| Budget Amount *help |
¥23,400,000 (Direct Cost: ¥18,000,000、Indirect Cost: ¥5,400,000)
Fiscal Year 2018: ¥6,890,000 (Direct Cost: ¥5,300,000、Indirect Cost: ¥1,590,000)
Fiscal Year 2017: ¥6,890,000 (Direct Cost: ¥5,300,000、Indirect Cost: ¥1,590,000)
Fiscal Year 2016: ¥9,620,000 (Direct Cost: ¥7,400,000、Indirect Cost: ¥2,220,000)
|
|---|
| Outline of Final Research Achievements |
In this research, we have developed a zero-shot machine translation method which can be trained only with monolingual image-text data, without the help of parallel text corpus. This method is realized by the idea of using images as a hub to align texts in different languages. Moreover, we have improved the method in many aspects to enhance its practicality such as output diversification and speeding up. These results are accepted at many top-level international conferences such as ACL and ICLR, and awarded the best paper awards twice at the NLP domestic conference.
|
