Unsupervised induction of sentence compression rules
João Cordeiro, Gaël Dias, Pavel Brazdil · 2009
In this paper, we propose a new unsupervised approach to sentence compression based on shallow linguistic processing. For that purpose, paraphrase extraction and alignment is performed over web news stories extracted automatically from the web on a daily basis to provide structured data examples to the learning process. Compression rules are then learned through the application of Inductive Logic Programming techniques. Qualitative and quantitative evaluations suggests that this is a worth following approach, which might be even improved in the future.