Tool retrieval is a necessary first step for tool-augmented LLMs once the set of available tools grows to thousands or more, and itscentral difficulty stems from two gaps. A user query states what the user wants to accomplish in task-centric language, whereas tooldocumentation describes functionality, parameters, and calling conditions in function-centric terms. This low lexical overlap produces aQuery-Documentation Mismatch. Complex queries also bundle several functional needs at once, while a single tool usually covers only one,producing a Multi-Intent Query Mismatch. We present ToolFuse, a training-free framework that addresses each gap on a different side ofretrieval. On the tool side, Complementarity-based Semantic Fusion treats tool documentation and LLM-generated synthetic intent as adocumentation view and an intent view of the same tool, encodes them separately, and assigns each tool a fusion weight from the geometriccomplementarity between its two views in vector space, without fine-tuning the frozen dense encoder. On the query side, QueryDecomposition splits a multi-intent query into focused sub-queries, retrieves for each, and aggregates the results with rank-aware round-robin. On ToolBench, the full ToolFuse raises the Per-Split average NDCG@5 of Qwen3-Embedding-0.6B from 64.72 to 75.49, with consistentgains across three additional dense encoder backbones.

Retrieval and reading in Table QA are typically optimized in isolation, leaving no principled way to balance retrieval precision against reader capacity. We ask: how much retrieval depth does a reader actually need? Across three benchmarks (E2E-WTQ, OTT-QA, and FeTaQA) and four LLMs, we conduct a controlled single-table stress test that guarantees the inclusion of the gold table while increasing retrieval depth to K = 50 using soft or hard distractors. This design isolates reader noise tolerance from retrieval failure. We find that the Noise Ceiling is set jointly by task type, reader, and distractor condition, and is not predictable from reader class alone. With hard negatives, every reader has a Noise Ceiling of K ≤ 5 for long-form answer generation, favoring precision-oriented retrieval. By contrast, several readers remain robust through K = 50 on short-answer and multi-hop QA, depending on the distractor condition. A reader-side Cost per Success analysis further shows that the trade-off between robustness and inference cost varies across readers and distractor conditions. Overall, these results establish the reader’s condition-specific Noise Ceiling as a concrete operating target for balancing retrieval depth against precision in controlled single-table QA.

Recent advances in open-domain question answering over tables have widely adopted large language models (LLMs) under the Retriever-Reader architecture. Prior works have effectively leveraged LLMs to tackle the complex reasoning demands of the Reader component, such as text-to-text, text-to-SQL, and multi-hop reasoning. In contrast, the Retriever component has primarily focused on optimizing the query representation—training retrievers to retrieve relevant tables based on questions, or to select keywords from questions for matching table segments. However, little attention has been given to enhancing how tables themselves are represented in embedding space to better align with questions. To address this, we propose QGpT (Question Generation from Partial Tables), a simple yet effective method that uses an LLM to generate synthetic questions based on small portions of a table. These questions are generated to simulate how a user might query the content of the table currently under consideration. The generated questions are then jointly embedded with the partial table segments used for generation, enhancing semantic alignment with user queries. Without the need to embed entire tables, our method significantly improves retrieval performance across multiple benchmarks for both dense and late-interaction retrievers.

This study presents a systematic method for identifying 22 persuasive techniques used in multilingual memes. We explored various fine-tuning techniques and classification strategies, such as data augmentation, problem transformation, and hierarchical multi-label classification strategies. Identifying persuasive techniques in memes involves a multimodal task. We fine-tuned the XLM-RoBERTA-large-twitter language model, focusing on domain-specific language modeling, and integrated it with the CLIP visual model’s embedding to consider image and text features simultaneously. In our experiments, we evaluated the effectiveness of our approach by using official validation data in English. Our system in the competition, achieving competitive rankings in Subtask1 and Subtask2b across four languages: English, Bulgarian, North Macedonian, and Arabic. Significantly, we achieved 2nd place ranking for Arabic language in Subtask 1.

Using large language models, this paper presents techniques to improve extremely low-resourced indigenous language translations. Our approaches are grounded in the use of (1) the presence of a datastore consisting of a limited number of parallel translation examples, (2) the inherent capabilities of LLMs like GPT-3.5, and (3) a word-level translation dictionary. We harness the potential of LLMs and in-context learning techniques in such a setting for using LLM as universal translators for extremely low-resourced languages. Our methodology hinges on utilizing LLMs as language compilers for selected language pairs, hypothesizing that they could internalize syntactic structures to facilitate accurate translation. We introduce three techniques: KNN-Prompting with Retrieved Prompting Context, Chain-of-Thought Prompting, and Learning-from-Mistakes Prompting, with the last method addressing past errors. The evaluation results suggest that, even with limited corpora, LLMs, when paired with proper prompting, can effectively translate extremely low-resource languages.

