Apparatus and method for spoken language understanding by using semantic role labeling

• US 7,742,911 B2
• Filed: 03/31/2005
• Issued: 06/22/2010
• Est. Priority Date: 10/12/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Hakkani-Tur, Dilek Z., Chotimongkol, Ananlada, Tur, Gokhan
• Primary Examiner(s): Opsasnick, Michael N
• Application Number:
  US11/095,299
• Publication Number:
  US 20060080101A1
• Time in Patent Office: 1,909 Days
• Field of Search: 704/4, 704/9, 704/251, 704/254
• US Class Current: 704/4
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

An apparatus and a method are provided for using semantic role labeling for spoken language understanding. A received utterance semantically parsed by semantic role labeling. A predicate or at least one argument is extracted from the semantically parsed utterance. An intent is estimated based on the predicate or the at least one argument. In another aspect, a method is provided for training a spoken language dialog system that uses semantic role labeling. An expert is provided with a group of predicate/argument pairs. Ones of the predicate/argument pairs are selected as intents. Ones of the arguments are selected as named entities. Mappings from the arguments to frame slots are designed.

Claims

1. A spoken language understanding method, the method causing a computing device to perform steps comprising:

semantically parsing, via a processor of the computing device, a received utterance from a user by semantic role labeling to generate at least one semantic role label for only head words of phrases in the received utterance;

extracting a predicate from the semantically parsed utterance;

identifying a syntactic path between head words and their respective extracted predicate;

estimating an intent of the user based on the extracted predicate and based on predicate/argument pairs, each predicate argument pair comprising a predicate verb and an argument which completes the predicate verb; and

responding to the received utterance based on the estimated intent.

2. The method of claim 1, further causing the computing device to perform steps comprising extracting at least one argument from the semantically parsed utterance and estimating at least one associated named entity using the at least one argument.

3. The method of claim 1, further causing the computing device to output syntactically preprocess the utterance before using semantic role labeling to semantically parse the utterance.

4. The method of claim 1, further causing the computing device to perform steps comprising:

mapping the estimated intent to a frame;

extracting at least one argument from the semantically parsed utterance; and

mapping the at least one argument to at least one frame slot of the frame.

5. The method of claim 1, further causing the computing device to perform feature extraction of the utterance.

6. The method of claim 5, wherein the performing of feature extraction is performed using part-of-speech tagging, syntactic parsing, and named entity extraction.

7. The method of claim 1, further causing the computing device to output the estimated intent.

8. The method of claim 2, further causing the computing device to output the at least one associated named entity.

9. The method of claim 7, further causing the computing device to perform an action based on the output estimated intent.

10. A method for training a spoken language dialog system that uses semantic role labeling, the method causing a computing device to perform steps comprising:

providing an expert with a plurality of predicate/argument pairs, each predicate argument pair comprising a predicate verb and an argument which completes the predicate verb;

selecting, via a processor of the computing device, ones of the predicate/argument pairs as intents;

selecting, via the processor, ones of the arguments as named entities;

designing mappings from the arguments to frame slots;

generating at least one semantic role label for only head words of phrases in the predicate/argument pairs; and

identifying a syntactic path between head words and their respective extracted predicate.

11. The method of claim 10, further causing the computing device to perform steps comprising:

grouping the arguments; and

defining the grouped ones of the arguments as values of a named entity.

12. The method of claim 10, further causing the computing device to use a clustering approach to determine potential named entities from the arguments.

13. The method of claim 10, further causing the computing device to use a clustering approach to determine potential intents from the predicate/argument pairs.

14. The method of claim 10, wherein the provided predicate/argument pairs are from a domain of interest.

15. An apparatus comprising:

a processor; and

storage for storing data and a plurality of instructions for the processor, wherein the apparatus is configured to execute the plurality of instructions to control the processor to;

semantically parse a received utterance by semantic role labeling to generate at least one semantic role label for only head words of phrases in the received utterance;

extract a predicate from the semantically parsed utterance,

identify a syntactic path between head words and their respective extracted predicate, and

estimate an intent based on the extracted predicate and based on predicate/argument pairs, each predicate argument pair comprising a predicate verb and an argument which completes the predicate verb.

16. The apparatus of claim 15, wherein the apparatus is further configured to:

syntactically preprocess the utterance before semantically parsing the utterance by semantic role labeling.

17. The apparatus of claim 15, wherein the apparatus is further configured to:

map the estimated intent to a frame;

extract at least one argument from the semantically parsed utterance; and

map the at least one argument to at least one frame slot of the frame.

18. The apparatus of claim 15, wherein the apparatus is further configured to:

perform feature extraction of the utterance.

19. The apparatus of claim 18 wherein the apparatus is configured to perform the feature extraction using part-of-speech tagging, syntactic parsing, and named entity extraction.

20. The apparatus of claim 15, wherein the apparatus is further configured to extract at least one argument from the semantically parsed utterance and estimate at least one associated named entity using the at least one argument.

21. The apparatus of claim 20, where in the apparatus is further configured to output the associated named entities.

22. The apparatus of claim 15, wherein the apparatus is further configured to output the estimated intent.

23. The apparatus of claim 22, wherein the apparatus is further configured to:

perform an action based on the output estimated intent.

24. An apparatus comprising:

a processor;

means for semantically parsing, via the processor, a received utterance by semantic role labeling to generate at least one semantic role label for only head words of phrases in the received utterance;

means for extracting a predicate from the semantically parsed utterance;

means for identifying a syntactic path between head words and their respective extracted predicate

means for estimating an intent based the extracted predicate and based on predicate/argument pairs, each predicate argument pair comprising a predicate verb and an argument which completes the predicate verb; and

means for responding to the received utterance based on the estimated intent.