Automatic detection and application of editing patterns in draft documents

• US 7,640,158 B2
• Filed: 11/08/2005
• Issued: 12/29/2009
• Est. Priority Date: 11/08/2005
• Status: Active Grant

• Current Assignee: Solventum Intellectual Properties Company (Solventum Corp.)
• Original Assignee: Multimodal Technologies Incorporated (3M Company)
• Inventors: Finke, Michael, Fritsch, Juergen, Detlef, Koll
• Primary Examiner(s): Armstrong; Angela A
• Application Number:
  US11/269,094
• Publication Number:
  US 20070106494A1
• Time in Patent Office: 1,512 Days
• Field of Search: 704/9, 704/235
• US Class Current: 704/235
• CPC Class Codes: G06F 40/232 Orthographic correction, e.g. spell checking or vowelisation
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

An error detection and correction system extracts editing patterns and derives correction rules from them by observing differences between draft documents and corresponding edited documents, and/or by observing editing operations performed on the draft documents to produce the edited documents. The system develops classifiers that partition the space of all possible contexts into equivalence classes and assigns one or more correction rules to each such class). Once the system has been trained, it may be used to detect and (optionally) correct errors in new draft documents. When presented with a draft document, the system identifies first content (e.g., text) in the draft document and identifies a context of the first content. The system identifies a correction rule based on the first content and the first context. The system may use a classifier to identify the correction rule. The system applies the correction rule to the first content to produce second content.

Claims

1. A method performed by a computer processor executing computer program instructions tangibly stored on a first computer-readable medium, the method comprising steps of:

(A) tangibly storing, on a second computer-readable medium, a data structure representing a plurality of editing patterns of the form T=(D,E,C), wherein each of the plurality of editing patterns relates particular content D in an original document corpus to corresponding content E in an edited document corpus in a context C shared by contents D and E, wherein the original document corpus and the edited document corpus are tangibly stored on a third and fourth computer-readable medium, respectively;

(B) deriving a plurality of correction rules, tangibly stored on a fifth computer-readable medium, from the plurality of editing patterns; and

(C) deriving a classifier, tangibly stored on a sixth computer-readable medium, for particular content D based on the data structure representing the plurality of editing patterns, the classifier defining decision criteria for selecting one of the plurality of correction rules to apply to content D based on a context C* of content D.

2. The method of claim 1, wherein the step (A) comprises steps of:

(A)(1) for each of a plurality of units of content D in the original document corpus, performing steps of;

(A)(1)(a) identifying a context C of content D in the original document corpus; and

(A)(1)(b) if the edited document corpus contains content E in the context C, wherein content E is not equivalent to content D, identifying an editing pattern T relating content D to content E in context C and storing a tangible representation of the editing pattern T within the data structure.

3. The method of claim 2, further comprising a step of:

(A)(2) for each of the editing patterns T generated in step (A)(1), performing steps of;

(A)(2)(a) identifying the content D and context C of the editing pattern T; and

(A)(2)(b) if the original document corpus contains content D in a context C′ and

the edited document corpus contains content D in a context equivalent to context C′

, identifying an editing pattern T′

relating content D to content D in context C′ and

storing a tangible representation of the editing pattern T′

within the data structure.

4. The method of claim 1, wherein step (C) comprises steps of, for each distinct content E_n represented in the plurality of editing patterns T=(D,E,C):

(C)(1) identifying a subset of the plurality of editing patterns for which E=E_n;

(C)(2) identifying at least one common feature of the contexts of the subset of the plurality of editing patterns; and

(C)(3) defining an equivalence class having decision criteria based on the at least one common feature.

5. The method of claim 1, further comprising a step of:

(D) applying the at least one correction rule to an instance of content D in a further document to edit the instance of content D in the further document.

6. The method of claim 1, further comprising a step of:

(D) using contents D of the plurality of editing patterns to identify at least one candidate error region in a further document.

7. The method of claim 1, wherein the step (A) comprises steps of:

(A)(1) identifying editing actions performed on the original document corpus to produce the edited document corpus; and

(A)(2) generating the plurality of editing patterns to reflect the editing actions.

8. The method of claim 1, further comprising a step of:

(C) prior to the step (A), transcribing speech to produce the original document corpus.

