What if Julius Caesar could read your blog?

You can now read my website in Latin, and for that matter a few dozen other languages. Should we be excited? Yes and no.

This new capability came as the result of an email from WordPress, my website platform, telling me that the Google Translate widget was now available. In minutes, it was installed and I picked Latin from the drop-down menu. Why Latin? Because I could!

I confess to a professional interest here: I taught Arabic-English translation for some years, and spent half a career researching what might go on in translator’s brains.

But for those with only a passing acquaintance with translating, a caveat or two might be in order. Firstly, the ‘translate’ button on my site is not necessarily the same thing as the alternative language versions (sometimes indicated with little flags) on the websites of big institutions. For example, the content on non-English websites of a university will usually by pre-authored in French, Chinese, etc., and not translated on the spot.

So how good are web-based translation tools? Well, it depends on what you’re translating and between which languages. The reason for the variation is to do with the process behind web-based tools.

In the pre-internet days, automatic translation (then known as ‘machine translation’) worked on the same principle employed by Miss Bollard, my terrifying Latin teacher in pre-Beatles England. The machine translation technique is to analyse the input text into word meanings and grammatical rules, and then reassemble those meanings and rules into another language. The computer’s real task is to attempt to replicate what went on in Miss Bollard’s classically trained brain. The advantage of the computer is that it doesn’t have to be sent snivelling to the headmaster’s study to be punished if it makes too many mistakes. If you’re interested in a beautifully drawn picture of early machine translation, check this excellent article by John Hutchins.

Web-based translation tools use a quite different process: They read the input text, and then scour the internet to match it with similar texts that have existing translations. The output text is then compiled from those translations. The system learns from new translations appearing on the web, and from improvements posted by users.

Which brings me to Julius Caesar. I suspect that the internet is not exactly bursting with English-Latin translations, which will explain why French web-based translations are better than Latin ones. But how can you check the quality if you don’t know the language? The quick and dirty way is to ‘back-translate’, i.e. use the tool to translate a chunk from English to, e.g. Welsh, and then translate the Welsh back into English.

Here is the first line of Pride and Prejudice back-translated through several languages using a web-based translation tool:

English-French-English: It is a universally recognized truth that only one man in possession of good fortune must be in need of a woman.

English-Welsh-English: It is truth universally acknowledged, that a single man must possess good fortune to be in want of a wife.

English-Latin-English: This is of a truth recognized as such, known, and will have the same good fortune to be his wife.

The differences are striking, and the quality seems to be in approximate proportion to the quantity of bilingual content likely to be available for the tool to work with: The French version has one serious error, confusing Austin’s single (unmarried) man for a unique man. The Welsh version is logically incorrect, and the Latin version is nonsense.

But before we send this web-based tool snivelling to the headmaster’s study, we should realise that we have unrealistic expectations for the Latin version. In fact the web-based tool is a bit like me in 1961 – too little material to work with! Come back, Miss Bollard. All is forgiven.

Find out about Stuart Campbell’s books here.