Basically everyone can translate. Translation steps are:
- Understanding of the text, including any special grammatical structures
- Transpose the text into target language words considering all relevant grammatical structures using the proper words according to the textual context.
In case of machine translation, the computer performs the understanding of the text, the grammatical analysis, finding of any specific structures, use in the target language the matching words according to their environment and finally presents the results .
The story of machine translation
Translators were written already in the 1950-60s, but up to the 1980s they all suffered essentially under the same pitfalls, that the appropriate grammatical apparatus did not yet exist, there were no POS taggers, lemmatizers or grammar analysts (parsers), and then the low storage capacity also hampered the language development. POS taggers first appeared in the late 1980s and in the 1990's they became generally known and accessible. POS taggers and lemmatizers are still the cornerstone of translation and show a gradually improving quality. At first, rule-based POS taggers dominated the race, but gradually appeared statistical methods, that do not require knowledge of either the initial or the target language, and yet high-quality analysis can be produced. If statistical methods and rules are used together, a very good quality can be achieved, that covers almost all hard identifiable cases. In addition the number of members taken into account also grew. Originally, only one prefix was taken into account, (2-gram tagger), the considered prefix count gradually expanded and today, in 2011, the five-member (5-gram tagger), that takes four prefixes into account is no more rare.
Today the direction of development is to develop usable grammar analysts (parsers), and there are promising results.
Machine translation basically starts with POS tagging and lemmatizing, then identifies the unknown grammatical structures, which require special care, then recognizes groups of words in the sentences that can be translated together, and then translates into the target language, considering the textual environment by appropriate rules. Finally, the target version is formally checked and repairs are carried out if necessary.
It is essential to have a good quality and large enough vocabulary dictionary, otherwise it is not worth to begin to write a translator.
The motivation to write the Cygnus translator was hundreds of pages of hand-translation of foreign texts, using the high-quality, high vocabulary dictionaries the translator gathered. The question was: if the human translator can look for the structures and words, why could not the machine? Or at least the machine could do mostly the "dirty work", and this may significantly increase the speed the translation. The result shows that the machine can be helpful, and such a program really helps a lot to perform translation tasks.
The Cygnus translator's individual modules work together as follows: