Many of those who read this blog are translators, so I hope the others will forgive a brief digression about MTPE: machine translation post-editing, i.e. fixing computer-generated translations.
I hadn’t done a sizeable MTPE for a while, because the rates you can get aren’t yet commensurate with the time it takes. But once in a while it’s interesting to see if they’ve got significantly better…
The most striking thing is that it (DeepL in this case) makes lots of the same mistakes as Dutch people when they write English – virtually everything I’ve been listing lately. At first sight, that’s well weird.
But on reflection, it suggests to me quite strongly that the engine’s content is being populated with large numbers of non-native NL-EN translations. So there’s a substantial GIGO effect and it will be a while yet before it gets anywhere near mother-tongue quality.
- It can’t handle Dutch usage of brackets, hyphens and other punctuation.
- Dutch sentence structures and comma splices remain untouched.
- Excessively literal when there’s little to go on, in particular in e.g. noun stack headings.
- Overuse of all the favourite Dutch overused words.
- An annoying penchant for “the + (verbal noun) + of”.
- Inappropriate style (e.g. legalese where accessible was needed, or contracted forms in formal text, or old-fashioned usages such as the “one” verb form).
- It’s also not got a clue about UK-US issues, not surprisingly.
- And oh yes, the text was about branding (company image), which it managed to translate at regular intervals variously as firing, burning and surf. Clearly not something even an incompetent human would do.
In short: it’s iterating towards Dunglish rather than English, in a kind of self-fulfilling prophecy. Some way short of competing with the mother tongue, but non-native translators will have to watch out soon, though!
As the majority of EN-NL translations are done by NL natives, it perhaps also suggests that computerized EN-NL is going to be better than NL-EN.