The parts of speech in a particular language can be drawn up on the basis of syntactic properties, morphological properties, and/or (perhaps most problematically) semantic properties.
What if we just want to classify lexemes in the MorphGNT based on what morphosynactic and morphosemantic features they have?
Minimally, we might get something like this:
|−||−||−||conjunctions, adverbs, interjections, prepositions, particles, indeclinable nouns and adjectives|
|+||−||−||nouns, adjectives, pronouns, articles|
We could consider voice, but it co-occurs with aspect, so its value is predictable.
Mood only appears in finite verbs, which means it’s also predictable (arguably, co-occurent with person but see below).
Number is predictable as it co-occurs with case or person.
As things stand above, gender is also predictable (it co-occurs with case).
However, let’s consider the distinction between the 1st/2nd person pronouns on the one hand and the proforms on the other.
(There are strong arguments beyond just morphology for distinguishing the (1st/2nd person) personal pronouns and proforms. See Bhat’s book Pronouns for cross-linguistic arguments for the distinction.)
The 1st/2nd person pronouns, unlike the proforms, don’t inflect for gender. So let’s add gender to the mix:
|−||−||−||−||conjunctions, adverbs, interjections, prepositions, particles, indeclinable nouns and adjectives|
|+||?||−||−||1st/2nd person personal pronouns|
|+||−||+||−||nouns, adjectives, proforms, articles|
The ? under person for the personal pronouns is because they don’t really inflect for person. Person is lexical in the personal pronouns.
Interestingly, though, if we do give it a + then we don’t need gender to distinguish the category.
You may wonder what about degree. I’m currently of the inclination that degree is better modeled derivationally rather than inflectionally, although that’s worthy of a separate post.