System and method of extracting clauses for spoken language understanding

• US 8,818,793 B1
• Filed: 05/28/2003
• Issued: 08/26/2014
• Est. Priority Date: 12/24/2002
• Status: Expired due to Fees

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Intellectual Property II LP (AT&T, Inc.)
• Inventors: Rahim, Mazin G, Bangalore, Srinivas, Gupta, Narendra K.
• Primary Examiner(s): SPOONER, LAMONT M
• Application Number:
  US10/446,489
• Time in Patent Office: 4,108 Days
• Field of Search: 704/9, 704/231, 704/251, 704/241, 704/25, 715/254, 715/255, 707 2- 6
• US Class Current: 704/9
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting

Abstract

A clausifier and method of extracting clauses for spoken language understanding are disclosed. The method relates to generating a set of clauses from speech utterance text and comprises inserting at least one boundary tag in speech utterance text related to sentence boundaries, inserting at least one edit tag indicating a portion of the speech utterance text to remove, and inserting at least one conjunction tag within the speech utterance text. The result is a set of clauses that may be identified within the speech utterance text according to the inserted at least one boundary tag, at least one edit tag and at least one conjunction tag. The disclosed clausifier comprises a sentence boundary classifier, an edit detector classifier, and a conjunction detector classifier. The clausifier may comprise a single classifier or a plurality of classifiers to perform the steps of identifying sentence boundaries, editing text, and identifying conjunctions within the text.

Claims

1. A method comprising:

annotating data by;

inserting, via a processor and via a discriminative classification approach independent of using n-grams, boundary tags at boundaries in a speech utterance text based on a training with weighted examples, wherein higher weights indicate more difficult examples, and wherein the boundary tags comprise one of a phrase boundary tag, a sentence boundary tag, and a paragraph boundary tag, to yield boundary marked speech utterance text;

thereafter, inserting an edit tag in the boundary marked speech utterance text, to yield edited text and unedited text, wherein the edit tag identifies a portion of the speech utterance text to be removed based on repeated words which do not contribute to language understanding, and wherein the speech utterance text has not been processed to identify clauses;

thereafter, inserting conjunction tags within the unedited text which identify, without relying on punctuation cues, coordinating conjunctions selected from a list comprising {and, but, for, nor, or, so, yet}, to yield conjunction tag text; and

identifying clauses within the speech utterance text based on the boundary marked speech utterance text, the edited text, and the conjunction tag text to yield annotated data; and

iteratively repeating the annotating of the data, where each successive iteration has a longer turn than an immediately preceding iteration and each successive iteration is used to retrain a model associated with the discriminative classification approach.

2. The method of claim 1, wherein different classifiers insert the boundary tags, insert the edit tag and insert the conjunction tags.

3. The method of claim 1, wherein a single classifier inserts the boundary tags, inserts the edit tag and inserts the conjunction tags.

4. The method of claim 1, wherein a plurality of classifiers inserts the boundary tags, inserts the edit tag and inserts the conjunction tags, wherein the plurality of classifiers is iteratively constructed.

5. The method of claim 1, wherein the boundary tags, the edit tag, and the conjunction tags are each annotated using unique tags.

6. The method of claim 5, wherein the unique tags associated with the boundary tags are “

<

” and

“

>

”

, the unique tags associated with the edit tag are “

[” and

“

]”

, and the unique tags associated with the conjunction tags are “

{” and

“

}”

.

7. A computer-readable storage device having instructions stored which, when executed by a computing device, cause the computing device to perform operations comprising:

annotating data by;

inserting, via a processor and via a discriminative classification approach independent of using n-grams, boundary tags at boundaries in a speech utterance text based on a training with weighted examples, wherein higher weights indicate more difficult examples, and wherein the boundary tags comprise one of a phrase boundary tag, a sentence boundary tag, and a paragraph boundary tag, to yield boundary marked speech utterance text;

thereafter, inserting an edit tag in the boundary marked speech utterance text, to yield edited text and unedited text, wherein the edit tag identifies a portion of the speech utterance text to be removed based on repeated words which do not contribute to language understanding, and wherein the speech utterance text has not been processed to identify clauses;

thereafter, inserting conjunction tags within the unedited text which identify, without relying on punctuation cues, coordinating conjunctions selected from a list comprising {and, but, for, nor, or, so, yet}, to yield conjunction tag text; and

identifying clauses within the speech utterance text based on the boundary marked speech utterance text, the edited text, and the conjunction tag text to yield annotated data; and

iteratively repeating the annotating of the data, where each successive iteration has a longer turn than an immediately preceding iteration and each successive iteration is used to retrain a model associated with the discriminative classification approach.

8. The computer-readable storage device of claim 7, wherein different classifiers insert the boundary tags, insert the edit tag and insert the conjunction tags.

9. The computer-readable storage device of claim 7, wherein a single classifier inserts the boundary tags, inserts the edit tag and inserts the conjunction tags.

10. The computer-readable storage device of claim 7, wherein a plurality of classifiers inserts the-boundary tags, inserts the edit tag and inserts the conjunction tags.

11. The computer-readable device of claim 7, wherein the boundary tags, the edit tag, and the conjunction tags are each annotated using unique tags.

12. The computer-readable device of claim 11, wherein the unique tags associated with the boundary tags are “

<

” and

“

>

”

, the unique tags associated with the edit tag are “

[” and

“

]”

, and the unique tags associated with the conjunction tags are “

{” and

“

}”

.

13. A system comprising:

a processor; and

a computer-readable storage medium having stored therein instructions which, when executed by the processor, cause the processor to perform operations comprising;

annotating data by;

inserting, via a processor and via a discriminative classification approach independent of using n-grams, boundary tags at boundaries in a speech utterance text based on a training with weighted examples, wherein higher weights indicate more difficult examples, and wherein the boundary tags comprise one of a phrase boundary tag, a sentence boundary tag, and a paragraph boundary tag, to yield boundary marked speech utterance text;

thereafter, inserting an edit tag in the boundary marked speech utterance text, to yield edited text and unedited text, wherein the edit tag identifies a portion of the speech utterance text to be removed based on repeated words which do not contribute to language understanding, and wherein the speech utterance text has not been processed to identify clauses;

thereafter, inserting conjunction tags within the unedited text which identify, without relying on punctuation cues, coordinating conjunctions selected from a list comprising {and, but, for, nor, or, so, yet}, to yield conjunction tag text; and

identifying clauses within the speech utterance text based on the boundary marked speech utterance text, the edited text, and the conjunction tag text to yield annotated data; and

iteratively repeating the annotating of the data, where each successive iteration has a longer turn than an immediately preceding iteration and each successive iteration is used to retrain a model associated with the discriminative classification approach.

14. The system of claim 13, wherein different classifiers insert the boundary tags, insert the edit tag and insert the conjunction tags.

15. The system of claim 13, wherein a single classifier inserts the boundary tags, inserts the edit tag and inserts the conjunction tags.

16. The system of claim 13, wherein a plurality of classifiers perform the steps of inserts the boundary tags, inserts the edit tag and inserts the conjunction tags.

17. The system of claim 13, wherein the boundary tags, the edit tag, and the conjunction tags are each annotated using unique tags.

18. The system of claim 17, wherein the unique tags associated with the boundary tags are “

<

” and

“

>

”

, the unique tags associated with the edit tag are “

[” and

“

]”

, and the unique tags associated with the conjunction tags are “

{” and

“

}”

.