Print

Print


On Mon, Nov 22, 2010 at 5:58 AM, Matthew Martin
<[log in to unmask]> wrote:
> re: word generation
>
> I'm not entirely sure I understood your question.  It certainly is a good idea to
> machine generate the various possibilities that a derivational system is
> capable of-- and it isn't very hard.  I've manually done the same thing with
> toki pona--GZB has some entries in it's dictionary that are marked "machine
> generated"

The question is, is there a generic piece of dictionary-building
software already in existence that would let me put in some words and
a set of derivational rules and auto-generate all of the derivations
for every word that I put in for me (and possibly tag the ones that
don't have human-verified definitions yet).

> The hard part is deciding what the heck the results mean.  In toki pona, you
> can attach all possible 100 odd modifiers to kala (fish) and about 20% seem
> immediately meaningful, implying things like oysters or flying fish. The other
> 80% are suggestive of nothing (what sort of fish is the smart-fish?).  On the
> otherhand, the same experiment with the word sona (knowlege) is much more
> productive and has fewer gaps.  There is probably a similar story with English,
> when you attach un-, -ly, -ment, -ish to words, some are much more
> productive than others, but there isn't a machine way to tell in advance if
> unfroglymentish mean anything or not.

The difficulty can vary depending on what kinds of derivations you
have; some are less ambiguous than others. But you'd reasonably retain
the option of having a human verify or clarify the computer's
suggestions.

> Btw, if you are going to do this on a computer, you can do do it with a single
> line of SQL using cross joins, one of the rare cases when a cross join is really
> useful.

Maybe I'm just being dense, but I don't see how that works. Does SQL
do regex replacements or something and I wasn't aware of it?

On Mon, Nov 22, 2010 at 10:05 AM, Gary Shannon <[log in to unmask]> wrote:
> It's not really a "derivational system", but I once compiled a list of
> ways one kind of word can be derived from another. It's at
> http://fiziwig.com/conlang/functions.txt I came up with about 250 or
> so possible derivational relationships between two words.

Oo, neat. That's be really useful for the "research idea"- building a
complete derivational table for one language for some standard set of
derivational operations, and comparing it to the equivalent tables for
a bunch of other languages, to see where they overlap and where they
don't.

-l.