Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
Wang, Kun and Zong, Chengqing and Su, Keh-Yih

Article Structure

Abstract

Since statistical machine translation (SMT) and translation memory (TM) complement each other in matched and unmatched regions, integrated models are proposed in this paper to incorporate TM information into phrase-based SMT.

Introduction

Statistical machine translation (SMT), especially the phrase-based model (Koehn et al., 2003), has developed very fast in the last decade.

Problem Formulation

Compared with the standard phrase-based machine translation model, the translation problem is reformulated as follows (only based on the best TM, however, it is similar for multiple TM sentences):

Proposed Models

Three integrated models are proposed to incorporate different features as follows:

Experiments

4.1 Experimental Setup

Conclusion and Future Work

Unlike the previous pipeline approaches, which directly merge TM phrases into the final translation result, we integrate TM information of each source phrase into the phrase-based SMT at decoding.

Topics

TER

Appears in 14 sentences as: TER (16)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Furthermore, integrated Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction in comparison with the pure SMT system.
    Page 1, “Abstract”
  2. Compared with the pure SMT system, the proposed integrated Model-III achieves 3.48 BLEU points improvement and 2.62 TER points reduction overall.
    Page 2, “Introduction”
  3. In this work, the translation performance is measured with case-insensitive BLEU-4 score (Papineni et al., 2002) and TER score (Snover et al., 2006).
    Page 6, “Experiments”
  4. In the tables, the best translation results (either in BLEU or TER ) at each interval have been marked in bold.
    Page 7, “Experiments”
  5. It can be seen that TM significantly exceeds SMT at the interval [0.9, 1.0) in TER score, which illustrates why professional translators prefer TM rather than SMT as their assistant tool.
    Page 7, “Experiments”
  6. Compared with TM and SMT, Model-I is significantly better than the SMT system in either BLEU or TER when the fuzzy match score is above 0.7; Model-II significantly outperforms both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.5; Model-III significantly exceeds both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.4.
    Page 7, “Experiments”
  7. Across all intervals (the last row in the table), Model-III not only achieves the best BLEU score (56.51), but also gets the best TER score (33.26).
    Page 7, “Experiments”
  8. If intervals are evaluated separately, when the fuzzy match score is above 0.4, Model-III outperforms both Model-II and Model-I in either BLEU or TER .
    Page 7, “Experiments”
  9. Model-II also exceeds Model-I in either BLEU or TER .
    Page 7, “Experiments”
  10. The only exception is at interval [0.5, 0.6), in which Model-I achieves the best TER score.
    Page 7, “Experiments”
  11. This might be due to that the optimization criterion for MERT is BLEU rather than TER in our work.
    Page 7, “Experiments”

See all papers in Proc. ACL 2013 that mention TER.

See all papers in Proc. ACL that mention TER.

Back to top.

BLEU

Appears in 11 sentences as: BLEU (14)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Furthermore, integrated Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction in comparison with the pure SMT system.
    Page 1, “Abstract”
  2. Compared with the pure SMT system, the proposed integrated Model-III achieves 3.48 BLEU points improvement and 2.62 TER points reduction overall.
    Page 2, “Introduction”
  3. In the tables, the best translation results (either in BLEU or TER) at each interval have been marked in bold.
    Page 7, “Experiments”
  4. Compared with TM and SMT, Model-I is significantly better than the SMT system in either BLEU or TER when the fuzzy match score is above 0.7; Model-II significantly outperforms both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.5; Model-III significantly exceeds both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.4.
    Page 7, “Experiments”
  5. SMT 8.03 BLEU points at interval [0.9, 1.0), while the advantage is only 2.97 BLEU points at interval [0.6, 0.7).
    Page 7, “Experiments”
  6. Across all intervals (the last row in the table), Model-III not only achieves the best BLEU score (56.51), but also gets the best TER score (33.26).
    Page 7, “Experiments”
  7. If intervals are evaluated separately, when the fuzzy match score is above 0.4, Model-III outperforms both Model-II and Model-I in either BLEU or TER.
    Page 7, “Experiments”
  8. Model-II also exceeds Model-I in either BLEU or TER.
    Page 7, “Experiments”
  9. This might be due to that the optimization criterion for MERT is BLEU rather than TER in our work.
    Page 7, “Experiments”
  10. The experiments show that the proposed Model-III outperforms both the TM and the SMT systems significantly (p < 0.05) in either BLEU or TER when fuzzy match score is above 0.4.
    Page 9, “Conclusion and Future Work”
  11. Compared with the pure SMT system, Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction on a Chinese—English TM database.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention BLEU.

See all papers in Proc. ACL that mention BLEU.

Back to top.

