| { |
| "dataset_info": { |
| "name": "BHT25", |
| "version": "1.0", |
| "release_date": "2024-12-19", |
| "total_triplets": 27149, |
| "languages": ["bn", "hi", "te"], |
| "license": "CC BY 4.0", |
| "doi": "[To be assigned upon publication]" |
| }, |
| |
| "corpus_statistics": { |
| "bengali": { |
| "language_code": "bn", |
| "language_family": "Indo-Aryan", |
| "script": "Bengali (Bangla)", |
| "total_sentences": 27149, |
| "total_tokens": 420212, |
| "total_characters": 2369637, |
| "vocabulary_size": 47856, |
| "type_token_ratio": 0.1138, |
| "sentence_length_tokens": { |
| "mean": 15.48, |
| "median": 14.0, |
| "std_dev": 8.31, |
| "min": 3, |
| "max": 147, |
| "25th_percentile": 10.0, |
| "75th_percentile": 19.0, |
| "95th_percentile": 29.0 |
| }, |
| "sentence_length_characters": { |
| "mean": 87.31, |
| "median": 78.0, |
| "std_dev": 48.24, |
| "min": 15, |
| "max": 452, |
| "25th_percentile": 56.0, |
| "75th_percentile": 108.0, |
| "95th_percentile": 174.0 |
| } |
| }, |
| |
| "hindi": { |
| "language_code": "hi", |
| "language_family": "Indo-Aryan", |
| "script": "Devanagari", |
| "total_sentences": 27149, |
| "total_tokens": 387548, |
| "total_characters": 2228937, |
| "vocabulary_size": 43214, |
| "type_token_ratio": 0.1115, |
| "sentence_length_tokens": { |
| "mean": 14.27, |
| "median": 13.0, |
| "std_dev": 7.58, |
| "min": 3, |
| "max": 132, |
| "25th_percentile": 9.0, |
| "75th_percentile": 18.0, |
| "95th_percentile": 27.0 |
| }, |
| "sentence_length_characters": { |
| "mean": 82.09, |
| "median": 74.0, |
| "std_dev": 45.67, |
| "min": 14, |
| "max": 605, |
| "25th_percentile": 52.0, |
| "75th_percentile": 102.0, |
| "95th_percentile": 165.0 |
| } |
| }, |
| |
| "telugu": { |
| "language_code": "te", |
| "language_family": "Dravidian", |
| "script": "Telugu", |
| "total_sentences": 27149, |
| "total_tokens": 444965, |
| "total_characters": 2598418, |
| "vocabulary_size": 52143, |
| "type_token_ratio": 0.1172, |
| "sentence_length_tokens": { |
| "mean": 16.39, |
| "median": 15.0, |
| "std_dev": 9.12, |
| "min": 3, |
| "max": 156, |
| "25th_percentile": 10.0, |
| "75th_percentile": 21.0, |
| "95th_percentile": 32.0 |
| }, |
| "sentence_length_characters": { |
| "mean": 95.71, |
| "median": 87.0, |
| "std_dev": 52.63, |
| "min": 17, |
| "max": 633, |
| "25th_percentile": 62.0, |
| "75th_percentile": 119.0, |
| "95th_percentile": 189.0 |
| } |
| } |
| }, |
| |
| "language_pair_statistics": { |
| "bengali_hindi": { |
| "token_length_ratio": 1.084, |
| "character_length_ratio": 1.063, |
| "semantic_consistency_score": 0.81, |
| "notes": "Both Indo-Aryan languages with similar morphology" |
| }, |
| "bengali_telugu": { |
| "token_length_ratio": 0.944, |
| "character_length_ratio": 0.912, |
| "semantic_consistency_score": 0.873, |
| "notes": "Cross-family pair; Telugu agglutinative morphology" |
| }, |
| "hindi_telugu": { |
| "token_length_ratio": 0.871, |
| "character_length_ratio": 0.858, |
| "semantic_consistency_score": 0.71, |
