This paper presents **CheckSent-BN** (Claim **Check**worthiness and **Sen**timent Classification in **B**engali **N**ews Headline), a novel multi-task dataset in Bengali comprising approximately 11.5K news headlines annotated for two critical natural language processing (NLP) tasks: claim checkworthiness detection and sentiment classification. To address the lack of high-quality annotated resources in Bengali, we employ a cost-effective annotation strategy that utilizes three large language models (GPT-4o-mini, GPT-4.1-mini, and Llama-4), followed by majority voting and manual verification to ensure label consistency. We provide benchmark results using multilingual and Bengali-focused transformer models under both single-task and multi-task learning (MTL) frameworks. Experimental results show that IndicBERTv2, BanglaBERT, and mDeBERTa model-based frameworks outperform other multilingual models, with IndicBERTv2 achieving the best overall performance in the MTL setting. CheckSent-BN establishes the first comprehensive benchmark for joint claim checkworthiness and sentiment classification in Bengali news headlines, offering a valuable resource for advancing misinformation detection and sentiment-aware analysis in low-resource languages. The CheckSent-BN dataset is available at: https://github.com/pritampal98/check-sent-bn

Code synthesis from natural language problem statements has recently gained popularity with the use of large language models (LLMs). Most of the available systems and benchmarks, however, are developed for English or other high-resource languages, and a gap exists for low-resource languages such as Bangla. Addressing the gap, the Bangla Language Processing (BLP) Workshop at AACL-IJCNLP 2025 featured a shared task on Bangla-to-Python code generation. Participants were asked to design systems that consume Bangla problem statements and generate executable Python programs. A benchmark data set of training, development, and test splits was provided, and evaluation utilized the Pass@1 metric through hidden test cases. We present here a system we developed, using the state-of-the-art LLMs through a zero-shot prompting setup. We report outcomes on several models, including variants of GPT-4 and Llama-4, and specify their relative strengths and weaknesses. Our best-performing system, based on GPT-4.1, achieved a Pass@1 score of 78.6% over the test dataset. We address the challenges of Bangla code generation, morphological richness, cross-lingual understanding, and functional correctness, and outline the potential for future work in multilingual program synthesis.

The present article introduces **IndicClaimBuster**, a novel multilingual claim verification dataset comprising ≈ 9K claims and their corresponding evidence in English, Hindi, Bengali, and Hindi-English CodeMixed texts. The data set covers three key domains: politics, law and order, and health, to address the challenges of verifiable facts. Each claim was sourced from reputable Indian news portals and is accompanied by three pieces of evidence, two LLM-generated and one manually curated. Additionally, a separate attempt was conducted to generate refuted claims by employing an LLM. We further develop two frameworks: an unsupervised baseline and a two-stage pipeline that comprises evidence retrieval and veracity prediction modules. For retrieval, we fine-tuned SBERT models, with e5-base demonstrating superior average performance across languages, whereas for veracity prediction, multilingual transformers (mBERT, XLM-R, MuRIL, IndicBERTv2) were fine-tuned. Results indicate MuRIL and IndicBERTv2 excel in Indian languages, while XLM-R performs the best for CodeMix. Our work contributes a high-quality multilingual dataset and strong baseline methodologies, offering valuable resources for advancing automated claim verification in linguistically diverse and low-resource settings for Indian languages. The IndicClaimBuster dataset is available at: https://github.com/pritampal98/indic-claim-buster

Traditional machine learning and deep learning models have demonstrated remarkable performance across various NLP tasks in multiple languages. However, these conventional models often struggle with languages with complex linguistic structures and nuanced contexts, such as Bengali. Recent advancements in quantum computing offer promising solutions for tackling complex, computationally challenging problems, providing faster, more efficient processing than classical systems. This research aims to address the challenges posed by the intricate linguistic structure of the less-resourced Bengali language by developing a quantum-enhanced framework for sentiment classification and claim-checkworthiness identification. We created a classical LSTM framework and proposed novel 2-qubit and 4-qubit classical-quantum frameworks, evaluating their effectiveness for sentiment classification and claim-checkworthiness identification tasks in Bengali. An entirely new dataset comprising ≈3K samples was developed by curating Bengali news headlines from prominent sources. We tagged these headlines with sentiment and claim checkworthy labels using state-of-the-art LLMs. Our findings indicate that the quantum-enhanced frameworks outperform the traditional models in both tasks. Notably, the 4-qubit-based framework achieved the highest F1-score in sentiment classification, while the 2-qubit-based framework demonstrated the best F1-score in claim checkworthiness identification.

