Fully Automatic High-Quality Translation

(faah-quit)

The is the holy grail of machine translation: a completely automatic high-quality translation of a foreign text into the target language.

For most of the history of machine translation the software developments concentrated on syntax rules which dictated what translation applied to a particular use of a word or phrase. These rules formed a highly complex base from which translations were derived. Because of the large database and extensive processing, machine translation has largely been fairly slow and fairly inaccurate. Any word out of place or such things a proper names would drive the machine translation into fits and produce largely trash as output.

In short, the best machine translators have been able to come up with so far are gists, that is some idea of what the translated text might mean in the original language. High-quality translation is still very much a goal to be strived for.

Newer translation methods include a statistical approach to translation. While past attempts have used more or less rigid rules, the statistical approach consists of entering multiple translated texts and these are then analyzed for phrase matching for the two languages. Statistics is then used to measure how often and where words occur in a given phrase in both languages in order to create a template for word reordering. When a new document is input to the system it may differ significantly from already processed text but the translator can draw from the prior documents and pick a translation with the highest statistical probability of being the correct translation.

Developments are on-going.

More Information

• The History of Machine Translation in a Nutshell