| "notes": "Greatest typological distance; cross-family translation" |
| } |
| }, |
| |
| "quality_metrics": { |
| "alignment": { |
| "method": "Hybrid Gale-Church + CLWE", |
| "alignment_accuracy": 0.758, |
| "validation_sample_size": 500, |
| "perfect_alignment_percentage": 75.8, |
| "partial_alignment_percentage": 18.4, |
| "poor_alignment_percentage": 5.8 |
| }, |
| "fluency": { |
| "rating_scale": "1-5 (1=incomprehensible, 5=perfectly fluent)", |
| "annotators_per_language": 3, |
| "composite_fluency_score": { |
| "mean": 4.08, |
| "std_dev": 0.67, |
| "median": 4.17, |
| "5th_percentile": 2.67, |
| "95th_percentile": 4.83 |
| }, |
| "fluency_by_language": { |
| "bengali": {"mean": 4.15, "std_dev": 0.63}, |
| "hindi": {"mean": 4.02, "std_dev": 0.69}, |
| "telugu": {"mean": 4.07, "std_dev": 0.68} |
| } |
| }, |
| "inter_annotator_agreement": { |
| "metric": "Fleiss_kappa", |
| "semantic_equivalence": 0.89, |
| "fluency_ratings_icc": 0.84, |
| "interpretation": "Almost perfect agreement" |
| }, |
| "semantic_consistency": { |
| "method": "Cross-Lingual Word Embeddings (FastText)", |
| "bn_hi_similarity": 0.81, |
| "bn_te_similarity": 0.873, |
| "hi_te_similarity": 0.71, |
| "notes": "Lower hi-te reflects greater family divergence" |
| } |
| }, |
| |
| "genre_distribution": { |
| "narrative_fiction": { |
| "count": 12271, |
| "percentage": 45.2, |
| "description": "Short stories and novel excerpts" |
| }, |
| "poetry_verse": { |
| "count": 5077, |
| "percentage": 18.7, |
| "description": "Traditional and modern poetry" |
| }, |
| "folk_literature": { |
| "count": 4235, |
| "percentage": 15.6, |
| "description": "Folk tales and oral traditions" |
| }, |
| "contemporary_prose": { |
| "count": 3339, |
| "percentage": 12.3, |
| "description": "Modern literary essays and articles" |
| }, |
| "classical_literature": { |
| "count": 2227, |
| "percentage": 8.2, |
| "description": "Traditional Sadhu Bhasha texts" |
| } |
| }, |
| |
| "vocabulary_coverage": { |
| "bengali": { |
| "top_100_words_coverage": 0.423, |
| "top_1000_words_coverage": 0.687, |
| "top_5000_words_coverage": 0.891, |
| "hapax_legomena_percentage": 42.3 |
| }, |
| "hindi": { |
| "top_100_words_coverage": 0.441, |
| "top_1000_words_coverage": 0.703, |
| "top_5000_words_coverage": 0.897, |
| "hapax_legomena_percentage": 39.8 |
| }, |
| "telugu": { |
| "top_100_words_coverage": 0.389, |
| "top_1000_words_coverage": 0.652, |
| "top_5000_words_coverage": 0.876, |
| "hapax_legomena_percentage": 45.7 |
| } |
| }, |
| |
| "data_collection": { |
| "source_period": "1910-2023", |
| "primary_authors": [ |
| "Rabindranath Tagore (Bengali)", |
| "Sarat Chandra Chattopadhyay (Bengali)", |
| "Premchand (Hindi)", |
| "Mahadevi Verma (Hindi)", |
| "Gurajada Apparao (Telugu)", |
| "Viswanatha Satyanarayana (Telugu)" |
| ], |
| "collection_period": "2022-2024", |
| "digitization_method": "OCR + Manual correction", |
| "ocr_accuracy": { |
| "bengali": 0.972, |
| "hindi": 0.968, |
| "telugu": 0.956 |
| } |
| }, |
| |
| "preprocessing": { |
