Swarm Doc Allspice Mark David Edinburgh VHFC Results Swarm Primer Simulation Last Week |
Summer 2003: Modelling the emergence and subsequent dispersion of vowel harmony in artificial agent populations, with an impetus from homophony. See Previous Work 25 August 2003 This morning I put together a spreadsheet of vowel frequencies in all the harmony corpora. It has the figures for each vowel across all available languages, as well as a front/back/neutral breakdown. I hope to have this ready in html by tomorrow. I've spent all afternoon trying to get a harmony asymptote at 0.5. No luck. Now I'm attempting to justify my consistent failure with evidence from the simulation. Since probMisspeak is set to 0, the only motivation agents have for deharmonizing is homophony avoidance. But when the lexicon gets to the point where all the homophony has been deharmonised out, there is no longer any reason to change the lexicon. Hence the asymptote. The next step is to see if the final value for harmony is constant for all values within the reasonable range for parameters. If it is, this could indicate that the lexicon, thirty.dat, is responsible for this value. As it stands, the current starting lexicon is 3.33% homophonic, 43.33% harmonic, and exhibits 30% potential homophony and 51.33% expected harmony. It is neither harmonic nor homophonic, but I have as yet no way to predict the harmony levels in the absence of homophony. This is because agents pick from this lexcion at random and there is no way to determine what the actual lexica will be. Daily total hours: 4 Paid hours remaining: 5.5 ___________________________________________________________ 26 August 2003 Craziness all day... backed up all my files on penguin, printed vowel spreadsheet, looked for patterns in simulation data Daily total hours: 2 Paid hours remaining: 3.5 ____________________________________________________________ 28 August 2003 Updated the vowel chart, started compiling data for backup CD. Daily total hours: 2 Paid hours remaining: 1.5 |