classification), comprising 5,007 annotated Bengali social media posts classified as commercial and noncommercial. A comprehensive annotation guideline accompanying the dataset is included to aid future dataset creation in resource-constrained languages. Furthermore, we performed extensive experiments on MDC³ considering both unimodal and multimodal domains. Specifically, the late fusion of textual (mBERT) and visual (ViT) models (i.e., ViT+mBERT) achieves the highest F1 score of 90.91, significantly surpassing other baselines.

We introduce DateLogicQA, a human-curated benchmark of 190 questions specifically designed to understand temporal bias in Large Language Models (LLMs). Covering seven date formats across past, present, and future contexts, DateLogicQA examines four reasoning types: commonsense, factual, conceptual, and numerical. Through human-led evaluations of 12 state-of-the-art LLMs, we identify Representation-Level Bias, arising from suboptimal embeddings that distort date semantics, and Logical-Level Bias, manifesting when correct date tokens yield flawed temporal reasoning. Our findings underscore persistent challenges in handling various date formats and temporal contexts, revealing the need for more robust pretraining data, targeted post-training methods, and precise tokenization strategies. By illuminating these biases, we provide actionable insights to guide the development of LLMs for accurate temporal reasoning across diverse real-world applications.

Existing in-context learning (ICL) methods for relation extraction (RE) often prioritize language similarity over structural similarity, which may result in overlooking entity relationships. We propose an AMR-enhanced retrieval-based ICL method for RE to address this issue. Our model retrieves in-context examples based on semantic structure similarity between task inputs and training samples. We conducted experiments in the supervised setting on four standard English RE datasets. The results show that our method achieves state-of-the-art performance on three datasets and competitive results on the fourth. Furthermore, our method outperforms baselines by a large margin across all datasets in the more demanding unsupervised setting.

The ways in which natural language processing (NLP) can inform how veterans can improve effectiveness in translating military experience to workforce utility is underexplored. We design NLP experiments to evaluate the degree of explanation in veteran job interview responses as a proxy for perceived hireability. We examine linguistic and psycholinguistic features, context, and participant variability to investigate the mechanics of effective communication in employee selection. Results yield good performance when distinguishing between varying degrees of explanation in responses using LIWC features, indicating robustness of linguistic feature integration. Classifying Over- and Under-explained responses reflects challenges of class imbalance and the limitations of tested NLP methods for detecting subtleties in overly verbose or concise communication. Our findings have immediate applications for assistive technologies in job interview settings, and broader implications for enhancing automated communication assessment tools and refining strategies for training and interventions in communication-heavy fields.

This paper introduces a new dataset for classifying memes by their template and communicative intent.It includes a broad selection of meme templates and examples scraped from imgflip and a smaller hand-annotated set of memes scraped from Reddit.The Reddit memes have been annotated for meta-category using a novel annotation scheme that classifies memes by the structure of the perspective they are being used to communicate.YOLOv11 and ChatGPT 4o are used to provide baseline modeling results.We find that YOLO struggles with template classification on real-world data but outperforms ChatGPT in classifying meta-categories.

Content moderation is essential in preventing the spread of harmful content on the Internet. However, there are instances where moderation fails and it is important to understand when and why that happens. Workflows that aim to uncover a system’s weakness typically use clustering of the data points’ embeddings to group errors together. In this paper, we evaluate the K-Means clustering of four text representations for the task of offensive language detection in English and Levantine Arabic. We find Sentence-BERT (SBERT) embeddings give the most human-interpretable clustering for English errors and the grouping is mainly based on the targeted group in the text. Meanwhile, SBERT embeddings of Large Language Model (LLM)-generated linguistic features give the most interpretable clustering for Arabic errors.

Recent advances in video-text retrieval (VTR) have largely relied on supervised learning and fine-tuning. In this paper, we introduce , a novel zero-shot VTR framework that leverages off-the-shelf video captioners, large language models (LLMs), and text retrieval methods—entirely without additional training or annotated data. Due to the limited power of captioning methods, the captions often miss important content in the video, resulting in unsatisfactory retrieval performance. To translate more information into video captions, we first generates initial captions for videos, then enhances them using a relevance-boosted captioning strategy powered by LLMs, enriching video descriptions with salient details. To further emphasize key content, we propose structural information extraction, organizing visual elements such as objects, events, and attributes into structured templates, further boosting the retrieval performance. Benefiting from the enriched captions and structuralized information, extensive experiments on several video-text retrieval benchmarks demonstrate the superiority of over existing fine-tuned and pretraining methods without any data. They also show that the enriched captions capture key details from the video with minimal noise. Code and data will be released to facilitate future research.

