aTbRef - improved support for Google Translate

[mwra (Mark Anderson)]

Thanks to helpful reports by community members, it is clear that simply translating whole pages via Google Translate (GT) is not as helpful as hoped. In investigating this is also shows that whilst GT ostensibly supports word or HTML element ‘no translate’ markers, it struggles to render the result. So if the resulting visual render is garbage, that’s little help to the reader. Still, we are all paying $0 for the service so we can’t complain too much. It’s a reminder that computers don’t understand human-written language. Translation is a probability-based best-guess: impressive, but all to often wrong.

But I can still do something to help non-English speaking users of aTbRef…

$NoTitleTranslate. I now (in the source TBX) have boolean for notes where translating $Name is not helpful. A good example is notes on action code operators or system attributes. Notes with this set true should not translate their titles:

Interestingly, GT completely fails to translate the table in the table in the above grab. It is only asked not to translate the word ‘Attribute’ in the left column of the table. This shows how poorly the process understands what the human reader understands in a momentary glimpse. Still, what we get is free.

In the sitemap now doesn’t translate quite a few note’s title ($Name value).

Likewise a lot of the big listings of code & attributes don’t translate items in the list as the aim is to find a note about $ChildCount rather than a listing using a translation of the name. In the app, ChildCount is the label used, not the translation. Having found the note about that attribute, hopefully GT translation will assist the reader in understanding what ‘ChildCount’ means in their chosen language.

Anyway, I hope this at least a little better than before and the changes are now live online. It is less comprehensive than I’d like but the service is free and, on digging deeper, GT has significant limitations in what it can do.

If you use the GT feature on aTbRef pages, or tried the feature and gave up, I’m interested to hear if the above improvements help. I can’t improve issues of which I am ignorant.

With over >2k source notes making the website, word-by-word markup of source for translation is but a dream. But, if there are achievable things I can do to help those non-native-English users, I’d love to hear. What I’m striving for is to make the resource more accessible to those without strong english language knowledge.

HTH

[mwra (Mark Anderson)]

Aha, I fixed the issue (my fault!) with the table:

But note the issues with whitespace remain, which I can’t fix.

[eastgate]

Idea: what happens if you try something like <code translate="no" style="padding-left: 1em;"> ?

The CSS element ::before might be useful. You could emit <code translate="no" class="noTranslate"> where

.noTranslate::before {
   content: " ";
   }

[mwra (Mark Anderson)]

Yes, I think such remediations are possible First, I’m trying to figure out the logic of which/why space is being lost.

I think the problem is that what I really want/mean is translate this text but where marked, re-use the source word instead of the translated word. So:

There is an <span translate="no">attribute</span> called <span translate="no">WordCount</span>

translates as:

Il existe un attribut appelé Nombre de mots

now return the marked words:

Il existe unattributeappeléWordCount

OK, we get an in elegance around the definite article and the buggy word spacing. Regardless, I sense GT doesn’t roll that way and likely translates text strings between exclusions before reassembling them with little regard to the HTML structure/render.

[WAKAMATSU (WAKAMATSU kunimitsu)]

Dear Dr.Mark Anderson,
Thank you for your consideration and efforts,
for changing the display of Tinderbox Reference File.
The translated pages are now much easier to read than before.

The only disappointing thing is that when I try to go back to the page,
the translated text reverts to the original text.
The window for changing the translation remains hidden.
I do not think it is just a problem with the Japanese environment.

I will list the things I noticed.
(I apologize for worrying about trivial matter.)
#01 : Strange translation(?):
(Secondary windows)
When you click on ‘torn-off’, the Secondary windows’ page where appears,
(why past participle torn is used?), on the refreshed page
The title is correctly displayed as “Tear-off windows”.

#02 : About Google strange translation:
These are primarily ‘torn-off’ windows.

This translation in Japanese is an interesting translation
of the word “ripped” window.
(This is a literal translation based on a natural expectation.)
I will read further and let you know if I find anything else.
Yours, WAKAMATSU

[mwra (Mark Anderson)]

@WAKAMATSU

The only disappointing thing is that when I try to go back to the page,
the translated text reverts to the original text.

Remember, you are translating the page you are no, not tall pages on the site. When moving between pages you may need to re-trigger translation. We can only use the service as provided by Google.

@WAKAMATSU

(why past participle torn is used?), on the refreshed page
The title is correctly displayed as “Tear-off windows”.

I’m afraid you need to ask Google that question. We can only use the translation they provide.

I appreciate the feedback, not least as it shows how machine translation is still poor for anything more than the most trivial. So, it is great for finding the nearest McDonalds in a strange country, less so for more intellectual needs.