Junnan
Chen,
Master’s
candidate
David
R.
Cheriton
School
of
Computer
Science
Conversations depend on information from the context. To go beyond one-round conversation, a chatbot must resolve contextual information such as: 1) co-reference resolution, 2) ellipsis resolution, and 3) conjunctive relationship resolution.
There are simply not enough data to avoid these problems by trying to train a sequence-to-sequence model for multi-round conversation similar to that of one-round conversation.
The contributions of this paper are: 1) We formulate the problem of context resolution for conversation; 2) We present deep learning models, including an end-to-end network for context resolution; 3) We propose a way of creating a huge amount of realistic data for training such models with good experimental results.