Capturing Lexical Variance with a Mixed Model : Verb in Ordering Variation
Hye-Won Choi · 2010
Based on the dative data collected from Sejong Corpus (Kim 2000), Choi (2010) has analyzed the word order variation and proposed a statistical model of logistic regression that identifies all predictive variables that influence word order and predicts which order (between 'dative before accusative' (DA) and 'accusative before dative' (AD)) is more likely to occur, given the predictor variables. Building up on the Choi's (2010) analysis, this paper investigates whether a lexical variance caused by word is also affecting the choice of dative word order. In particular, variation caused by the verbs of dative construction is examined in depth, and verb is indeed proved to be a significant random variable. To capture the new random effect by verbs, as well as the fixed effects identified in Choi (2010), this paper proposes a new mixed model that considers fixed and random effects together (Bresnan et al. 2007, 2008, 2009, Baayen 2008, Johnson 2008). The mixed model turns out to be a more powerful model that secures higher accuracy in its predictions.