Saeed Almheiri, Bilal Elbouardi, Salsabila Zahirah Pranida, Irina Nikishina, Ashwath Rao B, Parameswari Krishnamurthy, Muhammad Cendekia Airlangga, Rifo Ahmad Genadi, Nguyen Phan Gia Bao, Amir Hossein Yari, Hawau Olamide Toyin, Nurdaulet Mukhituly, Mena Attia, Besher Hassan, Ahmad Fathan Hidayatullah, Tatsuki Kuribayashi, Haonan Li, Suma Bhat, Fajri Koto. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Discourse-level translation remains a major challenge for machine translation (MT) systems, particularly for translation from Indian languages to English. This difficulty arises due to factors such as rich morphology, diverse syntactic structures, unmarked gender distinctions in pronouns, and the limited availability of discourse-aware training data. Existing evaluation benchmarks primarily focus on sentence-level translation quality and fail to capture important discourse phenomena such as pronoun resolution and lexical cohesion. To address this gap, we introduce IndicDISCO-MT, a parallel benchmark dataset covering translations from eight Indian languages such as Bengali, Gujarati, Hindi, Marathi, Kannada, Tamil, Telugu and Urdu to English. On top of this dataset, we are first to introduce DiscoAlign, a human annotated word to word alignment benchmark dataset, that captures correspondences between source and target words across languages. In addition, we propose two evaluation benchmarks, ProAlign and LexiAlign, designed to specifically assess the ability of large language models (LLMs) and MT systems to handle personal pronouns and lexical cohesion. Our evaluation of recent LLMs and MT systems on these benchmarks shows that although models achieve high overall translation quality, they still struggle to accurately preserve discourse-level phenomena. The proposed benchmarks provide a systematic framework for evaluating discourse aware translation and can facilitate the development of MT systems that generate more coherent and contextually consistent translations.

The extent to which large language models (LLMs) can perform culturally grounded reasoning across non-English languages remains underexplored. This paper examines the reasoning and self-assessment abilities of LLMs across seven major Indian languages-Bengali, Gujarati, Hindi, Kannada, Malayalam, Tamil, and Telugu. We introduce a multilingual riddle dataset combining traditional riddles with context-reconstructed variants and evaluate five LLMs-Gemini 2.5 Pro, Gemini 2.5 Flash, Mistral-Saba, LLaMA 4 Scout, and LLaMA 4 Maverick-under seven prompting strategies. In the first stage, we assess riddle-solving performance and find that while Gemini 2.5 Pro performs best overall, few-shot methods yield only marginal gains, and accuracy varies notably across languages. In the second stage, we conduct a self-evaluation experiment to measure reasoning consistency. The results reveal a key finding: a model's initial accuracy is inversely correlated with its ability to identify its own mistakes. Top-performing models such as Gemini 2.5 Pro are overconfident (4.34% True Negative Rate), whereas lower-performing models like LLaMA 4 Scout are substantially more self-aware (42.09% True Negative Rate). These results point to clear gaps in multilingual reasoning and highlight the need for models that not only reason effectively but also recognize their own limitations.

This study addresses the critical challenge of data scarcity in machine translation for Indian languages, particularly given their morphological complexity and limited parallel data. We investigate an effective strategy to maximize the utility of existing data by generating negative samples from positive training instances using a progressive perturbation approach. This is used for aligning the model with preferential data using Kahneman-Tversky Optimization (KTO). Comparing it against traditional Supervised Fine-Tuning (SFT), we demonstrate how generating negative samples and leveraging KTO enhances data efficiency. By creating rejected samples through progressively perturbed translations from the available dataset, we fine-tune the Llama 3.1 Instruct 8B model using QLoRA across 16 language directions, including English, Hindi, Bangla, Tamil, Telugu, and Santali. Our results show that KTO-based preference alignment with progressive perturbation consistently outperforms SFT, achieving significant gains in translation quality with an average BLEU increase of 1.84 to 2.47 and CHRF increase of 2.85 to 4.01 compared to SFT for selected languages, while using the same positive training samples and under similar computational constraints. This highlights the potential of our negative sample generation strategy within KTO, especially in low resource scenarios.

Generative Large Language Models (LLMs) have achieved remarkable advances in various NLP tasks. In this work, our aim is to explore the multilingual capabilities of large language models by using machine translation as a task involving English and 22 Indian languages. We first investigate the translation capabilities of raw large-language models, followed by exploring the in-context learning capabilities of the same raw models. We fine-tune these large language models using parameter-efficient fine-tuning methods such as LoRA and additionally with full fine-tuning. Through our study, we have identified the model that performs best among the large language models available for the translation task.Our results demonstrate significant progress, with average BLEU scores of 13.42, 15.93, 12.13, 12.30, and 12.07, as well as chrF scores of 43.98, 46.99, 42.55, 42.42, and 45.39, respectively, using two-stage fine-tuned LLaMA-13b for English to Indian languages on IN22 (conversational), IN22 (general), flores200-dev, flores200-devtest, and newstest2019 testsets. Similarly, for Indian languages to English, we achieved average BLEU scores of 14.03, 16.65, 16.17, 15.35 and 12.55 along with chrF scores of 36.71, 40.44, 40.26, 39.51, and 36.20, respectively, using fine-tuned LLaMA-13b on IN22 (conversational), IN22 (general), flores200-dev, flores200-devtest and newstest2019 testsets. Overall, our findings highlight the potential and strength of large language models for machine translation capabilities, including languages that are currently underrepresented in LLMs.