Start a new topic

Machine Translation: yes or no?

My statement of today is:


Blindly depending on MT engines is unacceptable, but not using valuable suggestions of several available MT engines isn't acceptable either.


Or: As a professional translator one has to take suggestions offered by MT system into account, in order to improve quality of translations.


Am I opening a can of worms with this statement?


Argument in favour of MT: You can enrich your personal glossary by storing suggestions from the MT systems. Variations in style etc.

So, dear colleagues, how (if at all) do you use MT?


First you build the target segment in your mind and then you have a look at the different MT suggestions?


Or rather the other way around?

Although I don't use MT for translation (it's almost completely useless when translating from Japanese), some times I query it to find a particular term or idiomatic expression, just as I would do with a traditional dictionary but probably in less time. At times it helps, but most of the time not, or it is even dangerous. The only pity—and the most disgusting thing these days—is that many Japanese translation companies and end clients have learned to rely upon Google Translate to check freelance translators' work, and you should see how they believe it!


Even when translating from English I often end up rewriting everthing because the word flow is not the way I would write from scratch.


And frankly, the few times I try Google Translate (for Japanese) or DeepL (for English) I never see something for which I would say "Oh, this is a really nice translation and I'm definitely going to use it". In other words, MT is just a huge collection of mediocre translation with the occasional spark of light in the dark.


This, at least, is my experience from very little use, but it could also be that if I used MT more often I would get surprised.



>MT for translation (it's almost completely useless when translating from Japanese


There is this CafeTran user from Japan who switched from Mac to Windows because of a superb MT system for Japanese. You can search the forums.

Well, I speak only for myself (I translate from Japanese into Italian, not English). My wife, who is a translator like me and sometimes uses Google Translate from Italian/English to Japanese, says that it has improved but it's still very unreliable. We don't know what other MTs are available for Japanese which can be called "superb", but if one actually exists we would like to know. I searched the forum and found nothing.

I've contacted you privately about this topic.

Yes, thank you. If Masato replies I will have a look at that MT software.

I find it totally fascinating how the four MT systems that I'm using constantly jump between 5 different translations for the same mechanical part, every next segment.


In segment 234 MT system 1 uses A, 2 uses B, 3 uses C and 4 uses D.

In segment 235 MT system 1 uses B, 2 uses C, 3 uses D and 4 uses A.

In segment 236 MT system 1 uses C, 2 uses D, 3 uses A and 4 uses B.

etc.


Are you using all four MT systems to decide which one to use? If so, isn't this another waste of time?

>Are you using all four MT systems to decide which one to use? 


No. I take from every one of them what I think what makes a good suggestion, a better one than I already get from my TM or glossary or that I have in my mind.


>If so, isn't this another waste of time?


Define: 'waste of time'?


No, I don't think it is. I think that the use of these 4 MT systems highly improves my style, knowledge of the subject and points me to the specific jargon that is used in the field of mechanical construction.


Not to forget the extra fun that they generate (and here I don't mean the ridiculous suggestions like 'seals') by offering interesting words. I really enjoy this! It's a kind of a puzzle. Which meaning does suggestion A for a mechanical part stress and which one does suggestion B stress. And which one do I want to use when and where, while striving for consistency?

alwayslockyourbike > Define: 'waste of time'?


Sorry, bad wording. I meant to say "If so, isn't this taking too much time?". But, since you are having fun with MT I presume that it's not. Maybe I should think about having some fun too.


By the way, from what I understood so far the MT system that Masato is using is not of the statistical type like Google and DeepL, right?

>By the way, from what I understood so far the MT system that Masato is using is not of the statistical type like Google and DeepL, right?


If I remember correctly, not. He'll respond, I'm sure.

Brief question: I think I already know the answer but I wanted to be sure: MT cannot be operated for individual translation units only, right? My previous CAT tool has this function, which I find very practical for querying MT only occasionally.

Login to post a comment