Can we fully rely on translations prepared with no human involvement?
Any text or utterance is set within a tangled network of stylistic devices, cultural references, semantics and ways in which sentence elements influence each other. A translator faces a great challenge, having to possess expert knowledge in grammar, vocabulary, semantics, syntax and culture of not only one, but two languages. Preparing a translation, especially a reliable one which will meet the client’s expectations, is definitely not just a mere word-for-word substitution. With all of the intangible hints and references, removing the human from the picture seems impossible. Technological advances, however, constantly challenge our beliefs.
How is machine translation prepared?
Basically, machine translation is either rule-based or statistical. The so-called “classical approach”, developed as a first solution within the field of MT, is based on numerous grammar and syntax rules, as well as bi- and multilingual dictionaries available for every language pair. Such system derives possible translation results via the syntactic analysis of a text and further creation of representations which can be transitioned into the target language. Efficient use of rule-based machine translation requires extensive bilingual database which could serve a source of knowledge on lexical items, syntax and grammar rules, requiring, therefore, significant input before even starting the usage. Obviously, the more source-target language pairs there are, the better translation results we are able to obtain – so constant updating of our database is vital for its further development.
Statistical machine translation, as the name may suggest, uses statistical models extracted from the analysis of bilingual content. It requires far less input than the abovementioned solution, exceeding it also in a number of different ways: it is possible to build it based on what translators or translation agencies usually already have, that is translation memories and term bases. Even monolingual texts can be utilised, as they provide reliable data on text fluency, enabling the machine to improve the quality of the target text. This approach can be also transferred to different languages, which is not possible in the case of the first solution – mainly due to the fact that while uploading new source-target texts, these are tailored to fit a certain language and their ability to provide reliable translation in different languages is limited.
Who can benefit from the use of machine translation?
In many cases it is creativity that people seek in the translations – especially when it comes to advertising, marketing or literature. A loyal client, who is, let’s say, a CEO of a manufacture plant, would their translation be rather concise and consistent instead of creative; and it is hard to blame them for such a choice. Persistent use of chosen lexical items ensures full repeatability and provides clear instructions for the employees. Machine translation makes it possible to translate large volumes of text without entrusting such a task to two or more professionals, therefore risking discrepancies in vocabulary and syntax. Such systems guarantee a significant increase in translation throughput, shortening the preparation time in case of translating manuals or handbooks for employees waiting for a revolutionary launch at one’s production plant. In such cases, MT is a perfect solution to all of these demands.
Choosing the best solution – rule-based vs. statistical machine translation
In many aspects, statistical MT is superior to the classical approach; it does not require that much effort when it comes to development and maintenance (however, it is worth to remember than the amount of text resources required for SMT to work is tremendous – but ensures an unparalleled fluency), doesd not require the help of professional linguists to take care of proper application of the rules and it basically adapts on its own to the ever-changing nature of every language. There is one significant obstacle that a beginner entrepreneur might face – SMT systems require remarkable equipment support, demanding state-of-the-art computers capable of handling the constant processing, as well as all of the necessary databases.