Reflections on documentary corpora

Abstract

For decades, language documentation proponents have argued for the separability of LD as its own sub-discipline. Many corpus linguists have made this same claim; thus, corpus linguistics shares the ethos of data over theorizing, whereby primary data represent authentic, connected discourse that is natural (not elicited), broadly sampled (across speakers, generations, dialects), and balanced (reflecting different usage contexts and genres). Nevertheless, many misconceptions remain about what a language corpus is, how it is formatted, how big or balanced it needs to be, and most importantly, how it is queried. In this reflection, I dispel some of these misconceptions, while reassuring community members and field linguists alike that a corpus is an exceedingly powerful tool for guiding the expansion of the documentary record, keeping precious language data in circulation, and helping to produce the classic descriptive by-products of LD such as dictionaries, phrasebooks, and grammars. Above all, the less-familiar but more direct by-products of corpus interrogation, such as word lists, frequency counts, concordance lines, N-grams, collocations, distribution, and dispersion plots, are so immediately interpretable and useful by speakers, learners, and linguists, that LD should give corpus linguistic training the same attention as project planning, ethics, recording, transcription, annotation, metadata, and archiving.