How do speech patterns spread through a community? Or, “Oh no, not another linguistics model”
Purpose of Models Suppose you have two populations: native English speakers and native German speakers Speakers vary in how they produce certain sounds If all speakers are part of same community what factors determine if a particular speech style will spread to everyone? –What style would it be?
Two related models Spatial model - agents wander around Network model - agents are connected in a network When people interact with each other will depend on the model How they interact will be the same in both models..
Rules for interaction - based on exemplar model Idea of the sound of a word isn’t one ideal pronunciation Instead, a word (or category) is represented as a list of possible instances of that word (exemplars) –[word 1 word 2 word 3 ]
Exemplar model of production and perception Produce a word - select one of the exemplars of that category –[word 1 word 2 word 3 ] Perceive a word - try to match that exemplar to existing exemplars in your categories, then add it to the list –categoryA [word 1 word 2 word 3 ] –categoryB [word 2 word 2 word 2 ]
Speech production in this model People know 6 word types –Short words that end in p,b,t,d,k, or g –Ex. “bat”, “mop” Final consonant can be pronounced 3 different ways –0 - unreleased –1 - voiced & released –2 - voiceless & released
Each of the 6 categories is made up of 10 exemplars – /__p/ [ ] To speak - choose random exemplar from the category –From /__p/ choose “0” Speech production in this model
Speech perception in this model To listen - try to match spoken exemplar to its corresponding category & its voiced or voiceless counterpart –Category pairs: p/b, t/d, k/g –/__p/ [ ] –/__b/ [ ] –Chance of assigning to a category based on square of # of matches “0” has 3 matches for p, 6 matches for b Four times as likely to be assigned to b Spoken exemplar of “1” would always be assigned to p
Speech perception in this model To assign a spoken exemplar to a category –Kick out a random exemplar –Replace with the spoken exemplar –/__b/ [ ] –/__b/ [ ] If there were no matches, spoken exemplar does not get assigned to any category
Summary Agents have 6 categories made up of 10 exemplars Categories are initialized from an input text file, with data for 8 agents –But 8 isn’t very many agents, so you have the option of creating extras –Best way to see is to take a look at the model!