Overview of Proposed Machine Translation System for Decoding

The decoding process is as follows.

1. Test Corpus

   We prepare Japanese test sentences.

2. Pattern-Based Machine Translation

   We use a Japanese-English Pattern-Based Machine Translation. If an input Japanese sentence matches the Japanese patterns, we can obtain a translated ``English'' test sentence.

3. ``English''-English Statistical Machine Translation

   Using an ``English''-English phrase table, $N$ -gram model, and Moses [Koehn
   
   2007], we decode the ``English'' test sentence. This involves ``English''-English translation, resulting in an English sentence.

   Figure: Flowchart of Decoding Process

   

4. Japanese-English Statistical Machine Translation System

   If an input Japanese sentence does not match the Japanese patterns, we conduct a standard Japanese-English SMT using a Japanese-English phrase table and $N$ -gram model to obtain an English sentence.

Figure 5 shows a flowchart of the decoding process.