Machine Translation (MT) is widely employed to address resource scarcity in low-resource languages by translating data from high-resource languages. While sentiment preservation in translation has long been studied, a critical but underexplored factor is the role of cultural alignment between source and target languages. In this paper, we hypothesize that semantic labels drift or are altered during MT due to cultural divergence. Through a series of experiments across culturally sensitive and neutral domains, we establish three key findings: (1) MT systems, including modern Large Language Models (LLMs), induce label drift during translation, particularly in culturally sensitive domains; (2) unlike earlier statistical MT tools, LLMs encode cultural knowledge, and leveraging this knowledge can amplify label drift; and (3) cultural similarity or dissimilarity between source and target languages is a crucial determinant of label preservation. Our findings highlight that neglecting cultural factors in MT not only undermines label fidelity but also risks misinterpretation and cultural conflict in downstream applications. We release our codebase to facilitate future research in cross-cultural translation: https://github.com/mohsinulkabir14/label_drift

In this paper, we introduce a novel weighted co-training approach that is guided by Large Language Models (LLMs). Namely, in our co-training approach, we use LLM labels on unlabeled data as target labels and co-train two encoder-only based networks that train each other over multiple iterations: first, all samples are forwarded through each network and historical estimates of each network’s confidence in the LLM label are recorded; second, a dynamic importance weight is derived for each sample according to each network’s belief (or confidence) in the quality of the LLM label for that sample; finally, the two networks exchange importance weights with each other—each network back-propagates all samples weighted with the importance weights coming from its peer network and updates its own parameters. By strategically utilizing LLM-generated guidance, our approach significantly outperforms conventional SSL methods, particularly in settings with abundant unlabeled data. Empirical results show that it achieves state-of-the-art performance on 4 out of 5 benchmark datasets and ranks first among 14 compared methods according to the Friedman test. Our results highlight a new direction in semi-supervised learning—where LLMs serve as knowledge amplifiers, enabling backbone co-training models to achieve SOTA performance efficiently.

This paper presents an overview of the BLP 2025 shared task Code Generation in Bangla, organized with the BLP workshop co-located with AACL. The task evaluates Generative AI systems capable of generating executable Python code from natural language prompts written in Bangla. This is the first shared task to address Bangla code generation. It attracted 152 participants across 63 teams, yielding 488 submissions, with 15 system-description papers. Participating teams employed both proprietary and open-source LLMs, with prevalent strategies including prompt engineering, fine-tuning, and machine translation. The top Pass@1 reached 0.99 on the development phase and 0.95 on the test phase. In this report, we detail the task design, data, and evaluation protocol, and synthesize methodological trends observed across submissions. Notably, we observe that the high performance is not based on single models; rather, a pipeline of multiple AI tools and/or methods.

High-resource languages, such as English, have access to a plethora of datasets with various question-answer types resembling real-world reading comprehension. However, there is a severe lack of diverse and comprehensive question-answering datasets in under-resourced languages like Bangla. The ones available are either translated versions of English datasets with a niche answer format or created by human annotations focusing on a specific domain, question type, or answer type. To address these limitations, this paper introduces BanglaRQA, a reading comprehension-based Bangla question-answering dataset with various question-answer types. BanglaRQA consists of 3,000 context passages and 14,889 question-answer pairs created from those passages. The dataset comprises answerable and unanswerable questions covering four unique categories of questions and three types of answers. In addition, this paper also implemented four different Transformer models for question-answering on the proposed dataset. The best-performing model achieved an overall 62.42% EM and 78.11% F1 score. However, detailed analyses showed that the performance varies across question-answer types, leaving room for substantial improvement of the model performance. Furthermore, we demonstrated the effectiveness of BanglaRQA as a training resource by showing strong results on the bn_squad dataset. Therefore, BanglaRQA has the potential to contribute to the advancement of future research by enhancing the capability of language models. The dataset and codes are available at https://github.com/sartajekram419/BanglaRQA