Abstract
In this paper we describe the participation of the Natural Language Engineering Lab (NLEL) - Universitat Polit`ecnica de Valencia and Autoritas Consulting team in the Discrimination between Similar Languages (DSL) 2015 shared task. We have participated both in open and close submissions. Our system for the open submission performs in two steps. Firstly, we apply a language detector to identify the distinct groups corresponding to families of languages/dialects, and then we distinguish between varieties with a probabilistic method. For the close submission, we implemented our probabilistic method in a multi-class classifier for all the language varieties together. Although our results on the development set were quite promising (93.07% and 86.08% respectively), a software bug (that we have detected only after the submission) dropped considerably our results in the final testing.
