Bazinga! A Dataset for Multi-Party Dialogues Structuring

Read paper | Watch video description

This dataset provides audio soundtracks and time-coded manual transcripts of episodes of the following TV and movies series:

24, Battlestar Galactica, Breaking Bad, Buffy The Vampire Slayer, ER, Friends, Game Of Thrones, Homeland, Lost, Six Feet Under, The Big Bang Theory, The Office, The Walking Dead, Harry Potter, Star Wars, and The Lord of the Rings.

Citation

@InProceedings{bazinga,
  author    = {Lerner, Paul  and  Bergoënd, Juliette  and  Guinaudeau, Camille  and  Bredin, Hervé  and  Maurice, Benjamin  and  Lefevre, Sharleyne  and  Bouteiller, Martin  and  Berhe, Aman  and  Galmant, Léo  and  Yin, Ruiqing  and  Barras, Claude},
  title     = {Bazinga! A Dataset for Multi-Party Dialogues Structuring},
  booktitle      = {Proceedings of the Language Resources and Evaluation Conference},
  month          = {June},
  year           = {2022},
  address        = {Marseille, France},
  publisher      = {European Language Resources Association},
  pages     = {3434--3441},
  url       = {https://aclanthology.org/2022.lrec-1.367}
}

Usage

import datasets

dataset = datasets.load_dataset(
    "bazinga/bazinga", 
    series="TheBigBangTheory", 
    audio=True,
    use_auth_token=True)

# iterate over test episodes ("validation" and "train" are also available)
for episode in dataset["test"]:
    
    identifier = episode["identifier"]          # TheBigBangTheory.Season01.Episode01

    # knowledge base
    kb = episode["knowledge_base"]
    kb["title"]                                 # Pilot
    kb["imdb"]                                  # https://www.imdb.com/title/tt0775431/
    kb["characters"]                            # ['leonard_hofstadter', 'sheldon_cooper', ..., 'kurt']

    # annotated transcript
    for word in episode["transcript"]:
        word["token"]                           # your
        word["speaker"]                         # leonard_hofstadter
        word["forced_alignment"]["start_time"]  # 14.240
        word["forced_alignment"]["end_time"]    # 14.350
        word["forced_alignment"]["confidence"]  # 0.990
        word["entity_linking"]                  # sheldon_cooper
        word["named_entity"]                    # None
        word["addressee"]                       # sheldon_cooper

    # audio (when dataset is initialized with audio=True)
    audio = episode["audio"]                    # path to wav file on disk

    # annotation status (GoldStandard, SilverStandard, or NotAvailable)
    status = episode["status"]
    status["speaker"]                           # GoldStandard
    status["token"]                             # GoldStandard
    status["forced_alignment"]                  # SilverStandard
    status["entity_linking"]                    # ...
    status["named_entity"]                      # ...
    status["addressee"]                         # ...

# get list of available series
datasets.get_dataset_config_names("bazinga/bazinga")
# [ "24", "BattlestarGalactica", ..., "TheBigBangTheory", ... ]