Start a new topic

Working with DeepL

It's really amazing what DeepL delivers.

I've deactivated the binding of the external editor, because it caused duplications in CafeTran's target editor when using macros to duplicate terms (they where inserted multiple times). Not a problem, just a strange effect.

While using DeepL to translate simple manuals, I note how bloody good its suggestions are.

My prediction is that freelancers will start to care for all the extra editing features that CafeTran offers, compared with other CAT tools, just to take full advantage of the currently available MT systems.

For instance, Kilgray can should about "The items listed below are productivity boosters for translation work and user experience enhancements", but personally I haven't seen any useful feature for my work as independent technical translator.

A wise man harvests the precious alternative translations when DeepL presents them, since there's no guarantee that they'll be presented in n segments from now, too.

Meaning: I'm often very impressed by the high-quality of terminological suggestions and note that after n segments they completely miss the sense of the same term.

Yep, I know, context.

One could even argue that the use of MT engines isn't optional anymore but an obligation if one wants to use all available resources.

One argument in favour of MT: You cannot know everything as a translator. And when you run each and every segment that you translate through (currently) 4 MT systems, you often get interesting terms suggested, that make you scratch your head.

You'd better investigate them and sometimes will discover that they are indeed the "official" translation in a certain context, of a certain regulation.

That's what MT is for, not for "perfectly translated sentences". But for term hints.

Nightmare scenario: the API of DeepL requires you to deliver your PEMT results back, before you can query DeepL again.

DeepL is here to stay.

Shouting at it or ridiculing it, won't make it go away.

We'd better prepare for it and try to integrate it as good as we can in our workflows. 

Meanwhile it's absolutely clear to me, that DeepL has indexed comparable sources of the texts that I'm translating. Often times it acts like a professional reviewer with expert knowledge of the topic, e.g. by placing dashes where needed, indicating that it 'understands' what it's talking about:

der Luftgeschwindigkeit oder der Temperatur > luchtsnelheid of -temperatuur

I can only say: wow.

What I don't like about DeepL is that it starts making up words, like here, based on the one not intended meaning of 'washer':


afstandswasser and afstandswasmachine (distance washing machine) are bogus.

So I'll have to stay away during work ...

Even funnier, for this:


I get:


Hat der Karneval in Jöln schon angefange:

Don't dig to DeepL

This one is too beautiful not to share, it's almost poetic:




I vote that this will become a classic example, just like the flesh is willing.

"The whisky was invisible", or Persistent myths of MT 

What I find absolutely frightening, is DeepL's ability to 'guess' the meaning of pronouns and replace them with the correct noun that they are referring too. In complex, long sentences.


The web interface often times stops accepting input via ALT+RETURN. Strange.

I really think that we should start hiding our IPs and mask all confidential info (non-translatables etc.) when querying MT systems.

The hiding of the IPs is not task of CafeTran, the other part is, in my opinion.

Login to post a comment