Crosswalks, which map one classification system to another, are critical tools for harmonizing data across time, countries, or frameworks. However, constructing crosswalks is labor-intensive and often requires domain expertise. This paper investigates the potential of Large Language Models (LLMs) to assist in creating crosswalks, focusing on two Danish occupational classification systems from different time periods as a case study. We propose a two-stage, prompt-based framework for this task, where LLMs perform similarity assessments between classification codes and identify final mappings through a guided decision process. Using four instruction-tuned LLMs and comparing them against an embedding-based baseline, we evaluate the performance of different models in crosswalks. Our results highlight the strengths of LLMs in crosswalk creation compared to the embedding-based baseline, showing the effectiveness of the interactive prompt-based framework for conducting crosswalks by LLMs. Furthermore, we analyze the impact of model combinations across two interactive rounds, highlighting the importance of model selection and consistency. This work contributes to the growing field of NLP applications for domain-specific knowledge mapping and demonstrates the potential of LLMs in advancing crosswalk methodologies.

To improve the performance of sentence pair modeling tasks, we propose an additional pre-training method, also known as transfer fine-tuning, for pre-trained masked language models.Pre-training for masked language modeling is not necessarily designed to bring semantically similar sentences closer together in the embedding space.Our proposed method aims to improve the performance of sentence pair modeling by applying contrastive learning to pre-trained masked language models, in which sentence embeddings of paraphrase pairs are made similar to each other.While natural language inference corpora, which are standard in previous studies on contrastive learning, are not available on a large-scale for non-English languages, our method can construct a training corpus for contrastive learning from a raw corpus and a paraphrase dictionary at a low cost.Experimental results on four sentence pair modeling tasks revealed the effectiveness of our method in both English and Japanese.

This paper examines how well visual language models (VLMs) understand video question answering (VideoQA) tasks and generate responses accordingly. Recently, VLMs based on Large Language Models (LLMs) have shown remarkable performance, but the processes of understanding and reasoning in VLMs remain under-explored. To tackle this challenge, we propose Video Understanding and Response Consistency Assessment, VURCA, a framework that incorporates a fine-grained question generation and answering process to measure how well the responses generated by VLMs align with what the model understands. In addition, we introduce an extended benchmark dataset, FgNExT-QA, which builds upon NExT-QA by incorporating more fine-grained VideoQA tasks. FgNExT-QA is designed to evaluate fine-grained understanding in video question answering. Through experiments, we found that despite the strong overall QA performance of VLMs, their understanding of both the video content and the question remains limited. In particular, they exhibit poor video comprehension in fine-grained VideoQA tasks.

Text style transfer (TST) is the task of transforming a text to reflect a particular style while preserving its original content. Evaluating TSToutputs is a multidimensional challenge, requiring the assessment of style transfer accuracy, content preservation, and naturalness. Us-ing human evaluation is ideal but costly, as is common in other natural language processing (NLP) tasks; however, automatic metrics forTST have not received as much attention as metrics for, e.g., machine translation or summarization. In this paper, we examine both set ofexisting and novel metrics from broader NLP tasks for TST evaluation, focusing on two popular subtasks—sentiment transfer and detoxification—in a multilingual context comprising English, Hindi, and Bengali. By conducting meta-evaluation through correlation with hu-man judgments, we demonstrate the effectiveness of these metrics when used individually and in ensembles. Additionally, we investigatethe potential of large language models (LLMs) as tools for TST evaluation. Our findings highlight newly applied advanced NLP metrics andLLM-based evaluations provide better insights than existing TST metrics. Our oracle ensemble approaches show even more potential.

In multi-turn dialogues, large language models face a critical challenge of ensuring coherence while adapting to user-specific information.. This study introduces the persona knowledge gap, the discrepancy between a model’s internal understanding and the knowledge required for coherent, personalized conversations. While prior research has recognized these gaps, computational methods for their identification and resolution remain underexplored. We propose Conversation Preference Elicitation and Recommendation (CPER), a novel framework that dynamically detects and resolves persona knowledge gaps using intrinsic uncertainty quantification and feedback-driven refinement. CPER consists of three key modules: a Contextual Understanding Module for preference extraction, a Dynamic Feedback Module for measuring uncertainty and refining persona alignment, and a Persona-Driven Response Generation module for adapting responses based on accumulated user context. We evaluate CPER on two real-world datasets: CCPE-M for preferential movie recommendations and ESConv for mental health support. Using A/B testing, human evaluators preferred CPER’s responses 42% more often than baseline models in CCPE-M and 27% more often in ESConv. A qualitative human evaluation confirms that CPER’s responses are preferred for maintaining contextual relevance and coherence, particularly in longer (12+ turn) conversations.

