Listening to speakers: Crowdsourcing the language resources for non-standardized languages

Citizen science, in particular voluntary crowdsourcing, is still little experimented solution to produce language resources for less-resourced languages with enough connected speakers. We present here experiments we led on part-of-speech annotation for non standardized languages, namely Alsatian and Guadeloupean Creole. We detail the methodology we used and show that it is adaptable to other languages, then we present the results we obtained. An analysis of the limits of this platform led us to develop a new one, that allows the creation of raw corpora and part-of-speech annotations, and the construction of a multivariant lexicon. The created platforms, language resources and tagging models are all freely available.