number marking

Last night, while I was dreaming my little dreamy dreams, I
suddenly hit on an interesting way of expressing number, which
I might use in a sketch for a new conlang that I've been working
on.

Nouns have up to three forms:

(1) If the noun is plural, indefinite singular, generic, or non-
referential, or if it's a mass noun, then it's unmarked.

(2) If it's a definite singular count noun, it's marked with the
suffix "-ma".

(3) If it's a noun of high animacy (people, spirits, higher
animals), then the definite plural may be optionally marked
with the suffix "-ngan".

Under this system, mass nouns have a single form:

tsyb    "salt, some salt, the salt"

Count nouns of low animacy have two forms:

umu     "a rock, rocks (in general), some rocks, the rocks"
umu-ma  "the rock"

Count nouns of high animacy have three forms:

kila    "a child, children (in general), some children"
kila-ma "the child"
kila-ngan       "the children"

Definiteness and number would also be fully cross-referenced on
the verb:

-0      "3rd pers. indefinite object"
-na     "3rd pers. definite singular object"
-met    "3rd pers. definite plural/mass object"

Examples:

tsyb i-tsiaba           "I saw some salt"
tsyb i-tsiaba-met               "I saw the salt"

umu i-tsiaba            "I saw a rock/rocks"
umu i-tsiaba-met                "I saw the rocks"
umu-ma i-tsiaba-na              "I saw the rock"

kila i-tsiaba           "I saw a child/children"
kila-ma i-tsiaba-na             "I saw the child"
kila-ngan i-tsiaba-met  "I saw the children"

Question:  Has anybody else thought of a system like this?  Are
there any natlangs that work this way?

Another possibility I'm exploring would be to have different
definite singular and plural suffixes for different semantic classes
(e.g. one for humans, one for animals, one for round things, etc.),
which would drag the system closer to a Chinese-style classifier
system...

What do you think?

Matt.