In this choices file, there is a predicative adjective lexical rule which is not directly specified as such, but rather inherits that property from a supertype. This leads the customization script to do the wrong thing with the specified PNG information (here NUM) and constrain the adjective's own INDEX rather than its subject's. Since the adjectives INDEX is of type event, this leads to a non-loading grammar.
Meanwhile, selecting "predicative" on this rule as well as its supertype also leads to a non-loading grammar, because the relevant supertype (stative-pred-lex-rule) is added to both.
It seems that the expected user behavior is to NOT create a mini-hierarchy like this, but rather just specify predicative on all such rules. I suspect the best fix here is in the customization behavior, rather than validation.
In this choices file, there is a predicative adjective lexical rule which is not directly specified as such, but rather inherits that property from a supertype. This leads the customization script to do the wrong thing with the specified PNG information (here NUM) and constrain the adjective's own INDEX rather than its subject's. Since the adjectives INDEX is of type event, this leads to a non-loading grammar.
Meanwhile, selecting "predicative" on this rule as well as its supertype also leads to a non-loading grammar, because the relevant supertype (stative-pred-lex-rule) is added to both.
It seems that the expected user behavior is to NOT create a mini-hierarchy like this, but rather just specify predicative on all such rules. I suspect the best fix here is in the customization behavior, rather than validation.
choices.txt