The reluctant oracle: Using strategic annotations to add value to, and extract value from, a signed language corpus

In this paper, I discuss the ways in which multimedia annotation software is being used to transform an archive of Auslan recordings into a true machine-readable language corpus. After the basic structure of the annotation files in the Auslan corpus is described and the exercise differentiated from transcription, the glossing and annotation conventions are explained. Following this, I exemplify the searching and pattern-matching at different levels of linguistic organisation that these annotations make possible. The paper shows how, in the creation of signed language corpora, it is important to be clear about the difference between transcription and annotation. Without an awareness of this distinction - and despite time consuming and expensive processing of the video recordings - we may not be able to discern the types of patterns in our corpora that we hope to. The conventions are designed to ensure that the annotations really do enable researchers to identify regularities at different levels of linguistic organisation in the corpus and, thus, to test, or build on, existing descriptions of the language.