Classifiers

[sylvainkahane]

In the UD=>SUD conversion, clf must be converted into comp:clf with the classifier as the head.

Example
UD annotation: N -[nummod]-> NUM -[clf]-> CLF
SUD annotation: N -[mod]-> CLF -[comp:clf]-> NUM

[sylvainkahane]

I change my mind. In language with classifiers, numerals can only modify classifiers, they cannot modify directly a common noun. But the numeral or determiner must be analyzed as usual, as mod or det.
In consequence, we don't really need a clf or comp:clf relation. Classifiers van be indicated by a category NounType=Clf. Moreover in some languages (e.g. Thai), the classifier can be used alone, as an anaphoric pronoun. In this case they won't be any place for a clf or comp:clf relation.

New proposition:
UD annotation: N -[nummod]-> NUM -[clf]-> CLF
SUD annotation: N -[mod]-> CLF[NounType=Clf] -[mod]-> NUM

[bguil]

Note that the relation clf never appear in the context you mention in UD_Thai-PUD whereas it appears in most Chinese treebanks: https://universal.grew.fr/?custom=68c98817d142a. Maybe UD_Thai-PUD should be updated.
I will updated the conversion as you proposed and test it on Chinese.

[sylvainkahane]

Yes UD_Thai-PUD do not follow the UD analysis of classifiers. I do not think it will be refused by the UD validator.