Paper: Fast Easy Unsupervised Domain Adaptation with Marginalized Structured Dropout

ACL ID P14-2088
Title Fast Easy Unsupervised Domain Adaptation with Marginalized Structured Dropout
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2014
Authors

Unsupervised domain adaptation often re- lies on transforming the instance represen- tation. However, most such approaches are designed for bag-of-words models, and ignore the structured features present in many problems in NLP. We propose a new technique called marginalized struc- tured dropout, which exploits feature structure to obtain a remarkably simple and efficient feature projection. Applied to the task of fine-grained part-of-speech tagging on a dataset of historical Por- tuguese, marginalized structured dropout yields state-of-the-art accuracy while in- creasing speed by more than an order-of- magnitude over previous work.