5

We have google translated docs which are almost error free and require only minor (about 10%) corrections, and even these corrections are redundant which could be learned by an intelligent machine learning software.

What would be nice is to have low granularity learning system that can recognize and correct minor in text errors.

Example:

google translated English(with errors):Make sure the road Changes features in Word/OpenOffice.org is disable

Good English\ corrected text:Make sure the Track Changes feature in Word/OpenOffice.org is disabled

So we see corrections in 3 words: road,features,disable

As we see the corrections are trivial, redundant, custom terminology related and easily learnable in our settings. Moreover they are in a relatively unpopular target language. Here I only gave examples in English and not the target language.

Is there a software tool or plugin which would learn and correct these errors utilizing translation memory in the word processor.

Improve this question
13
  • 5
    Don't you think that if such system existed, Google would have integrated it into their Translate already?
    – SF.
    Commented Apr 10, 2013 at 12:15
  • As we see the corrections are trivial , redundant , custom terminology related and easily learnable in our settings, more ever they are in a relatively unpopular target language, i only gave examples in English and not the target language.
    – Ali
    Commented Apr 10, 2013 at 13:56
  • I can assure you they are trivial for a human but they are extremely challenging from programming point of view. Natural language processing and generation is one of the more difficult and underdeveloped branches of AI research.
    – SF.
    Commented Apr 10, 2013 at 14:08
  • 1
    I'm a little unclear on what you're looking for. An out-of-the-box solution? A library of translation routines you can use to roll your own solution? Something else?
    – Goodbye Stack Exchange
    Commented Apr 10, 2013 at 17:15
  • 3
    The problem is that, if this is a question about finding a tool, it's fine here; if this is about creating a tool, it's straddling the line between whats on-topic here and what isn't. You're also more likely to get answers about constructing a coding solution on Stack Overflow, and maybe about assembling a solution from components on SuperUser. (Software recommendations are off-topic on the latter site, I checked.) Are you willing to edit this so it's more of a tool recommendation? If not, we'll be closing this for now.
    – Goodbye Stack Exchange
    Commented Apr 10, 2013 at 18:34

2 Answers 2

Reset to default
3

In my opinion as a computer technician, this is impossible at least outside the strictly technical language... And, even so, I don't think it's something you can expect to be error free.

Google translator is far from accurate, at least from what I see when interacting with people that depends on it to speak in English. It can translate something in an understandable way, but never in a perfect way. Specially because there are some localized word usages that won't make sense for a mere translator.

Somethings, at least at this time, will still need to be done by humans.

Sometimes I use Google Translator, since I'm not a native English speaker but never as a tool to translate, just to help me with some words. What I do is to get it's translator and review it, fixing what is not right.

Of course there should be better tools than Google translator, but they will for sure be expansive. Besides, they will never be error free since it's quite hard to make such complex IA algorithm.

At last, but not at least, there are sentences that we - as humans - have a hard time to comprehend in our own languages. Now picture a software dealing with that.

Improve this answer
2

I believe this can work fine in translation of non-literary, standardized documents, e.g. technical documentation, medical reports, and the like. The more both format and language (terms) are standardized, the better this will work.

Preparation

You must write with automatic translation in mind. Use the same words and phrases always with the same meaning and syntactical structure. For example, do not use "road" to refer both to a hard flat surface for vehicles, people, and animals to travel on and a process or a course of action that leads to a certain result. If you use "road" with both meanings, chances are that those meanings are expressed with two different words in your target language, and a translation programm must make a decision it is not equipped for.

If you keep meaning distinct and grammatical structure simple, the errors from automatic translation will always be caused by the same phenomenon that you can then correct automatically.

Autocorrection

If you always need to replace "road" with "Track", just do a search and replace. You can do this manually in any word editor, or write a script in any programming language.

If you use a script, create a matrix or two arrays, one with the errors to be replaced, one with the correct replacements, and let an inbuilt function such as string replacement or regular expression replacement scan the document for you.

As I said, this, of course, only works if there are no instances of "road" that you need to keep or that need to be replaced with something else.

Human Control

Always let a human being familiar with the meaning you want to convey read your document and correct final mistakes (or add new words to the two arrays).

Improve this answer

Not the answer you're looking for? Browse other questions tagged or ask your own question.