BHT25: Bengali-Hindi-Telugu Parallel Corpus for Literary Machine Translation

Overview

BHT25 is a high-quality trilingual parallel corpus comprising 25,000 sentence triplets across Bengali (BN), Hindi (HI), and Telugu (TE) languages. This dataset fills a critical gap in resources for cross-family Indian language machine translation, particularly for literary and culturally rich content.

Key Features:

• 🌐 Three Indian Languages: Bengali (Indo-Aryan), Hindi (Indo-Aryan), Telugu (Dravidian)
• 📚 Literary Domain: Sourced from works of renowned authors including Rabindranath Tagore and Sarat Chandra Chattopadhyay
• ✨ Archaic Varieties: Includes traditional Bengali Sadhu Bhasha
• ✅ Human-Verified: 75.8% semantic alignment accuracy with expert validation
• 🔍 Unique Identifiers: Each triplet has a unique ID for reproducibility
• 📖 High Quality: Composite fluency score of 4.08/5.0

Dataset Description

Languages

• Bengali (bn): Eastern Indo-Aryan language, ~265M speakers
• Hindi (hi): Central Indo-Aryan language, ~600M speakers
• Telugu (te): South-Central Dravidian language, ~95M speakers

Dataset Statistics

Metric	Bengali	Hindi	Telugu
Total Sentences	25,000	25,000	25,000
Total Tokens	~XXX,XXX	~XXX,XXX	~XXX,XXX
Avg. Tokens/Sentence	~XX	~XX	~XX
Vocabulary Size	~XX,XXX	~XX,XXX	~XX,XXX
Min Sentence Length	X tokens	X tokens	X tokens
Max Sentence Length	XXX tokens	XXX tokens	XXX tokens

Content Characteristics

The corpus covers diverse literary genres:

• Short stories and narratives
• Poetry and prose
• Folk literature
• Contemporary literary works
• Traditional Sadhu Bhasha (classical Bengali)

Quality Metrics

• Alignment Accuracy: 75.8% (human-validated)
• Semantic Consistency: 0.873 (Bengali-Telugu), 0.71 (Hindi-based pairs via CLWE)
• Translation Fluency: 4.08/5.0 (composite expert score)
• Inter-Annotator Agreement: κ = 0.89

Dataset Structure

Data Format

The dataset is provided in Parquet format with the following schema:

{
    'id': string,        # Unique identifier (BHT25_00001 to BHT25_25000)
    'bn': string,        # Bengali sentence
    'hi': string,        # Hindi sentence
    'te': string         # Telugu sentence
}

Example

{
    'id': 'BHT25_00001',
    'bn': 'আমি তোমাকে ভালোবাসি।',
    'hi': 'मैं तुमसे प्यार करता हूँ।',
    'te': 'నేను నిన్ను ప్రేమిస్తున్నాను।'
}

Data Splits

The dataset is provided as a single unified corpus without pre-defined train/dev/test splits. This design choice allows researchers maximum flexibility to partition the data according to their specific experimental requirements. The unique identifiers enable deterministic and reproducible splitting across different studies.

Suggested Split Strategy (for standardization):

• Train: 80% (20,000 triplets)
• Development: 10% (2,500 triplets)
• Test: 10% (2,500 triplets)

Users can implement splits using the unique IDs with their preferred strategy (random, stratified, etc.).

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("sudeshna84/BHT25")

# Access data
print(f"Total samples: {len(dataset['train'])}")
print(f"First example: {dataset['train'][0]}")

Creating Train/Dev/Test Splits

from datasets import load_dataset

# Load dataset
dataset = load_dataset("sudeshna84/BHT25", split="train")

# Create 80-10-10 split
train_test = dataset.train_test_split(test_size=0.2, seed=42)
train_dataset = train_test['train']
temp_dataset = train_test['test']

# Further split test into dev and test
dev_test = temp_dataset.train_test_split(test_size=0.5, seed=42)
dev_dataset = dev_test['train']
test_dataset = dev_test['test']

print(f"Train: {len(train_dataset)}, Dev: {len(dev_dataset)}, Test: {len(test_dataset)}")

Accessing Specific Language Pairs

# Extract Bengali-Hindi pairs
bn_hi_pairs = [(item['bn'], item['hi']) for item in dataset['train']]

# Extract Bengali-Telugu pairs
bn_te_pairs = [(item['bn'], item['te']) for item in dataset['train']]

# Extract Hindi-Telugu pairs
hi_te_pairs = [(item['hi'], item['te']) for item in dataset['train']]

Integration with Translation Models

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from datasets import load_dataset

# Load dataset
dataset = load_dataset("sudeshna84/BHT25", split="train")

# Load model (example: IndicTrans2)
tokenizer = AutoTokenizer.from_pretrained("ai4bharat/indictrans2-bn-hi")
model = AutoModelForSeq2SeqLM.from_pretrained("ai4bharat/indictrans2-bn-hi")

