Signi cant amount of work has been devoted recently to develop learning techniques that can be used to generate partial (shallow) analysis of natural language sentences rather than a full parse. In this work we set out to evaluate whether this direction is worthwhile by comparing a learned shallow parser to one of the best learned full parsers on tasks both can perform | identifying phrases in sentences. We conclude that directly learning to perform these tasks as shallow parsers do is advantageous over full parsers both in terms of performance and robustness to new and lower quality texts. 1

area. Learning Theory: developing the foundations for learning in the context of large scale real world problems. Learning and Knowledge Representation: a study of intermediate knowledge representations that facilitate learning in complex domains. Learning and Inference: a study of inference with outcomes of classiers. Learning in Natural Language: an experimental paradigm that builds on progress made in the above areas to address several key problems in natural language from a unied point of view. A number of key NLP problems have been solved this way and software has been made available to the community. Perhaps the most signicant aspect of this research program is in presenting the centrality of learning in areas traditionally thought of as inferential. In doing that, we have helped placing theoretical work in learning in the context of realistic inference problems and, along with developing the theoreti