In the Spoken Syntax Lab meeting, Friday March 21 at 1PM in Cordura Hall 100, John Rickford will present “Relativizer Omission, the Independence of Linguistic and Social Constraints, and Variationist “Comparative Reconstruction”.
Abstract: A linguistic variable that has been the focus of many quantitative, variationist analyses of English over the past two decades (cf. Guy and Bailey 1995, Lehmann 2001) is the omission of the relativizer (that or WH-forms like what, who, or which] in restrictive relative clauses, as in “That’s the man Ø (who/ that/what) I saw.” In recent work, (Rickford 2011) I’ve examined the occurrence of this variable in the vernacular/creole varieties of Barbados, Guyana, Jamaica, African America, Appalachia and Southwest England (Dorset, cf. Piercy et al 2011), considering, among other things, the evidence it provides on the creole origins hypothesis of AAVE. In this paper, I extend the data set to a total of nine varieties, including relativizer data from “Northern English” varieties in Ireland, Scotland and Northern England kindly made available by Sali Tagliamonte (cf. Tagliamonte et al 2005).
A key change in my analysis was the use of logistic regression with R, rather than Goldvarb or Varbrul, as it offers several advantages, including mixed effects modeling, and better ability to detect interactions in the data. Additionally, I included social variables: class and/or gender. This allowed me to look for interactions between social factors and linguistic constraints and test Labov’s important generalization about the independence of linguistic and social constraints. Finally, I combined the data from all nine varieties in a big mixed effects regression analysis, controlling for differences by variety by entering them as factors in a “Language Variety” factor group.
The results were intriguing. To begin with, the values (“factors” in variable rule terminology) that turned out to be most significant for relativizer omission across all nine individual language varieties were those that matched Wasow et al’s (2011) predictability hypothesis, like Superlative NP antecedents and occurrence in existential, possessive, or cleft structures, all of which have general processing explanations that make them less useful for recovering historical relationships. Moreover, although gender and/or class turned out to have significant effects on the rate of relativizer omission in several cases, they did not show any interaction with the effect of linguistic constraints, confirming Labov’s more general (2010) hypothesis about the independence of linguistic and social constraints. Finally, in the big regression runs, there were very few significant interactions with language
variety, suggesting that the widely separated language varieties I compared were essentially behaving alike with respect to relativizer omission. This calls into question the viability of variationist “Comparative Reconstruction” (Poplack 2000) for detecting prior diachronic relationships, especially when, as in this case, the variation is governed by general sentence processing constraints.