To Normalize, or Not to Normalize: The Impact of Normalization on Part-of-Speech Tagging

摘要

Does normalization help Part-of-Speech (POS) tagging accuracy on noisy,non-canonical data? To the best of our knowledge, little is known on the actualimpact of normalization in a real-world scenario, where gold error detection isnot available. We investigate the effect of automatic normalization on POStagging of tweets. We also compare normalization to strategies that leveragelarge amounts of unlabeled data kept in its raw form. Our results show thatnormalization helps, but does not add consistently beyond just word embeddinglayer initialization. The latter approach yields a tagging model that iscompetitive with a Twitter state-of-the-art tagger.