Skip to Main Content
Table 12 
Percentages of the destinations of the attention mechanism to source words (Wrd) or phrases (Phr) by using the proposed tree-to-sequence NMT models described in Sections 5 and 6. The percentages are obtained on the Chinese-to-Japanese and English-to-Japanese development data, respectively.
 Chinese-to-JapaneseEnglish-to-Japanese
Wrd (%)Phr (%)Small Data setLarge Data set
Wrd (%)Phr (%)Wrd (%)Phr (%)
α > 0.2 18.9 5.5 11.2 10.8 9.1 8.1 
α > 0.4 10.6 2.4 5.3 3.6 3.4 2.3 
α > 0.6 6.1 1.0 2.9 2.5 1.5 0.5 
α > 0.8 2.6 0.3 1.3 0.8 0.3 0.0 
 Chinese-to-JapaneseEnglish-to-Japanese
Wrd (%)Phr (%)Small Data setLarge Data set
Wrd (%)Phr (%)Wrd (%)Phr (%)
α > 0.2 18.9 5.5 11.2 10.8 9.1 8.1 
α > 0.4 10.6 2.4 5.3 3.6 3.4 2.3 
α > 0.6 6.1 1.0 2.9 2.5 1.5 0.5 
α > 0.8 2.6 0.3 1.3 0.8 0.3 0.0 
Close Modal

or Create an Account

Close Modal
Close Modal