# US 11,908,460 B2 · Using a generative adversarial network to train a semantic parser of a dialog system > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,908,460 B2 - Document ID: US11908460B2 - Status: Granted - Kind: B2 - Application number: 16/992,306 - Filed: 2020-08-13 - Published: 2021-03-11 - Granted: 2024-02-20 - Earliest claimed: 2019-09-11 - Assignees: Oracle International Corporation - Applicants: Oracle International Corporation - Inventors: Thanh Long Duong; Mark Edward Johnson - Web page: https://patentgrep.com/patent/US11908460B2 ## CPC - G06F 40/205: Parsing - G10L 15/1815: {Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning} - G06F 40/226: Validation - G06F 40/30: Semantic analysis - G10L 15/16: using artificial neural networks - G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue - G10L 15/26: Speech to text systems (G10L15/08 takes precedence) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract Disclosed herein are techniques for using a generative adversarial network (GAN) to train a semantic parser of a dialog system. A method described herein involves accessing seed data that includes seed tuples. Each seed tuple includes a respective seed utterance and a respective seed logical form corresponding to the respective seed utterance. The method further includes training a semantic parser and a discriminator in a GAN. The semantic parser learns to map utterances to logical forms based on output from the discriminator, and the discriminator learns to recognize authentic logical forms based on output from the semantic parser. The semantic parser may then be integrated into a dialog system. ## Claims ### Claim 1 A method for training a semantic parser of a dialog system, the method comprising: accessing seed data comprising seed tuples, each seed tuple of the seed data comprising a respective seed utterance and a respective seed logical form corresponding to the respective seed utterance; and training a semantic parser and a discriminator in a generative adversarial network (GAN), wherein, in the GAN, the semantic parser learns to map utterances to logical forms and the discriminator learns to recognize authentic logical forms, wherein training the semantic parser and the discriminator in the GAN comprises: training the discriminator with a first portion of the seed data based on output from the semantic parser to determine an updated discriminator; and training the semantic parser with a second portion of the seed data based on output from the updated discriminator to determine an updated semantic parser; wherein the semantic parser is useable as part of a dialog system. ### Claim 2 (depends on 1) The method of claim 1, wherein training the semantic parser and the discriminator in the GAN comprises: inputting random utterances to the semantic parser; determining, by the semantic parser, generated logical forms based on the random utterances; inputting, to the discriminator, tuples comprising generated tuples and a subset of seed tuples selected from the seed data, wherein the generated tuples comprise the random utterances and the generated logical forms; and determining, by the discriminator, predicted authenticities of the tuples. ### Claim 3 (depends on 2) The method of claim 2, wherein training the semantic parser and the discriminator in the GAN further comprises: comparing the predicted authenticities to actual authenticities of the tuples; and comparing the predicted authenticities to the actual authenticities of the tuples. ### Claim 4 (depends on 3) The method of claim 3, wherein an actual authenticity of a tuple, in the tuples, is based on the tuple being a seed tuple in the seed data. ### Claim 5 (depends on 3) The method of claim 3, wherein training the semantic parser and the discriminator in the GAN further comprises: inputting seed utterances to the semantic parser, wherein the seed utterances are selected from the seed tuples; determining, by the semantic parser, additional generated logical forms based on the seed utterances; inputting, to the discriminator, additional generated tuples comprising the seed utterances from the seed data and the additional generated logical forms; and determining, by the discriminator, additional predicted authenticities of the additional generated tuples. ### Claim 6 (depends on 5) The method of claim 5, wherein training the semantic parser and the discriminator in the GAN further comprises: comparing the additional predicted authenticities to selected authenticities of the additional generated tuples; and comparing the additional predicted authenticities to the selected authenticities of the additional generated tuples. ### Claim 7 (depends on 6) The method of claim 6, wherein the selected authenticities of the additional generated tuples indicate that the additional generated tuples are authentic. ### Claim 8 (depends on 1) The method of claim 1, wherein the semantic parser and the updated discriminator are static during the training. ### Claim 9 (depends on 8) The method of claim 8, wherein the semantic parser is a sequence-to-sequence neural network, and the discriminator is a binary classifier. ### Claim 10 (depends on 1) The method of claim 1, further comprising: receiving, by the dialog system, a speech input; applying an automatic speech recognition subsystem to the speech input to determine an utterance; applying the semantic parser to the utterance to determine a logical form; analyzing the logical form to generate a response text responsive to the speech input; and applying a text-to-speech subsystem to the response to text to determine audio output for the dialog system. ### Claim 11 A training system comprising: a data-input subsystem configured to access seed data comprising seed tuples, each seed tuple of