Why automated translation produces gobbledigook
July 4th, 2008 Amir Posted in Machine Translation, Website tips | 1 Comment »
Anyone who has done business over the internet with people in another country has likely encountered the various free online machine translations or their paid software variants. When faced with the daunting prospect of communicating in another language, either to read a missive or article in a foreign tongue, or to reply to a person who does not speak your language, resorting to the free online translation services can be irresistible.
The main problem occurs when you realize that much of the translation is not only imprecise but can result in outright gobbledigook!
One of the most popular online translators is the Google system and it clearly shows the limitations of the current state of machine translation. Just key in the Italian “Le signore sono nel caffe” which is accurately translated as “The ladies are in the cafe” and you will receive a translation of “The ladies are in coffee,” which has a very different meaning!
Some language pairs almost completely evade the capabilities of machine translation. A Korean web page dealing with Windows 2000 and XP operating systems is recently translated by Google as:
“From the xp the tearoom waitress su tree it searches for from the place 2000 where there is a method which it amends the bedspread which is strenuous.”
The software packages which retail for hundreds of dollars generally don’t score much better than the free online services: The essence of the conundrum is that accurate machine translation is simply not available at any price.
Machine translation is generally based on methods of linguistic rules where the words entered in the original language are translated according to a pre-programmed set of rules. These rules apply a parsing algorithm to the text and generate an intermediate representation of the general structure of the final translation in the target language. The devil is always in the details, and in machine translation, there are myriad syntactic, semantic, and morphological details which have to be considered by the machine translation system and a virtually overwhelming set of rules.
The major challenges facing machine translation are:
- Ambiguity
- Structural and lexical differences between languages
- Multi-word units such as idioms and collocations
Those factors are the major reasons why software finds it so difficult to translate common phrases such as “the cat’s meow,” “wolf down a meal” or “stool pigeon.” The software invariably structures the translation literally around the animal and misses the meaning entirely.
As software becomes more complex and computers become more capable, machine translation is slowly improving but its capacity will still remain far below that of a professional human translator for decades to come (listen to the great talk by Iain Mccowan). The essential problem is that computers still cannot synthesize the functions of the human brain. Although the computational speed of machines is many orders of magnitude greater than the human brain, there is still no computer that can process information and be cognizant of its meaning to the level of a three year old child! When something as dynamic and multi-layered as language is considered by a computer, it is unable to analyze the effective meanings as it cannot share the experience of a human being. It is that experience which is the critical aspect in human comprehension and the natural extension of comprehension: communication.
Given the current state of machine translation it is likely that none of us will live to see a “Star Trek Universal Translator” which not only accurately translates between one language and another, but does it by speech. If software has enormous problems translating ASCII characters, the obstacles faced in recognizing spoken words, translating them accurately and speaking them back out may be well nigh insurmountable.
For clear, precise, lucid translation there is no alternative to a professional certified human translator with ample fluency and experience in both languages of the desired pair. It is only through that process that translations can be accurate and convey the full meaning that was intended in the original language.
August 3rd, 2009 at 1:27 pm
I am a linguist and I could agree with you to some point.
My first argument is that machine transation can be good but this depends on what you want to use it for. The input text is quite important for the good output. If there are too many slang words, or bad grammar, or wrong spelling, there is no way to receive good translation.
Machine translators are not expected to be able to translate Homer and the fact that all translators work very bad can remain in the past now. I am not saying that machines are to move human translators out of the way. They are to be used by every day people to be able to read things on th einternet, write or read a short message, and things of this kind but not to take the role of human translators (that couldn’t happen).
An innovatin that has been made and that makes machines translate better is the use of topical dictionaries. One such application is Moztrans by Interlecta (this is an add-on for Mozilla). The fact that Moztrans uses dictionaries is just one of the features that makes it work well but not the only one. After all, google is not the only translator that exists online , is it?
About ambiguities, idioms, pragmatic meaning, marked structures…as I said before, machines are not humans. Some translators translate idioms, if the idiom is common enough and the machine has it in its database.
In conclusion, I want to repeat what the author has said in the last paragraph “For clear, precise, lucid translation there is no alternative to a professional certified human translator”. And I want to add, don’t underestimate machine translators for everyday use (translating pages and short texts) and try several, there is not only google, remember that! Google is not the best and not the only example of MT.