Writing assistance using machine translation techniques

• US 7,752,034 B2
• Filed: 10/07/2005
• Issued: 07/06/2010
• Est. Priority Date: 11/12/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Brockett, Christopher, Dolan, William
• Primary Examiner(s): Dorvil; Richemond
• Assistant Examiner(s): He; Jialong
• Application Number:
  US11/246,979
• Publication Number:
  US 20060053001A1
• Time in Patent Office: 1,733 Days
• Field of Search: 704 1- 10
• US Class Current: 704/9
• CPC Class Codes: G06F 40/253 Grammatical analysis; Style critique
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
• CPC Class Codes: Y10S 707/99931 Database or file accessing

Abstract

A system is configured to provide writing assistance within a monolingual input environment based on statistical machine translation techniques typically utilized to translate from an input language to a different target language.

Claims

1. A computer-implemented method for providing writing assistance to a user, the method comprising:

automatically generating a mapping between two strings of text that are in a same language, wherein automatically generating comprises automated generation by a computing device based on statistical or heuristic analysis of the two strings of text relative to a collection of training data, wherein each of the two strings of text is a phrase consisting of more than one word, and wherein analysis of the two strings of text comprises analysis of the two strings of text relative to a collection of native speaker data, the collection of native speaker data comprising pre- and post-editing text samples, pairs of well-formed and deliberately lesioned sentences, aligned sentence pairs collected from paraphrase data, and hand-created correction examples;

providing, based at least in part on the mapping, an indication of a style transformation that could be applied to modify at least a portion of input received from the user, wherein the input received from the user comprises words that are completely in the same language as the two strings of text;

wherein providing an indication of a style transformation comprises providing a list of suggested substitutions that each represent a potential shift from a first style in a first domain to a second style in a second domain;

wherein providing a list comprises providing a list ranked so as to reflect analysis relative to a language model trained based on a collection of well-formed data;

wherein each of the suggested substitutions comprises a string of words that is presented completely in the same language as the two strings of text and as the input received from the user; and

wherein providing an indication of a style transformation comprises evaluating at least one candidate style transformation relative to a target language model.

2. The method of claim 1, wherein automatically generating a mapping between two strings of text comprises automatically generating a mapping between a first string of text and a second string of text, wherein one of the first and second strings is in a technical domain and the other is in a non-technical domain, and wherein providing an indication of a style transformation comprises utilizing the mapping as a basis for analyzing content of the input received from the user and providing an indication of at least one substitution that can be applied to transform at least a portion of the input.

3. The method of claim 1, further comprising:

building a model of the style transformation based at least in part on the mapping;

automatically detecting the user input based at least in part on the user input diverging from the target language model; and

automatically highlighting the user input.

4. The method of claim 1, wherein providing an indication of a style transformation comprises automatically applying a modification so as to transform at least a portion of the input and wherein automatically applying the modification comprises replacing without the user'"'"'s review, approval, or selection.

5. The method of claim 4, further comprising:

flagging the automatically applied modification as an editorial change.

6. The method of claim 4, wherein the user is not notified of the automatically applied modification.

7. The method of claim 1, wherein providing a list comprises providing a list to the user so as to support a user selection process, wherein automatically generating a mapping between two strings of text comprises automatically generating a mapping between a first string of text and a second string of text, wherein one of the first and second strings is in one dialect of a language and the other is in another dialect of the same language, and wherein evaluating said at least one candidate style transformation comprises evaluating relative to a target language model trained to reflect a collection of well-formed data.

8. The method of claim 1, wherein providing a list comprises providing as input to an automated selection algorithm, and wherein providing an indication of a style transformation comprises providing an indication of a transposition of adjacent words.

9. The method of claim 1, further comprising:

the user highlighting a segment of text to form the user input; and

wherein providing an indication comprises providing the indication in response to the user'"'"'s highlighting.

10. The method of claim 9, further comprising:

the user expanding the segment of text to view further suggested replacements.

11. The method of claim 9, further comprising:

the user contracting the segment of text to view further suggested replacements.

12. A computer-implemented writing assistance system comprising:

an alignment component that automatically generates a model of a style transformation based on a comparison of two strings of text that are in a same language, each of the two strings of text being a phrase comprising more than one word, wherein automatically generating comprises automated generation by a computing processor based on statistical or heuristic analysis of the two strings of text relative to a collection of training data also in the same language, wherein automatically generating comprises automatically generating without processing text in a language other than the same language as the two strings of text, wherein automatically generating comprises automatically generating based on a statistical or heuristic alignment of the two strings, said statistical or heuristic alignment being alignment of more than one word phrases, and wherein the analysis of the two string of text relative to the collection of training data comprises analysis of the two strings of text relative to a collection of non-native speaker data, the collection of non-native speaker data comprising pre- and post-editing text samples of typical non-native speaker errors, pairs of well-formed sentences and sentences deliberately lesioned sentences to replicate non-native speaker patterns, aligned sentence pairs lesioned with content words in the native language, and aligned sentence pairs where one member of the pair is low-quality machine translation; and

a decoder component configured to process an input string that includes words or phrases and to generate, based at least in part on information received from the alignment component, a list of proposed substitutions that potentially could be applied to transform the style of at least a portion of the input string into a well-formed string, wherein the list of proposed substitutions comprises a list ranked so as to reflect analysis relative to a language model trained based on a collection of well-formed data.

13. The system of claim 12, wherein the decoder component is further configured to switch between native speaker and non-native speaker models.

14. The system of claim 13, wherein automated generation by a computing processor based on statistical or heuristic analysis comprises utilizing an n-gram language model with optional smoothing.

15. A computer-implemented method for providing writing assistance to a user, the method comprising providing writing assistance, using a processor, based on an automated statistical comparison of a collection of source texts reflecting a first style in a first domain relative to associated target texts reflecting a second style in a second domain, the source texts being in a same language as the target texts, wherein the automated statistical comparison is a monolingual process in that only data in the same language is processed without reference to any corresponding data in a different language, wherein the automated statistical comparison includes a statistical alignment of phrases in the same language, each of the phrases comprising more than one word, wherein the source texts and the target texts contain no grammatical or spelling errors, and wherein the automated statistical comparison of the source texts and the target texts comprises analysis of the source texts and the target texts relative to a collection of non-native speaker data, the collection of non-native speaker data comprising pre- and post-editing text samples of typical non-native speaker errors, pairs of well-formed sentences and sentences deliberately lesioned sentences to replicate non-native speaker patterns, aligned sentence pairs lesioned with content words in the native language, and aligned sentence pairs where one member of the pair is low-quality machine translation.

16. The method of claim 15, wherein providing writing assistance based on a comparison comprises providing writing assistance based on a comparison of texts received through a computer network from a community of users, and wherein the source texts are in a technical domain and the target texts are in a non-technical domain.

17. The method of claim 15, wherein providing writing assistance further comprises switching between native speaker and non-native speaker models.

18. The method of claim 17, further comprising:

utilizing a processor that is a functional component of the computer to apply the writing assistance to an output of a speech to text engine.