The limits of computer-generated translation

Computational linguists and experts in artificial intelligence are producing increasingly sophisticated models of language and, used wisely, computer-generated translation can be useful. Free web-based translation services offer a simple, confidential and inexpensive way to get the gist of documents or web pages in another language. This might be adequate by itself and save you the effort of having a text translated properly, or it might give you enough information to decide whether it was worth doing so. The accuracy of these services also varies depending on the languages in question: a translation between, for instance, Spanish and Italian, which have a high degree of structural and lexical equivalence, would likely be of higher quality than a translation between German and Chinese.

However, computer-generated translation is usually poorly equipped to take account of ambiguity, context, register or any of the idiosyncrasies of human language that a competent translator would be aware of and looking out for. Legal documents, for example, require very precise language that conforms to strict conventions and the wording of contracts or statements must be able to stand up in court and be subjected to detailed scrutiny. In scientific or technical writing, words can have very specialised meanings which differ from everyday use, such as 'resistance' or 'work'.

Furthermore, words and grammatical features in one language do not equate exactly to those in another, and translation often resembles rewriting the source text using the structures and vocabulary of the target language more than it does searching for direct equivalents. 'Cup' and 'mug' would both be translated as 'tasse' in French, while 'like' and 'love' would both be translated as 'aimer'. Normally, this would not pose a problem; either it would not matter or a translator would make a choice as to which was more appropriate based on context. However, in the case of sentences such as, 'I don't want a cup, I want a mug,' or, 'I don't like it, I love it,' it becomes necessary to find a way of conveying the distinction, which a speaker would make automatically in English, in French, where it can only be done by circumlocution.

I translated those two sentences into French using four different free internet translation services. Three out of the four rendered 'I don't like it, I love it,' as, 'Je ne l'aime pas, je l'aime,' which (assuming 'it' refers to the same thing) literally means, 'I don't like it, I like it,' (or, 'I don't love it, I love it,') which is contradictory. One translation service returned, 'Je ne l'aime pas, je l'adore,' which does capture the difference of intensity implicit in the English sentence.

'I don't want a cup, I want a mug,' was twice translated as, 'Je ne veux pas une tasse, je veux une tasse.' The literal English meaning of this would be, 'I don't want a cup, I want a cup,' (or, 'I don't want a mug, I want a mug,') which again is nonsensical. Two translations differentiated between 'cup' and 'mug', offering, 'Je ne veux pas une tasse, je veux une grande tasse,' and, 'Je ne veux pas de tasse, je veux une grande tasse,' although the second ('Je ne veux pas de tasse') implies, 'I don't want [any cup at all]'. Machine translations are also not sensitive to areas where cultural customs or other non-linguistic factors affect the choice of words. Where one would close a formal letter in English with 'Yours sincerely' or 'Yours faithfully', French would require a more elaborate construction along the lines of: Je vous prie d'agréer l'expression de mes sentiments les meilleurs. One free internet translator rendered this into English as: I you taken to accept, mister, the expression of my feelings the better ones. It is not difficult to imagine the impression this might make on a potential client.

However, a competent translator, working into their native language, would immediately recognise that a non-literal translation was needed, along with any specialist material or nuances of context and register that would affect the translation. While there is a place for free web-based translation services, most are a poor substitute for a translation done by a talented professional.