Trying to make the best out of the least: experiments on optimizing supervision in machine learning approaches for Spanish NLP
Abstract: I will present experiments and lines of research we are pursuing in the NLP group at the University of Córdoba, Argentina. Our work is mostly focused on empirical methods for under-resourced NLP problems, specially those benefitting Spanish resources for NLP. I will describe our method to optimize supervision in semi-supervised machine learning approaches. We have carried out experiments mixing semi-supervised and active learning, with only partial success. We are currently dealing with semi-supervised deep learning. We are specially interested in how to introduce a minimal amount of supervision that helps deep learning methods avoid local minima. We are applying these to the problem of verbal sense disambiguation of Spanish and to Named Entity recognition.