Decoding Pattern

The decoding process is as follows.

1. Input French Sentence

   We prepare input French sentences. An example sentence is shown in Table 5.

   
   

   表 5: French Sentence

   O se trouve le poste de police ?

   
   

2. Search French Pattern and Output English Pattern

   We search for a French pattern that is matched with the input French sentence using French patterns and a high-probability phrase table (section 4.1). And we obtain English patterns. Example French-English patterns are listed in Table 6. Also, an example French-English high-probability phrase table is shown in Table .

   
   

   表 6: French-English Patterns

   1	French Pattern	O se trouve le X1 de X2 ?
   	English Pattern	Where's the X2 X1 ?
   2	French Pattern	O se trouve le poste de X2 ?
   	English Pattern	Where is a X2 ?

   
   

   
   

   表 7: French-English High-Probability Phrase Table

   1	poste |||post |||0.5 0.07 0.5 0.21
   2	police |||police |||0.4 0.031 0.2 0.11
   3	police |||police station post |||0.1 0.07 0.1 0.01

   
   

3. Generate English Sentences

   We generate English sentences using the English pattern and French-English High-Probability phrase tables. Note that all possible English sentences are generated. Therefore, plural English sentences are generated from an input French sentence. Example English sentences are listed in Table 8.

   
   

   表 8: Generated English Sentences

   1	Where's the police post ?
   2	Where's is a police station post ?

   
   

4. Select One English Sentence.

   We select one English sentence from generated plural English sentences using $N$ -gram. An example of selected one English sentence is shown in Table 9.

   
   

   表 9: One English Sentence

   Where's the police post ?

   
   

Figure 2 shows the process of decoding English sentences for PBMT.

図 2: Decoding of Pattern-Based Machine Translation