SMT system

Appears in 11 sentences as: SMT system (9) SMT systems (4)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Furthermore, integrated Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction in comparison with the pure SMT system .
    Page 1, “Abstract”
  2. Especially, there is no guarantee that a SMT system can produce translations in a consistent manner (Ma et al., 2011).
    Page 1, “Introduction”
  3. Afterwards, they merge the relevant translations of matched segments into the source sentence, and then force the SMT system to only translate those unmatched segments at decoding.
    Page 1, “Introduction”
  4. Compared with the pure SMT system , the proposed integrated Model-III achieves 3.48 BLEU points improvement and 2.62 TER points reduction overall.
    Page 2, “Introduction”
  5. For the phrase-based SMT system , we adopted the Moses toolkit (Koehn et al., 2007).
    Page 6, “Experiments”
  6. We first extract 95% of the bilingual sentences as a new training corpus to train a SMT system .
    Page 6, “Experiments”
  7. Scores marked by “*” are significantly better ([9 < 0.05) than both the TM and the SMT systems .
    Page 7, “Experiments”
  8. Compared with TM and SMT, Model-I is significantly better than the SMT system in either BLEU or TER when the fuzzy match score is above 0.7; Model-II significantly outperforms both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.5; Model-III significantly exceeds both the TM and the SMT systems in either BLEU or TER when the fuzzy match score is above 0.4.
    Page 7, “Experiments”
  9. However, SMT system still inserts another “you”, regardless of “you do” has already existed.
    Page 9, “Experiments”
  10. The experiments show that the proposed Model-III outperforms both the TM and the SMT systems significantly (p < 0.05) in either BLEU or TER when fuzzy match score is above 0.4.
    Page 9, “Conclusion and Future Work”
  11. Compared with the pure SMT system , Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction on a Chinese—English TM database.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention SMT system.

See all papers in Proc. ACL that mention SMT system.

Back to top.

phrase-based

Appears in 9 sentences as: phrase-based (9)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Since statistical machine translation (SMT) and translation memory (TM) complement each other in matched and unmatched regions, integrated models are proposed in this paper to incorporate TM information into phrase-based SMT.
    Page 1, “Abstract”
  2. Statistical machine translation (SMT), especially the phrase-based model (Koehn et al., 2003), has developed very fast in the last decade.
    Page 1, “Introduction”
  3. On a Chinese—English computer technical documents TM database, our experiments have shown that the proposed Model-III improves the translation quality significantly over either the pure phrase-based SMT or the TM systems when the fuzzy match score is above 0.4.
    Page 2, “Introduction”
  4. Compared with the standard phrase-based machine translation model, the translation problem is reformulated as follows (only based on the best TM, however, it is similar for multiple TM sentences):
    Page 2, “Problem Formulation”
  5. mula (3) is just the typical phrase-based SMT model, and the second factor P(Mk|Lk, 2:) (to be specified in the Section 3) is the information derived from the TM sentence pair.
    Page 3, “Problem Formulation”
  6. Therefore, we can still keep the original phrase-based SMT model and only pay attention to how to extract
    Page 3, “Problem Formulation”
  7. For the phrase-based SMT system, we adopted the Moses toolkit (Koehn et al., 2007).
    Page 6, “Experiments”
  8. conducted using the Moses phrase-based decoder (Koehn et al., 2007).
    Page 7, “Experiments”
  9. Unlike the previous pipeline approaches, which directly merge TM phrases into the final translation result, we integrate TM information of each source phrase into the phrase-based SMT at decoding.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention phrase-based.

See all papers in Proc. ACL that mention phrase-based.

Back to top.

proposed models

Appears in 8 sentences as: proposed models (8)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Unlike previous multistage pipeline approaches, which directly merge TM result into the final output, the proposed models refer to the corresponding TM information associated with each phrase at SMT decoding.
    Page 1, “Abstract”
  2. Besides, the proposed models also outperform previous approaches significantly.
    Page 1, “Abstract”
  3. Furthermore, the proposed models significantly outperform previous pipeline approaches.
    Page 2, “Introduction”
  4. To estimate the probabilities of proposed models , the corresponding phrase segmentations for bilingual sentences are required.
    Page 6, “Experiments”
  5. In order to compare our proposed models with previous work, we re-implement two XML-Markup approaches: (Koehn and Senellart, 2010) and (Ma et al, 2011), which are denoted as Koehn-10 and Mall, respectively.
    Page 7, “Experiments”
  6. More importantly, the proposed models achieve much better TER score than the TM system does at interval [0.9, 1.0), but Koehn-10 does not even exceed the TM system at this interval.
    Page 9, “Experiments”
  7. Therefore, it can be concluded that the proposed models outperform the pipeline approaches significantly.
    Page 9, “Experiments”
  8. In addition, all possible TM target phrases are kept and the proposed models select the best one during decoding via referring SMT information.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention proposed models.

See all papers in Proc. ACL that mention proposed models.

