Selecting a response in a multi-turn interaction between a user and a conversational bot
A method to select a response in a multi-turn conversation between a user and a conversational bot. The conversation is composed of a set of events, wherein an event is a linear sequence of observations that are user speech or physical actions. Queries are processed against a set of conversations that are organized as a set of inter-related data tables, with events and observations stored in distinct tables. As the multi-turn conversation proceeds, a data model comprising an observation history, together with a hierarchy of events determined to represent the conversation up to at least one turn, is persisted. When a new input (speech or physical action) is received, it is classified using a statistical model to generate a result. The result is then mapped to an observation in the data model. Using the mapped observation, a look-up is performed into the data tables to retrieve a possible response.
1. A method for imitating a human conversational response, comprising:
in association with an automated bot executing in a computing system:
receiving a data model associated with a conversation, the data model comprising an observation history, together with a set of events;
querying against a set of annotated conversations to generate a first set of one or more candidate interpretations for an input that is an utterance or action, wherein a candidate interpretation is a reference to a specific utterance or action in the set of annotated conversations that shares a same meaning as the input;
for each candidate interpretation of the first set of candidate interpretations, creating an updated data model reflecting a new state of the conversation resulting from incorporating the candidate interpretation;
coherency filtering the set of updated data models derived from the first set of candidate interpretations against one or more first tests of coherence to identify a particular updated data model, and persisting the particular updated data model as a current state of the conversation;
using the particular updated data model, querying against the set of annotated conversations to generate a second set of one or more candidate interpretations for one or more observations that represent a possible response to the input;
for each candidate interpretation of the second set of candidate interpretations, creating an updated data model reflecting a potential future state of the conversation resulting from appending the candidate interpretation; and
coherency filtering the set of one or more updated data models derived from the second set of candidate interpretations against one or more second tests of coherence to identify a coherent response to the input for the automated conversational bot;
the automated bot returning the coherent response to the input.
2. The method as described in claim 1 wherein the particular updated data model is one that passes all of the first tests of coherence.
3. The method as described in claim 1 wherein the conversation is one of: chat, e-mail, messaging and a voice-based interaction.
4. The method as described in claim 1 wherein the candidate interpretation comprises a list of one or more events as identified from the set of annotated conversations that are associated to the input.
5. The method as described in claim 4 wherein the one or more events correspond to conversation fragments retrieved from the set of annotated conversations.
6. The method as described in claim 1 wherein a turn in the conversation captures all consecutive utterances from a same conversational entity.
7. The method as described in claim 1 wherein the conversation is received as a data stream in real-time or near real-time.
8. The method as described in claim 1 wherein the set of annotated conversations comprise an historical corpus of conversational transcripts.