Semi supervised word alignment model for parallel corpus

本研究的目的在于提高平行语料库中的词对齐精度。在本研究中，不仅使用了单词信息，还使用了词性信息和句子结构。由于难以对语料库中的所有句子添加信息，因此采用半监督方法进行训练。对日语、英语和汉语平行语料库进行了使用词性标签和专有名词意义标签的半监督添加方法，并验证了其有效性。其次，使用句子结构信息进行对齐，并验证其有效性。

The porous of this research is to improve word alignment accuracy in parallel corpus. In this research, not only word information, but also part-of-speech information and sentence structure are used. Semi-supervised approach is used for training, since it is difficult to additional information to all of sentence in corpus. For Japanese, English, and Chinese parallel corpus, semi-supervised aliment method using POS tag, and meaning tag for proper noun is conducted, and its effectiveness is confirmed. Next, sentence structure information is used for alignment, and its effectiveness is also confirmed.