Cloze multiple-choice questions (MCQs) are essential for assessing comprehension in educational settings, but manually designing effective distractors is time-consuming. Addressing this, recent research has automated distractor generation, yet such methods often neglect to adjust the difficulty level to the learner’s abilities, resulting in non-personalized assessments. This study introduces the Personalized Cloze Test Generation (PCGL) Framework, utilizing Large Language Models (LLMs) to generate cloze tests tailored to individual proficiency levels. Our PCGL Framework simplifies test creation by generating both question stems and distractors from a single input word and adjusts the difficulty to match the learner’s proficiency. The framework significantly reduces the effort in creating tests and enhances personalized learning by dynamically adjusting to the needs of each learner.

Creating high-quality True-False (TF) multiple-choice questions (MCQs), with accurate distractors, is a challenging and time-consuming task in education. This paper introduces True-False Distractor Generation (TFDG), a pipeline that leverages pre-trained language models and sentence retrieval techniques to automate the generation of TF-type MCQ distractors. Furthermore, the evaluation of generated TF questions presents a challenge. Traditional metrics like BLEU and ROUGE are unsuitable for this task. To address this, we propose a new evaluation metric called Retrieval-based Accuracy Differential (RAD). RAD assesses the discriminative power of TF questions by comparing model accuracy with and without access to reference texts. It quantitatively evaluates how well questions differentiate between students with varying knowledge levels. This research benefits educators and assessment developers, facilitating the efficient automatic generation of high-quality TF-type MCQs and their reliable evaluation.

In this paper, we tackle the task of distractor generation (DG) for multiple-choice questions. Our study introduces two key designs. First, we propose the concept of retrieval augmented pretraining, which involves refining the language model pretraining to align it more closely with the downstream task of DG. Second, we explore the integration of knowledge graphs and language models to further enhance the performance of DG. Our study unveils promising directions for further development in DG by showcasing the efficacy of knowledge augmentation and task-specific pretraining. These findings demonstrate the potential for leveraging both strategies to enhance the quality and performance of DG systems.

In this paper, we address the task of cloze-style multiple choice question (MCQs) distractor generation. Our study is featured by the following designs. First, we propose to formulate the cloze distractor generation as a Text2Text task. Second, we propose pseudo Kullback-Leibler Divergence for regulating the generation to consider the item discrimination index in education evaluation. Third, we explore the candidate augmentation strategy and multi-tasking training with cloze-related tasks to further boost the generation performance. Through experiments with benchmarking datasets, our best perfomring model advances the state-of-the-art result from 10.81 to 22.00 (p@1 score).

Manually designing cloze test consumes enormous time and efforts. The major challenge lies in wrong option (distractor) selection. Having carefully-design distractors improves the effectiveness of learner ability assessment. As a result, the idea of automatically generating cloze distractor is motivated. In this paper, we investigate cloze distractor generation by exploring the employment of pre-trained language models (PLMs) as an alternative for candidate distractor generation. Experiments show that the PLM-enhanced model brings a substantial performance improvement. Our best performing model advances the state-of-the-art result from 14.94 to 34.17 (NDCG@10 score). Our code and dataset is available at https://github.com/AndyChiangSH/CDGP.

In this paper, we investigate the following two limitations for the existing distractor generation (DG) methods. First, the quality of the existing DG methods are still far from practical use. There are still room for DG quality improvement. Second, the existing DG designs are mainly for single distractor generation. However, for practical MCQ preparation, multiple distractors are desired. Aiming at these goals, in this paper, we present a new distractor generation scheme with multi-tasking and negative answer training strategies for effectively generating multiple distractors. The experimental results show that (1) our model advances the state-of-the-art result from 28.65 to 39.81 (BLEU 1 score) and (2) the generated multiple distractors are diverse and shows strong distracting power for multiple choice question.

In this study, we investigate the employment of the pre-trained BERT language model to tackle question generation tasks. We introduce two neural architectures built on top of BERT for question generation tasks. The first one is a straightforward BERT employment, which reveals the defects of directly using BERT for text generation. And, the second one remedies the first one by restructuring the BERT employment into a sequential manner for taking information from previous decoded results. Our models are trained and evaluated on the question-answering dataset SQuAD. Experiment results show that our best model yields state-of-the-art performance which advances the BLEU4 score of existing best models from 16.85 to 18.91.

In this study, we investigate the employment of the pre-trained BERT language model to tackle question generation tasks. We introduce three neural architectures built on top of BERT for question generation tasks. The first one is a straightforward BERT employment, which reveals the defects of directly using BERT for text generation. Accordingly, we propose another two models by restructuring our BERT employment into a sequential manner for taking information from previous decoded results. Our models are trained and evaluated on the recent question-answering dataset SQuAD. Experiment results show that our best model yields state-of-the-art performance which advances the BLEU 4 score of the existing best models from 16.85 to 22.17.