Developments in the translation business: a brief history
The days when Saint Jerome translated the Bible from Greek and Hebrew into Latin – a daunting task if ever there was one – have long gone. The first dictionaries appeared in the Middle Ages, to be followed by other important sources of information for translators, such as encyclopaedias, thesauruses and specialist dictionaries in the legal, financial, medical, technical and other fields.
The advent of the PC in the 1980s introduced the automation of word processing, and thanks to the launch of the World Wide Web in the 1990s and its subsequent development it has become considerably easier to consult external sources. Dictionaries, encyclopaedias and information of every kind can be found and used with a few clicks of the mouse.
The late 1990s also saw the first large-scale introduction of computer-aided translation tools (or CAT tools) in the translation sector. Software such as SDL Trados, Déjà-Vu and Wordfast now allow translators to easily build and expand translation memories that offer immense support in the translation process. These tools alert the translator to existing translations of the same or similar texts that were included in the memory, allowing him to more or less copy relevant segments. Not only does this prevent a lot of double work, it is also great for consistency. However, in computer-aided translation the translator has retained his central role as the translation-generating party. The computer merely serves as a writing pad and as a memory from which previously translated sections from identical or similar source texts can be extracted. So while these CAT tools are a great help in terms of speeding up the translation process, promoting consistency and exploiting translation archives, the translator remains in charge.
In contrast to computer-aided translation, machine translation (MT) centres not on the person but on the machine. Here it is the computer that does the actual translating. If required, human translators can still be put to work to correct and embellish the machine-made translation – and in that sense they are still in charge – but the machine does most of the work. The first translation machines date from the 1950s, but machine translations only began to be taken seriously with the introduction of Google Translate in 2007. Since then, machine translations have become a common phenomenon to anyone who uses the Internet.
Types of machine translation
Roughly speaking there are two types of machine translations: those based on linguistic rules (rule-based machine translations, or RBMT) and statistical machine translations (SMT). Rule-based machine translations are produced essentially on the basis of grammar rules, algorithms and terminology databases, all of which serve as input to teach the software to produce its own translations. In the case of statistical machine translation, statistical models are used based on analyses of huge volumes of bilingual texts. For example, the statistical models that underpin Google Translate re based on the analysis of millions of pre-translated documents. Software developers today strive to create hybrid forms that incorporate both RBMT and SMT, as a combination of the two approaches – known as Hybrid Machine Translation (HMT) – has been shown to yield the best results.
Advantages and disadvantages
The main question today is whether – and if so, to what extent – users actually benefit from machine translation. For private individuals this question can probably be answered in the affirmative: MT enables them to get a general idea of the contents of a text much faster – and much cheaper – than would be the caser if they hired a translator. Machine translation would serve this purpose in particular in the context of fairly simple online texts, such as messages on Internet forums, in social media and online newspapers.
The usability of machine translations for professional business users is considerably more doubtful. The benefits are the same as for private users: MT offers a general idea of the contents of a text at low costs and in very little time. Confidentiality is not an issue, since no third parties need to be engaged to produce the translation. The machine does all the work. However, the use of MT in business contexts does involve substantial risks.
Let’s start with the most obvious and most serious of these risks: machine-made translations tend to be inaccurate and extremely unreliable in their representation of the source text. Machines can produce sentence after sentence of acceptable translation and then, all of a sudden, make the most incredible mistakes – mistakes of a type that human translators would never make. If you need to know the contents of, say, a court ruling, a contract or a manual, machine translations may offer help but may just as well result in total confusion. Human language is complex and can only be captured in rules to a certain extent. Machine translation is also hardly suitable for texts in which style and nuance are important , such as texts to be used for marketing purposes. The same applies to texts in which the accuracy of the content is essential , such as medical information. Technical manuals and helptexts tend to be more suitable for machine translation, especially when characterised by relatively straightforward and repetitive structures, and also because stylistic refinement is less of a priority in this segment.
Perhaps the greatest flaw of translation computers – and an aspect on which they will probably continue to be beaten by human translators for decades to come – is their inability to read (and write) between the lines. Translation machines are impervious to double meanings, innuendo or persuasion. They have no problem with irony, simply because they do not recognise it. And they have very little sense of humour. People are supremely capable of enriching their texts with subtleties of this nature – indeed, this is how they use language most of the time. But computers don’t. Their translations represent the surface of the text, but do not reflect any of its hidden purport. They are sadly incapable of choosing any other but the most obvious terms, which is why their translations tend to be sterile and artificial. This may prove a serious drawback, not just in personal documents but in business texts too.
Whatever the case may be, it is clear that machine translations will tend to be useless for professional users unless a post-editing (PE) phase. And that is where the human translator enters the scene once again. A linguist is able to correct interpretation errors, make sentence structures sound more logical, more comprehensible or more attractive, enhance connections between sentences, improve consistency and adapt the tone of voice to the intended target audience. The party commissioning the translation may decide on the desired intensity of this post-editing phase, depending on his own requirements. Some clients simply want to know what a text is about if they cannot read it themselves; others have other uses for their texts in mind and also value things such as consistency and register. The higher the desired quality of the translation, the longer the post-editing phase will take and the higher the costs.
So, despite the various advantages of machine translation, customers should not expect the eventual quality of these translations to equal the level of human-made translations – not even after post-editing. For texts whose effect in part depends on tone of voice or which demand 100% accuracy and clear, consistent wording, therefore, will continue to rely on old-fashioned “manual” translation skills. There is no reason to assume that this will change in the foreseeable future.