繁體中文依存句法剖析器

Traditional Chinese Dependency Parser

近年來依存句法分析樹一直是CoNLL比賽的重要項目，其重要性可見一班，但其中參賽的隊伍並非以繁體中文為母語，若遇到繁體中文的語料通常會轉為簡體中文再做更進一步處理。因此，本研究以建立繁體中文依存句法分析樹為主要目的，透過目前相當熱門的深度學習模型，多層感知器及遞迴神經網路，與現有的斷詞器結合，得到繁體中文語句之句法分析樹。這樣的模型對於句法分析樹有相當好的成效，並比過去類似Task相比有高達70%的成長。

Recently, dependency parser has been paid highly attention in CoNLL conference, in this case we can find how important it is. However, there is no native Taiwanese attend to this competition. Therefore, we aim to build a traditional Chinese dependency parser in this paper. Through different famous neural network structure , such as multilayer perceptron and recurrent neural network along with traditional Chinese segmentation, we are able to construct an efficient parser that could transfer a raw traditional Chinese sentences to a dependency tree. We have much better result than similar task that has been proposed at 2012.