Low-level Markup and Large-scale LFG Grammar Processing

It is commonly believed that shallow mark-up techniques such as part-of-speech disambiguation or low-level phrase chunking provide useful information that can improve the performance of natural language processing systems, even those that ultimately require deeper levels of analysis. In this paper, we discuss three types of shallow mark-up: part of speech tagging, named entities, and labeled bracketing. We show how they were integrated into the ParGram LFG English grammar and report on the results of parsing the PARC700 sentences with each type of mark-up. We observed that named-entity mark-up improves both speed and accuracy and labeled brackets also can be beneficial, but that part-of-speech tags are not particularly useful.