Quotation Extraction for Portuguese

Quotation extraction consists of identifying quotations and their authors. In this work, we present a Quotation Extraction system for Portuguese that is based on Entropy Guided Transformation Learning, a supervised Machine Learning algorithm. This is the first system that uses a Machine Learning approach for Portuguese. In order to train and evaluate the proposed system, we build the GLOBOQUOTES corpus, with news extracted from the GLOBO.COM portal. Our system obtains an Fβ=1 score of 79.02% for the subtask of associating a quotation to its author. For the whole Quotation Extraction task, the observed Fβ=1 score value is 66.03%. These findings indicate that the overall extraction quality is highly dependant on the quotation identification subtask.