Translating natural language utterances to keyword search queries

• US 10,878,009 B2
• Filed: 07/24/2018
• Issued: 12/29/2020
• Est. Priority Date: 08/23/2012
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Hakkani-Tur, Dilek Zeynep, Tur, Gokhan, Iyer, Rukmini, Heck, Larry Paul
• Primary Examiner(s): Leroux, Etienne P
• Application Number:
  US16/043,304
• Publication Number:
  US 20180329918A1
• Time in Patent Office: 889 Days
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3332 Query translation
• CPC Class Codes: G06F 16/3334 Selection or weighting of terms from queries, including natural language queries
• CPC Class Codes: G06F 16/3338 Query expansion
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/84 Mapping; Conversion
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

Natural language query translation may be provided. A statistical model may be trained to detect domains according to a plurality of query click log data. Upon receiving a natural language query, the statistical model may be used to translate the natural language query into an action. The action may then be performed and at least one result associated with performing the action may be provided.

Claims

1. A method for providing natural language query translation, the method comprising:

receiving a spoken natural language query;

translating the spoken natural language query into a search query according to a statistical model, wherein the statistical model is trained according to log data for a plurality of historical query result selections, and wherein the log data is mined to train the statistical model for domain detection in the absence of in-domain data;

performing the translated search query; and

providing at least one result associated with the translated search query.

2. The method of claim 1, wherein the statistical model is trained by identifying a plurality of domain independent salient phrases.

3. The method of claim 2, wherein each of the plurality of domain independent salient phrases comprises at least one word indicating that an associated search query comprises a natural language search query.

4. The method of claim 1, further comprising identifying a plurality of search queries associated with the log data that comprise natural language search queries according to the plurality of domain independent salient phrases.

5. The method of claim 4, further comprising generating a query pair by correlating at least one of the plurality of natural language search queries to at least one keyword-based search query.

6. The method of claim 5, wherein the correlation between the at least one of the plurality of natural language search queries and the at least one keyword-based search query is associated with a Uniform Resource Locator (URL) distribution.

7. The method of claim 6, wherein performing the search query comprises:

searching the log data for a query pair corresponding to the search query; and

identifying a domain associated with the search query according to the URL distribution.

8. A system comprising:

a memory storage; and

a processing unit coupled to the memory storage, wherein the processing unit is operable to;

receive a spoken natural language query from a user,

map the natural language query into a keyword-based query according to a statistical model, wherein the statistical model is trained according to log data for a plurality of historical query result selections, and wherein the log data is mined to train the statistical model for domain detection in the absence of in-domain data;

perform a search according to a query pair comprising the spoken natural language query and the keyword-based query; and

provide a plurality of results associated with the search to the user.

9. The system of claim 8, wherein the processor is further operative to:

detect a domain associated with the spoken natural language query; and

strip at least one domain-independent word from the spoken natural language query.

10. The system of claim 9, wherein the processor is further operative to:

identify a subset of a plurality of possible domains according to at least one feature of the spoken natural language query.

11. The system of claim 10, wherein the at least one feature of the spoken natural language query comprises at least one of the following:

a lexical feature, a contextual feature, a semantic feature, a syntactic feature, and a topical feature.

12. The system of claim 8, wherein the statistical model is trained according to a plurality of mined query pairs each comprising a previous natural language query and an associated previous keyword-based query.

13. The system of claim 12, wherein the previous natural language query is identified according to a domain independent salient phrase.

14. The system of claim 12, wherein the previous natural language query and the previous keyword-based query are associated according to a weighted Uniform Resource Locator (URL) graph.

15. The system of claim 12, wherein the mined query pairs each comprise a statistical weighting according to a semantic correlation between the previous natural language query and the previous keyword-based query.

16. A system for providing natural language query translation, the system comprising:

a processor; and

memory storing computer-readable instructions that when executed cause the processor to;

receive a spoken natural language query;

translate the spoken natural language query into a search query according to a statistical model, wherein the statistical model is trained according to log data for a plurality of historical query result selections, and wherein the log data is mined to train the statistical model for domain detection in the absence of in-domain data;

perform the translated search query; and

providing at least one result associated with the translated search query.

17. The system of claim 16, wherein the statistical model is trained according to a plurality of mined query pairs each comprising a previous natural language query and an associated previous keyword-based query.

18. The system of claim 17, wherein the previous natural language query is identified according to a domain independent salient phrase.

19. The system of claim 17, wherein the previous natural language query and the previous keyword-based query are associated according to a weighted Uniform Resource Locator (URL) graph.

20. The system of claim 17, wherein the plurality of mined query pairs each comprise a statistical weighting according to a semantic correlation between the previous natural language query and the previous keyword-based query.