The Role of Outliers in Linguistic Change in Progress
Bill Labov
University of Pennsylvania

One of the major problems in the mechanism of linguistic change is to explain how variables are steadily incremented in the same direction over many generations. Recent studies show that children acquire the social evaluation of stable community variables at 3-5 years of age, matching the adult means of style shifting. Probability matching to the mean cannot however lead to linguistic change. It is proposed that phonetic change is facilitated by the social evaluation of advanced outliers; as the frequency of such outliers rises, mean values shift in their direction.
Data to test this hypothesis are drawn from the recently completed acoustic measurement of vowel systems of 300 subjects of the Phonological Atlas of North America. The 85,000 measurements are normalized and classified by phonological environment, age of speaker, population size, and dialect area as determined by the recent national map of the Atlas. Among the active sound changes, the back upgliding vowels /uw, ow, aw/ show a uniform direction of fronting throughout North America. These changes permit an examination of various stages of parallel processes as reflected in values for 11 different dialects. While most variables approximate a normal distribution, there is usually some skewing towards the higher or lower tail of the distribution. The path of development for each variable is displayed by plotting mean second formant [F2] against skewness, where positive skewness reflects an excess of higher tokens, and negative skewness the reverse.
Early stages of change, as nuclei become more centralized, show high positive skewness. The large majority of the positive outliers are simple monosyllables which can accept heavy stress; negative outliers concentrate words with one or more following syllables, which shorten the nucleus and lead to an undershoot of the target. Such distributions reverse the normal expectation for vowels with peripheral targets. For dialects with more advanced changes, skewing falls in a linear fashion, with a -.81 correlation of mean F2 and skewness. This linear progression gives way to a curvilinear pattern in the most developed change, the fronting of free /uw/, where for the most advanced dialects, the South and Mid-Atlantic States, skewness regresses towards zero. Maximal regression coefficients for age are found in early and middle stages of change; as skewness reaches its maximum negative value, such age coefficients disappear.
The retardation and termination of these changes are not the result of structural or phonetic limits, but appears to represent the decline of social evaluation. The skewing progression is interpreted to support the hypothesis that in the early stages of change, social evaluation is focused on heavily stressed outliers which exceed the mean target; as change progresses, the focus shifts to the mean or expected value, while outliers are primarily nuclei with insufficient duration to reach the target.

< Back to List of Abstracts | Back to PLC23 Home Page >

About the PLC23 Committee
Previously held Penn Linguistics Colloquium: PLC22 (1998), PLC21 (1997)

Penn Department of Linguistics
University of Pennsylvania