Written text data is erroneous
Hi,
I am looking at samples of the training data for the Kikuyu language(kik/tts) and the text is not properly written in A LOT of cases. First of all the accent marks in the different texts vary . eg in some places I see "ù, û,ū" when it should be "ũ" same applies for ĩ. with these 2 characters being our main vowel characters with accented marks. In addition, we lack the "l" letter in our latinized alphabet wheras I have seen it in some of the text data. This also tires to the "b" english sound which for us is written as "mb" eg mbembe, instead of bebe.
I wonder where else I can raise such basic concerns.
I am a Masakhane contributor for Gĩgĩkũyũ/Kikuyu language and it is for reasons like the above I am part of that intentional NLP community.
Thanks
Hi Wangari,
Thanks for trying out our dataset and for flagging this issue. Looping in @diack for visibility to share this with our data collection partners. We will update the dataset once we have feedback.
Best,
Perry
Hi Perry,
If possible, could you keep track of the errors and fixes in WAXAL? I'm working on a gold parallel dataset for grammatical error correction for African languages, and some of these common Kikuyu mistakes would be very valuable for me.
Thanks!
Matt
@Wangari4 Thank you for flagging the errors, would you mind sharing specific examples where you see an issue so that we can raise it to the teams who generated the text?
@mattmalone212 Thank you for your question, we'll first assess the nature of these errors and get back to you on tracking.
Best regards,
Abdoulaye
hi @diack
I flagged some errors from the test set. From first glance it is safe to say from that set more text examples are erroneous in one way or the other than those which are fine. Here are 13 examples: https://docs.google.com/document/d/1r6HzcWvqymXSbEc-TMpYebSRYqwQKBnw-ph2nFEnOgI/edit?usp=sharing