Speech and language translation of an utterance

• US 8,914,277 B1
• Filed: 09/20/2011
• Issued: 12/16/2014
• Est. Priority Date: 09/20/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Liu, Ding
• Primary Examiner(s): Chawan, Vijay B
• Assistant Examiner(s): Shin, Seong-Ah A
• Application Number:
  US13/237,510
• Time in Patent Office: 1,183 Days
• Field of Search: 704 1- 10, 704/235, 704/240, 704/246, 704/251, 704/256, 704/260, 704/265, 704/277, 707/703, 707/711
• US Class Current: 704/4
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 15/34 Adaptation of a single recogniser for parallel processing, e.g. by use of multiple processors or cloud computing

Abstract

According to example configurations, a speech-processing system parses an uttered sentence into segments. The speech-processing system translates each of the segments in the uttered sentence into candidate textual expressions (i.e., phrases of one or more words) in a first language. The uttered sentence can include multiple phrases or candidate textual expressions. Additionally, the speech-processing system translates each of the candidate textual expressions into candidate textual phrases in a second language. Based at least in part on a product of confidence values associated with the candidate textual expressions in the first language and confidence values associated with the candidate textual phrases in the second language, the speech-processing system produces a confidence metric for each of the candidate textual phrases in the second language. The confidence metric can indicate degree to which the candidate textual phrase in the second language is an accurate translation of a respective segment in the utterance.

Claims

1. A method comprising:

performing, by computer processing hardware, operations of;

receiving an utterance spoken in a first language;

partitioning a spoken sentence in the utterance into multiple segments, a given segment of the multiple segments including multiple words spoken in the first language;

converting the given segment of the multiple segments into multiple candidate textual phrases in a second language, further comprising;

performing a speech-to-text translation of the given segment into a set of candidate textual expressions in the first language by translating the given segment into at least a first candidate textual expression and a second candidate textual expression in the first language; and

wherein performing the language translation includes;

identifying that the first candidate textual expression translates into a first candidate textual phrase and a second candidate textual phrase; and

identifying that the second candidate textual expression translates into a third candidate textual phrase and a fourth candidate textual phrase, the first candidate textual phrase being identical to the third candidate textual phrase; and

for each respective candidate textual expression in the set;

performing a language translation of the respective candidate textual expression into multiple candidate textual phrases in the second language;

producing a confidence metric for each respective candidate textual phrase of the multiple candidate textual phrases in the second language, the confidence metric indicating a confidence that the respective candidate textual phrase is an accurate translation of the given segment of the utterance into the second language;

producing a confidence value for each of the candidate textual expressions in the first language;

producing a confidence value for each of the candidate textual phrases in the second language; and

generating a confidence metric for the first candidate textual phrase based on a sum of a first term and a second term, the first term being a product of a confidence value for the first candidate textual expression multiplied by a confidence value for the first candidate textual phrase, the second term being a product of a confidence value for the second candidate textual expression multiplied by a confidence value for the third candidate textual phrase.

2. The method as in claim 1 further comprising:

producing the confidence metrics based on a sum of products of confidence values associated with translations of the given segment into candidate textual expressions in the first language and confidence values associated with translations of the candidate textual expressions into the candidate textual phrases in the second language.

3. The method as in claim 2, wherein the candidate textual phrases in the second language are derived from the candidate textual expressions in the first language.

4. The method as in claim 1, wherein each of the first candidate textual phrase, the second candidate textual phrase, and the fourth candidate textual phrase are unique with respect to each other.

5. The method as in claim 1, wherein partitioning the spoken sentence in the utterance comprises producing the given segment to include a phrase of multiple words in the first language but fewer than all words spoken in the sentence.

6. The method as in claim 1, wherein the confidence metric indicates a degree to which the respective candidate textual phrase in the second language is a best candidate translation of the given segment of the utterance into the second language.

7. A method comprising:

performing, by computer processing hardware, operations of;

parsing an uttered sentence into segments;

translating each of the segments into candidate textual expressions in a first language;

translating each of the candidate textual expressions into candidate textual phrases in a second language; and

producing, based at least in part on a product of confidence values associated with the candidate textual expressions in the first language and confidence values associated with the candidate textual phrases in the second language, a confidence metric for each of the candidate textual phrases in the second language, producing the confidence metric including;

executing separate translation paths in which a given segment of the utterance translates into a common candidate textual phrase in the second language, the separate translation paths including a first translation path and a second translation path;

the first translation path including;

a translation of the given segment of the utterance into a first candidate textual expression in the first language and a subsequent translation of the first candidate textual expression in the first language to the common candidate textual phrase in the second language; and

the second translation path including;

a translation of the given segment of the utterance into a second candidate textual expression in the first language and a subsequent translation of the second candidate textual expression in the first language to the common candidate textual phrase in the second language.

8. The method as in claim 7 further comprising:

producing a respective confidence metric of translating the given segment of the utterance in the first language into the common candidate textual phrase in the second language based on a sum of a first product and a second product, the respective confidence metric indicating a confidence that the common candidate textual phrase in the second language is an accurate translation of the given segment of the utterance in the first language;

producing a first confidence value, the first confidence value indicating a respective confidence that the first candidate textual expression is an accurate translation of the given segment of the utterance;

producing a second confidence value, the second confidence value indicating a respective confidence that the common candidate textual phrase in the second language is an accurate translation of the first candidate textual expression;

producing a third confidence value, the third confidence value indicating a respective confidence that the second candidate textual expression is an accurate translation of the given segment of the utterance;

producing a fourth confidence value, the fourth confidence value indicating a respective confidence that the common candidate textual phrase in the second language is an accurate translation of the second candidate textual expression;

the first product generated via multiplication of the first confidence value by the second confidence value; and

the second product generated via multiplication of the third confidence value by the fourth confidence value.