Transforming conversational training data for different machine learning models
A method includes receiving a command to transform a first statement in a first conversational training set into training data for a second training set, where the first conversational training set trains a first machine learning model in a first knowledge area, and where the second conversational training set trains a second machine learning model in a second knowledge area. The method also includes analyzing language of the first statement using a third machine learning model, where the third machine learning model is trained to recognize patterns of language variation between the first and second knowledge areas. The method also includes, responsive to the analyzing, selecting an original segment of the first statement. The method also includes, responsive to the selecting, transforming the first statement into a second statement, for the second conversational training set, that includes a substitute segment that at least partially replaces the original segment.
1. A method of transforming conversational training data, the method comprising, by a computer system:
receiving a command to transform a first statement in a first conversational training set into training data for a second conversational training set, wherein the first conversational training set trains a first machine learning model in a first knowledge area, and wherein the second conversational training set trains a second machine learning model in a second knowledge area;
analyzing language of the first statement using a third machine learning model, wherein the third machine learning model is trained to recognize patterns of language variation between the first and second knowledge areas;
responsive to the analyzing, selecting an original segment of the first statement as a candidate for substitution; and
responsive to the selecting, transforming the first statement into a second statement for the second conversational training set, the second statement comprising a substitute segment that at least partially replaces the original segment.
2. The method of claim 1 , wherein the analyzing comprises:
searching the first statement for one or more patterns; and
locating an instance of at least one pattern of the one or more patterns in the first statement.
3. The method of claim 2 , wherein the selecting comprises selecting a segment of the first statement that corresponds to the located instance of the at least one pattern.
4. The method of claim 2 , comprising:
determining whether the located instance of the at least one pattern satisfies criteria for substitution based, at least in part, on a result of the analyzing; and
wherein the selecting is performed responsive to a determination that the located instance of the at least one pattern satisfies the criteria for substitution, the selecting comprising selecting the located instance as the candidate for substitution.
5. The method of claim 4 , wherein the determining whether the located instance of the at least one pattern satisfies criteria for substitution comprises determining whether a sufficiently prominent combination of patterns has been recognized in the first statement.
6. The method of claim 4 , comprising, responsive to a determination that the located instance of the at least one pattern does not satisfy the criteria for substitution, leaving the first statement unchanged.
7. The method of claim 2 , wherein the one or more patterns are specified, at least on part, in a knowledge base associated with the third machine learning model.
8. The method of claim 2 , wherein the transforming comprises determining the substitute segment via a fourth machine learning model that is trained to identify substitute segments for transformations from the first knowledge area to the second knowledge area.
9. The method of claim 8 , comprising:
determining whether the substitute segment satisfies transformation criteria; and
responsive to a determination that the transformation criteria are satisfied, generating the second statement.
10. The method of claim 9 , comprising, responsive to a determination that the transformation criteria are not satisfied, updating a configuration such that a future instance of the at least one pattern is not processed for transformation.
11. The method of claim 9 , wherein the transformation criteria are specified, at least in part, in terms of a measured similarity of the substitute segment to the located instance of the at least one pattern.
12. The method of claim 9 , wherein the transformation criteria are specified, at least in part, in terms of a measured similarity of the substitute segment to the at least one pattern.
13. The method of claim 1 , wherein the second knowledge area is a specialization of the first knowledge area.
14. The method of claim 1 , wherein the second knowledge area is a generalization of the first knowledge area.
15. The method of claim 1 , wherein the original segment and the substitute segment each comprise a plurality of tokens.
16. The method of claim 1 , comprising outputting the second statement.
17. A method of transforming conversational training data, the method comprising, by a computer system:
receiving a command to transform a first conversational training set into a second conversational training set, wherein the first conversational training set trains a first machine learning model in a first knowledge area, and wherein the second conversational training set trains a second machine learning model in a second knowledge area;
for each statement of a plurality of statements in the first conversational training set:
analyzing language of the statement using a third machine learning model, wherein the third machine learning model is trained to recognize patterns of language variation between the first and second knowledge areas;
responsive to the analyzing, selecting an original segment of the statement as a candidate for substitution; and
responsive to the selecting, transforming the statement into a second statement for the second conversational training set, the second statement comprising a substitute segment that at least partially replaces the original segment; and
outputting the second conversational training set.
18. The method of claim 17 , comprising, for at least one statement of the first conversational training set, leaving the at least one statement unchanged, the second conversational training set comprising the unchanged at least one statement.
19. The method of claim 17 , comprising initiating training of the second machine learning model using the second conversational training set.
20. A computer system comprising a processor and memory, wherein the processor and the memory in combination are operable to implement a method comprising:
receiving a command to transform a first statement in a first conversational training set into training data for a second conversational training set, wherein the first conversational training set trains a first machine learning model in a first knowledge area, and wherein the second conversational training set trains a second machine learning model in a second knowledge area;
analyzing language of the first statement using a third machine learning model, wherein the third machine learning model is trained to recognize patterns of language variation between the first and second knowledge areas;
responsive to the analyzing, selecting an original segment of the first statement as a candidate for substitution; and
responsive to the selecting, transforming the first statement into a second statement for the second conversational training set, the second statement comprising a substitute segment that at least partially replaces the original segment.