Google Translate + TectoMT
1 like755 views
The document discusses combining Google Translate with TectoMT, a modular NLP software system. It provides background on each tool, noting Google Translate supports 34 languages via API and handles large amounts of text using computing power, while TectoMT takes a more linguistic approach. Examples are given showing translations from both Google Translate alone and combined with TectoMT, with the combination approach generating more accurate translations by conditioning on governing and dependent words rather than linear context. The document concludes by noting results from initial experiments combining the tools, and outlines plans to improve the approach.
Google Translate + TectoMT
- 1. Google Translate & TectoMT Martin Majliš
- 2. Plan What is Google Translate?
- 4. Why combine?
- 5. Results
- 6. TODO
- 7. Google Translate Google service
- 8. 34 languages
- 9. AJAX API
- 11. Google Translate API http://ajax.googleapis.com/ajax/services/language/translate?v=1.0&q=it's lunchtime!&langpair=en%7Ccs
- 12. {"responseData": {"translatedText":"přestávka na oběd, je to!"}, "responseDetails": null, "responseStatus": 200}
- 13. TectoMT a highly modular extendable NLP software system
- 14. composed of numerous (mostly previously existing) NLP tools integrated into a uniform infrastructure
- 15. aimed at (not limited to) developing MT system
- 17. Why combine?
- 18. Why combine? (cont.) Google Great amount of text
- 19. Computing power - 450k servers (2006) TectoMT Linguistic approach
- 20. Hypothesis: translation choices are conditioned rather by governing/dependent words than by linear predecessors/followers
- 21. Translation
- 22. Examples This time the fall in stocks on Wall Street is responsible for the drop. GT: Tato doba je pokles zásob na Wall Street je zodpovědný za pokles.
- 23. GT+TMT: Tato doba je pokles zásob na wall street, je zodpovědný za pokles.
- 24. Examples (cont.) Stocks fall in Asia GT: Zásoby pád v Asii
- 25. GT+TMT: Zásoba pád v asie.
- 26. Examples (cont.) The stock exchange in Taiwan dropped by 3.6 percent according to the local index. GT: Burze, na Tchaj-wanu klesla o 3,6 procent v závislosti na místní rejstřík.
- 27. Burze tchaj - wan klesl o 3, 6 procent na místní rejstřík.
- 28. Examples (cont.) Despite the fact that it is a part of China, Hong Kong determines its currency policy separately, that is, without being dependent on the Chinese Central Bank. GT: Navzdory skutečnosti, že je součástí Číny, Hong Kong určuje její měny politice samostatně, tedy bez závislosti na čínské centrální bance.
- 29. GT+TMT: Navzdory skutečnosti, že je součástí číny, hong kong určuje její měny politice samostatně, tedy bez na závislosti čínské centrální bance.
- 30. Errors Name entity Kevin Rudd => Kevin Rudd => kevin rudd
- 31. Numbers!
- 32. Results BLEU NIST Google Translate 0.1543 5.2542 GT + Parahrasing 0.0953 4.4241
- 33. TODO Somehow improve :) Analyse english sentence
- 35. Merge subtrees
- 36. Analyze + Parahrase czech sentence
- 37. Questions? ?
- 38. Thank you!