A corpus linguistics perspective on language documentation, data, and challenge of small corpora

Abstract

This paper deals with issues of corpus design that might prove problematic for the study of under-resourced languages, e.g. in language documentation. It argues that it is not yet well understood which linguistic and extra-linguistic (predictor) variables cause linguistic variation (i.e. the response variable), which means that the scope of a linguistic finding cannot always be assessed. In order to deal with this problem, it is argued that we need a flexible corpus architecture with the option of adding meta-data to corpora/sub-corpora at any point in time.