(Note that redundant remarks on distinct fields are purposefully separated in dedicated threads so they can be discussed separately)
I think that the lexical category should also be tracked as statement in this model. This is because a lexical category only have legitimacy within a grammar theory, and a language can be analyzed with different, possible conflicting, grammatical theories.
To give a concrete example, look at lojban grammar, which uses endogenous terms to describe its own grammar with concepts which doesn't completely match those of classical scholarship grammars. But it doesn't mean that you couldn't try to interpret sentences using vocabulary of such a grammar. More generally, it shouldn't be assumed that there is a single theory that make general consensus as of its ability to describe every utterance of every language that could ever be produced. Some go as far as questioning "Are there any languages that appear to have no grammar?". And what about if someone come with a theory where grammatical categories of lexemes are dynamic?
So at a minimum, this field should allow multiple values which are as flexible as Wikidata claim structure. But then, why would it be not rendered as a specific field rather than as a statement?