The term ” machine translation ” means translation in the strict sense of a text (or audio chat, live or delayed) made entirely by one or more computer programs, without a human translator has to intervene.
Sometimes this term includes also computer assisted translation, where the translation can be done partly manually or interactively with the machine.
Despite its weaknesses, it can render services in areas such as international surveillance (including technology watch), where it can take in a superficial way, but rapid, large amounts of text.
In recent years, it has grown considerably over the web, with several online systems that can translate automatically and within seconds web pages or texts longer and longer. It is a help very popular with the general public because it helps decipher – crudely – the theme of a web page in a completely unknown language, and key facts or information that it contains. For easier navigation, more search engines like Google, Altavista or Yahoo! allow its use.
Prerequisites depend on the intended approach: based translation rules (word-for- word, transfer, pivot), translation by example, statistical translation.
Based rules automatic translation requires:
- dictionary entries
- linguistic rules
Translation by example and statistical translation require:
- translation memory (all translated texts)
It may, in addition, require tools for language analysis such as:
- identifiers (tokens)
- morpho- syntactic tagger
- possibly: chunker, parser
Direct approach :
- word for word translation of the source text to the target text
- reformulation of the order of words translated into the target text
- construction of the interlingua representation of the source text
- construction of target text representation
- lexical and syntactic analysis of the source text
- transfer of the translated lemmas and structures in the target language
Translated and adapted from Wikipedia.