Language-specific encoding in endangered language corpora

The paper addresses problems of corpus building and retrieval resulting from codeswitching, which is a characteristic feature of endangered language recordings. The typical appearance of code-switching phenomena is first outlined on the basis of data collected in the DoBeS ‘ECLinG’ project, which dealt with three endangered Caucasian languages spoken in Georgia: Tsova-Tush (Batsbi), Udi, and Svan. The problem of language-specific retrieval is illustrated with examples showing the usage of the word da in Tsova-Tush contexts, which represents, as a homonym, either a native copula form (‘it is’) or the Georgian conjunction ‘and’. The subsequent section discusses the annotation requirements that are necessary to automatically distinguish the languages involved in code-switching, with a focus on the emerging ISO standard 639-6. It is argued that the fine-grained distinction of varieties and subvarieties and their interrelationship – as aimed at in this standard – requires a thorough reconsideration if it is to be applied in the markup of corpus data.