Part-of-speech tagging models for parsing Rebecca Watson
This event took place on 9th March 2006 at 9:00am (09:00 GMT)
Knowledge Media Institute, Berrill Building, The Open University, Milton Keynes, United Kingdom, MK7 6AA
We investigate the accuracy of alternative part-of-speech tag models and their impact on parser performance. In addition to considering single-tag and multipletag per word input, tag selection models which draw on information available from the parser are applied. Results indicate that given a 'good' PoS tagger, parserbased tag selection models are unable to improve on the low tag error rates of the tagger. Multiple-tag per word input can improve on parser accuracy but at a cost to eficiency. A dynamic tag selection model is also applied, which only increases the number of tags considered for sentences if a full parse could not be found. This achieves the best accuracy and provides a means to overcome the trade-off between tag error rates and increased parse ambiguity introduced by multiple-tag per word input.
We investigate the accuracy of alternative part-of-speech tag models and their impact on parser performance. In addition to considering single-tag and multipletag per word input, tag selection models which draw on information available from the parser are applied. Results indicate that given a 'good' PoS tagger, parserbased tag selection models are unable to improve on the low tag error rates of the tagger. Multiple-tag per word input can improve on parser accuracy but at a cost to eficiency. A dynamic tag selection model is also applied, which only increases the number of tags considered for sentences if a full parse could not be found. This achieves the best accuracy and provides a means to overcome the trade-off between tag error rates and increased parse ambiguity introduced by multiple-tag per word input.
Click below to play the event (32 minutes)