Comparing Constraints on Contraction Using Bayesian Regression Modeling

Laurel MacKenzie

    Research output: Contribution to journalArticlepeer-review

    Abstract

    This paper has three goals: (1) to document the factors shaping is-contraction in Mainstream American English; (2) to assess the extent to which these factors also shape contraction of has; (3) to use shared patterns of contraction across the two verbs to draw conclusions about how the varying forms are represented grammatically. While is has two distinct phonological forms in variation, has has three. This necessitates regression modeling which can handle non-binary response variables; I use Bayesian Markov chain Monte Carlo modeling. Through this modeling, I (1) uncover a number of novel predictors shaping contraction of is, and (2) demonstrate that many of the patterns shown by is are also in evidence for has. I also (3) argue that modeling has-variation as the product of two stages of binary choices—a common treatment of three-way variation in variationist sociolinguistics—cannot adequately explain the quantitative patterns, which are only compatible with a grammatical model under which three distinct forms vary with each other. The findings have theoretical and methodological consequences for sociolinguistic work on ternary variables.

    Original languageEnglish (US)
    Article number58
    JournalFrontiers in Artificial Intelligence
    Volume3
    DOIs
    StatePublished - Aug 12 2020

    Keywords

    • Bayesian modeling
    • English
    • contraction
    • copula
    • linguistic variable
    • multinomial regression

    ASJC Scopus subject areas

    • Artificial Intelligence

    Fingerprint

    Dive into the research topics of 'Comparing Constraints on Contraction Using Bayesian Regression Modeling'. Together they form a unique fingerprint.

    Cite this