Abstract
This paper describes experiments in detecting and annotating code-switching in a large multilingual diachronic corpus of Swiss Alpine texts. The texts are in English, French, German, Italian, Romansh and Swiss German. Because of the multilingual authors (mountaineers, scientists) and the assumed multilingual readers, the texts contain numerous code-switching elements. When building and annotating the corpus, we faced issues of language identification on the sentence and sub-sentential level. We present our strategy for language identification and for the annotation of foreign language fragments within sentences. We report 78% precision on detecting a subset of code-switches with correct language labels and 92% unlabeled precision.
Volk, Martin