What can a corpus tell us about phonetic and phonological variation?

Abstract

Studies in phonetic and phonological variation oriented in CL often encounter significant challenges right from the outset. The advantages of utilising the large datasets which CL promises are frequently offset by the often tedious processes involved in readying such large datasets for effective phonetic and/or phonological analysis. Lieberman (2019) predicts that the trend towards increased use of corpora in phonetic research is set to continue to grow, whilst Durand (2017) claims that as long as a range of criteria are met when spoken corpora are constructed and developed, their utility in accounting for both intra- and inter-speaker variation will be inestimable. The growing subfields of corpus phonetics and corpus phonology also hold the key to ensuring that researchers working on different languages – and particularly under-resourced ones – increasingly get access to data which do more than simply tick the “spoken component” box. The complexity of the decisions needing to be taken is much more onerous in the case of speech given the much greater amount of pre-processing work necessary to ensure usability of the data (Ernestus and Warner 2011; Cole and Hasegawa-Johnson 2012). Additionally, it is easy to forget that the digital age favours well-resourced languages such as English, French, German or Spanish, while less resourced languages are often still faced with having to invest heavily in manual processing of data, which inevitably results in a limited digital presence and a bit of a vicious cycle (Besacier et al. 2014; Mena et al. 2020). [excerpt]