Interpreting nonstandard terms in language processing using text-based communications

• US 10,102,845 B1
• Filed: 02/25/2013
• Issued: 10/16/2018
• Est. Priority Date: 02/25/2013
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Abu Zaki, Weam, Shi, Ying, Ramachandran, Rajiv
• Primary Examiner(s): Yehl, Walter
• Application Number:
  US13/776,594
• Time in Patent Office: 2,059 Days
• Field of Search: 704 9, 704257, 704275
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

Features are disclosed for determining a definition or value of a nonstandard term. A user utterance may be processed into one or more candidate transcriptions. An interpretation of the utterance can be generated from the transcriptions. If the transcription includes a word, phrase, or term that is not recognized or is used in a nonstandard way, one or more data stores may be queried regarding the proper value or definition of the term. If a definition or value is not available in the data stores, the user may be prompted to provide one. The user-supplied definition can be saved for future use, and may be used as a general definition of the term for other users.

Claims

1. A system comprising:

a computer-readable memory storing executable instructions; and

one or more processors in communication with the computer-readable memory, wherein the one or more processors are programmed by the executable instructions to at least;

receive first data regarding a first utterance of a first user of a plurality of users;

generate a transcription of the first utterance using an automatic speech recognition (“

ASR”

) subsystem, the transcription comprising a plurality of terms;

identify a first term of the plurality of terms as a named entity based at least partly on a part of speech of the first term and a co-occurrence in the first utterance of the first term and a second term, wherein the second term corresponds to an executable action, and wherein the executable action is to be executed based at least partly on the named entity;

determine, using a natural language understanding (“

NLU”

) subsystem, that the named entity is a nonstandard named entity based at least partly on a definition of the named entity not being resolved to a threshold degree of confidence;

determine that the first term is not included as an alias for a different term in a first set of aliases stored in a user data store, wherein the first set of aliases stored in the user data store is personalized to the first user;

after determining that the first term is not included in the first set of aliases personalized for the first user, determine that the first term is not included in a second set of aliases stored in a shared data store, wherein the second set of aliases stored in the shared data store is common to the plurality of users; and

after determining that the first term is not included in the first set of aliases personalized for the first user and is not included in the second set of aliases common to the plurality of users;

analyze text-based communications of the first user to identify an occurrence of a third term;

determine that the first term is an alias for the third term based at least partly on a context of the occurrence of the third term;

include the first term as the alias for the third term in the first set of aliases; and

execute the executable action in response to the utterance based at least partly on the first term.

2. The system of claim 1, wherein the first term comprises one of:

a slang term, a user-created term, or a term comprising a value that varies depending at least partly on a user of the term.

3. The system of claim 1, wherein the one or more processors are further programmed by the executable instructions to:

receive third data regarding a third utterance of a second user, the third utterance comprising the first term; and

generate a response, based at least partly on the alias of the first term, to the third utterance.

4. A computer-implemented method comprising:

under control of one or more computing devices configured with specific computer-executable instructions,

receiving audio data representing a first utterance of a first user in a plurality of users;

generating a transcription of the first utterance using automatic speech recognition and the audio data, wherein the transcription comprises a first plurality of terms;

identifying a first term of the first plurality of terms as an intent that corresponds to an executable action;

identifying a second term of the first plurality of terms as a named entity, wherein the executable action is to be executed based at least partly on the named entity, and wherein the second term is identified based at least partly on a part of speech of the second term and a co-occurrence of the second term with the first term in the transcription;

determining that the second term is a nonstandard term based at least on a definition of the nonstandard term not being determined to a threshold degree of confidence;

determining that the definition of the nonstandard term is not included in information personalized for the first user;

after determining that the definition of the nonstandard term is not included in the information personalized for the first user, determining that the definition of the nonstandard term is not included in information common to the plurality of users;

after determining that the definition of the nonstandard term is not included in the information common to the plurality of users;

analyzing text-based communications of the first user for information regarding the nonstandard term;

determining the definition of the nonstandard term, the definition based at least partly on a context of an occurrence of the nonstandard term in the text-based communications;

storing definition data regarding the definition of the nonstandard term in a data store;

receiving a representation of a second utterance of the first user, the representation of the second utterance comprising a second plurality of terms;

identifying the nonstandard term in the second plurality of terms; and

executing the executable action in response to the second utterance based at least partly on the definition data.

5. The computer-implemented method of claim 4, wherein the nonstandard term comprises one of:

a known term used in a nonstandard context, or an unknown term.

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

determining at least one of;

a part of speech of the nonstandard term, or a category relating to the nonstandard term, wherein the category comprises a person, a place, or a thing.

7. The computer-implemented method of claim 4, wherein determining that the second term is the nonstandard term is further based at least partly on failing to locate the nonstandard term in a data store of named entities.

8. The computer-implemented method of claim 4, wherein determining that the second term is the nonstandard term is further based at least partly on failing to obtain a definition for the nonstandard term in response to a search of unstructured data.

9. The computer-implemented method of claim 4, wherein identifying the nonstandard term in the second plurality of terms comprises:

querying the data store for the definition data associated with the nonstandard term, the definition data associated with the first user.

10. The computer-implemented method of claim 4, wherein storing the definition data comprises:

generating a phonetic representation of the nonstandard term; and

storing the phonetic representation.

11. The computer-implemented method of claim 10, wherein identifying the nonstandard term in the second plurality of terms comprises performing a query for the definition data, the query based at least partly on the phonetic representation.

