Using dialog context to improve language understanding

A system comprising a processor and computer executable instructions, that when executed by the processor, cause the system to perform operations comprising:

• receiving a plurality of example utterances;
• identify at least one example prior dialog context for each example utterance of the plurality of example utterances;
• creating a plurality of utterance-context pairs, each pair comprising an example utterance and an example dialog context;
• extracting a first set of features from the utterance-context pairs;
• training a first language model with training data comprising the first set of features;
• receiving an utterance;
• extracting a second set of features from the plurality of example utterances;
• training a second language model with training data comprising the second set of features;
• determining whether prior dialog context was received along with the utterance;
• responsive to determining that prior dialog context was received along with the utterance:
• determining at least one intent based on the utterance and prior dialog context using the first language model;
• or
• responsive to determining that no prior dialog context was received along with the utterance:
• determining at least one intent based on the utterance using the second language model;
• and
• returning the at least one intent.

A hardware computer storage medium comprising executable instructions that, when executed by a processor of a machine, cause the machine to perform operations comprising:

• obtaining a set of prompts from a dialog system;
• identifying at least one prompt from the set of prompts from the dialog system that the dialog system uses to elicit the at least one example utterance from a user of the dialog system;
• identifying example utterances that the user may enter when prompted by the at least one prompt;
• identify an example intent for each of the example utterances;
• identify at least one example entity associated with each example intent;
• creating a plurality of example utterance-dialog context pairs, each example utterance-dialog context pairs comprising an example utterance and either the example intent, an example entity associated with the example intent along with an associated example entity, or the example intent and the example entity associated with the example intent;
• extracting a set of features from the plurality of example utterance-dialog context pairs;
• and
• training a language understanding model with training data comprising the set of features.