Proceedings of COLING 2012: Technical Papers, pages 1681–1698, COLING 2012, Mumbai, December 2012. A Separately Passive-Aggressive Training Algorithmfor Jo int POS Tagging and Dependency Parsing Zhenghua Li 1, Min Zhang 2, Wanxiang Che 1, Ting Liu 1?(1) Research Center for puting and Information Re trieval, Harbin Institute of Technology, China (2) Institute for mResearch, Singapore {lzh,car,tliu}***@., ******@-. sg A BSTRACT Recent study shows that parsing accuracy can be largely impr oved by the joint optimization of part-of-speech (POS) tagging and dependency parsing. Ho wever, the POS tagging task does not bene?t much from the joint framework. We argue that t he fundamental reason behind is because the POS features are overwhelmed by the syn tactic features during the joint optimization, and the joint models only prefer such PO S tags that are favourable solely from the parsing viewpoint. To solve this issue, we propose a separately passive-aggressive learning algorithm (SPA), which is designed to separately u pdate the POS features weights and the syntactic feature weights under the joint optimizat ion framework. The proposed SPA is able to take advantage of previous joint optimization str ategies to signi?cantly improve the parsing accuracy, but also e their shortages to s igni?cantly boost the tagging accuracy by effectively solving the syntax-insensitive PO S ambiguity issues. Experiments on the Chinese Penn Treebank (CTB5) and the English Penn Tre ebank (PTB) demonstrate the effectiveness of our proposed methodology and empirica lly verify our observations as discussed above. We achieve the best tagging and parsing acc uracies on both datasets, % in tagging accuracy and % in parsing accuracy on CTB5, a nd % and % on PTB. K EYWORDS: Part-of-speech Tagging, Dependency Parsing, Joint Models , Separately Passive- aggressive Algorithm. Corresponding author 1681 1 Introduction Given an input sentence of n words, denoted
汉诺坝玄武岩板块构造环境及岩浆成因机理的分析1 来自淘豆网www.taodocs.com转载请标明出处.