Granted Patent B2

US 12,217,744 · App. 17/226,594 · Granted Feb 4, 2025

Inventors: Naoki Otani (Pittsburgh, PA); Jun Araki (San Jose, CA); Hyeongsik Kim (San Jose, CA)

Assignee: Robert Bosch GmbH

G10L15/18

Abstract

Claims (75)

1. A computer-implemented method for training a dialogue framework, the method comprising:

receiving input data;

obtaining situational data that provides context for the input data;

creating a first dataset that includes at least concatenated data, the concatenated data including at least a concatenation of the input data and the situational data;

generating, via an encoder, an encoded representation of the first dataset, the encoder including (i) a token embedder and (ii) an encoding network of a first pre-trained generative machine learning model that relates to a generative knowledge graph, the token embedder being configured in parallel with the encoding network such that the token embedder and the encoding network receive the same concatenated data as input:

generating, via a decoder, response data based on the first dataset by decoding the encoded representation, the decoder including a decoding network of a second pre-trained generative machine learning model that includes a first language model head and a second language model head; and

generating, via the decoder, goal data based on the first dataset by decoding the encoded representation,

wherein,

the goal data is used in multi-hop reasoning to guide the input data to the response data via the generative knowledge graph,

the first language model head generates the response data, and

the second language model head generates the goal data.

2. The computer-implemented method of claim 1 , further comprising:

generating first encoded data by encoding the concatenated data via a first encoding scheme performed by the token embedder; and

generating second encoded data by encoding the concatenated data via a second encoding scheme performed by the encoding network,

wherein the encoded representation includes the first encoded data and the second encoded data.

3. The computer-implemented method of claim 2 , wherein:

the first encoding scheme includes token embedding in an embedding space; and

the second encoding scheme includes knowledge embedding with respect to the generative knowledge graph that includes event-based data.

4. The computer-implemented method of claim 1 , further comprising:

creating a second dataset that includes the response data and the situational data;

generating, via the encoder, an encoded representation of the second dataset;

generating, via another decoder, additional goal data using the encoded representation of the second dataset, the another decoder including a decoding network of a third pre-trained generative machine learning model; and

fine-tuning the encoder based on loss data associated with the goal data and the additional goal data.

5. The computer-implemented method of claim 1 , wherein the encoding network of the first pre-trained generative machine learning model is domain agnostic and language agnostic.

6. The computer-implemented method of claim 1 , wherein the decoding network of the second pre-trained generative machine learning model is domain agnostic and language agnostic.

7. A system comprising:

at least one non-transitory computer readable medium including computer readable data; and

a processor operably connected to the at least one non-transitory computer readable medium, the processor being configured to execute the computer readable data to perform a method that includes:

receiving input data;

obtaining situational data that provides context for the input data;

creating a first dataset that includes concatenated data, the concatenated data including at least a concatenation of the input data and the situational data;

generating, via an encoder, an encoded representation of the first dataset, the encoder including (i) a token embedder and (ii) an encoding network of a first pre-trained generative machine learning model that relates to a generative knowledge base, the token embedder being configured in parallel with the encoding network such that the token embedder and the encoding network receive the same concatenated data as input;

generating, via a decoder, response data based on the first dataset by decoding the encoded representation, the decoder including a decoding network of a second pre-trained generative machine learning model that includes a first language model head and a second language model head; and

generating, via the decoder, goal data based on the first dataset by decoding the encoded representation,

wherein

the input data and the response data are connected to the goal data via the generative knowledge base,

the first language model head generates the response data, and

the second language model head generates the goal data.

8. The system of claim 7 , further comprising:

generating first encoded data by encoding the concatenated data via a first encoding scheme; and

generating second encoded data by encoding the concatenated data via a second encoding scheme;

wherein the encoded representation includes the first encoded data and the second encoded data.

9. The system of claim 8 , wherein:

the first encoding scheme includes token embedding in an embedding space; and

the second encoding scheme includes knowledge embedding with respect to the generative knowledge base that includes event-based data.

10. The system of claim 7 , further comprising:

creating a second dataset that includes the response data and the situational data;

generating, via the encoder, an encoded representation of the second dataset;

generating, via another decoder, another goal data using the second dataset, the another decoder including a decoding network of a third pre-trained generative machine learning model; and

fine-tuning the encoder based on loss data relating to the goal data and the another goal data.

11. The system of claim 7 , wherein the encoding network of the first pre-trained generative machine learning model is domain agnostic and language agnostic.

12. The system of claim 7 , wherein the decoding network of the second pre-trained generative machine learning model is domain agnostic and language agnostic.

13. A computer-implemented method comprising:

receiving input data;

obtaining situational data that provides context for the input data;

generating concatenated data, the concatenated data including at least a concatenation of the input data and the situational data;

obtaining a set of candidate responses;

generating, via an encoder, an encoded representation of the input data and the situational data, the encoder including (i) a token embedder and (ii) an encoding network of a first pre-trained generative machine learning model that relates to a generative knowledge graph, the token embedder being configured in parallel with the encoding network such that the token embedder and the encoding network receive the same concatenated data as input:

generating, via a decoder, goal data by decoding the encoded representation of the input data and the situational data, the decoder including a decoding network of a second pre-trained generative machine learning model that includes a first language model head and a second language model head; and

generating, via the decoder, a likelihood score for each candidate response of the set of candidate responses based on the input data and the situational data,

wherein,

the first language model head generates the likelihood score data, and

the second language model head generates the goal data.

14. The computer-implemented method of claim 13 , further comprising:

selecting a best candidate response from among the set of candidate responses based on an evaluation of the likelihood scores of the set of candidate responses; and

outputting a system response that includes the best candidate response.

15. The computer-implemented method of claim 13 , further comprising:

generating first encoded data by encoding the concatenated data via a first encoding scheme; and

generating second encoded data by encoding the concatenated data via a second encoding scheme,

wherein the encoded representation includes the first encoded data and the second encoded data.

16. The computer-implemented method of claim 15 , wherein:

the first encoding scheme includes token embedding in an embedding space; and

the second encoding scheme includes knowledge embedding with respect to the generative knowledge graph that includes event-based data.

17. The computer-implemented method of claim 13 , wherein the encoding network of the first pre-trained generative machine learning model is domain agnostic and language agnostic.

18. The computer-implemented method of claim 13 , wherein the decoding network of the second pre-trained generative machine learning model is domain agnostic and language agnostic.

Continuity (1)

References Cited (16)