Update README.md

added Data collection and verification (aka "Workflow") section to the README

README.md

@@ -2222,6 +2222,12 @@ flores2smol = {
 'xho_Latn': 'xh', 'ydd_Hebr': 'yi', 'yor_Latn': 'yo', 'yue_Hant': 'yue', 'zho_Hans': 'zh', 'zho_Hant': 'zh-Hant', 'zsm_Latn': 'ms', 'zul_Latn': 'zu',}
 ```
 
+## Data collection and verification
+Several languages are contributed by volunteers; they are listed as co-authors on the [ArXiv paper](https://arxiv.org/pdf/2502.12301). For the other languages, the translation provider we contracted has worked with us for many years, and has a pre-existing relationship with professional translators for all languages in the SMOL datasets. The translators are paid a fair wage, and their identities are contractually kept anonymous to us.
+We checked the delivery for duplicate translations, anomalous source/target length ratios, and similarity with Google Translate outputs. Very few languages were flagged this way. Manual inspection turned up several issues with nonunicode fonts (e.g. ô for ɔ ) for West African languages, and nonstandard orthography for Santali. These issues were then fixed. Following this, we ran [FunLangID](https://github.com/google-research/url-nlp/tree/main/fun_langid) on all segments and found no issues; 95% of languages were identified as correct, and of the remaining 8, confusion was only with dialects (bm/dyu, kg/ktu, efi/ibb, ayl/arz).
+The largest missing check is for fluency, which is hard to measure without 1) trusted native speakers _outside of the translation agency_, or 2) trusted LLMs; neither of which exist for all SMOL languages. 
+
+
 ## Known issues
 
 There are several known issues of this dataset: