SLIDE 1
Jon W. Carr
Centre for Language Evolution School of Philosophy, Psychology and Language Sciences University of Edinburgh
Iterated learning optimizes for simplicity
Language
Pressures shaping language
Iterated learning optimizes for simplicity Jon W. Carr Centre for - - PowerPoint PPT Presentation
Iterated learning optimizes for simplicity Jon W. Carr Centre for - - PowerPoint PPT Presentation
Iterated learning optimizes for simplicity Jon W. Carr Centre for Language Evolution School of Philosophy, Psychology and Language Sciences University of Edinburgh Pressures shaping language Language Pressures shaping language e l p m i
SLIDE 2
SLIDE 3
Language
Learning
Pressures shaping language
S i m p l e
SLIDE 4
Language
Communication Learning
Pressures shaping language
S i m p l e I n f
- r
m a t i v e
SLIDE 5 English Northern Paiute
50 100 1 2 3 4
⬅ Informative ⬅ Simple
Kinship terms are simple and informative
Kemp & Regier (2012)
SLIDE 6
Lab experiments
Learning-only Learning + Communication
SLIDE 7
50 100 1 2 3 4
Learning and communication pressures
⬅ Informative ⬅ Simple
Learning + communication Communication-only Learning-only
newhomo kamone gaku hokako kapa gakho wuwele nepi pihino nemone piga kawake gamenewawu gamenewawa gamenewuwu gamene mega megawawa megawuwu wulagi egewawu egewawa egewuwu ege tuge tuge tuge tuge tuge tuge tuge tuge tuge tupim tupim tupim miniku miniku miniku tupin tupin tupin poi poi poi poi poi poi poi poi poiKirby, Cornish, & Smith (2008) Kirby, Tamariz, Cornish, & Smith (2015) Kirby, Tamariz, Cornish, & Smith (2015)
SLIDE 8
Iterated learning gives rise to informative spatial terms
Carstensen, Xu, Smith, Regier (2015)
SLIDE 9
Experiment 1
SLIDE 10
Stimuli
Angle Size
SLIDE 11
Stimuli
Angle Size
SLIDE 12
Stimuli
Angle Size
SLIDE 13
Stimuli
Angle Size
SLIDE 14
Conditions
Angle-only Size-only Angle & Size
Easy to learn but low informativeness Informative but hard to learn
SLIDE 15
Training phase
SLIDE 16
Test phase
SLIDE 17
Results
Angle-only
SLIDE 18
Results
Size-only
SLIDE 19
Results
Angle & Size
SLIDE 20
Result: Learnability advantage for the less informative systems
SLIDE 21
Experiment 2
SLIDE 22
Iterated learning
SLIDE 23
SLIDE 24
SLIDE 25
1 category 2 categories 4 categories 3 categories
SLIDE 26
Measuring simplicity
SLIDE 27
Measuring simplicity
3×3 1×1 1×8 1×2 1×4 1×6 1×1 3×4 1×1 2×6 2×3 1×2 4×8 4×8
SLIDE 28
Simplicity
SLIDE 29
Simplicity Informativeness
SLIDE 30
Simplicity Informativeness
SLIDE 31
Simplicity Informativeness
SLIDE 32
Compactness A system of compact categories is more informative than a system of noncompact categories Dimensionality A system that uses many dimensions is less (?) informative than a system that uses few dimensions
What makes language informative?
Balanced categories A system of equally sized categories is more informative than a system of unequally sized categories Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories Discreteness A system of discrete categories is more informative than a system of fuzzy categories
SLIDE 33
What makes language informative?
Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories
Learning
Pressure for simplicityCommunication
Pressure for informativeness SLIDE 34
What makes language informative?
Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories
Learning
Pressure for simplicityCommunication
Pressure for informativeness SLIDE 35
What makes language informative?
Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories
Learning
Pressure for simplicityCommunication
Pressure for informativeness SLIDE 36
What makes language informative?
Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories
Learning
Pressure for simplicityCommunication
Pressure for informativenessOur experiments
SLIDE 37
What makes language informative?
Expressivity A system of many categories is more informative than a system of few categories Convexity A system of convex categories is more informative than a system of nonconvex categories
Learning
Pressure for simplicityCommunication
Pressure for informativenessCarstensen et al.
SLIDE 38
Conclusions
The pressure from learning yields simple languages, but this can manifest itself through multiple semantic properties: Expressivity Loss of words/categories to aid learning Convexity Reorganization of the meaning space to aid learning The pressure from communication has the same effect on convexity as learning does, making it difficult to identify the causal mechanism. In iterated learning, the pressure for simplicity favours semantic category systems that are convex – informativeness comes along for the ride.
SLIDE 39