10.Combination_of_SMTnRBMT.pdf

Transcript

1. About this paper Introduction System Architecture Implmentation Results Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System দຊ޺ 2013 ೥ 3 ݄ 22 ೔ দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
2. About this paper Introduction System Architecture Implmentation Results About this paper Title:Using Moses to integrate multiple rule-based machine translation engines into a hybrid system Author: Eisele, Andreas and Federmann, Christian and Saint-Amand, Herv{é} and Jellinghaus, Michael and Herrmann, Teresa and Chen, Yu Title: Proceedings of the Third Workshop on Statistical Machine Translation pages: 179182 Year: 2008 Organization:Association for Computational Linguistics দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
3. About this paper Introduction System Architecture Implmentation Results Introduction Hybrid of SMT and RBMT Introduced in [Chen et al 2007] Architecture of SMT with multiple RBMT systems in a multi-engine setup দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
4. About this paper Introduction System Architecture Implmentation Results System Architecture SMT phrase table + RBMT phrase table Central Element: SMT with modied phrase table Moded = augmented with entries from RBMT SMT searches for best translation based on the combined p.t. দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
5. About this paper Introduction System Architecture Implmentation Results The Combined Phrase Table Combined phrase table Moses original phrase-table Each RBMT's phrase-table Combined phrase table দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
6. About this paper Introduction System Architecture Implmentation Results MT Systems and Knowledge Sources Used 6 RBMT Systems: Web-based Altavista Babelsh (based on Systran) SDL ProMT Lucy (ospring of METAL) Local-based OpenLogos translatePro by lingenio দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
7. About this paper Introduction System Architecture Implmentation Results Alignment of RBMT output Alignment of RBMT input and output Phrase table generated by RBMT is not big enough With the help of tool (Theison, 2007) Previously conducted training phase with bigger corpus দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
8. About this paper Introduction System Architecture Implmentation Results Combination of Phrase Tables The phrase table The Concatenation of the Moses and RBMT phrase tables One minor adjustment New column of indication from which system the data originated RBMT phrase tables are from scratch দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
9. About this paper Introduction System Architecture Implmentation Results Tuning In the process of Tuning To search for the best weight conguration In this hybrid, essential for additional columns To adjust the weight in the new columns 800 sentences from Europarl development data 800 sentences used for MERT to tune up for BLEU দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
10. About this paper Introduction System Architecture Implmentation Results BLEU Score দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System
11. About this paper Introduction System Architecture Implmentation Results Example List দຊ޺ Using Moses to Integrate Multiple Rule-Based Machine Translation Engines into a Hybrid System