Tamazight data is mislabeled
#7
by
ayymen
- opened
The "Tamazight" ("ber" and "ber-Latn") subsets of the data are entirely in Kabyle ("kab").
Please know that by continuing to provide these datasets and the translations on Google Translate where pure Kabyle is labeled as Tamazight, you are:
- Disenfranchising speakers of variants other than Kabyle, some of them ironically have historically always called their dialects "Tamazight", such as speakers of "Tarifit", Central Atlas Tamazight, and Nafusi. Kabylophones have until recently only called their variant "Taqbaylit".
- Confusing people who are working on Kabyle or are searching for datasets in Kabyle since they wouldn't be able to find this dataset by searching using the language code "kab".
I kindly suggest you either:
- Simply change the code to "kab" and change the name on Google Translate to "Kabyle".
- Or create datasets and models that would be worthy of a generic "ber" or "Tamazight" code/label, by taking a standardized approach or by combining data from multiple variants (not just Kabyle) at the same time and let neural/statistical/example-based MT models output their own "Standard Tamazight" or "Interamazigh".
The Tifinagh portion of the data does not follow any widely used (Northern Berber) Tifinagh writing standard I know of, neither the one used in Morocco nor the one used in Libya. It seems to be simply an inconsistent transliteration from Latin without taking into consideration any rules or norms.
And finally, where did the glottocode you're using for Tamazight ("cent2136") come from? it seems to be associated with Central Ojibwa.