Publications
Conference papers, journal articles and workshop papers.
conference papers
-
LREC 2026NepTam: A Nepali-Tamang Parallel Corpus and Baseline Machine Translation ExperimentsPublishedPDF
Bib
@inproceedings{ghimire-etal-2026-neptam, title = "{N}ep{T}am: A {N}epali-{T}amang Parallel Corpus and Baseline Machine Translation Experiments", author = "Ghimire, Rupak Raj and Subedi, Bipesh and Prasain, Balaram and Poudyal, Prakash and Acharya, Praveen and Karki, Nischal and Tiwari, Rupak and Sharma, Rishikesh Kumar and Poudel, Jenny and Bal, Bal Krishna", editor = "Piperidis, Stelios and Bel, N{\'u}ria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio", booktitle = "Proceedings of the Fifteenth Language Resources and Evaluation Conference", month = may, year = "2026", address = "Palma de Mallorca, Spain", publisher = "ELRA Language Resource Association", url = "https://aclanthology.org/2026.lrec-1.696/", doi = "10.63317/37edei5qcjb3", pages = "8850--8861", abstract = "Modern Translation Systems heavily rely on high-quality, large parallel datasets for state-of-the-art performance. However, such resources are largely unavailable for most of the South Asian languages. Among them, Nepali and Tamang fall into such category, with Tamang being among the least digitally resourced languages in the region. This work addresses the gap by developing NepTam20K, a 20K gold standard parallel corpus, and NepTam80K, an 80K synthetic Nepali{--}Tamang parallel corpus, both sentence-aligned and designed to support machine translation. The datasets were created through a pipeline involving data scraping from Nepali news and online sources, pre-processing, semantic filtering, balancing for tense and polarity (in NepTam20K dataset), expert translation into Tamang by native speakers of the language, and verification by an expert Tamang linguist. The dataset covers five domains: Agriculture, Health, Education and Technology, Culture, and General Communication. To evaluate the dataset, baseline machine translation experiments were carried out using various multilingual pre-trained models:mBART, M2M-100, NLLB-200, and a vanilla Transformer model. The fine-tuning on the NLLB-200 achieved the highest sacreBLEU scores of 40.92 (Nepali {\textrightarrow} Tamang) and 45.26 (Tamang {\textrightarrow} Nepali)." } -
ICAIL 2026Enhanced Legal Argument Mining through Parameter-Efficient Fine-tuning of Large Language Models and Retrieval Augmentation on ECHR CorpusAccepted First author
-
ICON 2026Trilingual Machine Translation for English, Nepali, and Tamang: Corpus Development and EvaluationAccepted
-
ICON 2026Domain Adaptation of Small Language Models through Continual Pre-Training and Supervised Fine-Tuning on Nepali Legal CorpusAccepted
-
RegICON 2025Benchmarking BERT-based Models for Sentence-level Topic Classification in Nepali LanguagePublished First authorPDF
Bib
@misc{karki2026benchmarkingbertbasedmodelssentencelevel, title={Benchmarking BERT-based Models for Sentence-level Topic Classification in Nepali Language}, author={Nischal Karki and Bipesh Subedi and Prakash Poudyal and Rupak Raj Ghimire and Bal Krishna Bal}, year={2026}, eprint={2602.23940}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2602.23940}, }
journal articles
-
NLP (CUP)From Promises to Accountability: Comparative AI Modeling and Visualization of Electoral Commitments in Nepali Political TextsarXivPDF
Bib
@misc{poudyal2026promisesaccountabilitycomparativeai, title={From Promises to Accountability: Comparative AI Modeling and Visualization of Electoral Commitments in Nepali Political Texts}, author={Prakash Poudyal and Nischal Karki and Amrit Dahal and Rajani Chulyadyo and Sameer Tamrakar and Bal Krishna Bal}, year={2026}, eprint={2610.32150}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2610.32150}, }
workshop papers
-
WMT 2026Bidirectional Finetuning of NLLB-200 for Low-Resource Indic Languages: Assamese-English and Bodo-English Machine TranslationPublishedPDF
Bib
@inproceedings{subedi-karki-2026-brs, title = "{BRS}-{NK}: Bidirectional Finetuning of {NLLB}-200 for Low-Resource {I}ndic Languages: {A}ssamese-{E}nglish and {B}odo-{E}nglish Machine Translation", author = "Subedi, Bipesh and Karki, Nischal", editor = "Haddow, Barry and Kocmi, Tom and Koehn, Philipp and Monz, Christof", booktitle = "Proceedings of the Eleventh Conference on Machine Translation", month = oct, year = "2026", address = "Budapest, Hungary", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2026.wmt-main.164/", doi = "10.18653/2026.wmt-main.164", pages = "2239--2246", ISBN = "979-8-89176-474-3", abstract = "This paper describes our submission to the WMT 2026 Low-Resource Indic Language Translation Shared Task for the English-Assamese and English-Bodo pairs in both translation directions. We fine-tune the pretrained NLLB-200 distilled 600M model for primary as well as contrastive tasks. Since Bodo is not supported by NLLB-200, we use the Hindi (hin{\_}Deva) language tag as a proxy for Bodo during fine-tuning and inference, based on an empirical evaluation of Devanagari-script language tags. The constrained systems are trained exclusively on the official task data, while the contrastive systems additionally incorporate publicly available parallel corpora, including AI4Bharat BPCC and Google Smol. A rigorous preprocessing pipeline, combined with the extra parallel data, yields consistent gains across all automatic metrics, with the largest improvements observed for the lower-resource English-Bodo pair." }