NepTam: A Nepali-Tamang Parallel Corpus and Baseline Machine Translation Experiments

Rupak Raj Ghimire; Bipesh Subedi; Balaram Prasain; Prakash Poudyal; Praveen Acharya; Nischal Karki; Rupak Tiwari; Rishikesh Kumar Sharma

cs.CLMar 14, 2026

NepTam: A Nepali-Tamang Parallel Corpus and Baseline Machine Translation Experiments

Authors: Rupak Raj Ghimire, Bipesh Subedi, Balaram Prasain, Prakash Poudyal, Praveen Acharya, Nischal Karki, Rupak Tiwari, Rishikesh Kumar Sharma,

Organizations: ILPRL, Kathmandu University, Nepal · Tribhuvan University, Nepal · Dublin City University, Ireland

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-Tamang) and 45.26 (Tamang-Nepali).

Figures & tables

Figure 1: Overall procedure of the proposed Nepali–Tamang Machine Translation resource development

Before	\nep काठमाडौं: आजको मौसम राम्रो छ (Nepali)
	kathmandu: aajako mausam ramro chha
	Kathmandu: Today’s weather is good
After	\nep आजको मौसम राम्रो छ। (Nepali)
	aajako mausam ramro chha.
	Today’s weather is good.
Before	\nep _घरमा खाना पकाउँदैछ , । (Nepali)

Table 1: Example of Text Before and After pre-processing

Tense	Pattern (Verb)
Non Past	\nep छु, छौँ, छस्, छेस्, छौ, छ, छे, छन्, छिन्, छौं, दिनँ, दैनौँ, दैनस्, दैनौ, दिनौ
Past	\nep एँ, यौँ, यौं, इस्, यौ, यो, ई, ए, इन्, इनँ, एनौ, इनस्, इनौ, एन, इन, एनन्, इनन्
Unknown	Does not end with the pattern

Table 2: Tense Classification based on Verb Patterns Prasain (2011)

Polarity	Pattern (Verb)
Affirmative	\nep छु, छौँ, छस्, छेस्, छौ, छ, छे, छन्, छिन्, एँ, यौँ, यौं, इस्, यौ, यो, ई, ए
Negative	\nep दिनँ, दैनौँ, दैनस्, दैनौ, दिनौ, इनौ, एन, इन, एनन्, इनन्
Unknown	Does not end with the pattern

Table 3: Polarity Classification based on Verb Patterns Prasain (2011)

Category	NepTam20K		NepTam80K
	Train	Test
Sentence Length
Short	5,863 (39.1%)	1,954 (39.1%)	15,941 (19.9%)
Medium	6,793 (45.3%)	2,266 (45.3%)	52,122 (65.2%)
Long	1,552 (10.3%)	518 (10.4%)	10,640 (13.3%)
Very Long	823 (5.5%)	262 (5.2%)	1,396 (1.7%)

Table 4: Category-wise Data Distribution in NepTam20K and NepTam80K Dataset.

Model	Hyperparameter	Value
$NepTam_{mBART}$	Epochs	7
	Batch size	8
	Grad. accumulation	2
	Learning rate	7e-5
	Weight decay	0.01
$NepTam_{M2M}$	Epochs	5

Table 5: Training and fine-tuning configurations for all models on NepTam20K.

Model	Parameter Size	Direction	sacreBLEU	chrF	chrF++	METEOR	COMET
$NepTam_{M2M}$	418M	ne $\rightarrow$ tam	40.24	73.30	69.27	60.81	75.67
		tam $\rightarrow$ ne	42.73	72.21	68.74	61.07	79.13
$NepTam_{NLLB}$	600M	ne $\rightarrow$ tam	40.92	73.98	69.94	61.44	75.89
		tam $\rightarrow$ ne	45.26	73.71	70.30	62.31	80.19
$NepTam_{mBART}$	610.9M	ne $\rightarrow$ tam	40.14	72.91	68.92	60.46	75.60
		tam $\rightarrow$ ne	42.96	71.65	68.25	60.43	79.04