# Prepare data for training
def preprocess_function(examples):
    inputs = examples['bn']
    targets = examples['hi']
    model_inputs = tokenizer(inputs, max_length=128, truncation=True, padding='max_length')
    labels = tokenizer(targets, max_length=128, truncation=True, padding='max_length')
    model_inputs["labels"] = labels["input_ids"]
    return model_inputs

tokenized_dataset = dataset.map(preprocess_function, batched=True)

Data Analysis Example

import pandas as pd
from datasets import load_dataset

# Load dataset
dataset = load_dataset("sudeshna84/BHT25", split="train")

# Convert to pandas for analysis
df = pd.DataFrame(dataset)

# Analyze sentence lengths
df['bn_length'] = df['bn'].str.split().str.len()
df['hi_length'] = df['hi'].str.split().str.len()
df['te_length'] = df['te'].str.split().str.len()

print("Average sentence lengths:")
print(f"Bengali: {df['bn_length'].mean():.2f} tokens")
print(f"Hindi: {df['hi_length'].mean():.2f} tokens")
print(f"Telugu: {df['te_length'].mean():.2f} tokens")

# Length distribution
import matplotlib.pyplot as plt
df[['bn_length', 'hi_length', 'te_length']].hist(bins=50, figsize=(15, 5))
plt.suptitle('Sentence Length Distribution')
plt.show()

Applications

This corpus can be used for various NLP tasks:

1. Machine Translation

   • Neural machine translation training and evaluation
   • Cross-family translation (Indo-Aryan ↔ Dravidian)
   • Low-resource language pair modeling

2. Cross-Lingual Analysis

   • Cross-lingual word embeddings evaluation
   • Semantic similarity studies
   • Syntactic divergence analysis

3. Multilingual NLP

   • Multilingual language model fine-tuning
   • Zero-shot translation experiments
   • Transfer learning for Indian languages

4. Linguistic Research

   • Comparative morphology and syntax
   • Literary translation analysis
   • Archaic language preservation (Sadhu Bhasha)

5. Quality Estimation

   • Automatic alignment algorithm evaluation
   • Translation quality metrics benchmarking

Methodology

Data Collection

The corpus was constructed through:

1. Source Selection: Literary texts from renowned Indian authors
2. Sentence Extraction: Language-specific boundary detection
3. Alignment: Hybrid Gale-Church algorithm enhanced with cross-lingual word embeddings (CLWE)
4. Validation: Manual verification by native speakers

Quality Assurance

Each sentence triplet underwent:

• Automated preprocessing and alignment scoring
• Manual verification for semantic equivalence
• Expert review for fluency and naturalness
• Inter-annotator agreement measurement (κ = 0.89)

Alignment Algorithm

A hybrid approach combining:

• Gale-Church length-based alignment
• Cross-lingual word embedding (FastText) similarity
• Threshold-based filtering (score ≥ 0.7)
• Human expert correction

Citation

If you use this dataset in your research, please cite:

@article{sani2024bht25,
  title={A Bengali-Hindi-Telugu Parallel Corpus for Enhanced Literary Machine Translation},
  author={Sani, Sudeshna and Gangashetty, Suryakanth V and Samudravijaya, K and Nandi, Anik and Priya, Aruna and Kumar, Vineeth and Dubey, Akhilesh Kumar},
  journal={Data in Brief},
  year={2024},
  publisher={Elsevier}
}

Related Research:

@article{sani2024esa,
  title={Emotion-Semantic-Aware Neural Machine Translation for Bengali-Hindi-Telugu},
  author={Sani, Sudeshna and others},
  journal={IEEE Access},
  year={2024},
  doi={10.1109/ACCESS.2024.XXXXXXX}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

You are free to:

• Share — copy and redistribute the material in any medium or format
• Adapt — remix, transform, and build upon the material for any purpose, even commercially

Under the following terms:

• Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made

Ethical Considerations

• All source texts are either in the public domain or used with explicit permission
• Author attributions are maintained in metadata
• No personal or sensitive information is included
• The dataset supports linguistic diversity and cultural preservation

Authors and Affiliations

Sudeshna Sani¹, Suryakanth V Gangashetty¹, Samudravijaya K¹*, Anik Nandi², Aruna Priya³, Vineeth Kumar³, Akhilesh Kumar Dubey¹

¹ Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Guntur, Andhra Pradesh, India
² School of Business, Woxsen University, Hyderabad, Telangana, India
³ School of Technology, Woxsen University, Hyderabad, Telangana, India

*Corresponding author

Contact

For questions, issues, or collaborations, please:

• Open an issue in this repository
• Contact: [your.email@klef.edu.in]

Acknowledgments

We thank the native speakers who participated in the manual verification process and the literary estates for permission to use source materials.

Dataset Card

For more detailed information about the dataset construction, evaluation metrics, and validation procedures, please refer to our paper in Data in Brief.

---

Last Updated: December 2024
Version: 1.0
DOI: [Will be added upon publication]