Granted Patent B1

US 10,891,435 · App. 15/900,687 · Granted Jan 12, 2021

Inventors: Nicholas Ruiz (Berkeley Heights, NJ); John Chen (Millburn, NJ); Srinivas Bangalore (Morristown, NJ)

Assignee: INTERACTIONS LLC

G06F40/30G06F40/58

Abstract

Claims (112)

1. A computer-implemented method comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding intent label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the intent label associated with the expression with the translation of the expression into the second natural language;

generating an intent model for the second natural language from the translated expressions and the intent labels using a machine learning algorithm;

receiving a user expression from a user;

determining that the user expression is in the second natural language;

using the generated intent model for the second natural language to obtain an intent label for the user expression;

transitioning from a current state to a next state of a workflow for a telephonic conversation with the user, based on the intent label for the user expression; and

providing a prompt to the user to elicit information for the next state.

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

determining that a confidence score associated with the intent label is below a minimum confidence threshold;

responsive to the determination, providing the expression to a human agent different from the user.

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

receiving an intent label for the user expression from the human agent; and

regenerating the intent model for the second natural language using the intent label received from the human agent.

4. A computer-implemented method comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding semantic label;

receiving, from a user, a user expression in a second natural language;

translating the user expression to an equivalent expression in the first natural language;

obtaining, for the user expression in the second natural language, a semantic label by applying a semantic model for the first natural language to the translated equivalent expression;

transitioning from a current state to a next state of a workflow for a conversation with the user, based on the semantic label for the user expression; and

providing a prompt to the user to elicit information for the next state.

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

determining that a confidence score associated with the semantic label is below a minimum confidence threshold;

responsive to the determination, providing the expression to a human agent.

6. A computer-implemented method comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding intent label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the intent label associated with the expression with the translation of the expression into the second natural language;

generating an intent model for the second natural language from the translated expressions and the intent labels using a machine learning algorithm;

receiving a user expression from a user;

determining that the user expression is in the second natural language;

using the generated intent model for the second natural language to obtain an intent label for the user expression;

determining that a confidence score associated with the intent label is below a minimum confidence threshold; and

responsive to the determination, providing the user expression to a human agent different from the user.

7. The computer-implemented method of claim 6 , further comprising:

receiving an intent label for the user expression from the human agent; and

regenerating the intent model for the second natural language using the intent label received from the human agent.

8. A computer-implemented method comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding semantic label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the semantic label associated with the expression with the translation of the expression into the second natural language;

generating a semantic model for the second natural language from the translated expressions and the semantic labels using a machine learning algorithm;

receiving a user expression from a user;

determining that the user expression is in the second natural language;

using the generated semantic model for the second natural language to obtain a semantic label for the expression;

transitioning from a current state to a next state of a workflow for a conversation with the user, based on the semantic label for the expression; and

providing a prompt to the user to elicit information for the next state,

wherein the conversation with the user is via textual messages.

9. A computer-implemented method comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding semantic label;

receiving an expression in a second natural language;

translating the expression to an equivalent expression in the first natural language;

obtaining, for the user expression in the second natural language, a semantic label by applying a semantic model for the first natural language to the translated equivalent expression;

determining that a confidence score associated with the semantic label is below a minimum confidence threshold; and

responsive to the determination, providing the expression to a human agent.

10. A computer system comprising:

a computer processor; and

a non-transitory compute-readable storage medium storing instructions that when executed by the computer processor perform actions comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding intent label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the intent label associated with the expression with the translation of the expression into the second natural language;

generating an intent model for the second natural language from the translated expressions and the intent labels using a machine learning algorithm;

receiving an expression from a user;

determining that the expression is in the second natural language;

using the generated intent model for the second natural language to obtain an intent label for the expression;

transitioning from a current state to a next state of a workflow for a telephonic conversation with the user, based on the intent label for the expression; and

providing a prompt to the user to elicit information for the next state.

11. A computer system comprising:

a computer processor; and

a non-transitory compute-readable storage medium storing instructions that when executed by the computer processor perform actions comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding intent label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the intent label associated with the expression with the translation of the expression into the second natural language;

generating an intent model for the second natural language from the translated expressions and the intent labels using a machine learning algorithm

receiving an expression from a user;

determining that the expression is in the second natural language;

using the generated intent model for the second natural language to obtain an intent label for the expression;

determining that a confidence score associated with the intent label is below a minimum confidence threshold; and

responsive to the determination, providing the expression to a human agent different from the user.

12. A computer system comprising:

a computer processor; and

a non-transitory compute-readable storage medium storing instructions that when executed by the computer processor perform actions comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding intent label;

generating, for a second natural language, a translated corpus comprising, for each expression in the corpus, a translation of the expression into the second natural language;

associating, for each expression, the intent label associated with the expression with the translation of the expression into the second natural language;

generating an intent model for the second natural language from the translated expressions and the intent labels using a machine learning algorithm;

receiving an expression from a user;

determining that the expression is in the second natural language;

using the generated intent model for the second natural language to obtain an intent label for the expression;

transitioning from a current state to a next state of a workflow for a conversation with the user, based on the intent label for the expression; and

providing a prompt to the user to elicit information for the next state,

wherein the conversation with the user is via textual messages.

13. A computer system comprising:

a computer processor; and

a non-transitory computer-readable storage medium storing instructions that when executed by the computer processor perform actions comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding semantic label;

receiving, from a user, an expression in a second natural language;

translating the expression to an equivalent expression in the first natural language;

obtaining, for the user expression in the second natural language, a semantic label by applying a semantic model for the first natural language to the translated equivalent expression;

transitioning from a current state to a next state of a workflow for a conversation with the user, based on the semantic label for the expression; and

providing a prompt to the user to elicit information for the next state.

14. A computer system comprising:

a computer processor; and

a non-transitory compute-readable storage medium storing instructions that when executed by the computer processor perform actions comprising:

accessing a corpus comprising a plurality of expressions in a first natural language, each of the expressions having a corresponding semantic label;

receiving an expression in a second natural language;

translating the expression to an equivalent expression in the first natural language;

obtaining, for the user expression in the second natural language, a semantic label by applying a semantic model for the first natural language to the translated equivalent expression;

determining that a confidence score associated with the semantic label is below a minimum confidence threshold; and

responsive to the determination, providing the expression to a human agent.

Assignments (8)

Cited By (2)

US 12,210,848 US 12,475,333