The impression I get when looking at how conlangers deal with parts of speech is that the main method in existence is this:
- Take the English set of parts of speech
- Conflate some of them (typically <verbs and adjectives>, <nouns and adjectives> or <adjectives and adverbs>).
- Break even.
Let's break even in some other manner. (NB: I claim very few conlangers will profit!)
What distinguishes word classes? The following seem to be reasonable characteristics, ranked from strong to weak:
- syntactical properties (strong)
- certain, but not all information structural properties (strong, but diffuse!)
This one's somewhat unclear, and that's good, because it gives us some flexibility. Clearly there's information-structural differences between words within the same word class sometimes.
- morphological properties (somewhat strong)
- semantic features (weak)
Strength should be seen as correlating with how easily applicable it is. Semantic features is weak because, well, "life", "live", "alive" have a really strong semantic overlap - they seem to refer to the same underlying concept, but they provide different information-structural and syntactic "interfaces" for that meaning. If a word satisfies either of the two top requirements, or two of the weaker features, or partially one of the top features and wholly one of the weaker features, I think that should be sufficient.
So, some ideas here.
Three Parts of Speech
These do not quite fit in the same language.
Titles of address
Consider a language in which titles - sir, mister, reverend, etc - deviate sufficiently from nouns and adjectives as far as syntax and morphology go not to qualify as either.
These particles can go anywhere in a noun phrase - including the edges - have no morphosyntactical markers, but there may be unique morphemes that go on them. These may correlate with grammatical subsystems of the language - gender and number and such. They cannot stand by themselves as head of an NP, however.
This class is not entirely closed, and there may be ways of turning adjectives, nouns and verbs into this class. There are, however, some titles that do not have corresponding nouns, verbs or adjectives. Many of those that have no nominal, adjectival or verbal cognates are also morphologically very simple.
Particles of Social Relations
Similar to the previous category, but these mark social relations of humans. In particular they (optionally?) mark the relation of nouns to the higher ranking noun (either by some rank hierarchy or by some syntactical notion of rank). A quirk is that they can also mark the relation of vocatives to the speaker - as long as the vocative is morphologically distinct from the object case, which it isn't for all nouns and names.
Ways of manipulating the rank - voice transformations or other tricks - may permit for marking social relations centered on a person of lower status, or who occupies a syntactical role with lower syntactical status.
There could maybe also be a way of introducing persons who have no semantic role other than being the syntactic center. Maybe having an auxiliary (or some voice construction) whose subject is the social hub and which demotes all other nouns to lower status? Maybe topics always are social hubs, and hanging topics are permitted? Maybe there's some adposition or case marker that raises the rank of a person. Also, the social hubs material possessions can take a similar, inanimate marker.
Particle of Reference
A referential particle is a postposition-like optional word that goes after an NP to which a third person pronoun in the same clause or nearby will refer. There are also a separate particle of possessive reference, which goes on the possessor of a possessum, if these also have separate syntactical roles in the sentence. Unlike a proper postposition, they cannot "outrank" a conjunction: man PARTICLE and his house : a man and his own house (not a man and some other man's house). Multiple particles can be on separate NPs that are co-referred to, even if these are different syntactical entities or possibly even separated by subclause boundaries: the man PARTICLE listened while the woman PARTICLE played the piano PARTICLE and they formed a beautiful scene.
I have been thinking about this post for a while now, but the three types of particles I came up with seem to be about borderline for whether they make sense as word classes. Clearly, I have provided them all with rather unique semantics and made sure to give them unique syntactical behaviors, but it still seems a bit much to call them word classes.