Machine Translation: Is it useful or not?
 

If you try any of the free machine translation systems that are available on the Internet, you will either die of laughter or be appalled (or both).

However, to conclude from this that Machine Translation (MT) is useless would be a dangerous misjudgment. For over 10 years, we have continually been achieving cost savings of 30% by using post-edited MT translations.

Why has MT-supported translation not become more widespread?

Cost-savings of the scale mentioned above are only possible in controlled environments with optimized processes. To use a process of just "text in — translation out" delivers the previously-mentioned poor quality.

2 Systems

There are currently two types of MT systems available: those that break down the source text sentence by sentence based on grammatical rules and then put it back together in the target language, and those that use a large bilingual body of text and use statistical methods to identify sentences and parts of sentences, along with the corresponding translation fragments, and create a translation on the basis of a "puzzle".

Scientific studies show that the qualitative results are roughly compatible. In other words, about the same level of effort is required to raise the MT translation to a suitable level.

If you have clearly formulated source texts, the grammatical rules may be an asset. On the other hand, if you use a large amount of texts that have already been translated and that all cover the same topic, the statistics deliver better results.

Potential for Cutting Costs/Example

If there is a sufficient amount of texts for translation from one area, MT systems can lead to cost and time savings in a well-thought-out work environment. At the end of 2009, we translated approximately 300,000 words from English into German for one customer and we were able to grant a discount of 20% for the sections of the text that were pre-translated using MT and that were brought to a "human standard" by translators, without having had the time to optimize the process.

You are here: Home FAQ MT

Theses

  1. MT may lead to cost savings.
  2. Unprocessed MT source texts are usually unsuitable.
  3. The post-processing by trained translators may deliver high-quality translations or "used translations" for a lower price.
  4. Initial investment is required and this investment must prove to be profitable.
  5. The larger the quantity of texts and the greater their homogeneity, the more you should consider MT.
  6. MT should be used as a part of an integrated environment (along with TM)
  7. MT is neither a cure-all remedy, nor the devil's handiwork.

Links

LucyLucy Software

Rule-based MT system

Language WeaverLanguage Weaver

Statistical MT system

©2009 Eule Lokalisierung GmbH