On Fri, 08 Feb 2008, Peres wrote: > Hello crew, > > I subscribed to tp-devel some time ago but haven't introduced myself yet. > My name is peres, and I am working (slacking) on the random planet names > task. You can randomly find me on #tp :)
Welcome peres! > Here is a brief explanation of how my generator should work. My design > involves 3 stages, as follows: > > 1 - generation of pronounceable sequences, based on valid syllables for a > selected language [currently English]. Reading the Wikipedia pages for syllable and phoneme removed the need for me to ask a question here. > 2 - filtering, so that bad sequences are discarded. This is based on > sequence-level features like repetition of sounds and the like. > > 3 - rendering: here the surviving sequences are transformed into words, by > mapping the sounds they represent into a spelling [this should be done > according to rules for a language pronounciation]. > > At the moment I have an implementation for stage 1, I'm working on #2, and > I have very rough ideas about #3. Conceptually, this sound very compelling. > If you look at the attached files then please consider that stage 3 is so > trivial right now, that it renders words using intermediate representations > of their sound (and we all know that English is not exactly a phonetic > language). I guess that explains the frequency of the occurrence of the letter 'j'. It's funny how many English (language) place names are just made of simple words, though more exotic names are kind of appropriate in space. > So, while part of the weirdness in those lists can be attributed to a > missing rendering stage, the main problem is that the filters in stage 2 > are not good enough. In fact, most of the words you see there should be > discarded instead of being rendered. Rejection ratio is now about 20%, but > I am expecting to get 85-90% in order of obtain good results. > > My kind request to the mailing list is the following: I would love if some > of you could look into the lists, and delete anything but the names they > deem usable or with some potential (please consider the rendering problem > here). Your feedback will be useful to add further filtering rules. I would have thought you would be more interested in the bad ones and why we think they are bad. Actually, that is a question: are the filters positive (things to keep) or negative (things to remove)? > Thank you :) > peres Looks like a good start. Later Lee
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ tp-devel mailing list [email protected] http://www.thousandparsec.net/tp/mailman.php/listinfo/tp-devel
