Re: A method of generating "flavored" words

>Here's an interesting procedure I stumbled upon while playing with word
>generation. Take words from a lnaguage whose flavor you wish to capture.
>This
>can be your own conlang to generate "matching" words, or any natlang you
>would
>like to copy the flavor of. Or it could even be a combination of two or
>more
>different natlangs to get a blended flavor.
>
>Collect a few dozen or more words. Divide those words into VCV groups where
>the
>V's represent ALL the adjacent vowels in a word and the C represents ALL
>the
>adjacent consonants. (...)
>These groups will be assembled together with other groups from other words
>with
>the rule being that the complete set of adjacent ending vowels must match
>the
>complete set of starting vowels for the attached group.
(examples snipped)

>Try it. You'll like it.
>
>--gary
Neat! But... in its current state, it only works with certain kind of
phonotactics. The method does well in getting clustering correctly, but
rules working farther than the next syllable end up ignored. Exagerrated
example from Finnish follows:
ruokki "auk" + värikkyys "colorfulness" + pyylevä "chubby (person)"
-> "uokkikkyyle", which violates about 3 rules of Finnish phonotactics...

It is fixable, however. To fix the distribution of initial/final vowels,
just treat the word boundary /#/ as a consonant phoneme - so you'd have to
find -0#V- and -V#0- segments for the end and the beginning. And with
harmony systems, underspecify phonemes where applicable. With these, my
example would turn into maybe "luokkikkuulen" - which is better, but still
wouldn't pass as a single word. The only way to fix the "no consecutiv
plosiv geminates" rule, I'm afraid, would be to use segments longer than
VCV; maybe CVCV with 50% overlap.

Oh, and I now recall coming sum' years ago across a name generator using a
similar system; it used N-grapheme segments with N-1-grapheme overlap
required...

John Vertical