I couldn't find how in general sortal classifiers work in compound words but I notice the following myself. By sortal classifiers I mean those that actualize shape boundaries, in contrast with mensural classifiers that create shape boundaries.
Some specific and originally exclusive classifiers can be used to compose compound words. This is probably because in ancient Chinese the structure is N+NUM+CL. With the number deleted the structure is reanalyzed as a compound noun. For example,
书 + 一 + 本 → 书本
马 + 一 + 匹 → 马匹； 布 + 一 + 匹 → 布匹
花 + 一 + 朵 → 花朵； 云 + 一 + 朵 → 云朵
人 + 一 + 口 → 人口
枪 + 一 + 支 → 枪支； 箭 + 一 + 支 → 箭支
Some classifiers with strong implication of shapes or features can also work in this way.
纸 + 一 + 张 → 纸张
米 + 一 + 粒 → 米粒； 沙 + 一 + 粒 → 沙粒
叶 + 一 + 片 → 叶片； 肉 + 一 + 片 → 肉片； 药 + 一 + 片 → 药一片
房 + 一 + 间 → 房间
土 + 一 + 块 → 土块； 肉 + 一 + 块 → 肉块
The above compounding doesn't change the lexical meaning, but generally the compound words are used collectively; they refer to a collective concept or type and most of them can be used with + 一 + 些 (some). They rarely can be individualized, exceptions are notably 一个房间、一把枪支、一个土块.
However, some compound words have altered meaning.
These are generally individualizable. For example, 一个信封、一把扫把.
These seem a bit ad hoc. 1) Is there any systematic view on the ability of sortal classifiers to form compound words in Chinese and its usage?
This is not a feature exclusive to Chinese. For example, in Nuosu Yi, ꄮ (te) is the pine tree. ꁧ (bbo) is a tree. ꄮꁧ refers to the pine tree. Like Chinese, ancient Yi is also primarily monosyllabic. I guess it was originally
ꄮ + (ꋍ) + ꁧ → ꄮꁧ
ꋍ (one) usually omitted. Then it's reanalyzed as ꄮꁧ, which is a binome. It seems that the resulted compound words is used more individually than collectively. One pine tree is
But my knowledge of Yi doesn't allow me to dig deeper. Also, it could be diachronically wrong since ꁧ being used as a classifier could be later than the appearance of the binome ꄮꁧ.
Another Nuosu Yi example
ꁱꂷ bbur ma （文字, writing/characters）
ꁱ is the verb of to write, while ꂷ is the universal classifier corresponding to the Chinese 个.
ꁶꂷ nbit ma： n. （装饰品上錾刻出来的）花纹 decorative lines and drawings put on on clothes or dishes
ꁶ nbit： v. 錾（石头）chisel (rock)
The structure seems different though, probably a V+O lexicalized as a noun.
2) How is the ability of sortal classifiers to form compound words other Sino-Tibetan languages (like Yi and Tibetan)? Is it generally present? If present, is it active or inactive? How does it change the word meaning and usage?
It turned out to be quite common in the Lolo-Burmese languages, e.g. Loloish (Jino, Lisu, Lahu, Hani, Naxi, Nusu), Nungish (Derung), Burmish (Zaiwa, Achang), and also Kra–Dai languages, though less prevalent, e.g. Thai, Lao and Be.
e.g. Naxi： ndʑy³¹ (mountain) dɯ³³ (one) ndʑy³¹ (mountain)
Interesting it also existed in Chinese Oracle bone scripts and bronze inscriptions, and completely disappeared from the Han Dynasty.
人(people) 十有六(16) 人(people)。 (Oracle bone scripts)
In Oracle bone scripts, this structure is only used for 人、羌、骨. In bronze inscriptions, it expanded to 牛、羊、聝、旅、田.
From the Chinese classics, it seems that the grammaticalisation of classifiers and word disyllabification have happened at the same pace (汉语量词语法化动因研究). Instead of viewing classifiers as morphemes in compound words, I would agree that the fact that classifiers have a noun/verb origin contributes to the formation of both classifiers and noun-classifier compounds.