|
|
 |
|
|
White Papers |
|
| | Conversing with stochastic language models | | | | In this paper we look at the problem of conversation simulation via stochastic language modeling. In particular, we show how a stochastic language model may be used to discover constraints which are beneficial when using the model generatively in the context of conversation simulation, we introduce a goal-oriented language model which is conductive to having its generations constrained in this way, and we discuss how the UpWrite, a technique borrowed from the field of syntactic pattern recognition, may be used to abstract the model’s representation of user input by discovering higher-level structure in the observed data. | | | | | | | |
|
|
|