Logo

  Call us on +44 (1905) 676231 (UK & rest of World)
+1 (970) 665 2125 (US & Canada)

For all your translation needs

 

   Machine Translation Tips

Contact Us

 

Linked documents are in PDF format and you will need a viewer installed to read them. If you don't have a viewer already, please click here to visit Adobe's website and download a free copy (51.95mb) or here to download the Open Source Sumatra PDF Reader (9.7mb), from Google Code Project.

Give the computer a break!

In order to achieve optimum results using machine translation (MT), you should write your source text using clear, concise, grammatically correct language. There is little difference between writing a document for machine translation and writing a document for human translation. If a human cannot understand what you are trying to say, a computer will fail miserably.

The following tips will help you to produce much better results from your Translution System:

Keep the structure of your sentence clear, simple, and direct

You should divide long sentences containing several ideas into short sentences. If your sentence contains a semicolon, break it into two sentences. Try to keep to the rule: One idea = One sentence. Sentences of five to twenty words in length are the easiest for an MT system to translate, however, do not make sentences too short. Sentences of less than five words can be problematic and ambiguous.

If possible, avoid colloquialisms, idiomatic expressions and slang.

Expressions like: Now you're talking turkey, Gimme high–five, Dude! and even By Jove, she's a cracker! could pose major challenges for most machine translation systems. If you are not absolutely sure whether you are using an idiom, analyze the literal sense of the words in your text. If it proves to be nonsensical, there is a very good chance that it will be nonsensical in the translation. If this is the case, express the same idea in a different way. However, if you are still determined to use such an expression, you might get away with it by entering it in your dictionary as a Fixed Expression. This would tell the system not to bother trying to analyze and translate the text and the target–language equivalent would be delivered exactly as you have entered it in your Dictionary.

Spell correctly

Machine Translation dictionaries have difficulty finding misspelled words. And remember, words that require accented characters are misspelled if the accents are missing! Whenever possible, spell acronyms with upper case letters, e.g. API, not api; NATO, not nato. Maintain consistent spelling and hyphenation of terms throughout your document and, if possible, across documents.

Avoid ambiguity and vague references

At the risk of being repetitive, use words again when referring back to something. Keep pronouns to a minimum. For example, in English the pronoun it could be translated as he or she in the target language, depending on the gender of the noun to which it refers. Here is a very simple example to illustrate this:

For the English: I bought a hat and a car yesterday. It is very beautiful.
The French translation would be: J'ai acheté un chapeau et une voiture hier. Il est très beau.

The machine translation system has a problem working out whether the word "It" in the second sentence refers to hat or car. You may be very clear that it refers to the car and not the hat, but anyone else would have to guess, and computers are no more clairvoyant than people are. In this example, the system has delivered a masculine pronoun, rather than the correct, feminine, pronoun.

Repeating the noun in question:
"I bought a hat and a car yesterday. The car is very beautiful."
results in a perfectly accurate French translation:

J'ai acheté un chapeau et une voiture hier. La voiture est très belle.

When listing or enumerating, repeat the noun being modified, "Take the red, green and yellow apples" will probably not be translated correctly. "Take the red apples, green apples and yellow apples" will almost certainly be translated correctly.

Use standard, formal language

Clear, formal, grammatically correct writing is easiest for a reader to understand. Clear, formal, grammatically correct writing is easiest for a machine translation system to understand. Make sure your verbs agree in number with the subject. Correct punctuation will help as well. If you want your emails translated correctly, you may well need to change your writing habits.

Use the definite article even when you don't want to

When a machine translation system analyses English it is often difficult for the system to distinguish between verbs and nouns. Most short nouns can be verbs as well, e.g. bank, log, lodge. Instructions often tend to omit definite articles, for example:

This would be better like this:

Twist cap off — Twist the cap off
Remove seal — Remove the seal
Insert applicator — Insert the applicator

Summary

All the above tips will help you achieve improved quality in your machine translation. However, remember that the most important tool to improve quality is your Translution dictionary. Use your Dictionary Manager to build up your own personal terminology. This will correct any errors that the machine translation system may make in translating terms.


Back to top

 

 

 

 

 

 

 

 

 

 

 

Company DetailsPrivacy Policy
Terms and ConditionsLicense Agreement
© Translution 2005 – 2012

Valid XHTML 1.1