Back to top.

sentence pair

Appears in 6 sentences as: sentence pair (4) sentence pairs (2)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. They first determine whether the extracted TM sentence pair should be adopted or not.
    Page 1, “Introduction”
  2. is the final translation; [tm_s,tm_t,tm_f,s_a,tm_a] are the associated information of the best TM sentence-pair; tm_s and tm_t denote the corresponding TM sentence pair ; tm_f denotes its associated fuzzy match score (from 0.0 to 1.0); 8_a is the editing operations between tm_8 and s; and tm_a denotes the word alignment between tm_s and tmi.
    Page 2, “Problem Formulation”
  3. mula (3) is just the typical phrase-based SMT model, and the second factor P(Mk|Lk, 2:) (to be specified in the Section 3) is the information derived from the TM sentence pair .
    Page 3, “Problem Formulation”
  4. useful information from the best TM sentence pair to guide SMT decoding.
    Page 3, “Problem Formulation”
  5. We randomly selected a development set and a test set, and then the remaining sentence pairs are for training set.
    Page 6, “Experiments”
  6. The remaining 28.3% of the sentence pairs are thus not adopted for generating training samples.
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention sentence pair.

See all papers in Proc. ACL that mention sentence pair.

Back to top.

BLEU points

Appears in 4 sentences as: BLEU points (5)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Furthermore, integrated Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction in comparison with the pure SMT system.
    Page 1, “Abstract”
  2. Compared with the pure SMT system, the proposed integrated Model-III achieves 3.48 BLEU points improvement and 2.62 TER points reduction overall.
    Page 2, “Introduction”
  3. SMT 8.03 BLEU points at interval [0.9, 1.0), while the advantage is only 2.97 BLEU points at interval [0.6, 0.7).
    Page 7, “Experiments”
  4. Compared with the pure SMT system, Model-III achieves overall 3.48 BLEU points improvement and 2.62 TER points reduction on a Chinese—English TM database.
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention BLEU points.

See all papers in Proc. ACL that mention BLEU points.

Back to top.

machine translation

Appears in 4 sentences as: machine translation (4)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. Since statistical machine translation (SMT) and translation memory (TM) complement each other in matched and unmatched regions, integrated models are proposed in this paper to incorporate TM information into phrase-based SMT.
    Page 1, “Abstract”
  2. Statistical machine translation (SMT), especially the phrase-based model (Koehn et al., 2003), has developed very fast in the last decade.
    Page 1, “Introduction”
  3. Compared with the standard phrase-based machine translation model, the translation problem is reformulated as follows (only based on the best TM, however, it is similar for multiple TM sentences):
    Page 2, “Problem Formulation”
  4. Last, some related approaches (Smith and Clark, 2009; Phillips, 2011) combine SMT and example-based machine translation (EBMT) (Nagao, 1984).
    Page 9, “Conclusion and Future Work”

See all papers in Proc. ACL 2013 that mention machine translation.

See all papers in Proc. ACL that mention machine translation.

Back to top.

segmentations

Appears in 4 sentences as: segmentations (4)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. To estimate the probabilities of proposed models, the corresponding phrase segmentations for bilingual sentences are required.
    Page 6, “Experiments”
  2. As we want to check what actually happened during decoding in the real situation, cross-fold translation is used to obtain the corresponding phrase segmentations .
    Page 6, “Experiments”
  3. Afterwards, we generate the corresponding phrase segmentations for the remaining 5% bi-
    Page 6, “Experiments”
  4. Having repeated the above steps 20 times4, we obtain the corresponding phrase segmentations for the SMT training data (which will then be used to train the integrated models).
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention segmentations.

See all papers in Proc. ACL that mention segmentations.

Back to top.

development set

Appears in 3 sentences as: development set (3)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. We randomly selected a development set and a test set, and then the remaining sentence pairs are for training set.
    Page 6, “Experiments”
  2. Furthermore, development set and test set are divided into various intervals according to their best fuzzy match scores.
    Page 6, “Experiments”
  3. All the feature weights and the weight for each probability factor (3 factors for Model-III) are tuned on the development set with minimum-error-rate training (MERT) (Och, 2003).
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention development set.

See all papers in Proc. ACL that mention development set.

Back to top.

word alignment

Appears in 3 sentences as: word alignment (2) word alignments (1)
In Integrating Translation Memory into Phrase-Based Machine Translation during Decoding
  1. is the final translation; [tm_s,tm_t,tm_f,s_a,tm_a] are the associated information of the best TM sentence-pair; tm_s and tm_t denote the corresponding TM sentence pair; tm_f denotes its associated fuzzy match score (from 0.0 to 1.0); 8_a is the editing operations between tm_8 and s; and tm_a denotes the word alignment between tm_s and tmi.
    Page 2, “Problem Formulation”
  2. ), we can find its corresponding TM source phrase tm_sa(k) and all possible TM target phrases (each of them is denoted by tmiaw» with the help of corresponding editing operations 8_a and word alignment tm_a.
    Page 2, “Problem Formulation”
  3. The system configurations are as follows: GIZA++ (Och and Ney, 2003) is used to obtain the bidirectional word alignments .
    Page 6, “Experiments”

See all papers in Proc. ACL 2013 that mention word alignment.

See all papers in Proc. ACL that mention word alignment.

Back to top.

PDF available from www.aclweb.org.
Send us an email