TajPersParallelCorpus
Tajik–Persian Parallel Corpus is a large-scale parallel corpus for Tajik and Persian.
Description
This dataset contains aligned Tajik–Persian sentence pairs collected from multiple sources, including news, poetry, prose, lexical resources, and named-entity lists.
It is intended for machine translation, cross-lingual retrieval, linguistic analysis, tokenizer evaluation, and other NLP tasks.
Dataset Statistics
Overview
| Metric |
Value |
| Sentence pairs |
328,253 |
| Unique Tajik sentences |
328,253 |
| Unique Persian sentences |
328,253 |
| Total Tajik tokens |
1,888,484 |
| Total Persian tokens |
1,984,629 |
| Unique Tajik words |
253,320 |
| Unique Persian words |
169,174 |
| Avg. Tajik tokens / sentence |
5.75 |
| Avg. Persian tokens / sentence |
6.05 |
| Avg. Tajik chars / sentence |
33.35 |
| Avg. Persian chars / sentence |
28.0 |
| Max Tajik tokens / sentence |
157 |
| Max Persian tokens / sentence |
161 |
| Max Tajik chars / sentence |
973 |
| Max Persian chars / sentence |
838 |
Category Distribution
| Category |
Count |
Share |
| poetry_parts |
154,891 |
47.19% |
| masnavi |
38,972 |
11.87% |
| unique_tajik_words |
38,151 |
11.62% |
| shahnameh |
24,819 |
7.56% |
| prose_parts |
23,936 |
7.29% |
| paranames_per |
20,101 |
6.12% |
| words |
14,292 |
4.35% |
| paranames_loc |
9,586 |
2.92% |
| dr |
1,542 |
0.47% |
| jj |
803 |
0.24% |
| paranames_org |
722 |
0.22% |
| bbc |
438 |
0.13% |
Source Distribution
| Source |
Count |
Share |
| unknown |
328,253 |
100.00% |
Data Format
Each record has the following fields:
| Field |
Type |
Description |
farsi |
string |
Persian side of the aligned pair |
tajik |
string |
Tajik side of the aligned pair |
category |
string |
File-based category, such as poetry_parts, prose_parts, or words |
source |
string |
Optional source label if available |
Example
{
"farsi": "در اقصای عالم بگشتم بسی",
"tajik": "Дар ақсои олам бигаштам басе,",
"category": "poetry_parts"
}
Intended Uses
- Tajik–Persian machine translation
- Cross-lingual lexical studies
- Subword tokenization experiments
- Corpus linguistics
- Named-entity and domain adaptation research
Limitations
- The corpus is heterogeneous and domain-imbalanced.
- Some categories are much larger than others, especially poetic and literary text.
- Token counts are based on whitespace tokenization.
Citation
@dataset{tajpersparallelcorpus},
title = {Tajik–Persian Parallel Corpus},
author = {TajikNLPWorld},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/TajikNLPWorld/TajPersParallelCorpus}
}