Large language models (LLMs) like GPT-4 and LLaMA-3 offer transformative potential across industries, e.g., enhancing customer service, revolutionizing medical diagnostics, or identifying crises in news articles. However, deploying LLMs faces challenges such as limited training data, high computational costs, and issues with transparency and explainability. Our research focuses on distilling compact, parameter-efficient tailored language models (TLMs) from LLMs for domain-specific tasks with comparable performance. Current approaches like knowledge distillation, fine-tuning, and model parallelism address computational efficiency but lack hybrid strategies to balance efficiency, adaptability, and accuracy. We present ANON - an adaptive knowledge distillation framework integrating knowledge distillation with adapters to generate computationally efficient TLMs without relying on labeled datasets. ANON uses cross-entropy loss to transfer knowledge from the teacher’s outputs and internal representations while employing adaptive prompt engineering and a progressive distillation strategy for phased knowledge transfer. We evaluated ANON’s performance in the crisis domain, where accuracy is critical and labeled data is scarce. Experiments showed that ANON outperforms recent approaches of knowledge distillation, both in terms of the resulting TLM performance and in reducing the computational costs for training and maintaining accuracy compared to LLMs for domain-specific applications.

Debate education fosters critical thinking skills but often incurs high human costs. Recent advancements in Large Language Models (LLMs) show promise in automating counter-argument generation. However, it remains unclear how best to guide LLMs to target both implicit and critical premises. In this study, we systematically compare multi-step and one-step generation methods for counter-arguments across 100 debate topics. Our findings reveal that one-step approaches consistently outperform multi-step pipelines, owing to their better grasp of the “motion spirit,” minimized propagation of hallucinations, and avoidance of challenging intermediate tasks. Among premise-targeting methods, a one-step strategy that accounts for both implicit and explicit premises—Generated and Targeted Premise Attack (GTG)—emerges as the strongest performer in expert and automated evaluations. These results highlight the value of direct, integrated prompts for leveraging LLMs in complex argumentation tasks and offer insights for developing more effective automated debate agents.

The effectiveness of embedding methods is crucial for optimizing text classification performance in Automated Machine Learning (AutoML). However, selecting the most suitable pre-trained model for a given task remains challenging. This study introduces the Corpus-Driven Domain Mapping (CDDM) pipeline, which utilizes a domain-annotated corpus of pre-fine-tuned models from the Hugging Face Model Hub to improve model selection. Integrating these models into AutoML systems significantly boosts classification performance across multiple datasets compared to baseline methods. Despite some domain recognition inaccuracies, results demonstrate CDDM’s potential to enhance model selection, streamline AutoML workflows, and reduce computational costs.

The recent large-scale emergence of LLMs has left an open space for dealing with their consequences, such as plagiarism or the spread of false information on the Internet. Coupling this with the rise of AI detector bypassing tools, reliable machine-generated text detection is in increasingly high demand. We investigate the paraphrasing attack resilience of various machine-generated text detection methods, evaluating three approaches: fine-tuned RoBERTa, Binoculars, and text feature analysis, along with their ensembles using Random Forest classifiers. We discovered that Binoculars-inclusive ensembles yield the strongest results, but they also suffer the most significant losses during attacks. In this paper, we present the dichotomy of performance versus resilience in the world of AI text detection, which complicates the current perception of reliability among state-of-the-art techniques.

In recent years, stylometry has been investigated in many different fields. Hence, in this work, we are going to tackle this problem, detecting, generating, and evaluating textual documents according to the writing style by leveraging state-of-the-art models. In the first step, the sentences will be extracted from several different books, each belonging to a different author, to create a dataset. Then the selected models will be trained to detect the author of sentences in the dataset. After that, generator models are utilized to generate sentences based on the authors’ writing styles with unpaired samples in the dataset. Finally, to evaluate the performance of the generators, the previously trained models will be used to assess the generated sentences and to compare the distribution of various syntactic features between the original and generated sentences. We hope the result shows that models can be achieved to detect and generate textual documents for the given authors according to their writing style.

