Finding a boot to fit : acoustic measures of vowel quality in a real-time corpus of Glaswegian vernacular

Recent research has demonstrated fronting of high back vowels in several varieties of English related to a range of (socio)linguistic factors (e.g. Harrington 2007; Labov et al 2006; Maclagan et al 2009). The Scottish English vowel in the single lexical set BOOT, corresponding to English English GOOSE and FOOT, ha s been reported to be fronted, and/or central in the vowel space (/ʉ/), from auditory accounts since before the Second World War (e.g. McAllister 1938, Macaulay 19 77). Recent acoustic and articulatory analysis of Scottish English in the Eastern Central Belt suggests the vowel is fronted and lowered, and that the tongue position can be as front as that of /i/ and /e/ (Scobbie 2011). The intriguing question is whether there has also been any change over time in the acoustic realization of the Scottish English vowel. We have gathered recordings of Glaswegian vernacular over the past 40 years with which to try to answer this question. However previous research has shown that formant measures are sensitive to many factors, including recording equipment (e.g. Decker & Nycz 2011), microphone placement (e.g. Plichta 2004), measurement technique (e.g. McDougall & Nolan 2007). Determining acoustic change using a real-time corpus presents a number of methodological challenges. In this poster we compare acoustic measures using three different algorithms and two different techniques from 4 middle-aged speakers recorded at two different time points. We use standard algorithms implemented in Praat, EMU and SFS to take manual and automatic, and static and dynamic, measures of the first three formants of /i a ʉ/. All tokens of /ʉ/ were extracted, including both Standard Scottish English items and those belonging to the Scots variable set OUT (e.g. SSE out, Scots oot). Instances of /i/ and /a/ were counterbalanced with /ʉ/ for phonetic context and lexical item. Normalization by taking the log of Bark-transformed measures (Traunmueller 1997, cf. Harrington & Cassidy 1999), then allowed comparison of the speakers within and across time points. The corpus for this real-time project consists of existing sociolinguistic and oral history recordings from young, middle, and old, male and female, speakers recorded over four decades, from the 1970s to the 2000s. Here we concentrate on analysing data from middle-aged speakers recorded in the 1970s and in the 2000s. The recordings from 2000s were made for the Glasgow Media Project (Stuart-Smith 2006), and consist of spontaneous conversations from self-selected pairs of speakers. Those from the 1970s are from sociolinguistic interviews between fieldworker and informant (Macaulay 1977). The differences in the formant measures are discussed with respect to the different algorithms applied and the interpretation of the real-time change in the quality of /ʉ/. On this basis we make initial recommendations as to which procedures may be useful when comparing diverse, poor quality recordings in sociolinguistic research