Modeling the Translation of Predicate-Argument Structure for SMT
Xiong, Deyi and Zhang, Min and Li, Haizhou

Article Structure

Abstract

Predicate-argument structure contains rich semantic information of which statistical machine translation hasn’t taken full advantage.

Introduction

Recent years have witnessed increasing efforts towards integrating predicate-argument structures into statistical machine translation (SMT) (Wu and Fung, 2009b; Liu and Gildea, 2010).

Related Work

Predicate-argument structures (PAS) are eXplored for SMT on both the source and target side in some previous work.

Predicate Translation Model

In this section, we present the features and the training process of the predicate translation model.

Argument Reordering Model

In this section we introduce the discriminative argument reordering model, features and the training procedure.

Integrating the Two Models into SMT

In this section, we elaborate how to integrate the two models into phrase-based SMT.

Experiments

In this section, we present our eXperiments on Chinese-to-English translation tasks, which are trained with large-scale data.

Analysis

In this section, we conduct some case studies to show how the proposed models improve translation accuracy by looking into the differences that they make on translation hypotheses.

Conclusions and Future Work

We have presented two discriminative models to incorporate source side predicate-argument structures into SMT.

Topics

translation model

Appears in 29 sentences as: Translation Model (1) translation model (27) translation models (7)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. In this paper, we propose two discriminative, feature-based models to exploit predicate-argument structures for statistical machine translation: 1) a predicate translation model and 2) an argument reordering model.
    Page 1, “Abstract”
  2. The predicate translation model explores lexical and semantic contexts surrounding a verbal predicate to select desirable translations for the predicate.
    Page 1, “Abstract”
  3. This suggests that conventional leXical and phrasal translation models adopted in those SMT systems are not sufficient to correctly translate predicates in source sentences.
    Page 1, “Introduction”
  4. Thus we propose a discriminative, feature-based predicate translation model that captures not only leXical information (i.e., surrounding words) but also high-level semantic contexts to correctly translate predicates.
    Page 1, “Introduction”
  5. In Section 3 and 4, we will elaborate the proposed predicate translation model and argument reordering model respectively, including details about modeling, features and training procedure.
    Page 2, “Introduction”
  6. Our predicate translation model is also related to previous discriminative leXicon translation models (Berger et al., 1996; Venkatapathy and Bangalore, 2007; Mauser et al., 2009).
    Page 2, “Related Work”
  7. This will tremendously reduce the amount of training data required, which usually is a problem in discriminative leXicon translation models (Mauser et al., 2009).
    Page 2, “Related Work”
  8. Furthermore, the proposed translation model also differs from previous leXicon translation models in that we use both leXical and semantic features.
    Page 2, “Related Work”
  9. In this section, we present the features and the training process of the predicate translation model .
    Page 2, “Predicate Translation Model”
  10. Following the context-dependent word models in (Berger et al., 1996), we propose a discriminative predicate translation model .
    Page 2, “Predicate Translation Model”
  11. Given a source sentence which contains N verbal predicates , our predicate translation model Mt can be denoted as
    Page 2, “Predicate Translation Model”

See all papers in Proc. ACL 2012 that mention translation model.

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

Back to top.

semantic role

Appears in 8 sentences as: semantic role (5) semantic roles (4)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Therefore they either postpone the integration of target side PASs until the whole decoding procedure is completed (Wu and Fung, 2009b), or directly project semantic roles from the source side to the target side through word alignments during decoding (Liu and Gildea, 2010).
    Page 2, “Related Work”
  2. (2011) incorporate source language semantic role labels into a tree-to-string SMT system.
    Page 2, “Related Work”
  3. tic window, we use its semantic role (i.e., ARGO, ARGM-TMP and so on) A3; and head word A?
    Page 3, “Predicate Translation Model”
  4. In order to train the discriminative predicate translation model, we first parse source sentences and labeled semantic roles for all verbal predicates (see details in Section 6.1) in our word-aligned bilingual training data.
    Page 3, “Predicate Translation Model”
  5. On the source side, the features include the verbal predicate, the semantic role of the argument, the head word and the boundary words of the argument.
    Page 4, “Argument Reordering Model”
  6. its semantic role A7"
    Page 5, “Argument Reordering Model”
  7. To train the proposed predicate translation model and argument reordering model, we first parsed all source sentences using the Berkeley Chinese parser (Petrov et al., 2006) and then ran the Chinese semantic role labeler6 (Li et al., 2010) on all source parse trees to annotate semantic roles for all verbal predicates.
    Page 6, “Experiments”
  8. After we obtained semantic roles on the source side, we extracted features as described in Section 3.2 and 4.2 and used these features to train our two models as described in Section 3.3 and 4.3.
    Page 6, “Experiments”

See all papers in Proc. ACL 2012 that mention semantic role.

See all papers in Proc. ACL that mention semantic role.

Back to top.

BLEU

Appears in 7 sentences as: BLEU (7)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Statistical significance in BLEU differences
    Page 6, “Experiments”
  2. Our first group of experiments is to investigate whether the predicate translation model is able to improve translation accuracy in terms of BLEU and whether semantic features are useful.
    Page 6, “Experiments”
  3. 0 The proposed predicate translation models achieve an average improvement of 0.57 BLEU points across the two NIST test sets when all features (lex+sem) are used.
    Page 6, “Experiments”
  4. 0 When we integrate both lexical and semantic features (lex+sem) described in Section 3.2, we obtain an improvement of about 0.33 BLEU points over the system where only lexical features (lex) are used.
    Page 7, “Experiments”
  5. We obtain an average improvement of 0.4 BLEU points on the two test sets over the baseline when we incorporate the proposed argument reordering model into our system.
    Page 7, “Experiments”
  6. provement of up to 0.92 BLEU points over the baseline, which is shown in Table 5.
    Page 7, “Experiments”
  7. EXperimental results show that both models are able to significantly improve translation accuracy in terms of BLEU score.
    Page 8, “Conclusions and Future Work”

See all papers in Proc. ACL 2012 that mention BLEU.

