This paper presents the findings of the Lowresource Indic Languages Translation Shared Task organized at the Eleventh Conference on Machine Translation (WMT 2026). We evaluated machine translation systems across ten English–Indic language pairs spanning three major language families of Northeast India, grouped into moderately resourced and extremely low-resource categories. Built upon the expanded INDICNE-CORP 2.0 dataset, this edition introduced a rigorous, bidirectional multidomain benchmark covering the Healthcare, Political, Travel, Sports, and Entertainment domains to assess real-world generalizability. The task attracted 24 participating teams who deployed diverse methodologies, including parameter-efficient fine-tuning (PEFT) of multilingual foundations, retrieval-augmented generation (RAG) using frontier large language models (LLMs), and hybrid neuro-symbolic decoding constraints. The systems were evaluated using a comprehensive suite of lexical metrics (BLEU, METEOR, TER, ChrF++) and semantic metrics (BERTScore, COMET). Our analysis reveals a persistent performance cliff under extreme data scarcity and a pronounced directional asymmetry, demonstrating that while current architectures readily decode Indic representations into English, they struggle to generate morphologically complex Indic target text. By publicly releasing these datasets, evaluation protocols, and competitive baselines, this shared task establishes a foundation for advancing research in translation technologies for underrepresented and indigenous languages.

This study proposes the results of the lowresource Indic language translation task organized in collaboration with the Tenth Conference on Machine Translation (WMT) 2025. In this workshop, participants were required to build and develop machine translation models for the seven language pairs, which were categorized into two categories. Category 1 is moderate training data available in languages i.e English–Assamese, English–Mizo, English-Khasi, English–Manipuri and English– Nyishi. Category 2 has very limited training data available in languages, i.e English–Bodo and English–Kokborok. This task leverages the enriched IndicNE-corp1.0 dataset, which consists of an extensive collection of parallel and monilingual corpora for north eastern Indic languages. The participant results were evaluated using automatic machine translation metrics, including BLEU, TER, ROUGE-L, ChrF, and METEOR. Along with those metrics, this year’s work also includes Cosine similarity for evaluation, which captures the semantic representation of the sentence to measure the performance and accuracy of the models. This work aims to promote innovation and advancements in low-resource Indic languages.

This paper presents the results of the low-resource Indic language translation task, organized in conjunction with the Ninth Conference on Machine Translation (WMT) 2024. In this edition, participants were challenged to develop machine translation models for four distinct language pairs: English–Assamese, English-Mizo, English-Khasi, and English-Manipuri. The task utilized the enriched IndicNE-Corp1.0 dataset, which includes an extensive collection of parallel and monolingual corpora for northeastern Indic languages. The evaluation was conducted through a comprehensive suite of automatic metrics—BLEU, TER, RIBES, METEOR, and ChrF—supplemented by meticulous human assessment to measure the translation systems’ performance and accuracy. This initiative aims to drive advancements in low-resource machine translation and make a substantial contribution to the growing body of knowledge in this dynamic field.

This paper presents the results of the low-resource Indic language translation task organized alongside the Eighth Conference on Machine Translation (WMT) 2023. In this task, participants were asked to build machine translation systems for any of four language pairs, namely, English-Assamese, English-Mizo, English-Khasi, and English-Manipuri. For this task, the IndicNE-Corp1.0 dataset is released, which consists of parallel and monolingual corpora for northeastern Indic languages such as Assamese, Mizo, Khasi, and Manipuri. The evaluation will be carried out using automatic evaluation metrics (BLEU, TER, RIBES, COMET, ChrF) and human evaluation.

Unsupervised Machine Translation (MT) model, which has the ability to perform MT without parallel sentences using comparable corpora, is becoming a promising approach for developing MT in low-resource languages. However, majority of the studies in unsupervised MT have considered resource-rich language pairs with similar linguistic characteristics. In this paper, we investigate the effectiveness of unsupervised MT models over a Manipuri-English comparable corpus. Manipuri is a low-resource language having different linguistic characteristics from that of English. This paper focuses on identifying challenges in building unsupervised MT models over the comparable corpus. From various experimental observations, it is evident that the development of MT over comparable corpus using unsupervised methods is feasible. Further, the paper also identifies future directions of developing effective MT for Manipuri-English language pair under unsupervised scenarios.

Development of hand crafted rule for syllabifying words of a language is an expensive task. This paper proposes several data-driven methods for automatic syllabification of words written in Manipuri language. Manipuri is one of the scheduled Indian languages. First, we propose a language-independent rule-based approach formulated using entropy based phonotactic segmentation. Second, we project the syllabification problem as a sequence labeling problem and investigate its effect using various sequence labeling approaches. Third, we combine the effect of sequence labeling and rule-based method and investigate the performance of the hybrid approach. From various experimental observations, it is evident that the proposed methods outperform the baseline rule-based method. The entropy based phonotactic segmentation provides a word accuracy of 96%, CRF (sequence labeling approach) provides 97% and hybrid approach provides 98% word accuracy.