Exaptation of Dialect Diversity for Social Marking

1
Exaptation of Dialect Diversity for Social Marking

• Daniel Livingstone
• School of Computing
• University of Paisley
• daniel.livingstone_at_paisley.ac.uk

2
Dialect Diversity

• Why?
• Due to separation (spatial/social) of language
  learners and imperfect learning of language
• neutral evolution
• Due to application of dialect diversity for
  social marking

3
Social Marking Not Required

• Our previous work has argued that social marking
  is not required
• Contra Milroy (92), Nettle (99)
• Supported by range of other work on effect of
  spatial separation on signal diversity
• From birdsong to grammar
• Although not required, what effect will social
  marking have if included?
• Provides benefits to speakers, so through
  exaptation it could take on new role
• What effect might this have on further evolution
  of dialect diversity?

4
Socially Marked Dialect Differences
5
Henry Higgins states his position

• Where there is free and uniform intercourse
  between all members of a community the language
  will be uniform in the sense of not splitting
  into dialects.
• When the community is too large to permit of
  uniform communication dialects begin
• Henry Sweet, The History of English Sounds, 1888,
  Clarendon Press, Oxford, para. 189 (page 52)
• Sweet then proposes a thought experiment

6
Sweets Model

• suppose a large plain covered with villages of
  equal size and independence at equal distances,
  each village communicating directly only with its
  immediate neighbours, there will in a few
  generations be a distinctly different dialect in
  each village, and in course of time the dialects
  of the most northern, southern, eastern and
  western villages will become mutually
  unintelligible to one another and to that of the
  central village. But there will be no lines of
  division the dialects will shade insensibly into
  one another
• 2 dimensional plain with equidistant groups
• Intra-group contact only with immediate neighbours

7
Nettles Model

• Grid of Villages (7 by 7)
• 20 individuals in each village
• Individuals age in 5 steps
• Newborn agents learn from all others in same
  village
• Migration rate gives chance of an individual
  swapping with an individual in a neighbouring
  village after learning

8
Nettles Results

• Without social factors influencing language
  learning, even very low rates of contact prevent
  any significant dialect diversity from occurring
• BUT problematic phonological model
• Limiting learning to examples from high status
  elders enables the emergence of dialects
• Where high status elders exist, learners learn
  ONLY from such elders

9
de Boers Model of Emergent Phonology

• Well received model of phonological learning of
  vowel systems
• Detailed in previous Evolution of Language
  conferences
• Results have good match to observed results in
  human vowel systems
• Learning based on interactions between
  individuals no explicit averaging
• Two distinct versions of same vowel may be
  interpreted by a learner as two different vowels.
  Nettles learners will always learn the average
  vowel, even though it may not actually exist

10
(No Transcript)
11
Dialect in Emergent Vowel Systems

• Using a linear arrangement of learners
• Learning limited to neighbourhoods around each
  learner
• No hard boundaries
• Distinct dialects emerged
• Arbitrary sub-groups at extreme ends of line
  learned very different vowel systems
• Adjacent sub-groups learned similar vowel systems

12
Simple Linear Arrangement
13
Emergent Phonology in a Spatial Village Grid Model

• Use Nettles village grid model
• Same in principle as Sweets
• Nettles migration ageing rules
• Age 200 lifecycles
• de Boers emergent phonology model
• Initialise each group with same vowel system

14
Measuring Differences In Emergent Vowel Systems

• Problems
• How to compare?
• Clustering How many clusters? Number of vowels
  is not fixed
• Even where no. of vowels is the same which
  cluster to compare with which?
• Simple Solution Bin results
• Count bin differences
• Mean Square Error of bin differences

15
Bin Difference MSE

• Split vowel space into a number of areas (bins)
• Bin Difference
• For each vowel system, note whether each bin is
  occupied by at least one articulated vowel
• To compare two dialects, score 1 point for each
  bin occupied in one dialect but not in the other
• Higher scores imply greater differences
• MSE
• Square the difference in number of vowels
  occupying corresponding bins in two systems being
  compared
• NOT always accurate measures
• But over many comparisons, indicative of
  underlying differences

16
No Migration

• Baseline testNo migration
• How different might we expect dialects to be if
  the evolve with no contact at all?
• 100 runs with two villages
• No migration, 200 ageing cycles
• Mean Bin Difference 7.14
• Mean MSE 704.25

17
Comparing Dialects

• Taking lead diagonal from each of 100 sets of
  results
• Find Bin Difference/MSE values for pairs of
  villages diagonally adjacent, two apart, , six
  apart
• Repeat for migration rates of 10, 25, 50 and
  75
• Actual chance of an agent migrating is set to
  half migration rate as each migration moves 2
  agents
• Are vowel systems in neighbouring villages more
  similar than systems in distant ones?

18
Continuum of Dialects
8
7
6
5
Bin Difference
4
3
1
2
3
4
5
6
Distance (Diagonal)
19
Simple Social Marking

• After Nettle
• Introduce status as an influence on learning
  process
• In Nettle, learners ONLY learn from high status
  individuals. Here, learners learn from all
  around, but high status individuals provide more
  examples
• Simple socially marked learning
• Only natives of a group can hold high status

20
Increased Learning From High Status Elders
21
Non-Transferable Status
22
Comparison
23
Extrapolation

• The effect of social marking on dialect diversity
• Preference to learn from exemplars within own
  group will
• Strengthen inter-group linguistic influence
• Weaken intra-group influences
• Increase dialect diversity
• Impede formation of dialect chains ?

24
To do

• Wider exploration of parameter space
• Alternative models of influence, status and
  social selection
• Test robustness of results against changes to
  social structures
• Better measures of dialect difference
• In progress
• What if decreased similarity in dialect leads to
  decreased contact?