How to Automatically Detect who Someone is Talking To Using Transcripts

In a multi-party dialogue we attempt to determine which participant(s)
each utterance in the dialogue is directed towards. We build upon a
basic approach where each utterance is predicted to be directed both
towards the participant who spoke directly before and the participant
who spoke directly after the utterance. We use new approaches
attempting to improve accuracy of predictions by taking into account
the content of the utterance, and drawing meaning from names and
linguistics.