Semi-Supervised Learning of Sequence Models with Method of Moments

Zita Marinho1, AndrĂ© F. T. Martins2, Shay B. Cohen3, Noah A. Smith4
1Carnegie Mellon University / University of Lisbon, 2Priberam, Instituto de Telecomunicacoes, 3University of Edinburgh, 4University of Washington


Abstract

We propose a fast and scalable method for semi-supervised learning of sequence models, based on anchor words and moment matching. Our method can handle hidden Markov models with feature-based log-linear emissions. Unlike other semi-supervised methods, no decoding passes are necessary on the unlabeled data and no graph needs to be constructed---only one pass is necessary to collect moment statistics. The model parameters are estimated by solving a small quadratic program for each feature. Experiments on part-of-speech (POS) tagging for Twitter and for a low resource language (Malagasy) show that our method can learn from very few annotated sentences.