The bit that interests me is using "corpus linguistics"
Corpus Linguists tend to work in statistics rather than absolute rules and judgements.
So although they would say x is a delexicalised adjective it would often be governed by something like "97% of the times in 2000 occurrences based on a corpus of size 5000000 tokens taken from the specific domain of texts of legal representations in the USA"
Corpus linguistics is based on actual usage rather than abstract rules.
Domain of the corpus is core to this idea, the classic case is the difference in the use of adjectives to describe legal and non-legal drugs.