firefox-translations-training icon indicating copy to clipboard operation
firefox-translations-training copied to clipboard

Ensure numbers associated to units of measure are not translated as the same number with another unit of measure

Open marco-c opened this issue 9 months ago • 7 comments

This is likely a problem in cleaning.

E.g. Avec les frais, cette amende s'élève à 7500$. is translated to With the costs, this fine amounts to 7500 euros..

marco-c avatar May 08 '24 08:05 marco-c

num_mismatch won't help here I guess.

marco-c avatar May 21 '24 21:05 marco-c

I see all the new models translate this example and everything else related to numbers correctly. I think num_mismatch discarded all the training examples with currency conversions. Closing.

eu9ene avatar Jul 12 '24 23:07 eu9ene

@eu9ene isn't num_mismatch only filtering out cases where the numbers between the two sentences don't match?

marco-c avatar Jul 15 '24 10:07 marco-c

Yes, that's the case. For example, in the vi->en model, we often translate đồng to $.

marco-c avatar Jul 23 '24 17:07 marco-c

In https://vnexpress.net/co-nen-nghi-viec-co-thu-nhap-cao-de-tap-trung-phat-trien-ban-than-4772735.html, 300 triệu đồng is translated to $300 million.

marco-c avatar Jul 23 '24 17:07 marco-c

https://www.nettavisen.no/okonomi/na-kommer-det-store-bil-sjokket-avgiftseksplosjon/s/5-95-686476 has this issue, but sometimes the number is changed to some other random number also. Examples:

Original Translation
Det vil si at man må betale 25 prosent moms på bilens kjøpspris over 500.000 kroner. For en bil til 750.000 kroner utgjør det 62.500 kroner ekstra. That is, 25 percent tax is required on the car’s purchase price over $500,000. For a car of 750,000 kroner, it amounts to 62,500 kroner extra.
I forslaget til stasbudsjettet for 2023 foreslår derfor regjeringen en markant økning i engangsavgiften. Therefore, in the proposal for the $2023, the government proposes a marked increase in the one-time fee.
Denne er på på 12,5 kroner per kg over 500 kg. It is worth $100 per kilogram of over 500 kg.
Det vil gi en engangsavgift på om lag 17 500 kroner for en elbil med gjennomsnittlig vekt, om lag 1 900 kg. It will give a one-time fee of about $7,500 for an average weight electric car, about 1,900 kilograms (1,900 lb).
Ifølge Bilnytts avgiftskalkulator blir blant annet Toyotas populære ladbare hybrid RAV4 ca 37.000 kroner dyrere, According to Car Ny's calculator, Toyota's popular charged hybrid RAV4 is about $77,000 (U.S.) more expensive,

The same US-based unit/currency translations seem to be used when translating the page to Swedish as well.

zcorpan avatar Sep 30 '24 15:09 zcorpan

For what it's worth, you can see what models have robustness changes here: https://github.com/mozilla/firefox-translations-training/issues/757

gregtatum avatar Oct 15 '24 19:10 gregtatum