A Polylectal Linguistic Resource for Rromani

We describe the characteristics and problems of Rromani to develop linguistic resources and then examine existing applications using empirical and linguistic methods. Empirical methods can rapidly develop some resources, but they always contain several types of errors; the linguistic methods are more time-consuming, but more reliable. We then present a polylectal linguistic resource for Rromani that covers all basic dialectal variants, both at the lexical and morphological levels. This resource is composed of a lexicon, a grammar that describes inflectional and derivational morphology, and a grammar that describes agglutination. It can be used both for Rromani language studies and for developing Natural Language Processing (NLP) applications. We show that the same architecture can describe other low-resource languages.