The efficient identification of previously fact-checked claims across multiple languages is a challenging task. It can be time-consuming for professional fact-checkers even within a single language. It becomes much more difficult to perform manually when the claim and the fact-check may be in different languages. This paper presents a systematic approach for the retrieval of top-k relevant fact-checks for a given post in a monolingual and cross-lingual setup using two transformer-based fact-checked claim retrieval frameworks that share a common preprocessing pipeline but differ in their underlying encoder implementations: TIDE, a TensorFlow-based custom dual encoder applied to english-translated data, and PTEX, a PyTorch-based encoder operating on both english-translated and original-language inputs, and introduces a lightweight post-processing technique based on a textual feature: Keyword Overlap Count applied via reranking on top of the transformer-based frameworks. Training and evaluation on a large multilingual corpus show that the fine-tuned E5-Large-v2 model in the PTEX framework yields the best monolingual track performance, achieving an average Success@10 score of 0.8846 and the same framework model with post-processing technique achieves an average Success@10 score of 0.7393 which is the best performance in crosslingual track.

Traditional machine learning (ML) and deep learning (DL) models have shown effectiveness in natural language processing (NLP) tasks, such as sentiment analysis. However, they often struggle with complex linguistic structures, such as sarcasm and implicit claims. This paper introduces a Quantum Long Short-Term Memory (QLSTM) framework for detecting sarcasm and identifying claims in text, aiming to enhance the analysis of complex sentences. We evaluate four approaches: (1) classical LSTM, (2) quantum framework using QLSTM, (3) voting ensemble combining classical and quantum LSTMs, and (4) hybrid framework integrating both types. The experimental results indicate that the QLSTM approach excels in sarcasm detection, while the voting framework performs best in claim identification.

Claim span identification, a crucial task in Natural Language Processing (NLP), aims to extract specific claims from texts. Such claim spans can be further utilized in various critical NLP applications, such as claim verification, fact-checking, and opinion mining, among others. The present work proposes a multilingual claim span identification framework for handling social media data in English, Hindi, Bengali, and CodeMixed texts, leveraging the strengths and knowledge of transformer-based pre-trained models. Our proposed framework efficiently identifies the contextual relationships between words and precisely detects claim spans across all languages, achieving a high F1 score and Jaccard score. The source code and datasets are available at: https://github.com/pritampal98/claim-span-multilingual

Fact-checkers are often hampered by the sheer amount of online content that needs to be fact-checked. NLP can help them by retrieving already existing fact-checks relevant to the content being investigated. This paper presents a systematic approach for the retrieval of top-k relevant fact-checks for a given post in a monolingual and cross-lingual setup using transformer-based pre-trained models fine-tuned with a dual encoder architecture. By training and evaluating the shared task test dataset, our proposed best-performing framework achieved an average success@10 score of 0.79 and 0.62 for the retrieval of 10 fact-checks from the fact-check corpus against a post in monolingual and crosslingual track respectively.

With the advancement of natural language processing (NLP) and sophisticated Large Language Models (LLMs), distinguishing between human-written texts and machine-generated texts is quite difficult nowadays. This paper presents a systematic approach to classifying machine-generated text from human-written text with a combination of the transformer-based model and textual feature-based post-processing technique. We extracted five textual features: readability score, stop word score, spelling and grammatical error count, unique word score and human phrase count from both human-written and machine-generated texts separately and trained three machine learning models (SVM, Random Forest and XGBoost) with these scores. Along with exploring traditional machine-learning models, we explored the BiLSTM and transformer-based distilBERT models to enhance the classification performance. By training and evaluating with a large dataset containing both human-written and machine-generated text, our best-performing framework achieves an accuracy of 87.5%.