Parser Adaptation for Social Media by Integrating Normalization

This work explores normalization forparser adaptation. Traditionally, normalizationis used as separate pre-processingstep. We show that integrating the normalizationmodel into the parsing algorithmis beneficial. This way, multiple normalizationcandidates can be leveraged,which improves parsing performance onsocial media. We test this hypothesisby modifying the Berkeley parser; out-ofthe-boxit achieves an F1 score of 66.52.Our integrated approach reaches a significantimprovement with an F1 score of67.36, while using the best normalizationsequence results in an F1 score of only66.94