On Sat, 21 Jul 2012 17:17:18 +0100, And Rosta <[log in to unmask]> wrote:

>Logan Kearsley, On 13/07/2012 00:38:
>> On 12 July 2012 17:26, Alex Fink<[log in to unmask]>  wrote:
>> [...]
>>> One could perhaps generalize this so that there aren't just a set of disjoint domains and that's that, but rather a tree of domains, with a progressively longer form needed to switch to a word whose least common ancestor with the current one is farther up the tree.  Or a system like that but with multiple current topics.
>> This is what I call the namespace approach. [...]
>Alex & Logan, could you each give a couple of examples, so I can better grock what you're getting at? Logan's idea eludes me completely. I think I get Alex's, but it'd be nice to see a sketch of an actual morphological implementation.

Here's an example of how the thing I envisioned might look, a quickly-generated example rather than one well thought out in any sense.   Say that stems of words which fit within the taxonomy are all built of CV syllables, with one syllable for each level of the taxonomy.  For instance, stealing a piece of Wilkins' taxonomy (though not his forms) (why is _An Essay towards a Real Character and a Philosophical Language_ not on Project Gutenberg?), a part of the tree might be this, with leaf nodes starred:
- /zi/ mammals
- /zita/ canids
* /zitaba/ domestic dog
* /zitabe/ wolf
- /zika/ felids
* /zikabe/ lion
- /pi/ stones
- /pido/ vulgar stones
* /pidobe/ slate

Then the system would work thus: there is a pointer to some location (some node whose children are leaves, say) in the tree, marking the subject matter of the latest conversation, and words could be abbreviated to only the suffix which differs from the pointer.  (Restrict / decorate as needed for self-segregation etc.  Also I've left out the machinery for setting the pointer.)  So if the pointer was at the 'canids' node then /be/ would be 'wolf', /kabe/ would be 'lion', /pidobe/ would be 'slate'.  /be/ could be either of the latter two if the pointer were appropriately placed elsewhere.  Etc.  

Logan's example is closer than mine (Wilkins') to what a tree actually optimized for this use would look like, except that I wouldn't insist that the suffix "stellar" in one subtree have even metaphorical relation to the suffix "stellar" in another subtree (though all the better for ease of learning when it can).  I can't tell if Logan's is a tree with depth more than one, but it would help in mine: a word like "theory (as per the scientific method)" could be a leaf of a note which was parent to nodes for the terminology of particular sciences.  

>2. In my past work on Livagian, I found that the major causes of verbosity are (i) the morphological expression of predicate--argument structure far more than the morphological expression of predicate content, and (ii) semantically decompositional syntax (e.g. "cause to die" rather than "kill"), which proliferates predicates and requires the encoding of much concomitant predicate--argument structure. (Livagian is a spoken language; if it had a two-dimensional graphical phonology, like UNLWS, then the verbosity problem would disappear.)

Hm, my gut instinct would be that one could save a fair amount of length on this stuff if one were willing to make the choice of which construction was morphosyntactically unmarked based on the context in a sufficiently complicated manner (probably including lexicalizing various kinds of different defaults for particular lexemes, e.g.).  Especially if one had large corpora and could construct that rule in a usage-based way.