{
  "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"]
    }
  }
}