| "unicode_normalization": "NFKC", |
| "character_filtering": true, |
| "sentence_segmentation": "Rule-based + Statistical", |
| "minimum_sentence_length": 3, |
| "maximum_sentence_length": 150, |
| "duplicate_removal": true, |
| "language_detection_threshold": 0.95 |
| }, |
| |
| "ethical_compliance": { |
| "copyright_cleared": true, |
| "public_domain_percentage": 78.3, |
| "licensed_content_percentage": 21.7, |
| "personal_information_removed": true, |
| "offensive_content_filtered": true, |
| "excluded_sentences_ethical": 78, |
| "annotator_compensation": "₹500/hour", |
| "work_hours_limit": "4 hours/day" |
| }, |
| |
| "known_limitations": { |
| "alignment_issues": { |
| "percentage": 5.8, |
| "description": "Suboptimal alignment in complex literary passages" |
| }, |
| "genre_imbalance": { |
| "dominant_genre": "narrative_fiction", |
| "percentage": 45.2, |
| "mitigation": "Users can filter by genre metadata" |
| }, |
| "archaic_language": { |
| "sadhu_bhasha_percentage": 8.2, |
| "description": "Traditional Bengali may challenge modern systems" |
| }, |
| "residual_ocr_errors": { |
| "estimated_rate": 0.005, |
| "description": "Rare characters and conjuncts most affected" |
| } |
| }, |
| |
| "file_metadata": { |
| "format": "Apache Parquet", |
| "compression": "snappy", |
| "compressed_size_mb": 4.73, |
| "uncompressed_size_mb": 12.3, |
| "compression_ratio": 0.384, |
| "encoding": "UTF-8", |
| "schema_version": "1.0" |
| }, |
| |
| "citation": { |
| "bibtex": "@article{sani2024bht25,\n title={BHT25: A Bengali-Hindi-Telugu Parallel Corpus for Enhanced Literary Machine Translation},\n author={Sani, Sudeshna and Gangashetty, Suryakanth V and Samudravijaya, K and Nandi, Anik and Priya, Aruna and Kumar, Vineeth and Dubey, Akhilesh Kumar},\n journal={Data in Brief},\n year={2024},\n publisher={Elsevier}\n}", |
| "apa": "Sani, S., Gangashetty, S. V., Samudravijaya, K., Nandi, A., Priya, A., Kumar, V., & Dubey, A. K. (2024). BHT25: A Bengali-Hindi-Telugu Parallel Corpus for Enhanced Literary Machine Translation. Data in Brief, XX, XXXXX." |
| }, |
| |
| "contact": { |
| "corresponding_author": "Sudeshna Sani", |
| "email": "sudeshna.sani@klef.edu.in", |
| "affiliation": "Koneru Lakshmaiah Education Foundation, Guntur, India", |
| "github": "https://github.com/sudeshna84/BHT25", |
| "huggingface": "https://huggingface.co/datasets/sudeshna84/BHT25" |
| }, |
| |
| "version_history": { |
| "v1.0": { |
| "release_date": "2024-12-19", |
| "changes": "Initial public release with 27,149 triplets" |
| } |
| }, |
| |
| "planned_updates": { |
| "v1.1": { |
| "planned_date": "2025-Q1", |
| "features": ["Sentence-level emotion annotations"] |
| }, |
| "v2.0": { |
| "planned_date": "2025-Q3", |
| "features": ["Expansion to 50,000 triplets"] |
| }, |
| "v2.1": { |
| "planned_date": "2025-Q4", |
| "features": ["Parallel audio for multilingual speech research"] |
| }, |
| "v3.0": { |
| "planned_date": "2026", |
| "features": ["Extension to Tamil and Kannada"] |
| } |
| } |
| } |
|
|
