Datasets:
Mistakes in ar-en Translation
In the rows 15 and 16 (assuming the count starts at 0) for the the ar-en subset, there is an error in each row where in:
-Row 15: "Finland* . 17" is translated as "السويـد* . ٢١". However, it should be translated as "فنلنــدا* .١٧". This means that both the Arabic text and number were incorrectly translated.
-Row 16: Same thing applies here in which both the Arabic text and number were incorrectly translated. "Iceland* . 17" is translated as "فنلنــدا* . ٢١". However, it should be translated as "أيسلندا* .١٧".
While skimming through the Dataset Viewer, these were the mistakes that I noticed , so take care because there may be mistakes in the rest 12+ million rows.
Hi @AhmadYasser1 , thanks for reporting.
I guess this is an issue with the original data. Note that this script just downloads the original data and transforms it into a dataset.
The original data comes from: https://opus.nlpl.eu/download.php?f=UNPC/v1.0/moses/ar-en.txt.zip
And its content is supposed to be parallelly aligned sentences. The wrong parallel sentences from the original data files are:
Australia . 3
Denmark* . 17
Finland* . 17
Iceland* . 17
Mexico . 19
Norway . 23
United States of America . 25
and
استراليا . ٤
الدانمرك* . ٢١
السويـد* . ٢١
فنلنــدا* . ٢١
المكسيك . ٢١
النرويج . ٢٥
الوﻻيات المتحدة اﻷمريكية . ٢٧
Maybe, it would be worth informing the authors of the original dataset... https://opus.nlpl.eu/UNPC.php
Thank you for the clarification; I will be informing the authors of the original dataset.
Thanks @AhmadYasser1 .
And please, feel free to inform us if the authors are planning to fix this or not.
Sure, will do!
I could not find the authors because the link you sent is not working. Any other ways I can reach them?