next up previous
次へ: Experiment 上へ: Proposed Method 戻る: Notes

Database

We used simple sentence pairs and complex/compound sentence pairs extracted from the EPWING electronic dictionaries [8]. We used 100,000 English-Japanese simple sentence pairs for the training and 10,000 English sentences for the test. The average of simple sentence is 9.0 word/sentence for the training in English. And the average of simple sentence is 10.5 word/sentence for the training in Japanese. Similarly, we used 100,000 translation pairs from compound/complex sentence pairs for the training and 10,000 English sentences for the test. The average of compound/complex sentence is 11.2 word/sentence for the training in English. And the average of compound/complex sentence is 13.9 word/sentence for the training in Japanese. We used Mecab [7] as the morphological analyzer and the standing tokenizer of Moses [5] .
Table V shows example English-Japanese sentence pairs.

表 V: Example Sentences
\scalebox{1.0}[1.0]{
\begin{tabular}{\vert l\vert l\vert}
\hline
\multicolumn{2...
... は まず エンジン を\\
& かけ なけれ ば なり ませ ん 。\\ \hline
\end{tabular}}




平成25年9月17日