- Number
- 10691894
- Published
- 2020-06-23
- Filed
- 2018-05-01
- Assignee
- Disney Enterprises, Inc.
- Inventors
- Doggett; Erika
- CPC
- G06F16/3329; G06F16/3344; G06F40/216; G06F40/253; G06F40/268; G06F40/30; G06F40/35; G06F40/56; G06N20/00; G06N3/044; G06N3/0442; G06N3/08; G10L15/22
- Verdict
- Set aside generic NLG language tool, no specific creative hook
- Source
- Google Patents · FreePatentsOnline
Abstract
A process receives a user input in a human-to-machine interaction. The process generates, with a natural language generation engine, one or more response candidates. Further, the process measures, with the natural language generation engine, the semantic similarity of the one or more response candidates. In addition, the process selects, with the natural language generation engine, a response candidate from the one or more response candidates. The process measures, with the natural language generation engine, an offensiveness measurement and a politeness measurement of the selected response. The process determines, with the natural language generation engine, that the offensiveness measurement or the politeness measurement lacks compliance with one or more predefined criteria. The process selects, with the natural language generation engine, an additional response candidate from the one or more response candidates that has a higher semantic similarity measurement than remaining response candidates from the one or more response candidates.
Background
BACKGROUND1. Field(1) This disclosure generally relates to the field of computing systems. More particularly, the disclosure relates to artificial intelligence (“AI”) systems.2. General Background(2) Some current AI systems allow for the use of natural language generation (“NLG”) when interacting with users. NLG has been incorporated into conversations (written or oral) between a computerized system and a human user in a manner of speaking to which the human user is accustomed. Yet, such systems often provide a disincentive to human user participation when using what is deemed to be offensive language to many human users. For example, current generative language chatbots have been prone to being directed by some human users, whether purposefully or accidentally, toward producing offensive language. As a result, deployment of NLG AI systems for use with practical applications has been somewhat limited.SUMMARY(3) In one aspect, a computer program product comprises a non-transitory computer readable storage device having a computer readable program stored thereon. The computer readable program when executed on a computer causes the computer to receive, with a processor, a user input in a human-to-machine interaction. Further, the computer is caused to generate, with an NLG engine, one or more response candidates. In addition, the computer is caused to measure, with the NLG engine, the semantic similarity of the one or more response candidates. The computer is also caused to sele
Claims
1. A computer program product comprising a non-transitory computer readable storage device having a computer readable program stored thereon, wherein the computer readable program when executed on a computer causes the computer to: receive, with a processor, a user input in a human-to-machine interaction; generate, with a natural language generation engine, a plurality of response candidates; measure, with the natural language generation engine, a semantic similarity of the plurality of response candidates to the user input; select, with the natural language generation engine, a response candidate from the plurality of response candidates as a selected response candidate; measure, with the natural language generation engine, an offensiveness measurement and a politeness measurement of the selected response candidate; determine, with the natural language generation engine, whether the offensiveness measurement or the politeness measurement complies with one or more criteria; responsive to a determination that the offensiveness measurement and the politeness measurement comply with the one or more criteria, output, with the natural language generation engine, the selected response candidate; and responsive to a determination that the offensiveness measurement and the politeness measurement do not comply with the one or more criteria; select, with the natural language generation engine, an additional response candidate from the plurality of response candidates, as a selected additional response candidate, the selected additional response candidate having a higher semantic similarity measurement than remaining response candidates from the plurality of response candidates; measure, with the natural language generation engine, an additional offensiveness measurement and an additional politeness measurement of the selected additional response candidate; determine, with the natural language generation engine, that the additional offensiveness measurement and the additional politeness measurement comply with the one or more criteria; and output, with the natural language generation engine, the selected additional response candidate.
9. A method comprising: receiving, with a processor, a user input in a human-to-machine interaction; generating, with a natural language generation engine, a plurality of response candidates; measuring, with the natural language generation engine, a semantic similarity of the plurality of response candidates to the user input; selecting, with the natural language generation engine, a response candidate from the plurality of response candidates as a selected response candidate; measuring, with the natural language generation engine, an offensiveness measurement and a politeness measurement of the selected response candidate; determining, with the natural language generation engine, whether the offensiveness measurement or the politeness measurement complies with one or more criteria; responsive to a determination that the offensiveness measurement and the politeness measurement comply with the one or more criteria, outputting, with the natural language generation engine, the selected response candidate; and responsive to a determination that the offensiveness measurement and the politeness measurement do not comply with the one or more criteria: selecting, with the natural language generation engine, an additional response candidate from the plurality of response candidates, as a selected additional response candidate, the selected additional response candidate having a higher semantic similarity measurement than remaining response candidates from the plurality of response candidates; measuring, with the natural language generation engine, an additional offensiveness measurement and an additional politeness measurement of the selected additional response candidate; determining, with the natural language generation engine, that the additional offensiveness measurement and the additional politeness measurement comply with the one or more criteria; and outputting, with the natural language generation engine, the selected additional response candidate.
17. An apparatus comprising: a processor configured to: receive a user input in a human-to-machine interaction; generate a plurality of response candidates; measure a semantic similarity of the plurality of response candidates to the user input; select a response candidate from the plurality of response candidates as a selected response candidate; measure an offensiveness measurement and a politeness measurement of the selected response candidate determine whether the offensiveness measurement or the politeness measurement complies with one or more criteria; responsive to a determination that the offensiveness measurement and the politeness measurement comply with the one or more criteria, output, with the natural language generation engine, the selected response candidate; and responsive to a determination that the offensiveness measurement or the politeness measurement does not comply with the one or more criteria: select an additional response candidate from the plurality of response candidates, as a selected additional response candidate, the selected additional response candidate having a higher semantic similarity measurement than remaining response candidates from the plurality of response candidates; measure an additional offensiveness measurement and an additional politeness measurement of the selected additional response candidate determine, with the natural language generation engine, that the additional offensiveness measurement and the additional politeness measurement comply with the one or more criteria; and output the selected additional response candidate.