Publication

Social, geographical, and lexical influence on Dutch dialect pronunciations

Ko, V., Wieling, M., Wit, E., Nerbonne, J. & Krijnen, W., 2014, In : CLIN Journal. 4, p. 29-38 10 p.

Research output: Contribution to journalArticleAcademicpeer-review

Copy link to clipboard

Documents

  • KoEtAl2014

    Final author's version, 219 KB, PDF document

Wieling et al. (2011) combined generalized additive modeling (GAM) with mixed-effects regression modeling to identify the influence of social, lexical, and geographical variables on the variation of Dutch dialect pronunciations. The conclusion of their study was that the pronunciation distance from standard Dutch became greater for locations with a smaller population, a higher average age (of the inhabitants), words with a greater frequency, and words with relatively many vowels. When Wieling et al. (2011) performed their quantitative study in 2011, they were not able to analyze the dataset in a single generalized additive mixed-effects regression model due to the large size of the dataset. Instead, they first used a generalized additive model to represent geography and
included the fitted values of this non-linear model as a predictor in a linear mixed-effects regression model. As more advanced methods to fit generalized additive mixed-effects regression models have become available, we improve on their approach here by constructing a single generalized additive (i.e. non-linear) mixed-effects regression model in which the non-linear geographical influence is varied depending on word frequency and word category (i.e. verbs vs. non-verbs). Non-verbs and higher frequency words generally showed a higher pronunciation distance from standard Dutch than lower frequency words and verbs. In contrast to Wieling et al. (2011), we did not find enough
support to include the number of inhabitants and the average income in a location in our model. However, we did find a comparable effect of the vowel-consonant ratio. Our findings highlight the potential of using generalized additive modeling to uncover significant non-linear patterns, while
simultaneously allowing for the inclusion of regular (linear) predictors and an extensive random effects structure.
Original languageEnglish
Pages (from-to)29-38
Number of pages10
JournalCLIN Journal
Volume4
Publication statusPublished - 2014

Download statistics

No data available

ID: 14344821