Passively suggesting text in an electronic document

1. A method of providing replacement text for a text unit in an electronic input document, comprising:

examining an electronic input document presented on a display of a client to identify a text unit in the electronic input document and contextual information about the input document;

determining, for the identified text unit, a set of annotations for the text unit and the input document responsive to the contextual information, the set of annotations describing predictive characteristics of replacement text for the text unit;

representing one or more words of the identified text unit as one or more vectors that characterize meaning of the identified text unit;

evaluating, responsive to the set of annotations, a set of candidate texts from a corpus of documents that can replace the text unit, the evaluating comprising:

identifying a set of annotations for a corpus text unit, the set of annotations for the corpus text unit describing how well the corpus text unit satisfies the predictive characteristics for the replacement text,

performing a plurality of evaluations to generate an evaluation score, each evaluation comparing an annotation from the set of annotations for the corpus text unit with an annotation of the set of annotations for the text unit in the input document,

comparing one or more vectors associated with the corpus text unit with the one or more vectors representing the identified text unit to generate a meaning score,

determining a confidence score for the corpus text unit that indicates suitability of the corpus text unit as the replacement text for the text unit, wherein the confidence score is a combination of the evaluation score and the meaning score, and

adding the corpus text unit to the set of candidate texts responsive to determining the confidence score;

identifying a subset of the set of candidate texts as a set of replacement texts for the text unit; and

presenting at least one replacement text from the set of replacement texts on the display of the client.

2. The method of claim 1 , wherein there are a plurality of text units in the electronic input document and wherein the examining, determining, identifying, and evaluating are passively performed for ones of the plurality of text units.

3. The method of claim 1 , wherein identifying the candidate texts comprises:

adding the candidate text to the set of replacement texts responsive to comparison of the confidence score for the candidate text to a threshold.

4. The method of claim 1 , wherein presenting the replacement texts on the display of the client comprises:

graphically flagging the identified text unit on the display of the client, the graphical flag indicating that replacement texts are available for the text unit.

5. The method of claim 1 , wherein presenting the replacement texts on the display of the client comprises:

graphically flagging the identified text unit on the display of the client, the graphical flag indicating that replacement texts are available for the text unit; and

displaying one or more of the replacement texts from the set of replacement texts on the display of the client in association with the graphically-flagged identified text unit.

6. The method of claim 1 , further comprising:

replacing the text unit in the electronic input document with the presented at least one replacement text.

7. The method of claim 1 , wherein the one or more vectors representing the identified text unit include a coarse vector and a fine vector, the fine vector characterizing the meaning of the identified text unit at a higher precision than the coarse vector, and the method further comprises:

identifying the set of candidate texts from the corpus of documents as coarsely matching candidate texts by comparing the coarse vector for the identified text unit to coarse vectors for the set of candidate texts, and

wherein generating the meaning score further comprises comparing the fine vector for the identified text unit with a fine vector for the candidate text.

8. The method of claim 1 , wherein the set of annotations include at least one of text of the input document, position of the identified text unit in the input document, information on a user of the input document, and information on a purpose of the input document.

9. A system for passively suggesting text to replace one or more text units in an electronic input document, comprising:

a computer processor for executing computer program instructions; and

a non-transitory computer-readable storage medium storing computer program instructions executable by the processor to perform operations comprising:

examining an electronic input document presented on a display of a client to identify a text unit in the electronic input document and contextual information about the input document;

determining, for the identified text unit, a set of annotations for the text unit and the input document responsive to the contextual information, the set of annotations describing predictive characteristics of replacement text for the text unit;

representing one or more words of the identified text unit as one or more vectors that characterize meaning of the identified text unit;

evaluating, responsive to the set of annotations, a set of candidate texts from a corpus of documents that can replace the text unit, the evaluating comprising:

identifying a set of annotations for a corpus text unit, the set of annotations for the corpus text unit describing how well the corpus text unit satisfies the predictive characteristics for the replacement text,

performing a plurality of evaluations to generate an evaluation score, each evaluation comparing an annotation from the set of annotations for the corpus text unit with an annotation of the set of annotations for the text unit in the input document,

comparing one or more vectors associated with the corpus text unit with the one or more vectors representing the identified text unit to generate a meaning score,

determining a confidence score for the corpus text unit that indicates suitability of the corpus text unit as the replacement text for the text unit, wherein the confidence score is a combination of the evaluation score and the meaning score, and

adding the corpus text unit to the set of candidate texts responsive to determining the confidence score;

identifying a subset of the set of candidate texts as a set of replacement texts for the text unit; and

presenting at least one replacement text from the set of replacement texts on the display of the client.

