Proceedings of Methods XVI

A list of contributors

Part I: Innovation

Katharina Pabst, Lex Konnelly, Melanie Röthlisberger, and Sali A. Tagliamonte

Individual- vs. community-level variation
New evidence from variable (t,d) in Canadian English¹

1 Introduction

The simplification of consonant clusters involving the alveolar stops /t/ and /d/ in word-final clusters in English, also known as (t,d) deletion, is one of the most frequently studied variables in variationist sociolinguistics (Labov et al. 1968; Guy and Boyd 1990; Tagliamonte and Temple 2005, inter alia). This research has shown that uninflected or monomorphemic words, as in (1a), undergo deletion at a higher rate than regular past tense verbs, as in (1b), with semi-weak verbs, i.e., past tense forms that have both a stem-vowel change and a past tense suffix, patterning in between, as in (1c).

As Tamminga and Fruehwald (2013) note, this pattern has been explained from a variety of theoretical perspectives, ranging from variable rules (Labov et al. 1968) to lexical phonology (Guy 1991) and Optimality Theory (Kiparsky 1994). All of these models are based on the assumption that (t,d) deletion is a single, unified process, an assumption that Tamminga and Fruehwald question. Analyzing two corpora of American English, they found that semi-weak verbs have higher rates ←17 | 18→of inter-speaker and inter-lexical variation than monomorphemes and regular past tense verbs. Drawing on previous work suggesting that children do not attach the past tense suffix to semi-weak verbs (Guy and Boyd 1990; Roberts 1997; Smith et al. 2009), they argue that the surface variation between the different morphological forms is the result of three separate processes, two phonological (phonological deletion in monomorphemes and phonological deletion in all past tense forms), and one morphological (allomorphy in semi-weaks). Thus, while all monomorphemic, semi-weak, and regular past tense forms are subject to the more general phonological rule of consonant cluster reduction, the semi-weak verbs are also subject to allomorphy; for example, a semi-weak verb like kept also has a competing allomorph kep.

This paper replicates Tamminga and Fruehwald’s (2013) analysis using data from Toronto, Canada (Tagliamonte 2003–2006). Tamminga and Fruehwald (2013) hypothesized that if (t,d) deletion is one unified process, inter-speaker differences should show a consistent range across categories – that is, their rates of deletion should be relatively similar for each morphological class. They investigated this by fitting a mixed-effects model on the full dataset that included a random slope of individual speaker by morphological class. Results showed that the range of inter-speaker differences was not consistent across these categories for their data: Speakers were more tightly clustered for regular past tense and monomorphemes than for semi-weak verbs.

Testing Tamminga and Fruehwald’s (2013) hypotheses on a different data set allows us to address two issues: First, it allows us to test whether semi-weak verbs display higher rates of inter-speaker and inter-lexical variation in different varieties. If this were true, it would be further supportive to the hypothesis that there are indeed multiple distinct processes at work. Second, it allows us to further investigate “[o];ne of the enduring questions in linguistics” (Guy 1980:1): the relationship between individual- and community-level variation. Decades of research have shown that there is indeed such a thing as a “community grammar”, meaning that individuals who are part of the same community tend to display similar patterns of variation. However, relatively little is known about the extent to which individuals can and do deviate from community norms (for a more detailed discussion of these issues, see Walker & Meyerhoff 2003). Taking a closer look at inter-speaker variation will allow us to address this issue in the literature and augment our understanding of variation at the community level.

2 The data

The data we use to replicate Tamminga and Fruehwald’s (2013) findings stem from a subset of the Toronto English Archive, a corpus consisting of sociolinguistic ←18 | 19→interviews with speakers born and raised in Toronto, Canada. The subset consists of 56 speakers (28 men and 28 women) between the ages of 17 and 92. The speakers represent a variety of educational backgrounds, occupations, and ethnicities (for a more detailed description of the data set, see Tagliamonte 2012). The size and representation of social groups within the corpus make it the ideal tool with which to pursue the question of the role of the individual vs. the group, as it closely models the makeup of the broader speech community.

The data comprise 3,418 tokens: all lexical items with word-final consonant clusters ending in /t/ or /d/, excluding negative and interrogative constructions (e.g., isn’t and won’t), proper nouns, neutralization contexts, lexical items where the exact type of noun phrase could not be determined as well as the conjunction and (cf. Tagliamonte and Temple 2005). The data show that (t,d) deletion in the Toronto data has four variants: deletion (Ø), and three surface realization forms: [t];, [d], and the glottal stop [ʔ]. For the purposes of this study, all tokens that end in [t] or [d] will be treated as realized, while all tokens that end in Ø and [ʔ] will be treated as deleted (following Walker 2012).

The overall distribution of final consonant deletion and retention is presented in Tab. 1.

Tab. 1: Overall distribution of variable (t,d)

As the table shows, the rate of deletion in the Toronto data is 49.85%. This contrasts with the data from York English in the UK, which has a much lower deletion rate of 24% (Tagliamonte and Temple 2005:287).

3 Coding and analysis

Our coding schema follows existing precedent, focusing on three linguistic factors most commonly found to be significant in conditioning rates of (t,d) deletion: preceding phonological environment, following phonological environment, and the morphological class of the word (cf. Guy and Boyd 1990; Bayley 1994; Santa Ana 1996), summarized in the Tab. 2. Tokens were also coded for age, gender, occupation, and normalized corpus frequency; however, consistent with the analysis that we replicate here, we do not consider demographic factors in the present study.

Tab. 2: Summary of coding protocol