Online translation software: Quality should be paid attention to

Automated translation tools have been around for a long time, and new techniques are helping to boost their performance. However, quality is still a problem.

Go to Google's Language Tools page (http://www.google.com/language_tools) and try the following: Type in the text Text ' Automated language translation - is it an idea whose time has come? '(Automatic translation - Is it an idea that the time has come?) And has the program translate that question into French. Next, translate the results from French into German. And finally, translate the German sentence get back to English. You will have the result ' Did language translation automate - is it an idea, from which the time came? '(Automated translation - Is it an idea that time comes from?).

Repeat the same thing with the phrase ' Reboot your computer and try again .' (Restart your computer and try again) You will have the final result of ' Their computer and attempt still again load .' (Their computers and efforts are still entered again.) Oh my! It is true that there are 'three star stars' !!! Perhaps you will really be bored if the manuals of multi-language devices are translated like this.

However, to be fair, translation tools also have positive support in some aspects. And researchers say new approaches will bring about improvements in the way translation tools work.

Update the database

Online translation software: Quality should be paid attention to Picture 1 Ford has started using translation machine software since 1998, and it has now translated five million instructions for assembly from English to Spanish, German, and Portuguese. and Spanish for Mexicans. English manuals are updated every day, and their translations - about 5,000 pages a day - are quickly transferred to Ford factories around the world.

Nestor Rychtyckyj, Ford's artificial intelligence expert (Artificial Intelligence - AI), said: ' You can't do this whole job manually .'

The automaker uses Enterprise Global Server software from Systran Software, a company that provides translation software based in San Diego, California. In addition, Ford engineers also write an AI program to translate high-level, professional instructions accurately and clearly. Each instruction is stored in the database for translation.

Ford also has to develop specialized car glossaries. According to Rychtyckyj, maintaining such dictionaries is easier than finding a proficient translation expert in terms. He said: ' Most of our efforts are to build glossaries, and often change them. Your translation results will be much better if you have advance preparation . '

Systran's software uses a reliable technique called rule-based translation (rules-based translation). Such systems use bilingual dictionaries combined with a database on how to use words and grammatical rules. Commercial translators are often supplemented with glossaries, or databases that store translated translation sentences or paragraphs. These additions are usually created by the user during the translation process. As a result, translation software can give the most accurate or approximate words or sentences compared to the original word or sentence so that the translator can easily review.

Software training

Statistical machine translation is a new technique but not yet widely used. This technique uses original texts and their translations to 'train' the software. Over time, these systems will 'learn' how to create a good translation, then use probability and statistical calculations to choose the most approximate word or phrase based on context.

Statistical systems require a lot of text to train algorithms, but they don't need grammar rules, bilingual dictionaries or translation memory. They develop their own rules and continue to adjust those rules over time.

Google uses the Systran translation software, but they also develop systems based on their own statistics to translate English into Arabic, Russian, Chinese, Korean, Japanese, and opposite. According to Google's researcher Franz Josef Och, these languages ​​are difficult for translation software because they are different from Western languages ​​whose roots are the Latin (Vulgar Latin).

For years, Microsoft has put a parsing software based on grammar rules into their Word text editor. Recently, the company has incorporated translation memories, rules-based and statistical translation software with human support to translate documents for databases about their customer support information. . Steve Richardson, a senior Microsoft researcher, said: 'The new trend in the research community is to see people can combine these statistically pure techniques with some knowledge of language. how.'

When asked if there would be breakthroughs in translation, he replied: ' Breakthroughs have emerged during the research process. On the practical side, there will be breakthroughs in creating systems that are integrated into the business process . '

The future trend is to combine databases that have stored translated sentences or paragraphs with rule-based or statistical translation software, or both. Such hybrid systems will become more sophisticated and complex.

However, the current translation systems are not able to produce high-quality translations as 'superior' translators, especially for popular language texts. According to Rychtyckyj of Ford, how hybrid systems become smart is not a problem, the quality of translation depends on the construction of the original language correctly.

The translation process generally has two main stages :

- Analyze the original in the original language to 'understand' exactly what the author 'wants to say'.

- Recreate the meaning that the author wants to convey in the target language to get a text equivalent to the original.

At first glance, this process seems simple but behind it is a complex cognitive process. In order to 'decode' the meaning of the original text correctly, the translator must analyze and interpret all of its characteristics. This is a process that requires an extensive knowledge of grammar, semantics, syntax, idioms . of the original language, as well as the culture of the author of that text. At the same time, translators must have a similar knowledge about the language they will convert.

Therefore, the challenge for translation software is how to program the computer to 'understand' a document as a human being.