Deriving multiple meaning representations for an utterance in a natural language understanding (NLU) framework

Patent metadata

Field	Value
Publication number	US-11720756-B2
Application number	US-202117451405-A
Country	US
Kind code	B2
Filing date	Oct 19, 2021
Priority date	Jul 2, 2019
Publication date	Aug 8, 2023
Grant date	Aug 8, 2023

Abstract

The present approaches are generally related to an agent automation framework that is capable of extracting meaning from user utterances, such as requests received by a virtual agent (e.g., a chat agent), and suitably responding to these user utterances. In certain aspects, the agent automation framework includes a NLU framework and an intent-entity model having defined intents and entities that are associated with sample utterances. The NLU framework may include a meaning extraction subsystem designed to generate meaning representations for the sample utterances of the intent-entity model to construct an understanding model, as well as generate meaning representations for a received user utterance to construct an utterance meaning model. The disclosed NLU framework may include a meaning search subsystem that is designed to search the meaning representations of the understanding model to locate matches for meaning representations of the utterance meaning model.

First claim

What is claimed is:

1. An agent automation system, comprising:
a memory configured to store a natural language understanding (NLU) framework, wherein the NLU framework includes a part-of-speech (POS) component, a variability filter component, a parser component, and a final scoring and filtering component; and
a processor configured to execute instructions of the NLU framework to cause the agent automation system to perform actions comprising:
performing, via the POS component, part-of-speech (POS) tagging to generate a set of potential POS taggings for a set of utterances;
performing, via the variability filter component, variability filtering of the set of potential POS taggings to generate a set of final nominee POS taggings, wherein each of the set of final nominee POS taggings is distinct from one another;
parsing, via the parsing component, the set of final nominee POS taggings to generate a set of potential meaning representations for the set of final nominee POS taggings; and
selecting, via the final scoring and filtering component, a final set of meaning representations for the set of utterances from the set of potential meaning representations.

2. The agent automation system of claim 1 , wherein the NLU framework includes a vocabulary subsystem, and wherein the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
performing, via the vocabulary subsystem, vocabulary injection to generate the set of utterances based on an original utterance, wherein the set of utterances includes the original utterance and one or more re-expressions of the original utterance having different phraseology, different terminology, or a combination thereof.

3. The agent automation system of claim 2 , wherein the original utterance is a user utterance, and wherein the final set of meaning representations forms part of an utterance meaning model that defines a search key for a meaning search.

4. The agent automation system of claim 2 , wherein the original utterance is a sample utterance of an intent-entity model stored in the memory, and wherein the final set of meaning representations forms part of an understanding model that defines a search space for a meaning search.

5. The agent automation system of claim 1 , wherein the POS component, the variability filter component, the parser component, and the final scoring and filtering component are each implemented as plug-ins of the NLU framework.

6. The agent automation system of claim 1 , wherein the POS component, the parser component, or a combination thereof, is implemented as a machine-learning (ML)-based component that includes an artificial neural network.

7. The agent automation system of claim 1 , wherein, to perform POS tagging, the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
for each utterance of the set of utterances:
generating at least one potential POS tagging of the utterance;
determining a respective confidence score for the at least one potential POS tagging; and
selecting the at least one potential POS tagging for inclusion in the set of potential POS taggings in response to determining that the respective confidence score is greater than or equal to a predefined POS threshold value.

8. The agent automation system of claim 1 , wherein, to perform variability filtering, the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
calculating a variability score between a first potential POS tagging and a second potential POS tagging of the set of potential POS taggings; and
in response to the variability score being less than a predefined variation threshold value, selecting only one of the first potential POS tagging and the second potential POS tagging for inclusion in the set of final nominee POS taggings.

9. The agent automation system of claim 1 , wherein, to perform variability filtering, the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
calculating a tagging distance between each potential POS tagging of the set of potential POS taggings;
clustering the set of potential POS taggings into groups of potential POS taggings based on the tagging distance and a predefined variation threshold value; and
selecting a representative potential POS tagging for each of the groups of potential POS taggings for inclusion in the set of final nominee POS taggings.

10. The agent automation system of claim 1 , wherein, to parse the set of final nominee POS taggings, the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
for each POS tagging of the set of final nominee POS taggings:
generating at least one potential meaning representation based on the POS tagging;
determining a respective confidence score for the at least one potential meaning representation; and
selecting the at least one potential meaning representation for inclusion in the set of potential meaning representations in response to determining that the respective confidence score is greater than or equal to a predefined parse threshold value.

11. The agent automation system of claim 1 , wherein, to select the final set of meaning representations, the processor is configured to execute the instructions of the NLU framework to cause the agent automation system to perform actions comprising:
for each potential meaning representation of the set of potential meaning representations:
combining a POS tagging confidence score and a parsing confidence score associated with the potential meaning representation to calculate a respective final score of the potential meaning representation; and
selecting the potential meaning representation for inclusion in the final set of meaning representations in response to determining that the respective final score of the potential meaning representation is greater than or equal to a final scoring and filtering threshold value.

12. A method of operating a natural language understanding (NLU) framework, comprising:
performing part-of-speech (POS) tagging to generate a set of potential POS taggings for a set of utterances;
performing variability filtering of the set of potential POS taggings to generate a set of final nominee POS taggings, wherein each of the set of final nominee POS taggings is distinct from one another;
parsing the set of final nominee POS taggings to generate a set of potential meaning representations for the set of final nominee POS taggings; and
selecting a final set of meaning representations for the set of utterances from the set of potential meaning representations.

13. The method of claim 12 , wherein each potential meaning representation of the set of potential meaning representations is a tree data structure that includes:
a particular potential POS tagging from the set of potential POS taggings that was generated for a particular utterance of the set of utterance; and
at least one vector representing a semantic meaning of at least one portion of the particular utterance.

14. The method of claim 12 , comprising:
before performing POS tagging, performing vocabulary injection to generate the set of utterances based on an original utterance by:
identifying one or more tokens of the original utterance that are present in a vocabulary model;
selecting one or more alternative tokens from the vocabu

Assignees

• Servicenow Inc

Inventors

• Sapugay Edwin
• Sarda Gopal

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06F40/30 Primary

  Semantic analysis · CPC title

• G06F40/205

  Parsing · CPC title

• G06F40/211

  Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars · CPC title

• G06F40/216

  using statistical methods · CPC title