Data visualization is integral to any Exploratory Data Analysis (EDA) task. However, generating visualization requires expertise, presenting a steep learning curve and a significant cognitive load. Natural language interfaces for EDA aim to lower this barrier by allowing users to generate visualizations through natural language queries. However, complexity remains when EDA is performed collaboratively, requiring an environment to support multi-user interaction. In this thesis proposal, we discuss challenges in user-system interaction in a collaborative multi-user setup, such as errors in visualization generation due to misinterpretation of user requests. We hypothesize that a Conversational Assistant (CA) capable of understanding user-initiated clarification requests and generating accurate responses can improve user experience and support collaborative EDA tasks. To this end, we propose to develop such a CA (Figure tab:system_issues) and evaluate it through a user study, thus examining its impact on user experience in a collaborative environment for EDA.

Commonsense inference and domain-specific expertise are crucial for understanding and responding to emotional, cognitive, and topic-specific cues in counseling conversations with crime victims. However, these key evidences are often dispersed across multiple utterances, making it difficult to capture through single-hop reasoning. To address this, we propose MENDER, a novel Multi-hop commonsensE and domaiN-specific Chain-of-Thought (CoT) reasoning framework for knowleDge-grounded empathEtic Response generation in counseling dialogues. MENDER leverages large language models (LLMs) to integrate commonsense and domain knowledge via multi-hop reasoning over the dialogue context. It employs two specialized reasoning chains, viz. Commonsense Knowledge-driven CoT and Domain Knowledge-driven CoT rationales, which extract and aggregate dispersed emotional, cognitive, and topical evidences to generate knowledge-grounded empathetic counseling responses. Experimental evaluations on counseling dialogue dataset, POEM validate MENDER’s efficacy in generating coherent, empathetic, knowledge-grounded responses.

This paper proposes SkipCLM, a novel method for improving multilingual machine translation in Decoder Transformers. We augment contrastive learning for cross-lingual alignment with a trainable skip connection to preserve information crucial for accurate target language generation. Experiments with XGLM-564M on the Flores-101 benchmark demonstrate improved performance, particularly for en-de and en-zh direction translations, compared to direct sequence-to-sequence training and existing contrastive learning methods. Code is available at: https://github.com/s-nlp/skipclm.

Large language models (LLMs) have gained popularity in recent years for their utility in various applications. However, they are sensitive to non-semantic changes in prompt formats, where small changes in the prompt format can lead to significant performance fluctuations. In the literature, this problem is commonly referred to as prompt brittleness. Previous research on prompt engineering has focused mainly on developing techniques for identifying the optimal prompt for specific tasks. Some studies have also explored the issue of prompt brittleness and proposed methods to quantify performance variations; however, no simple solution has been found to address this challenge. We propose Mixture of Formats (MOF), a simple and efficient technique for addressing prompt brittleness in LLMs by diversifying the styles used in the prompt few-shot examples. MOF was inspired by computer vision techniques that utilize diverse style datasets to prevent models from associating specific styles with the target variable. Empirical results show that our proposed technique reduces style-induced prompt brittleness in various LLMs while also enhancing overall performance across prompt variations and different datasets.

The proportion of responses to a question and its options, known as the response distribution, enables detailed analysis of human society. Recent studies highlight the use of Large Language Models (LLMs) for predicting response distributions as a cost-effective survey method. However, the reliability of these predictions remains unclear. LLMs often generate answers by blindly following instructions rather than applying rational reasoning based on pretraining-acquired knowledge. This study investigates whether LLMs can rationally estimate distributions when presented with explanations of “artificially generated distributions” that are against commonsense. Specifically, we assess whether LLMs recognize counterintuitive explanations and adjust their predictions or simply follow these inconsistent explanations. Results indicate that smaller or less human-optimized LLMs tend to follow explanations uncritically, while larger or more optimized models are better at resisting counterintuitive explanations by leveraging their pretraining-acquired knowledge. These findings shed light on factors influencing distribution prediction performance in LLMs and are crucial for developing reliable distribution predictions using language models.

Large Language Models (LLMs) are primarily trained on high-resource natural languages, limiting their effectiveness in low-resource settings and in tasks requiring deep logical reasoning. This research introduces Rosetta-PL, a benchmark designed to evaluate LLMs’ logical reasoning and generalization capabilities in a controlled environment. We construct Rosetta-PL by translating a dataset of logical propositions from Lean into a custom logical language, which is then used to fine-tune an LLM (e.g., GPT-4o). Our experiments analyze the impact of the size of the dataset and the translation methodology on the performance of the model. Our results indicate that preserving logical relationships in the translation process significantly boosts precision, with accuracy plateauing beyond roughly 20,000 training samples. These insights provide valuable guidelines for optimizing LLM training in formal reasoning tasks and improving performance in various low-resource language applications.