Tennis Interviews

Transcripts for tennis singles post-match press conferences for major tournaments between 2007 to 2015 (6,467 post-match press conferences).

Distributed together with: Tie-breaker: Using language models to quantify gender bias in sports journalism. Liye Fu, Cristian Danescu-Niculescu-Mizil, Lillian Lee IJCAI workshop on NLP meets Journalism, 2016.

Dataset details

Speaker-level information

Speakers in this dataset are tennis professional players, represented by their real names. As this dataset do not contain information about individual reporters, we use a single pseudo user with username “REPORTER” to represent them.

For each player, additional metadata include:

  • gender: player gender

Utterance-level information

Each question or answer is viewed as an utterance. For each utterance, we provide:

  • id: index of the utterance
  • speaker: the speaker who authored the utterance
  • conversation_id: id of the first utterance in the conversation this utterance belongs to
  • reply_to: id of the utterance to which this utterance replies to (None if the utterance is not a reply)
  • timestamp: time of the utterance
  • text: textual content of the utterance

Metadata for each utterance include:

  • is_answer: whether the utterance is an answer from a player
  • is_question: whether the utterance is a question raised by a reporter
  • pair_idx: index of the question-answer pair
  • parsed: parsed version of the utterance text, represented as a SpaCy Doc

Conversational-level information

Each round of question-answer pair is considered as a conversation. Metadata associated with conversations include additional information about the match for which the post-match interview is held:

  • match_id: id of the match in the original dataset
  • opponent: opponent in the match (available only if the opponent has at least one interview recorded in our dataset)
  • result: outcome of the match (1 indicates the player being interviewed has won the match; 0 otherwise)
  • stage: stage of the tournament (e.g., ‘The Final’)
  • tournament: tournament name
  • tournament_type: type of the tournament, indicating tournament prestige
  • player_ranking: ranking of the player at the time of the match


To download directly with ConvoKit:

>>> from convokit import Corpus, download
>>> corpus = Corpus(filename=download("tennis-corpus"))

For some quick stats:

>>> corpus.print_summary_stats()
Number of Speakers: 359
Number of Utterances: 163948
Number of Conversations: 81974

Additional note


Please email any questions to: (Liye Fu).