Abstract

The present article takes a simple observation as a starting point for its subsequent reflection on the epistemological and methodological issues surrounding the resort to small corpora: the latter have survived when the rise of corpus linguistics and the development of big data have multiplied resources. To us, this survival is proof of the material and logistical necessity as well as the scientific relevance of small corpora, which we explore from the particular angle of spoken language and within the specific domain of phonology. We explain why this field has a unique connection to its linguistic object and consequently a privileged relation to corpora and rethink the viability of corpora not in terms of size but in terms of purpose. We then defend the methodology and ambitions put forward by the PAC program whose database on contemporary oral English exclusively relies on small specialized corpora built in the various English-speaking areas around the world. Not only do we demonstrate how relevant these small corpora are individually for the quantitative and qualitative study of specific phenomena, such as levelling in the Mancunian variety of northern English for instance, but also show how they collectively allow for the comparative and cumulative analysis as well as the theoretical modelling of such phenomena as rhoticity and r-sandhi in the different varieties of English spoken worldwide. Therefore, we illustrate how the results yielded by the in-depth study of the small PAC corpora help us take into account the system dynamics of the varieties of English and the sociolinguistic weight of local identities, especially along the north-south axis, in the description and understanding of their phonetic and phonological characteristics. In addition, by supporting small corpora, we implicitly defend corpus phonology and question the automatic superiority that large corpora enjoy as far as the in-depth analysis of spoken language and of the evolution of the varieties of English spoken worldwide is concerned.