SYSTEMS AND METHODS FOR RESPONDING TO NATURAL LANGUAGE SPEECH UTTERANCE

• US 20080235023A1
• Filed: 05/30/2008
• Published: 09/25/2008
• Est. Priority Date: 06/03/2002
• Status: Active Grant

• Current Assignee: Dialect, LLC
• Original Assignee: VoiceBox Technologies Corporation (Microsoft Corporation)
• Inventors: Kennewick, Robert A., Locke, David, Kennewick, Richard, Freeman, Tom, Kennewick, Michael R.
• Granted Patent: US 8,015,006 B2
• Time in Patent Office: Days
• US Class Current: 704/257
• 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
• CPC Class Codes: G10L 2015/228 of application context
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching

Abstract

Systems and methods for receiving natural language queries and/or commands and execute the queries and/or commands. The systems and methods overcomes the deficiencies of prior art speech query and response systems through the application of a complete speech-based information query, retrieval, presentation and command environment. This environment makes significant use of context, prior information, domain knowledge, and user specific profile data to achieve a natural environment for one or more users making queries or commands in multiple domains. Through this integrated approach, a complete speech-based natural language query and response environment can be created. The systems and methods creates, stores and uses extensive personal profile information for each user, thereby improving the reliability of determining the context and presenting the expected results for a particular question or command.

Claims

1-17. -17. (canceled)

18. A method responsive to a user generated natural language speech utterance, comprising:

receiving the user generated natural language speech utterance, the received user utterance containing at least one request;

determining a meaning of the utterance, wherein determining the meaning includes parsing information relating to the utterance to determine a context for the at least one request contained in the utterance;

formulating the at least one request contained in the utterance in accordance with a grammar used by a domain agent that receives, processes, and responds to requests associated with the determined context; and

presenting results of the processed request to the user, the presented results generated as a result of the domain agent processing the formulated request.

19. The method according to claim 18, further comprising recognizing words and phrases contained in the received utterance using information in one or more dictionary and phrase tables, wherein determining the meaning of the utterance further includes parsing the recognized words and phrases to determine the context for the at least one request.

20. The method according to claim 19, wherein recognizing the words and phrases contained in the received utterance includes:

dynamically updating the information in the dictionary and phrase tables based on a dynamic set of prior probabilities or fuzzy possibilities;

determining an identity of the user based on voice characteristics of the utterance; and

associating the recognized words and phrases and a pronunciation of the recognized words and phrases with the identity of the user, the determined context, and the at least one request when the recognized words and phrases satisfy a predetermined confidence level.

21. The method according to claim 19, wherein recognizing the words and phrases contained in the received utterance further includes:

requesting the user to spell one or more of the words and phrases when the recognized words and phrases fail to satisfy a predetermined confidence level; and

updating the information in the dictionary and phrase tables based on a phonetic alphabet spelling of the requested words and phrases, the phonetic alphabet spelling provided by the user in one or more subsequent natural language speech utterances.

22. The method according to claim 19, wherein determining the context for the at least one request contained in the utterance includes:

matching the recognized words and phrases to a set of keywords associated with a dynamic set of prior probabilities or fuzzy possibilities;

scoring one or more possible contexts for the matched keywords using one or more of a profile of the user, data content associated with the domain agent, or a context stack of one or more recent contexts; and

determining the context by selecting one or more of the possible contexts having a score satisfying a predetermined confidence level.

23. The method according to claim 22, wherein determining the context for the at least one request contained in the utterance further includes:

requesting the user to verify the recognized words and phrases when none of the possible contexts have a score satisfying the predetermined confidence level; and

determining the context based on information provided by the user in one or more subsequent natural language speech utterances.

24. The method according to claim 19, wherein formulating the at least one request in accordance with the grammar used by the domain agent includes:

determining one or more required values and one or more optional values for a request in the grammar used by the domain agent;

extracting one or more criteria and one or more parameters from one or more keywords contained in the recognized words and phrases, the criteria and parameters extracted using procedures sensitive to the determined context;

inferring one or more criteria and one or more parameters using a dynamic set of prior probabilities or fuzzy possibilities; and

transforming the extracted and inferred criteria and parameters into a format compatible with the grammar used by the domain agent, the transformed criteria and parameters including all of the required values and one or more of the optional values for the request.

25. The method according to claim 18, wherein the at least one request contained in the utterance includes at least one question and at least one command, wherein the domain agent processes the formulated request by:

directing the at least one command to local or remote processing devices; and

querying local or network information sources to resolve information requested in the at least one question.

26. The method according to claim 25, wherein querying the local or network information sources to resolve the information requested in the at least one question includes:

submitting a plurality of asynchronous queries to the information sources, the plurality of queries including a plurality of duplicate queries submitted to distinct information sources;

asynchronously receiving a plurality of results of the queries from the information sources; and

scoring a relevance of each of the asynchronously received results using a dynamic set of prior probabilities or fuzzy possibilities to determine one or more best responses to the at least one question.

27. The method according to claim 26, wherein determining the one or more best responses to the at least one question includes:

determining one or more tokens required to formulate a response to the question;

extracting one or more values for the determined tokens from the asynchronously received results, the one or more values extracted by applying one or more scraping criteria to the asynchronously received results;

evaluating the extracted values using the dynamic set of prior probabilities or fuzzy possibilities, wherein evaluating the extracted values includes resolving ambiguous, incomplete, or conflicting information; and

selecting one or more best values for the determined tokens from the evaluated values, wherein the one or more best responses include the one or more selected best values.

28. The method according to claim 27, wherein determining the one or more best responses to the at least one question further includes:

determining that the dynamic set of prior probabilities or fuzzy possibilities has failed to resolve the ambiguous, incomplete, or conflicting information; and

resolving the ambiguous, incomplete, or conflicting information using subsequently received results of one or more the plurality of asynchronous queries that remain pending.

29. The method according to claim 27, wherein determining the one or more best responses to the at least one question further includes:

determining that the dynamic set of prior probabilities or fuzzy possibilities has failed to resolve the ambiguous, incomplete, or conflicting information; and

resolving the ambiguous, incomplete, or conflicting information by submitting one or more additional queries to the information sources, wherein submitting the additional queries includes inferring the information sources to be queried based on results of one or more of the plurality of asynchronous queries that have already been received.

30. The method according to claim 27, wherein determining the one or more best responses to the at least one question further includes:

determining that the dynamic set of prior probabilities or fuzzy possibilities has failed to resolve the ambiguous, incomplete, or conflicting information; and

resolving the ambiguous, incomplete, or conflicting information by requesting the user to provide additional information for resolving the ambiguous, incomplete, or conflicting information.

31. The method according to claim 18, wherein presenting the results of the processed request to the user includes:

selecting a format template for presenting the results;

selecting a personality for presenting the results;

determining an order for presenting one or more tokens contained in the results; and

performing one or more variable substitutions and transformations on the tokens to vary a terminology of the results.

32. The method according to claim 31, wherein presenting the results of the processed request to the user further includes:

formatting the results in accordance with the selected format template, wherein a text to speech engine reads the formatted results in accordance with the selected personality to create a system generated speech utterance that presents the results to the user.