2.3.1 Dictionary-Based Translation
The first and oldest experiments in computational linguistics were executed with the Direct Machine Translation System. It creates simple word-by-word translations almost regardless of linguistic rules, semantic and syntactic sources of errors. Direct Translation Systems only need a bilingual dictionary and a few grammatical and linguistic rules for the target language. First it looks up each word from the input in the dictionary. Next, it tries to correct the word order and places, for example, the adjective before the noun in German translations before finally outputting the result. It does not analyse the source text in detail. That's why this system has never been able to present translations in the target language which conform to the source text as far as grammar or semantic formations. Word-by-word translation systems, the first generation of translation systems, do not require a lot of memory and hardware speed. Another reason for using this system is simply the fact that there has not been any alternative to Direct Translation Systems before the 1960s. That’s why one had to be satisfied with the results giving a rough idea of the text. Later, the system was improved to be at least able to detect the tenses used in the source text or to find out whether it has to translate a question or a clause of statement. Although some features were upgraded, it can not cope with longer sentences with language specific devices like the English gerund, for example, which has no equivalent in German. A traditional Dictionary-Based Machine Translation consists of three main steps: a minimum analysis, the bilingual dictionary look-up and a few corrections like word reordering in the target text. Some modern systems prepare a text before the analysis process starts and make the system detect the source query correctly by syntactic transformations which include for example word reordering. For the semantic transfer, the dictionary look-up, nowadays a lexical storage is used to find out which is the right translation for ambiguous words in the text. Therefore some books call it an “intelligent dictionary”. But often the (Intelligent) Dictionary-Based System makes fatal mistakes. German translations like “zu Abend essender Tisch” (like the IBM Personal Translation Manager suggests) or ”Speisen der Tabelle” (Direct Machine Translation by Reverso Translator) for the English ”dining table” raise the question: Are human translators cheaper and more productive?