Multi-Domain Natural Language Processing Architecture

• US 20140163959A1
• Filed: 12/12/2012
• Published: 06/12/2014
• Est. Priority Date: 12/12/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Hebert, Matthieu, Robichaud, Jean-Philippe, Parisien, Christopher M., Duta, Nicolae, Tremblay, Jerome, Almahairi, Amjad, Kaushik, Lakshmish, Boisvert, Maryse
• Granted Patent: US 10,282,419 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

An arrangement and corresponding method are described for multi-domain natural language processing. Multiple parallel domain pipelines are used for processing a natural language input. Each domain pipeline represents a different specific subject domain of related concepts. Each domain pipeline includes a mention module that processes the natural language input using natural language understanding (NLU) to determine a corresponding list of mentions, and an interpretation generator that receives the list of mentions and produces a rank-ordered domain output set of sentence-level interpretation candidates. A global evidence ranker receives the domain output sets from the domain pipelines and produces an overall rank-ordered final output set of sentence-level interpretations.

Claims

1. An arrangement for multi-domain natural language processing comprising:

a plurality of parallel domain pipelines for processing a natural language input, each domain pipeline representing a different specific subject domain of related concepts, and each domain pipeline including;

a. a mention module that processes the natural language input using natural language understanding (NLU) to determine a corresponding list of mentions, and

b. an interpretation generator that receives the list of mentions and produces a rank-ordered domain output set of sentence-level interpretation candidates; and

a global evidence ranker that receives the domain output sets from the domain pipelines and produces an overall rank-ordered final output set of sentence-level interpretations.

2. An arrangement according to claim 1, wherein the arrangement operates in real time to process the natural language input with minimum response patency.

3. An arrangement according to claim 1, wherein the mention module uses both rule-based NLU and statistical-based NLU to determine the corresponding list of mentions.

4. An arrangement according to claim 1, wherein the mention module further associates additional semantic attachments related to the mentions and includes the semantic attachments with the list of mentions to the interpretation generator.

5. An arrangement according to claim 4, wherein the global evidence ranker further includes mentions and semantic attachments with the final output set of sentence-level interpretations.

6. An arrangement according to claim 1, wherein the list of mentions includes a confidence value for each mention.

7. An arrangement according to claim 1, wherein the interpretation generator further includes a confidence value for each interpretation candidate.

8. An arrangement according to claim 1, wherein the interpretation generator determines a query intent representing a semantic meaning of the natural language input to produce the interpretation candidates.

9. An arrangement according to claim 1, wherein the interpretation generator extracts relationships between mentions to produce the interpretation candidates.

10. An arrangement according to claim 1, wherein the interpretation generator performs a semantic re-ranking of the interpretations candidates within the domain output set.

11. A method for multi-domain natural language processing comprising:

processing a natural language input in a plurality of parallel domain pipelines, each domain pipeline representing a different specific subject domain of related concepts, the processing including;

a. processing the natural language input using natural language understanding (NLU) to determine a corresponding list of mentions, and

b. processing the list of mentions to produce a rank-ordered domain output set of sentence-level interpretation candidates; and

processing the domain output sets from the domain pipelines to produce an overall rank-ordered final output set of sentence-level interpretations.

12. A method according to claim 11, wherein the method operates in real time to process the natural language input with minimum response patency.

13. A method according to claim 11, wherein both rule-based NLU and statistical-based NLU are used to determine the corresponding list of mentions.

14. A method according to claim 11, wherein processing the natural language input includes associating additional semantic attachments related to the mentions and including the semantic attachments with the list of mentions.

15. A method according to claim 14, wherein mentions and semantic attachments are included with the sentence-level interpretations.

16. A method according to claim 11, wherein the list of mentions includes a confidence value for each mention.

17. A method according to claim 11, wherein each interpretation candidate further includes an associated confidence value.

18. A method according to claim 11, wherein producing the interpretation candidates includes determining a query intent representing a semantic meaning of the natural language input.

19. A method according to claim 11, wherein producing the interpretation candidates includes extracting relationships between mentions.

20. A method according to claim 11, wherein producing the interpretation candidates includes performing a semantic re-ranking of the interpretations candidates within the domain output set.