Decoding

The decoding process is as follows.

1. Test Corpus

   We prepared French test sentences.

2. Pattern-Based Machine Translation

   We use a French-English PBMT. If an input French sentence matches with the French 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 [4], we decode the "English" test sentence. This involves "English"-English translation, resulting in an English sentence.

4. French-English Statistical Machine Translation System

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

Figure 5 shows a flowchart of the decoding process.

図 5: Flowchart of Decoding