Hierarchical natural language understanding systems

Patent metadata

Field	Value
Publication number	US-11880659-B2
Application number	US-202117162318-A
Country	US
Kind code	B2
Filing date	Jan 29, 2021
Priority date	Jan 29, 2021
Publication date	Jan 23, 2024
Grant date	Jan 23, 2024

Abstract

Methods and systems for hierarchical natural language understanding are described. A representation of an utterance is inputted to a first machine learning model to obtain information on the first utterance. According to the information on the utterance a determination that the representation of the utterance is to be inputted to a second machine learning model that performs a dedicated natural language task is performed. In response to determining that the representation of the utterance is to be inputted to a second machine learning model, the utterance is inputted to the second machine learning model to obtain an output of the dedicated natural language task.

First claim

What is claimed is:

1. A method comprising:
inputting a representation of a first utterance to a first machine learning model to obtain information on the first utterance;
according to the information on the first utterance, determining that the representation of the first utterance is to be inputted to a second machine learning model that performs a dedicated natural language task; and
in response to determining that the representation of the first utterance is to be inputted to a second machine learning model, inputting the representation of the first utterance to the second machine learning model to obtain an output of the dedicated natural language task.

2. The method of claim 1 further comprising:
inputting a representation of a second utterance to the first machine learning model to obtain information on the second utterance;
determining that the information on the second utterance is an output of a second natural language task; and
in response to determining that the information on the second utterance is the output of the second natural language task bypassing the second machine learning model.

3. The method of claim 1 , wherein the dedicated natural language task is one of natural language inference, paraphrasing, named entity recognition, and question answering.

4. The method of claim 1 , wherein the information on the first utterance includes one or a combination of a field associated with the first utterance, an entity associated with the first utterance, one or more filtered historical data associated with the first utterance, a stance of a user associated with the first utterance, a sentiment of the user associated with the first utterance.

5. The method of claim 4 , wherein one or more of the field associated with the first utterance, the entity associated with the first utterance, the one or more filtered historical data associated with the first utterance, the stance of the user associated with the first utterance, the sentiment of the user associated with the first utterance are inputted to the second machine learning model with the representation of the first utterance to obtain the output of the dedicated natural language task.

6. The method of claim 1 , wherein the first machine learning model is a zero-shot model that is operative to observe a representation of an utterance from a class that was not observed during a training phase of the first machine learning model and is operative to predict one or more classes that the representation of the first utterance belongs to.

7. The method of claim 1 , wherein the first machine learning model and the second machine learning model enable a conversational artificial intelligence system for communication with a user.

8. The method of claim 1 , wherein the second machine learning model is for use in a predetermined field.

9. The method of claim 1 , wherein in response to determining that the representation of the first utterance is to be inputted to a second machine learning model, further inputting the information on the first utterance to the second machine learning model to obtain the output of the dedicated natural language task.

10. The method of claim 1 , the determining that the representation of the first utterance is to be inputted to a second machine learning model includes:
selecting, based on the information on the first utterance, the second machine learning model from a plurality of second machine learning models.

11. A non-transitory machine-readable storage medium that provides instructions that, if executed by a processor, will cause said processor to perform operations comprising:
inputting a representation of a first utterance to a first machine learning model to obtain information on the first utterance;
according to the information on the first utterance, determining that the representation of the first utterance is to be inputted to a second machine learning model that performs a dedicated natural language task; and
in response to determining that the representation of the first utterance is to be inputted to a second machine learning model, inputting the representation of the first utterance to the second machine learning model to obtain an output of the dedicated natural language task.

12. The non-transitory machine-readable storage medium of claim 11 , wherein the operations further comprise:
inputting a representation of a second utterance to the first machine learning model to obtain information on the second utterance;
determining that the information on the second utterance is an output of a second natural language task; and
in response to determining that the information on the second utterance is the output of the second natural language task bypassing the second machine learning model.

13. The non-transitory machine-readable storage medium of claim 11 , wherein the dedicated natural language task is one of natural language inference, paraphrasing, named entity recognition, and question answering.

14. The non-transitory machine-readable storage medium of claim 11 , wherein the information on the first utterance includes one or a combination of a field associated with the first utterance, an entity associated with the first utterance, one or more filtered historical data associated with the first utterance, a stance of a user associated with the first utterance, a sentiment of the user associated with the first utterance.

15. The non-transitory machine-readable storage medium of claim 14 , wherein one or more of the field associated with the first utterance, the entity associated with the first utterance, the one or more filtered historical data associated with the first utterance, the stance of the user associated with the first utterance, the sentiment of the user associated with the first utterance, are inputted to the second machine learning model with the representation of the first utterance to obtain the output of the dedicated natural language task.

16. The non-transitory machine-readable storage medium of claim 11 , wherein the first machine learning model is a zero-shot model that is operative to observe a representation of an utterance from a class that was not observed during a training phase of the first machine learning model and is operative to predict one or more classes that the representation of the first utterance belongs to.

17. The non-transitory machine-readable storage medium of claim 11 , wherein the first machine learning model and the second machine learning model enable a conversational artificial intelligence system for communication with a user.

18. The non-transitory machine-readable storage medium of claim 11 , wherein the second machine learning model is for use in a predetermined field.

19. The non-transitory machine-readable storage medium of claim 11 , wherein in response to determining that the representation of the first utterance is to be inputted to a second machine learning model, further inputting the information on the first utterance to the second machine learning model to obtain the output of the dedicated natural language task.

20. The non-transitory machine-readable storage medium of claim 11 , the determining that the representation of the first utterance is to be inputted to a second machine learning model includes:
selecting, based on the information on the first utterance, the second machine learning model from a plurality of second machine learning models.

Assignees

• Salesforce Inc

Inventors

• Pentyala Shiva Kumar
• Soumet Jean-Marc
• Harinath Shashank
• Bhagavath Shilpa
• Liu Johnson
• Chadha Ankit

Classifications

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F40/295

  Named entity recognition · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F40/216 Primary

  using statistical methods · CPC title