10. The system of claim 9 , wherein there are a plurality of text units in the electronic input document and wherein the examining, determining, identifying, and evaluating are passively performed for ones of the plurality of text units.

11. The system of claim 9 , wherein identifying the candidate texts comprises:

adding the candidate text to the set of replacement texts responsive to comparison of the confidence score for the candidate text to a threshold.

12. The system of claim 9 , wherein presenting the replacement texts on the display of the client comprises:

graphically flagging the identified text unit on the display of the client, the graphical flag indicating that replacement texts are available for the text unit.

13. The system of claim 9 , wherein presenting the replacement texts on the display of the client comprises:

graphically flagging the identified text unit on the display of the client, the graphical flag indicating that replacement texts are available for the text unit; and

displaying one or more of the replacement texts from the set of replacement texts on the display of the client in association with the graphically-flagged identified text unit.

14. The system of claim 9 , further comprising:

replacing the text unit in the electronic input document with the presented at least one replacement text.

15. The system of claim 9 , wherein the one or more vectors representing the identified text unit include a coarse vector and a fine vector, the fine vector characterizing the meaning of the identified text unit at a higher precision than the coarse vector, and the method further comprises:

identifying the set of candidate texts from the corpus of documents as coarsely matching candidate texts by comparing the coarse vector for the identified text unit to coarse vectors for the set of candidate texts, and

wherein generating the meaning score further comprises comparing the fine vector for the identified text unit with a fine vector for the candidate text.

16. The system of claim 9 , wherein the set of annotations include at least one of text of the input document, position of the identified text unit in the input document, information on a user of the input document, and information on a purpose of the input document.

17. A non-transitory computer-readable storage medium storing computer program instructions executable by a processor to perform operations for providing suggested text for an electronic input document, the operations comprising:

examining an electronic input document presented on a display of a client to identify a text unit in the electronic input document and contextual information about the input document;

determining, for the identified text unit, a set of annotations for the text unit and the input document responsive to the contextual information, the set of annotations describing predictive characteristics of replacement text for the text unit;

representing one or more words of the identified text unit as one or more vectors that characterize meaning of the identified text unit;

evaluating, responsive to the set of annotations, a set of candidate texts from a corpus of documents that can replace the text unit, the evaluating comprising:

identifying a set of annotations for a corpus text unit, the set of annotations for the corpus text unit describing how well the corpus text unit satisfies the predictive characteristics for the replacement text,

performing a plurality of evaluations to generate an evaluation score, each evaluation comparing an annotation from the set of annotations for the corpus text unit with an annotation of the set of annotations for the text unit in the input document,

comparing one or more vectors associated with the corpus text unit with the one or more vectors representing the identified text unit to generate a meaning score,

determining a confidence score for the corpus text unit that indicates suitability of the corpus text unit as the replacement text for the text unit, wherein the confidence score is a combination of the evaluation score and the meaning score, and

adding the corpus text unit to the set of candidate texts responsive to determining the confidence score;

identifying a subset of the set of candidate texts as a set of replacement texts for the text unit; and

presenting at least one replacement text from the set of replacement texts on the display of the client.

18. The computer-readable storage medium of claim 17 , wherein there are a plurality of text units in the electronic input document and wherein the examining, determining, identifying, and evaluating are passively performed for ones of the plurality of text units.

19. The computer-readable storage medium of claim 17 , wherein identifying the candidate texts comprises:

adding the candidate text to the set of replacement texts responsive to comparison of the confidence score for the candidate text to a threshold.

20. The computer-readable storage medium of claim 17 , wherein presenting the replacement texts on the display of the client comprises:

graphically flagging the identified text unit on the display of the client, the graphical flag indicating that replacement texts are available for the text unit; and

displaying one or more of the replacement texts from the set of replacement texts on the display of the client in association with the graphically-flagged identified text unit.

21. The non-transitory computer-readable storage medium of claim 17 , wherein the one or more vectors representing the identified text unit include a coarse vector and a fine vector, the fine vector characterizing the meaning of the identified text unit at a higher precision than the coarse vector, and the method further comprises:

identifying the set of candidate texts from the corpus of documents as coarsely matching candidate texts by comparing the coarse vector for the identified text unit to coarse vectors for the set of candidate texts, and

wherein generating the meaning score further comprises comparing the fine vector for the identified text unit with a fine vector for the candidate text.

22. The non-transitory computer-readable storage medium of claim 17 , wherein the set of annotations include at least one of text of the input document, position of the identified text unit in the input document, information on a user of the input document, and information on a purpose of the input document.