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

receiving a representation of a third utterance of a second user in the plurality of users, the representation of the third utterance comprising one or more terms;

identifying the nonstandard term in the one or more terms of the representation of the third utterance; and

generating a response to the third utterance based at least partly on the definition data.

13. The computer-implemented method of claim 12, wherein generating the response to the third utterance comprises:

querying the data store for the definition data associated with the nonstandard term.

14. The computer-implemented method of claim 4, wherein the plurality of users share one or more characteristics.

15. The computer-implemented method of claim 14, wherein the one or more characteristics comprises at least one of a geographic location, an age, an ethnicity, or a gender.

16. The computer-implemented method of claim 4, wherein storing definition data regarding the definition of the nonstandard term in the data store comprises:

determining that another definition of the nonstandard term is included in the information common to the plurality of users;

determining that the definition of the nonstandard term is dissimilar to the other definition of the nonstandard definition; and

storing the definition of the nonstandard term in the data store such that the data store comprises the definition of the nonstandard term and the other definition of the nonstandard term.

17. The computer-implemented method of claim 4, further comprising determining that the definition of the nonstandard term has been associated with the nonstandard term for a threshold number of multiple users in the plurality of users that based at least partly on the definition of the nonstandard term being generic to the plurality of users.

18. The computer-implemented method of claim 4, wherein:

the information personalized for the first user comprises at least one of a glossary of terms personalized for the first user or a set of definitions personalized for the first user; and

the information personalized for the first user is stored in a user-specific data store.

19. The computer-implemented method of claim 4, wherein:

the information common to the plurality of users comprises at least one of a glossary or dictionary of nonstandard terms commonly used by the plurality of users; and

the information common to the plurality of users is stored in a shared data store.

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

determining that the first term comprises a verb, wherein the identifying the first term as the intent corresponding to the executable action is based at least partly on the determining that the first term comprises the verb; and

determining that the second term comprises a noun, wherein the identifying the second term as the named identity is based at least partly on the determining that the second term comprises the noun.

21. The computer-implemented method of claim 4, wherein analyzing the text-based communications of the first user for information regarding the nonstandard term comprises analyzing at least one of:

text messages or email messages.

22. The computer-implemented method of claim 4, wherein analyzing the text-based communications of the first user for information regarding the nonstandard term comprises:

analyzing a first text-based communication to identify an occurrence of the nonstandard term; and

determining that the occurrence of the non-standard term is associated with an addressee of the first text-based communication.

23. The computer-implemented method of claim 22, further comprising:

determining an identity of the addressee; and

determining that the nonstandard term refers to the identity of the addressee.

24. The computer-implemented method of claim 4, wherein executing the executable action comprises at least one of:

initiating playback of music;

generating a text-to-speech presentation representing a response to the second utterance;

dialing a phone number;

or initiating download of a file.

25. One or more non-transitory computer-readable media comprising executable code that, when executed, cause one or more computing devices to perform a process comprising:

receiving audio data representing a first utterance of a user of a plurality of users;

generating a transcription of the first utterance using automatic speech recognition and the audio data, the transcription comprising a first plurality of terms;

identifying a first term of the first plurality of terms as an intent that corresponds to an executable action;

identifying a second term of the first plurality of terms as a named entity, wherein the executable action is to be executed based at least partly on the named entity, and wherein the second term is identified based at least partly on a part of speech of the second term and a co-occurrence of the second term with the first term in the transcription;

determining that the second term is a nonstandard term based at least on a definition of the nonstandard term not being determined to a threshold degree of confidence;

determining that the definition of the nonstandard term is not included in information personalized for the user;

after determining that the definition of the nonstandard term is not included in the information personalized for the user;

analyzing text-based communications of the user for information regarding the nonstandard term;

determining the definition of the nonstandard term based at least partly on a context of an occurrence of the nonstandard term in the text-based communications;

storing definition data regarding the definition of the nonstandard term in a data store;

receiving a representation of a second utterance of the user, the second utterance comprising a second plurality of terms;

identifying the nonstandard term in the second plurality of terms; and

executing the executable action in response to the second utterance based at least partly on the definition data.

26. The one or more non-transitory computer-readable media of claim 25, wherein the process further comprises:

determining at least one of a part of speech or a category relating to the nonstandard term, wherein the category comprises a person, a place, or a thing.

27. The one or more non-transitory computer-readable media of claim 25, the process further comprising:

prompting the user for information regarding the nonstandard term, wherein prompting the user comprises one of;

generating a visual display, or generating a text-to-speech presentation.

28. The one or more non-transitory computer-readable media of claim 25, wherein determining that the second term is the nonstandard term is further based at least partly on failing to locate the nonstandard term in a data store of named entities.

29. The one or more non-transitory computer-readable media of claim 25, wherein determining that the second term is the nonstandard term is further based at least partly on failing to obtain a definition for the nonstandard term in response to a search of unstructured data.

30. The one or more non-transitory computer-readable media of claim 25, wherein identifying the nonstandard term in the second plurality of terms comprises:

querying the data store comprising the information common to the plurality of users for the definition data associated with the nonstandard term.

31. The one or more non-transitory computer-readable media of claim 25, wherein storing the definition data comprises:

generating a phonetic representation of the nonstandard term; and

storing the phonetic representation.

32. The one or more non-transitory computer-readable media of claim 31, wherein identifying the nonstandard term in the second plurality of terms comprises performing a query for the definition data, the query based at least partly on the phonetic representation.