Computational drama analysis from almost zero electronic text: The case of Alsatian theater

International audienceAt the MeThAl project, we are creating the first large TEI corpus of Alsatian theater; Alsatian refers to Germanic varieties spoken in Alsace (Eastern France). The corpus, covering mainly the 1870-1940 period, will have above 500,000 tokens (50 plays) for which no previous electronic text existed. We present our automatic TEI encoding workflow assisted by a Conditional Random Fields model based on OCR sources, followed by manual correction. As the corpus shows large orthographic variation (there is no standard spelling) and NLP resources for Alsatian are scarce, several text analyses are challenging; we discuss our approach to tackle this. We developed detailed character metadata using TEI feature structures, encoding characters' social variables like their socioprofessional group and social class. This provided an overview of the evolution of social groups in the plays across time, complementing earlier, smaller-sample studies of the tradition. Besides metadata-based analyses, with a view to performing emotion detection, we present an emotion lexicon which handles orthographic variation in the corpus, and how it can be used to train automatic variant detection. We also outline how we are publicly sharing the resources, on open data repositories, on DraCor, and via a corpus navigation interface. An overall goal is starting off computational literary research (CLS) on Alsatian theater and helping compare Alsatian theater with other traditions for which rich CLS results already exist, including the two dominant ones which influenced Alsatian theater most: German and French