Challenging listening conditions make understanding Glaswegian /r/ even harder
Non-ideal listening conditions come in many forms. One challenge to speech perception may exist when words are acoustically very similar, possibly resulting in the wrong message being conveyed to the listener. Another potential difficulty is when the listener is uncertain about the identity or the accent of the talker.
Both of these challenging listening conditions can be examined using Glaswegian speech as a case study. Previous research has found that Glaswegian can be hard to understand for unfamiliar listeners (Smith et al. 2014; Adank et al. 2009), but this paper focuses on the difficulty that may arise for native Glaswegian listeners when asked to identify which word from a pair, e.g. 'hut/hurt', the speaker produced. Due to extreme /r/ weakening over time (Lawson et al. 2017), many working class (WC) Glaswegians produce these words so that they are acoustically (Anon 2015) and perceptually (2016) almost identical. Middle class (MC) speakers, in contrast, produce e.g. 'hurt' with a very strong /r/ (Anon 2012, 2015; Lawson et al. 2011) so their 'hut-hurt' distinction is not difficult to perceive – however, if the listener is unaware that they are hearing a MC speaker, MC 'hut' could still be confusable with WC 'hurt'. To explore the perceptual consequences of uncertainty about speaker accent, this experiment manipulated whether words like 'hut' and 'hurt' were presented in single-talker or mixed-talker blocks.
Glaswegian listeners heard single words from 12 pairs presented in three blocks: Single MC, Single WC, and Mixed (MC and WC stimuli were randomised together). Their task was to use the mouse to click the word they heard, out of two onscreen options per trial – e.g. 'hut/hurt', 'bud/bird'. MouseTracker (Freeman&Ambady 2010) recorded cursor trajectories as participants moved towards their chosen word, in the top-left or top-right of the screen.
Trajectories were analysed using a suite of measures (RT; Area-Under-the-Curve (ibid); Discrete-Cosine-Transformation (Watson&Harrington 1999)), finding significant processing costs when listeners tried to distinguish WC 'hut-hurt'. Furthermore, listeners experienced more perceptual difficulty when word pairs were heard in the Mixed block, meaning that when words from two accents and speakers were randomised together the intended message was much harder to decode. Interestingly, even the ‘easier’ MC pairs were more difficult in the Mixed block. These results are discussed in terms of some general principles which underlie exemplar theories and Bayesian inference, concerning how listeners resolve simultaneous uncertainty about the linguistic unit being produced and the speaker.