MT Use in Lingosail, by Yongpeng Wei, Lingosail
Download as PPTX, PDF1 like299 views
1) MTPE improves translation efficiency and quality while allowing for more scalable management compared to traditional translation. 2) Vertical MT training is important to build confidence in results by focusing on specific domains. 3) Terminology plays a key role by being extracted from documents, applied to MT, and collected for future use.
MT Use in Lingosail, by Yongpeng Wei, Lingosail
- 1. Yongpeng WEI 2016.04.25 TAUS Executive Forum Beijing 2016
- 2. Services • Translation Technology • Corpus / Term-base Construction • Vertical MT Training • On-line CAT Customization • Translation (MTPE) Service • MTPE production • MTPE training Markets • Intellectual Properties • Cross-border E-Commerce • Internet Finance • …
- 3. Machine Translation: • Hybrid of statistical approaches and linguistic knowledge • Neural Machine Translation • Knowledge Graphs • … Human Translation: • Crowd sourcing • Data sharing • Workflow automation • Writing for Translation • …
- 4. Easier Entry • Lower requirement for domain knowledge • Lower cost of learning and self- improvement for translators Higher Efficiency • Speed-up search and verification process • Reduce repetitions • Reduce typing time • Reduce elementary mistakes More Scalable Management • Unify workflow • Increase standardization
- 5. Zhang 23 Libra Student Major in English Q: How long have you taken part in MTPE project? A: Two months. Q: What do you think MTPE is different from common translation? A: Efficiency has been improved in significant measure(from 500 words/h to 750-1000 words/h). It has also shaped my translation style and thinking mode. However, if the MT result and the terms could not be easily confirmed by the translator, he/she need more other resources to verify the translation, then the process looks like common translation. Q:Recent monthly income? A:3500 Yuan (As an intern)
- 6. Sun 24 Aquarius Full-time Major in Japanese Q: How long have you taken part in MTPE project? A: Five months. Q: What do you think MTPE is different from common translation? A: I think it is easier to start. Some sentences are hard to understand and organize. MT gives helpful clues for me. Q:Recent monthly income? A:About 10000 Yuan (As an in-house translator).
- 7. Han 25 Aries Part-time Major in Japanese Q: How long have you taken part in MTPE project? A: Nearly one year (part-time). Q: What do you think MTPE is different from common translation? A: The translation system and the term base are powerful. For some of the MT results, you need only to adjust the word orders. It largely increases the efficiency, as well as the accuracy. Q:Recent monthly income? A:About 9000 Yuan (As a part-time translator).
- 8. 1) MT is useful (in PE production), for ◦ Efficiency ◦ Quality ◦ Scalability 2) Confidence is the key, so vertical MT training is a must ◦ Domain specific ◦ Find out the reliable parts (Training needed for translators) 3) Terminology plays an important role ◦ Pre-translating: extract & check ◦ On-translating: applied into MT ◦ Post-translating: collection for future use
- 10. 1. Monolingual Term Extract ◦ MT used for: Term translation (in the context) ◦ MT used for: Term recognition ( EBMT-based)
- 11. 2. Bilingual Term Extract ◦ MT used for: Word/Phrase Alignment ◦ Term extracted for: reference term pairs (pre-trans) ◦ Term extracted for: term pairs collection (post-trans)
- 12. 1. For Clients: CLIR for Database (e.g. patents) ◦ MT used for: large volume of documents in the DB ◦ Accuracy of Keywords(Terminologies) ◦ Document similarity 2. For Translators: Find reference documents (in target language) ◦ MT used for: document to be translated ◦ Sentence similarity ◦ Document similarity
- 13. Lower deploying threshold ◦ Affordable computing capability (esp. for Neural MT) ◦ Less corpus requirement ◦ Acknowledged Quality Metrics Involving deeper knowledge base ◦ Terminology is the first step ◦ Knowledge Graph More interactive ◦ Real-time interactions (e.g. Input Method) ◦ Feedback from end-users ◦ Dynamic user behaviors