Print

Print


On Wed, Jul 8, 2009 at 8:46 PM, Mechthild Czapp<[log in to unmask]> wrote:

Brett said:
>> So if someone does make a program that produces a whole language
>> randomly, I might try generating a language and then studying it and
>> putting it to use, to see what that's like.

> I would like such a program as well though... for playing around.

Besides the features I mentioned in my last post, what would y'all
like to see in such a program?

My inclination would be to take the words.pl script that John Fisher
posted to the CONLANG list in 1997, expand it to be able to handle
more flexible input, and add additional scripts that match up the
word-forms generated by words.pl with meanings from additional input
files in various ways (with isolating, agglutinating and fusional
options).  The existing version of words.pl is on my website in the
conlang-scripts.zip file.

I have a vague idea about how to randomly create a case system: have a
data structure of some kind with theta roles (agent, experiencer,
force, patient, focus, source, recipient, location, instrument...)
arranged in a graph, such that splitting up the graph into contiguous
subsets would usually if not always result in a plausible case system.
 The simplest structure would be a linear continuum like:

agent
force
experiencer
focus
patient

and putting the first three into one case and the last two in another
would get your basic nominative-accusative system, putting the first
two in one case and the last three in another would be more like
ergative-absolutive, and splitting them up three or more ways would
get something like an active or tripartite system.  But probably
something with a more complex graph than this linear continuum,
working in the source, recipient, etc. theta roles as well, would give
more variable and interesting results.  There probably also needs to
be a generic "subject of intransitive verb" role, somewhere in the
middle of the graph with connections to all the other major roles, to
let the system generate true ergative-absolute and active systems.

Maybe the graph could be structured in such a way, with "subject of
intransitive verb" defined as the center, so that the distance of a
theta role from the center gives its probability (in a random language
generated by the software) of being represented by an adposition
rather than a case.  So the graph is split up into contiguous subsets,
and the subset that includes "subject of intransitive verb" will
generally get the least-marked case form, the neighboring subsets will
be matched to other case forms if the language has cases at all, and
the subsets that are further out will be more likely to be marked with
adpositions instead of cases.

Does anyone have any suggestions for creating structures like that for
generating random interesting and plausible tense, aspect, mood etc.
systems?

-- 
Jim Henry
http://www.pobox.com/~jimhenry/