The term machine translation (MT) is used with regards to computerised systems that are responsible for the generation of translations with or without human assistance. It needs to be emphasised that computer-based translation tools that support translators by giving access to remote terminology databanks, on-line dictionaries, transmission and reception of texts are excluded from the above definition. The borders among machine-aided human translation (MAHT) and human-aided machine translation (HAMT) are frequently ambiguous and the term computer-aided translation (CAT) can cover both of them, however the core of MT itself is the mechanisation of the whole process of translation.
Even though the ultimate objective of machine translation systems is to generate high-quality translation, in reality the output is typically revised (in other words it is post-edited). It needs to be emphasised that from this perspective the output received with the use of MT is the same as the output that is obtained of the majority of human translators, which is usually reviewed by a second translator before further distribution. At the same time, the types of errors that are produced by MT systems are quite different from these of human translators (incorrect articles, prepositions, pronouns, verb tenses and so on). In general, post-editing is the norm, however in specific situations MT output can be unedited or only slightly revised. The above is the case in these situations in which the translation is intended only for a group of specialists, who are familiar with the subject in question. In addition, output of MT can serve as a rough draft that can be used by a human translator (in the course of the pre-translation stage).
As to the quality of the translation produced with the use of MT systems, it can be improved in different manners. Firstly, the above can be achieved by developing more sophisticated methods of MT. Secondly, the translation may be improved by imposing specific limitations on the input. For instance, the system can be planned to handle texts restricted to the sublanguage (grammar and vocabulary) of a specific subject field (such as physics) and/or type of documents (such as patents). On the other hand, input texts can be written in a language that is controlled, which limits the variety of vocabulary, and enable avoiding polysemy and homonymy and multifaceted sentence structures. What is more, there is a third option, which necessitates input texts to be marked (pre-edited) with indicators of suffixes, prefixes, phrase and clause boundaries, word divisions, or of other grammatical categories (for example the noun cónvict and its homonymous verb convíct). Lastly, the system can be able to denote and indicate various problems of ambiguity and leave the final selection to humans (typically translators) for resolution in the course of the translation process in an interactive mode.