Automated chatbot linguistic expression generation
AssigneeSap SE
InventorPablo Roisman
Life of this patent
Estimating the term…
Abstract
Linguistic expressions for training a chatbot can be generated in an automated system via linguistic expression templates that are associated with intents. The pre-categorized linguistic expressions can then be used for training and validation. Chatbot development can thus be improved by having a large number of expressions for development, leading to a more robust chatbot. In practice, the process can iterate with modifications to the templates until a suitable benchmark is met. The technique can be applied across human languages to generate chatbots conversant in any number of languages and is applicable to a variety of domains.
Claims19
A computer-implemented method of automated linguistic expression generation comprising:
- based on a plurality of stored linguistic expression generation templates following a syntax, generating a plurality of generated linguistic expressions for developing a chatbot, wherein the generated linguistic expressions have respective pre-categorized intents determined from intents associated with the linguistic expression generation templates from which the generated linguistic expressions are generated;
- training the chatbot with selected training linguistic expressions of the generated linguistic expressions;
- and
- performing a validation on the trained chatbot with selected validation linguistic expressions of the generated linguistic expressions, wherein the validation generates a benchmark value indicative of performance of the chatbot.
A computing system comprising:
- one or more processors;
- memory storing a plurality of stored linguistic expression generation templates, wherein the templates are associated with respective intents;
- wherein the memory is configured to cause the one or more processors to perform operations comprising:
- based on a plurality of stored linguistic expression generation templates following a syntax supporting multiple alternative phrases, generating a plurality of generated linguistic expressions for developing a chatbot, wherein the generated linguistic expressions have respective pre-categorized intents determined from intents associated with the linguistic expression generation templates from which the generated linguistic expressions are generated and generating the plurality of generated linguistic expressions for developing the chatbot comprises choosing from among the multiple alternative phrases;
- and
- training the chatbot with selected training linguistic expressions of the training linguistic expressions and respective of the pre-categorized intents.
One or more non-transitory computer-readable media comprising computer-executable instructions that, when executed, cause a computing system to perform a method comprising:
- from a plurality of stored linguistic expression generation templates following a syntax supporting optional multiple alternative phrases and grouped by intent, generating a plurality of generated linguistic expressions for developing a chatbot, wherein the generated linguistic expressions have respective pre-categorized intents according to an intent group of a given linguistic expression generation template from which the generated linguistic expressions were generated;
- augmenting the generated linguistic expressions via machine translation;
- filtering the generated linguistic expressions based on an edit distance;
- selecting a first subset of the generated linguistic expressions for use in training;
- selecting a second subset of the generated linguistic expressions for use in validation;
- training the chatbot with the generated linguistic expressions selected for use in training;
- and
- performing a validation on the trained chatbot with the generated linguistic expressions selected for use in validation, wherein the validation generates one or more benchmark values indicative of performance of the chatbot.
Dependencies are inferred from the claim text.