Each of the five tasks has 1,000 dialogues each for training, validation, and testing. The file format for each task is as follows:

ID user_utterance [tab] bot_response
...

The IDs for a given dialog start at 1 and increase. When the IDs in a file reset back to 1, the following sentences are the start of a new dialog. This is an example of a raw dialog:

1 hi       hello what can i help you with today
2 can you make a restaurant reservation with french cuisine for four  
  people in an expensive price range      i'm on it
3 <SILENCE>       where should it be
4 tokyo please   ok let me look into some options for you
5 <SILENCE>       api_call french tokyo four expensive

The model must learn to predict the bot response for each user utterance. The response can be a sentence or an API call (starting with api_call).