9. The method of claim 1, wherein the step (C) comprises a step of using an automated transcription system to transcribe speech to produce the original document corpus.

10. The method of claim 1, wherein the step (C) comprises a step of using at least one human transcriptionist to produce the original document corpus.

11. A computer readable medium tangibly storing computer-readable signals representing the at least one correction rule derived in step (B) of claim 1.

12. A computer readable medium tangibly storing computer-readable signals representing the at least one classifier derived in step (C) of claim 1.

13. A method performed by a computer processor executing computer program instructions tangibly stored on a first computer-readable medium, the method comprising steps of:

(A) tangibly storing, on a second computer-readable medium, a data structure representing a plurality of editing patterns of the form T=(D,E,C), wherein each of the plurality of editing patterns relates particular content D in an original document corpus to corresponding content E in an edited document corpus in a context C shared by contents D and E, wherein the original document corpus and the edited document corpus are tangibly stored on a third and fourth computer-readable medium, respectively, wherein (A) comprises;

(A)(1) comparing documents in the original document corpus to corresponding documents in the edited document corpus to identify differences between them, the differences comprising at least one word that appears in the original document corpus in a particular context C₀ and that does not appear in the edited document corpus in the context C₀; and

(A)(2) generating the plurality of editing patterns to reflect the identified differences between the documents in the original document corpus and the documents in the edited document corpus; and

(B) deriving at least one correction rule, tangibly stored on a fifth computer-readable medium, from the data structure representing the plurality of editing patterns.

14. The method of claim 13, wherein the step (A)(1) comprises steps of:

(A)(1)(a) aligning a first document in the original document corpus and a corresponding document in the edited document corpus; and

(A)(2)(b) identifying the differences between the original document corpus and the edited document corpus based on the alignment.

15. A computer program product, comprising a first computer-readable medium having computer readable program code tangible embodied therein, said computer readable program code adapted to be executed by a processor to implement a method, the method comprising:

(A) tangibly storing, on a second computer-readable medium, a data structure representing a plurality of editing patterns of the form T=(D,E,C), wherein each of the plurality of editing patterns relates particular content D in an original document corpus to corresponding content E in an edited document corpus in a context C shared by contents D and E, wherein the original document corpus and the edited document corpus are tangibly stored on a third and fourth computer-readable medium, respectively;

(B) deriving a plurality of correction rules, tangibly stored on a fifth computer-readable medium, from the plurality of editing patterns; and

(C) deriving a classifier, tangibly stored on a sixth computer-readable medium, for particular content D based on the plurality of editing patterns, the classifier defining decision criteria for selecting one of the plurality of correction rules to apply to content D based on a context C* of content D.

16. The computer program product of claim 15, wherein (A) comprises, for each of a plurality of units of content D in the original document corpus:

(A)(1) identifying a context C of content D in the original document corpus; and

(A)(2) identifying an editing pattern T relating content D to content E in context C and storing a tangible representation of the editing pattern T within the data structure if the edited document corpus contains content E in the context C, wherein content E is not equivalent to content D.

17. The computer program product of claim 15, wherein the method further comprises:

(D) applying the at least one correction rule to an instance of content D in a further document to edit the instance of content D in the further document.

18. The computer program product of claim 15, wherein the method further comprises:

(D) using contents D of the plurality of editing patterns to identify at least one candidate error region in a further document.

19. The computer program product of claim 15, wherein (A) comprises:

(A)(1) comparing documents in the original document corpus to corresponding documents in the edited document corpus to identify differences between them, the differences comprising at least one word that appears in the original document corpus in a particular context C₀ and that does not appear in the edited document corpus in the context C₀; and

(A)(2) generating the plurality of editing patterns to reflect the identified differences between the documents in the original document corpus and the documents in the edited document corpus.

20. The computer program product of claim 19, wherein (A)(1) comprises:

(A)(1)(a) aligning a first document in the original document corpus and a corresponding document in the edited document corpus; and

(A)(1)(b) identifying the differences between the original document corpus and the edited document corpus based on the alignment.

21. The computer program product of claim 15, wherein (A) comprises:

(A)(1) identifying editing actions performed on the original document corpus to produce the edited document corpus; and

(A)(2) generating the plurality of editing patterns to reflect the editing actions.