I think that to create stems analogous to lujvo, the original idea for Xorban was along the lines of the following example: if grk- is a stem and zdn- is a stem, then a new stem can be formed by simple concatenation (grkzdn-) and then given any meaning. If this clashes with an existing defined stem grkzdn-, or if brevity is desirable, then it is possible to shorten one or both component stems: grzdn-, grkzd- or even grzd-. Alternately, for a less important stem, a "z" (< zei) or another C can be used as glue within an augmented form: grkzzdn-.
I have gradually come to see the light on this idea. As long as there is a dictionary editor keeping track of coinages and guaranteeing uniqueness, then the mnemonicity of the component stems will surely be sufficient for the needs of speakers, and the added machinery needed for unique compositionality of form is rather pointless. Unique compositionality of form within stems delivers little value for its cost, because it won't entail compositionality of meaning anyway, and even if it could, then it would just be a second way of encoding syntax, and therefore redundant.
However, aside from lujvo, another consideration is the analog of tanru. Right now, everything in Xorban is thoroughly compositional. So something like "John is shaped like a bear" has to be translated as:
la ma djna le crbe zlmjake.
... unless "crbzlmj", basically meaning "arctomorphic", were already in the dictionary. Imagining for a moment that computer viruses were a novelty:
la ma djna le je li smpi psike li vdri smseki vmcake
"John is removing computer-related virus-similars."
... which is moderately complicated, involving three bindings to approximate "computer virus" veridically. These sorts of things will probably be common. One way out its to permit people to make up their own stems on the fly, but ultimately that would invite people to ignore/override the dictionary entries coined official from compounds. It would be nice to have another mechanism, easy and flexible and nonveridical. What if we delegated "y" to indicate a *nonce* or *elliptical* compound stem. Unlike dictionary stems, any stem with "y" would be explicitly non-precise and glorked from context. In short, they'd be like tanru, not like lujvo.
So the above would be:
la ma djna crbyzlmja.
"John is bear-shaped"
la ma djna le smpyvdre vmcake.
"John is removing computer-viruses"
Later on, if the compounds acquire a useful standard meaning, the <y> could be dropped and the resultant new basic stem added to the dictionary.
The pronunciation of <y> would be as followed: Presumably in operators, it would simply be [j] as has been previously assumed. As compound-hyphen, <-y-> *could* be [ (@)j@ ]. We could also allow a secondary allophonic realization of <-y->, namely [y]. This is not too far fetched, since [j] is front close semivowel/approximant, and schwa is allowed to be rounded, and probably will be rounded as it is fronted. (Side note: <w> would not be allowed in stems, or if they were, they'd have to be pronounced [ (@)w@ ].) The production rules would be:
simple-formula := ( stem y )* stem VkV
stem := (C C C* | q ( C? V? )* q)
This would give us an answer to tanru and permit experimental stems that didn't cause a clash with dictionary definitions. Two simple examples with sm-:
grkysm- "dog-related things", "dog stuff"
smygrk- "some sort of dog", "analogous to a dog"
--
co ma'a mke
Xorban blog:
Xorban.wordpress.com
My LL blog:
Loglang.wordpress.com