On 2/12/2014 4:26 PM, Jim Henry wrote: > On Wed, Feb 12, 2014 at 2:21 PM, H. S. Teoh<[log in to unmask]> wrote: >> I have never felt comfortable with using automated tools to generate >> entire lexicons. It feels too much like a relex of something. It's one >> thing to have a program automatically fill out conjugated forms, it's >> another thing altogether to have the roots automatically invented as >> well. > > It doesn't seem like a relex to me, if you are using a carefully > hand-generated list of definitions to match the automatically created > word-forms to. I did that with säb zjeda, taking the NSM list and > some material from Rick Harrison's ULD and deleting, merging, > spllitting and adding a bunch of items before I ran my generator > script, and assigning guesstimates of frequency-weight which the > script would use to decide which definitions get short roots and which > get longer ones. I did something along those lines with Kisuna. That was really more of an experiment on whether a language with only 6 segmental phonemes (guess which ones) would be usable, though. Most of the Zharranh vocabulary also was randomly assigned to ULD entries, and as I've mentioned, I used to fill gaps in Jarda and Tirėlat that way. I've thrown away quite a few of those words, but some of them stayed in. I do wonder if something like Kisuna would ever develop as a spoken language, though. I can't find any actual text, but I've got the vocabulary list. It ends up looking something like this: nasan 619 just, fair, equitable [aj] nasana 311 stupid [aj] nasani 5C4 whistle [v] nasank 09C organ (of body) nasans 482 slowly (with little speed) [av] nasansi 4C7 potassium Of course, for this to work, word boundaries must be audible. But that can be done with fixed stress and allophones.