Datasets:
Problem in dataset
Hello. I am a foreigner and I don't speak or write Taiwanese Minnan.
We were training a system regarding transcription of this written language to phonetic IPA.
The quality of data was bad because chinese text wasn't exactly what was spoken.
example 1:
chinese text is too big:
一面抹壁雙面光。
用抹刀抹牆壁，不但牆壁光滑，連抹刀也光滑。形容人很會做人，雙方面都能討好。
Tsi̍t-bīn buah-piah siang-bīn kng.
example 2:
chinese text too big/ not exactly what spoken:
七除八扣
計算時一再折扣，自總數中盡量減除。
tshit-tû-peh-khàu
Regarding column 1 there are also probably mistakes:
宣布命令 in minnan roman -> suan1-poo3 bing7-ling7
But not sure if they speak like this in Taiwan because I never visited.
Hope that you will understand my concern and hope that you will understand because It's difficult for me to explain it.
Thank you.
The quality of data was bad because
chinesetext wasn't exactly what was spoken.
I’m not a member of the dataset team, but I’m Taiwanese, so I may be able to help clarify this from a local perspective.
The chinese column seems to be a Mandarin Chinese translation or explanation, not the exact spoken transcript of Taiwanese Taigi / Taiwanese Minnan. Therefore, it should not be used for IPA conversion, ASR evaluation, or phonetic alignment.
Also, pronunciation can vary significantly across different regions in Taiwan, so some romanized forms may reflect regional accents or transcription choices. In recent years, the name “Taiwanese Taigi” has also become more commonly used instead of “Taiwanese Minnan,” especially in Taiwan-centered contexts.