Measuring accent recognition and feature salience in real time: Which urban Northern Englishes are (still) distinct for listeners?

References –

Cardoso, Amanda, Erez Levon, Devyani Sharma, Dominic Watt & Yang Ye. 2019. Inter-Speaker Variation And The Evaluation Of British English Accents In Employment Contexts. In S. Calhoun, P. Escudero, M. Tabain & P. Warren (eds.), Proceedings of the 19th International Congress of Phonetic Sciences, 1615–1619. Melbourne, VIC.
Leemann, Adrian, Marie-José Kolly & David Britain. 2018. The English Dialects App: The creation of a crowdsourced dialect corpus. Ampersand 5. 1–17. https://doi.org/10.1016/j.amper.2017.11.001.
Montgomery, Chris & Emma Moore. 2018. Evaluating S(c)illy Voices: The effects of salience, stereotypes, and co-present language variables on real-time reactions to regional speech. Language 94(3). 629–661. https://doi.org/10.1353/lan.2018.0038.
Strycharczuk, Patrycja, Manuel López-Ibáñez, Georgina Brown & Adrian Leemann. 2020. General Northern English. Exploring Regional Variation in the North of England With Machine Learning. Frontiers in Artificial Intelligence 3. 48. https://doi.org/10.3389/frai.2020.00048.

 Abstract –

In the North of England, regional dialect levelling has resulted in what Cardoso et al. (2019) and Strycharczuk et al. (2020) have termed ‘General Northern English’ (GNE). Strycharczuk et al. (2020) find that random forest models struggle to discriminate between Leeds, Manchester, and Sheffield accents, and that the cues they do use are less well studied in sociolinguistic work (such as the KIT and NEAR lexical sets). This paper examines how listeners distinguish between five contemporary varieties of Northern English by addressing two questions:

How well do listeners recognise specific Northern accents?
What features do listeners attend to as they attempt to classify the accents?

100 respondents (50 from the South East of England, and 50 from Yorkshire and the Humber) were played samples representing Newcastle, Leeds, Manchester, Liverpool and Sheffield. Samples were young female speakers chosen from the English Dialects App corpus (Leemann, Kolly & Britain 2018), matching Strycharczuk et al.'s (2020) computational approach to the recognisability of Northern English accents.

Respondents were asked to listen to each sample and identify where they thought the speaker was from by choosing from a list of the five locations. As they listened, they were asked to indicate by clicking a button whenever they heard something that helped them to identify where the speaker was from, resulting in a set of time-aligned click data. Listeners then reviewed these click data using fragments of the transcript at the time of each click as well as a brief snippet of audio leading up to their click. They then annotated their clicks indicating what they had attended to (cf. Montgomery & Moore 2018).

We found that listeners were poor at identifying accents from Manchester, Leeds, and Sheffield (~33% accuracy for each location), but much more accurate at identifying Newcastle and Liverpool speakers (75% and 87% accuracy, respectively). In terms of feature attention, the BATH vowel was frequently noted for all speakers, but attention to other features differed according to speaker. For Liverpool, consonants such as /k/ and /t/ were noted. For Newcastle, the CURE vowel, along with /t/ and (ING) were attended to. For Sheffield, Leeds, and Manchester, noted features were more similar (the STRUT vowel, as well as (ING) and happY for at least two of the speakers), with some speaker-specific vowels noted in certain cases (e.g. the FACE vowel for the Sheffield speaker).

Our data demonstrate that when attempting to classify speakers, listeners attend to features well studied by sociolinguists. Listeners did not attend to the less-studied features used by random forest models in Strycharczuk et al. (2020). We also find that consonants are important for accent recognition. Our data show that listeners perceive some accent differences between speakers even when they cannot correctly identify where they are from. However, the low recognition rates for Manchester, Sheffield, and Leeds suggest that these varieties are not easily discriminable. We argue that this provides further evidence for a ‘General Northern English’ used by young female speakers in parts of the North of England.