The Perfect Approach to Adverbs: Applying Variation Theory to Competing Models

The question of adverbs and the meaning of the present perfect across varieties of English is central to sociolinguistic variationist methodologies that have approached the study of the present perfect (Winford, 1993; Tagliamonte, 1997; van Herk, 2008, 2010; Davydova, 2010; Tagliamonte, 2013). This dissertation attempts to disentangle the effect of adverbial support from the three canonical readings of the present perfect (Resultative, Experiential and Continuative). Canadian English, an understudied variety of English, is used to situate the results seen in the Early Modern English data. Early Modern English reflects the time period in which English has acquired the full modern use of the present perfect with the three readings. In order to address both these questions and current controversies over statistical models in sociolinguistics, different statistical models are used: both the traditional Goldvarb X (Sankoff, Tagliamonte and Smith, 2005) and the newer mixed-effects logistic regression (Johnson, 2009). What is missing from the previous literature in sociolinguistics that advocates logistic mixed-effects models, and provided in this dissertation, is a clear statement of where they are inappropriate to use and their limitations. The rate of adverbial marking of the present perfect in Canadian English falls between rates reported for US and British English in previous studies. The data show in both time periods that while adverbs are highly favored in continuative contexts, they are strongly disfavored in experiential and resultative contexts. In Early Modern English, adverbial support functions statistically differently for resultatives and experientials, but that difference collapses in the Canadian English sample. Both this and the other linguistic contexts support a different analysis for each set of data with respect to adverbial independence from the meaning of the present perfect form. Finally, when the focus of the analysis is on linguistic rather than social factors, both the traditional and newer models provide similar results. Where there are differences, however, these can be accounted for by the number of tokens and different estimation techniques for each model