Table 6: Performance of translation models trained on NepTam20K (Train), evaluated on NepTam20K (Test) set for Nepali–Tamang (ne

\rightarrow

tam) and Tamang–Nepali (tam

\rightarrow

ne) directions

Model	Direction	sacreBLEU	chrF	chrF++	METEOR	COMET
$NepTam_{M2M}$	ne $\rightarrow$ tam	40.56	73.70	69.62	60.55	75.68
	tam $\rightarrow$ ne	44.03	72.83	69.34	60.50	79.28
$NepTam_{NLLB}$	ne $\rightarrow$ tam	41.79	74.58	70.60	62.19	76.12
	tam $\rightarrow$ ne	48.28	75.57	72.31	64.25	81.26
$NepTam_{mBART}$	ne $\rightarrow$ tam	41.44	74.07	70.10	61.21	75.83
	tam $\rightarrow$ ne	46.68	74.59	71.17	61.96	80.70

Table 7: Performance of translation models trained on NepTam80K(Train), evaluated on NepTam20K Test set for Nepali–Tamang (ne

\rightarrow

tam) and Tamang–Nepali (tam

\rightarrow

ne) directions

Low-Resource MT

Low-resource machine translation develops translation systems for languages with scarce parallel text, tools, or standardized evaluation.

See all 85 papers

Samasāmayik: A Parallel Dataset for Hindi-Sanskrit Machine Translation
Mar 25, 2026N J Karthika, Keerthana Suryanarayanan, Jahanvi Purohit +3Machine Translation Low-Resource MT
Indian Institute of Technology Bombay · Geakminds Technologies Private Limited · Indian Institute of Technology Roorkee
Neural Machine Translation for Low-Resource Tangkhul--English
Jun 24, 2026Chormi Zimik Vashai, Agniva MaitiLow-Resource MT Low-Resource Language Processing
Independent Researcher · KIIT University
COILD: An Indic-Centric Parallel Corpus and Benchmark for Machine Translation Across Indian Languages
Sep 23, 2026Kshetrimayum Boynao Singh, Nitin Kumar Mishra, Palash Pratim Dutta +22Low-Resource MT Machine Translation
IIT Patna · IIT Delhi · IIT Guwahati +3

Mar 25, 2026J/K move · Enter open · S save

Samasāmayik: A Parallel Dataset for Hindi-Sanskrit Machine Translation

N J Karthika, Keerthana Suryanarayanan, Jahanvi Purohit, Ganesh Ramakrishnan+2

Indian Institute of Technology Bombay · Geakminds Technologies Private Limited · Indian Institute of Technology Roorkee

Machine Translation Low-Resource MT

Open paper AskPDF arXiv

We release Samasāmayik, a novel, meticulously curated, large-scale Hindi-Sanskrit corpus, comprising 92,196 parallel sentences. Unlike most data available in Sanskrit, which focuses on classical era text and poetry, this corpus aggregates data from diverse sources covering contemporary materials, including spoken tutorials, children's magazines, radio conversations, and instruction materials. We benchmark this new dataset by fine-tuning three complementary models - ByT5, NLLB and IndicTrans-v2, to demonstrate its utility. Our experiments demonstrate that models trained on the Samasamayik corpus achieve significant performance gains on in-domain test data, while achieving comparable performance on other widely used test sets, establishing a strong new performance baseline for contemporary Hindi-Sanskrit translation. Furthermore, a comparative analysis against existing corpora reveals minimal semantic and lexical overlap, confirming the novelty and non-redundancy of our dataset as a robust new resource for low-resource Indic language MT.

Abstract

Low-Resource MT

Explore similar work

Samasāmayik: A Parallel Dataset for Hindi-Sanskrit Machine Translation

Neural Machine Translation for Low-Resource Tangkhul--English

COILD: An Indic-Centric Parallel Corpus and Benchmark for Machine Translation Across Indian Languages