Cognitive Robotics

 

Generating Vowel Systems in a population of agents, De Boer, 1995

Page history last edited by Nicholas Davis 1 yr ago

The language game: 

 

There is an initiator and an imitator selected from a group of non-uniform agents. Each agent has the potential to create sounds from three parameters that range from 0-1.

  1. Tongue protrusion: 1=back 0=front
  2. Tongue height: 1=top 0=bottom
  3. Lip rounding: 1= round 0= unrounded

 

The initiator starts by randomly selecting a sound and the imitator tries to match this sound, if none in the repertoire, then he picks randomly. The distance betwen the initiator and imitator determines sucess. Initiator gives non-verbal feedback to indicate sucess (most likely a light or something) and each update their phoneme list as a result.

 

Each phoneme produced gets a quality score of (count score)/(sucess score)

 

Low quality phonemes are deleted below a threshold value. Close phonemes less than the noise level are merged.

 

Experiments show that a population generally converges on a set of vowels, but random creation still allows for flexibility.

Comments (0)

You don't have permission to comment on this page.