the seed data comprising a respective seed utterance and a respective seed logical form corresponding to the respective seed utterance; and a generative adversarial network (GAN) comprising: a discriminator configured to learn to recognize authentic logical forms; and a semantic parser configured to learn to map utterances to logical forms; and an error-minimization module configured to train the discriminator with a first portion of the seed data based on output from the semantic parser to determine an updated discriminator, and train the semantic parser with a second portion of the seed data based on output from the updated discriminator to determine an updated semantic parser, wherein the semantic parser is useable as part of a dialog system. ### Claim 12 (depends on 11) The training system of claim 11, wherein: the data-input subsystem is configured to input random utterances to the semantic parser; the semantic parser is configured to generate logical forms based on the random utterances; the data-input subsystem is configured to input, to the discriminator, tuples comprising generated tuples and a subset of seed tuples selected from the seed data, wherein the generated tuples comprise the random utterances and the generated logical forms; and the discriminator is configured to predict authenticities of the tuples. ### Claim 13 (depends on 12) The training system of claim 12, wherein the error-minimization module is further configured to: compare the predicted authenticities to actual authenticities of the tuples; and compare the predicted authenticities to the actual authenticities of the tuples. ### Claim 14 (depends on 13) The training system of claim 13, wherein an actual authenticity of a tuple, in the tuples, is based on the tuple being a seed tuple in the seed data. ### Claim 15 (depends on 13) The training system of claim 13, wherein: the data-input subsystem is further configured to input seed utterances to the semantic parser, wherein the seed utterances are selected from the seed tuples; the semantic parser is configured to determine additional generated logical forms based on the seed utterances; the data-input subsystem is configured to input, to the discriminator, additional generated tuples comprising the seed utterances from the seed data and the additional generated logical forms; and the discriminator is configured to determine additional predicted authenticities of the additional generated tuples. ### Claim 16 (depends on 15) The training system of claim 15, wherein the error-minimization module is further configured to: compare the additional predicted authenticities to selected authenticities of the additional generated tuples; and compare the additional predicted authenticities to the selected authenticities of the additional generated tuples. ### Claim 17 (depends on 16) The training system of claim 16, wherein the selected authenticities of the additional generated tuples indicate that the additional generated tuples are authentic. ### Claim 18 (depends on 11) The training system of claim 11, wherein the semantic parser is a sequence-to-sequence neural network, and the discriminator is a binary classifier. ### Claim 19 A computer-program product for training a semantic parser of a dialog system, the computer-program product comprising a computer-readable storage medium having program instructions embodied thereon, the program instructions executable by one or more processors to cause the one or more processors to perform a method comprising: accessing seed data comprising seed tuples, each seed tuple of the seed data comprising a respective seed utterance and a respective seed logical form corresponding to the respective seed utterance; training a discriminator of a generative adversarial network (GAN) based on a semantic parser of the GAN, wherein training the discriminator comprises: inputting random utterances to the semantic parser; determining, by the semantic parser, generated logical forms based on the random utterances; inputting, to the discriminator, tuples comprising generated tuples and a subset of seed tuples selected from the seed data, wherein the generated tuples comprise the random utterances and the generated logical forms; determining, by the discriminator, predicted authenticities of the tuples; comparing the predicted authenticities to actual authenticities of the tuples; and updating the discriminator based on comparing the predicted authenticities to the actual authenticities of the tuples; and training the semantic parser of the GAN based on the discriminator of the GAN, wherein training the semantic parser comprises: inputting seed utterances to the semantic parser, wherein the seed utterances are selected from the seed tuples; determining, by the semantic parser, additional generated logical forms based on the seed utterances; inputting, to the discriminator, additional generated tuples comprising the seed utterances from the seed data and the additional generated logical forms; determining, by the discriminator, additional predicted authenticities of the additional generated tuples; comparing the additional predicted authenticities to selected authenticities of the additional generated tuples; and updating the semantic parser based on comparing the additional predicted authenticities to the selected authenticities of the additional generated tuples, wherein the semantic parser is useable to map utterances to logical forms in a dialog system. ### Claim 20 (depends on 19) The computer-program product of claim 19, wherein: an actual authenticity of a tuple, in the tuples, is based on the tuple being a seed tuple in the seed data; and the selected authenticities of the additional generated tuples indicate that the additional generated tuples are authentic.