Measures%20of%20Variation%20Among%20English%20and%20American%20Dialects

1
Measures of Variation Among English and American
Dialects

• Robert Shackleton
• U.S. Congressional Budget Office

2
Goals

• Compare speech variants used by English and
  American speakers, using easily accessible data
• Use several different quantitative methods to
  assess variation among speakers
• Compare different quantitative methods
• Use results to gain some insight into English
  origins of American speech variants

3
Data

• Nearly all data from Kurath McDavids
  Pronunciation of English in the Atlantic States
  some from Kuraths Dialect Structure of Southern
  England
• All or nearly all data collected by Guy Lowman
• 82 phonemes classified into 285 variants by
  Kurath and McDavid

4
Data

• Four regions
• Southern England (59 informants) settled lt700
• Southeastern Massachusetts (22 informants)
  settled lt1650
• S.E. Virginia / N.E. North Carolina (31
  informants) settled lt1690
• S.W. Virginia / S. West Virginia (19 informants)
  settled 1750-1800
• Informants largely older, rural, long-settled
  families
• In some cases, more than one variant per
  informant
• Some missing data
• Some data arbitrarily attributed to one of two or
  three possible informants in a given locality

5
Methods

• Shared variants based on proportion of variants
  shared between two speakers
• Genetic distance based on relative frequencies
  of variants, treating variants of a given phoneme
  as analogous with allelles of a given gene
• Linguistic distance measured as a Euclidean
  distance between variants in an idealized
  geometric grid (e.g. ² and e are closer to each
  other than i and Þ)
• Each measure involves arbitrary assumptions
• Choice of phonemes to include
• Classification of responses into variants
• Quantification of distances among variants
• Important difference first two approaches assume
  that variants are discrete linguistic approach
  does not

6
Genetic Approach

• Nei's genetic distance D measures how closely
  related populations of pronunciation patterns are
  if
• Change is always to a completely new variant
• All phonemes have the same rate of change
• Population sizes remain constant over time
• Occurrence of variant 1 absence 0
• Occasionally, frequency of variant in a set of
  similar words (0 lt x lt 1)
• In some cases, more than one variant per speaker
• Each informant represented by a vector of 285
  numbers, each between 0 and 1
• In this sample
• D ranges from 0.00 to 1.70
• 50 shared pronunciations gt D 0.7

7
Linguistic Approach

• Variants are characterized by a set of numbers
  representing degrees of height, backing,
  rounding, rhoticity, length

8
Linguistic Approach

• Difference between variants measured as Euclidean
  distance
• Distance between two speakers LD measured as the
  average Euclidean distance between their variants
• Could also measure the dispersion of distances,
  etc.
• In this sample
• LD ranges from 0.00 to 1.68
• 50 shared variants gt LD 0.70 to 1.16

9
Cluster Analysis

• Methods of grouping informants on the basis of
  similarity of their speech patterns
• Many different approaches
• Different measures of similarityPearson
  correlations, Euclidean distances, cosines,
  genetic or linguistic distances
• Different methods of grouping similar
  observations into clusterssingle, average, and
  complete linkages, various algorithms for
  estimating phylogenetic relationships
• Results highly dependent on approach
• English speakers tend to group into five regions
  (East Midlands, East Anglia, Southeast,
  Southwest, Devonshire)
• North American regions tend to be distinct, and
  to cluster most closely with to Southeast England
• EVNC and SWVA consistently cluster together

10
Results

• Distance measures are generally correlated
• Neis distance and shared variants are very
  similar, despite nonlinearity
• Linguistic distance is least similarcontains
  different information about similarity of speech
  forms

11
Shared Variants
East Midlands
East Anglia
Southeast
Southwest
Devonshire
Massachusetts
EVNC
SWVA
12
Neis Genetic Distance
East Midlands
East Anglia
Southeast
Southwest
Devonshire
Massachusetts
EVNC
SWVA
13
Linguistic Distance
East Midlands
East Anglia
Southeast
Southwest
Devonshire
Massachusetts
EVNC
SWVA
14
Distribution of Variants

• Some variants are widespread others not
• 12 appear in all 8 regions
• 29 appear in 7 regions
• 42 appear in 6 regions
• 59 appear in 5 regions
• Even within regions, lots of variation
• Informants in a given region typically share 60
  to 75 of variants, but range is 33 to 90
• Degree of variation reflected in genetic and
  linguistic distance measures
• Southern England
• More diversity than in North America
• 91 of variants found somewhere
• 23 found in every region
• 20 found only in southern England
• Shared variants between English informants 22 to
  83
• Shared variants between English and American
  informants 18 to 63

15
Distribution of Variants

• North American regions
• Less diversity than in England22 of southern
  English variants absent
• 80 of variants found somewhere
• 37 found in every region
• 9 found only in North America (12 of North
  American variants)
• Nearly half of American innovations shared
  across all N. American regions
• Many innovations are known to have existed in
  southern England, but were not recorded
• North American distribution of southern English
  variants
• Slightly greater frequency of eastern (esp.
  southeastern) English variants in American
  regions
• Of 41 variants found in eastern but not in
  western England, 14 (34) appear in Massachusetts
  and in the South, 7 (17) in Massachusetts but
  not in the South, 13 (32) in the South but not
  in Massachusetts
• Of 33 variants found in western but not in
  eastern England, 5 (15) appear in Massachusetts
  and in the South, 2 (6) in Massachusetts but not
  in the South, 11 (33) in the South but not in
  Massachusetts

16
Distribution of Variants

• Massachusetts
• Both more and fewer shared variants with English
  informants than the South
• On average, more shared variants with the South
  than with English informants
• By all measures, MA informants show somewhat
  greater affinity with eastern English
• The South
• EVNC and SWVA comparatively homogeneous and
  similar
• Similar intra- and interregional variation
• Similar variation with MA and England
• Slightly greater affinity with western English
  than MA
• Southern American informants have greatest number
  of shared variants with Devonshire informants,
  but lowest linguistic distance with southeastern
  English informants
• Can illustrate differences using average values,
  or values for typical informants who have the
  greatest average number of shared variants or
  lowest average distance with all other speakers
  in region

17
Regional Comparison Averages
18
Regional Comparison Typical Informants
19
Conclusions

• Different measures yield somewhat different,
  complementary insights into linguistic variation
• By all measures, extensive variation in and among
  regions
• Patterns of variationincreasing in population
  and age of settlementare reminiscent of
  species-area relationship
• American settlement resulted in lower variation
  in American regions, leveling, and somewhat
  different populations of variants in different
  regions
• Slightly dominant influence from the metropolitan
  area
• Greater eastern influence in the north, western
  influence in the south
• Relatively little innovation
• Leveling process analagous to loss of species
  during reduction in habitat
• Results are largely consistent with the
  historical record of early English immigration to
  North America (except for absence of East Anglian
  influence in Massachusetts)