Program dialog by example

Patent metadata

Field	Value
Publication number	US-11211049-B2
Application number	US-201916502534-A
Country	US
Kind code	B2
Filing date	Jul 3, 2019
Priority date	Jul 3, 2019
Publication date	Dec 28, 2021
Grant date	Dec 28, 2021

Abstract

One embodiment provides a method including receiving authoring conversational training data. A machine learning based conversational agent is trained with the conversational training data. The training includes: creating and storing example transcripts of user utterances, creating and storing example transcripts of agent utterances, sequencing utterance transcripts using the example transcripts of user utterances and the example transcripts of agent utterances, forming a corpus from the sequenced utterance transcripts, marking speech patterns that represent social actions from tagging the sequenced utterance transcripts, and forming a patterned corpus from the marked speech patterns.

First claim

What is claimed is:

1. A method comprising:
receiving authoring conversational training data; and
training a fully machine learning based conversational agent with the conversational training data for learning sequential patterns of user utterances and of agent utterances and for learning natural language classification of the user utterances and the agent utterances, comprising:
creating and storing example transcripts of user utterances from a software editing tool;
creating and storing example transcripts of agent utterances from the software editing tool;
sequencing utterance transcripts using the example transcripts of user utterances and the example transcripts of agent utterances;
forming a corpus from the sequenced utterance transcripts;
marking dynamic entities of speech patterns that represent social actions from tagging the sequenced utterance transcripts, wherein the sequenced utterance transcripts embody desired speech patterns for the fully machine learning based conversational agent to exhibit;
training a supervised machine learning model of the software editing tool with the sequenced utterance transcripts; and
forming a patterned corpus from the marked speech patterns;

wherein the trained fully machine learning based conversational agent predicts a sequence of intents and actions, and generates responses based on the predictions.

2. The method of claim 1 , further comprising:
analyzing the patterned corpus to determine annotation entry positions; and
replacing the marked dynamic entities with context variable entries.

3. The method of claim 2 , further comprising:
annotating user and agent utterance transcripts based on the determined annotation entry positions.

4. The method of claim 2 , further comprising:
receiving the context variable entries in the example transcripts of user utterances and the example transcripts of agent utterances.

5. The method of claim 4 , further comprising:
receiving edits for the example transcripts of user utterances to correct behavior of the machine learning based conversational agent.

6. The method of claim 5 , further comprising:
receiving edits of the example transcripts of agent utterances to correct behavior of the machine learning based conversational agent.

7. The method of claim 6 , wherein:
the software editing tool provides an interface for:
receiving the context variable entries;
receiving the edits for the example transcripts of user utterances; and
receiving the edits for the example transcripts of agent utterances; and

the software editing tool uses the supervised machine learning model that takes features from a sequence of tokens and the sequence of intents and actions, the supervised machine learning model is trained to predict the sequence of intents and actions that generate the responses provided from the trained fully machine learning based conversational agent, and output from the software editing tool is a contextual transcript reflecting meta information relevant for a current conversation and forms next training data.

8. The method of claim 7 , further comprising:
filtering the patterned corpus by user intents and agent intents using tags from the tagging of the sequenced utterance transcripts; and
providing filtered results including example sequences of user and agent transcripts.

9. A computer program product for programming dialog by example, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
receive, by the processor, authoring conversational training data;
train, by the processor, a fully machine learning based conversational agent with the conversational training data for learning sequential patterns of user utterances and of agent utterances and for learning natural language classification of the user utterances and the agent utterances, comprising:
create and store, by the processor, example transcripts of user utterances from a software editing tool;
create and store, by the processor, example transcripts of agent utterances from the software editing tool;
sequence, by the processor, utterance transcripts using the example transcripts of user utterances and the example transcripts of agent utterances;
form, by the processor, a corpus from the sequenced utterance transcripts;
mark, by the processor, dynamic entities of speech patterns that represent social actions from tagging the sequenced utterance transcripts, wherein the sequenced utterance transcripts embody desired speech patterns for the fully machine learning based conversational agent to exhibit;
train, by the processor, a supervised machine learning model of the software editing tool with the sequenced utterance transcripts; and
form, by the processor, a patterned corpus from the marked speech patterns;

wherein the trained fully machine learning based conversational agent predicts a sequence of intents and actions, and generates responses based on the predictions.

10. The computer program product of claim 9 , wherein the program instructions executable by the processor further to cause the processor to:
analyze, by the processor, the patterned corpus to determine annotation entry positions;
replacing the marked dynamic entities with context variable entries; and
annotate, by the processor, user and agent utterance transcripts based on the determined annotation entry positions.

11. The computer program product of claim 10 , wherein the program instructions executable by the processor further to cause the processor to:
receive, by the processor, the context variable entries in the example transcripts of user utterances and the example transcripts of agent utterances.

12. The computer program product of claim 11 , wherein the program instructions executable by the processor further to cause the processor to:
receive, by the processor, edits for the example transcripts of user utterances to correct behavior of the machine learning based conversational agent; and
receive, by the processor, edits for the example transcripts of the example transcripts of agent utterances to correct behavior of the machine learning based conversational agent.

13. The computer program product of claim 12 , wherein:
the software editing tool provides an interface for:
receiving the context variable entries;
receiving the edits for the example transcripts of user utterances; and
receiving the edits for the example transcripts of agent utterances; and

the software editing tool uses the supervised machine learning model that takes features from a sequence of tokens and the sequence of intents and actions, the supervised machine learning model is trained to predict the sequence of intents and actions that generate the responses provided from the trained fully machine learning based conversational agent, and output from the software editing tool is a contextual transcript reflecting meta information relevant for a current conversation and forms next training data.

14. The computer program product of claim 13 , wherein the program instructions executable by the processor further to cause the processor to:
filter, by the processor, the patterned corpus by user intents and agent intents using tags from the tagging of the sequenced utterance transcripts; and
provide, by the processor, filtered results including example sequences of user and agent transcripts.

15. An apparatus comprising:
a memory configured to store instructions; and
a processor configured to execute the instru

Assignees

• IBM

Inventors

• Moore Robert J
• Chowdhary Pawan
• Jadav Divyesh
• Huang Lei
• Lee Sunhwan
• Liu Eric Young
• Mishra Saurabh

Classifications

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G10L15/063 Primary

  Training · CPC title