A Diversity-Promoting Objective Function for Neural Conversation Models
This addresses the issue of dull and repetitive outputs in conversational AI, making interactions more engaging for users, though it is an incremental improvement over existing methods.
The authors tackled the problem of neural conversation models generating generic responses by proposing Maximum Mutual Information (MMI) as an alternative objective function, resulting in more diverse and appropriate responses with gains in BLEU scores and human evaluations.
Sequence-to-sequence neural network models for generation of conversational responses tend to generate safe, commonplace responses (e.g., "I don't know") regardless of the input. We suggest that the traditional objective function, i.e., the likelihood of output (response) given input (message) is unsuited to response generation tasks. Instead we propose using Maximum Mutual Information (MMI) as the objective function in neural models. Experimental results demonstrate that the proposed MMI models produce more diverse, interesting, and appropriate responses, yielding substantive gains in BLEU scores on two conversational datasets and in human evaluations.