See all papers in Proc. ACL that mention BLEU.

Back to top.

phrase-based

Appears in 7 sentences as: phrase-based (7)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. The two models are integrated into a state-of-the-art phrase-based machine translation system and evaluated on Chinese-to-English translation tasks with large-scale training data.
    Page 1, “Abstract”
  2. We integrate these two discriminative models into a state-of-the-art phrase-based system.
    Page 1, “Introduction”
  3. In this section, we elaborate how to integrate the two models into phrase-based SMT.
    Page 5, “Integrating the Two Models into SMT”
  4. In particular, we integrate the models into a phrase-based system which uses bracketing transduction grammars (BTG) (Wu, 1997) for phrasal translation (Xiong et al., 2006).
    Page 5, “Integrating the Two Models into SMT”
  5. It is straightforward to integrate the predicate translation model into phrase-based SMT (Koehn et al.,
    Page 5, “Integrating the Two Models into SMT”
  6. (2)) is integrated into the whole log-linear model just like the conventional lexical translation model in phrase-based SMT (Koehn et al., 2003).
    Page 5, “Integrating the Two Models into SMT”
  7. The two models have been integrated into a phrase-based SMT system and evaluated on Chinese-to-English translation tasks using large-scale training data.
    Page 8, “Conclusions and Future Work”

See all papers in Proc. ACL 2012 that mention phrase-based.

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

Back to top.

BLEU points

Appears in 4 sentences as: BLEU points (4)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. 0 The proposed predicate translation models achieve an average improvement of 0.57 BLEU points across the two NIST test sets when all features (lex+sem) are used.
    Page 6, “Experiments”
  2. 0 When we integrate both lexical and semantic features (lex+sem) described in Section 3.2, we obtain an improvement of about 0.33 BLEU points over the system where only lexical features (lex) are used.
    Page 7, “Experiments”
  3. We obtain an average improvement of 0.4 BLEU points on the two test sets over the baseline when we incorporate the proposed argument reordering model into our system.
    Page 7, “Experiments”
  4. provement of up to 0.92 BLEU points over the baseline, which is shown in Table 5.
    Page 7, “Experiments”

See all papers in Proc. ACL 2012 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 Modeling the Translation of Predicate-Argument Structure for SMT
  1. Predicate-argument structure contains rich semantic information of which statistical machine translation hasn’t taken full advantage.
    Page 1, “Abstract”
  2. In this paper, we propose two discriminative, feature-based models to exploit predicate-argument structures for statistical machine translation : 1) a predicate translation model and 2) an argument reordering model.
    Page 1, “Abstract”
  3. The two models are integrated into a state-of-the-art phrase-based machine translation system and evaluated on Chinese-to-English translation tasks with large-scale training data.
    Page 1, “Abstract”
  4. Recent years have witnessed increasing efforts towards integrating predicate-argument structures into statistical machine translation (SMT) (Wu and Fung, 2009b; Liu and Gildea, 2010).
    Page 1, “Introduction”

See all papers in Proc. ACL 2012 that mention machine translation.

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

Back to top.

maXimum entropy

Appears in 4 sentences as: maXimum entropy (3) maximum entropy (2)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. The essential component of our model is a maXimum entropy classifier pt(e|C that predicts the target translation 6 for a verbal predicate 2} given its surrounding context C(v).
    Page 2, “Predicate Translation Model”
  2. This will increase the number of classes to be predicted by the maximum entropy classifier.
    Page 3, “Predicate Translation Model”
  3. Using these events, we train one maximum entropy classifier per verbal predicate (16,121 verbs in total) via the off-the-shelf MaxEnt toolkit3.
    Page 3, “Predicate Translation Model”
  4. After all features are extracted, we use the maXimum entropy toolkit in Section 3.3 to train the maXimum entropy classifier as formulated in Eq.
    Page 5, “Argument Reordering Model”

See all papers in Proc. ACL 2012 that mention maXimum entropy.

See all papers in Proc. ACL that mention maXimum entropy.

Back to top.

SMT system

Appears in 4 sentences as: SMT system (2) SMT systems (2)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Unfortunately they are usually neither correctly translated nor translated at all in many SMT systems according to the error study by Wu and Fung (2009a).
    Page 1, “Introduction”
  2. This suggests that conventional leXical and phrasal translation models adopted in those SMT systems are not sufficient to correctly translate predicates in source sentences.
    Page 1, “Introduction”
  3. (2011) incorporate source language semantic role labels into a tree-to-string SMT system .
    Page 2, “Related Work”
  4. The two models have been integrated into a phrase-based SMT system and evaluated on Chinese-to-English translation tasks using large-scale training data.
    Page 8, “Conclusions and Future Work”

See all papers in Proc. ACL 2012 that mention SMT system.

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

Back to top.

statistically significant

Appears in 4 sentences as: Statistical significance (1) statistically significant (3)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Statistical significance in BLEU differences
    Page 6, “Experiments”
  2. Such an improvement is statistically significant (p < 0.01).
    Page 6, “Experiments”
  3. Such a gain, which is statistically significant , confirms the effectiveness of semantic features.
    Page 7, “Experiments”
  4. The improvements on the two test sets are both statistically significant (p < 0.01).
    Page 7, “Experiments”

See all papers in Proc. ACL 2012 that mention statistically significant.

See all papers in Proc. ACL that mention statistically significant.

Back to top.

word alignments

Appears in 4 sentences as: word alignments (4)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Therefore they either postpone the integration of target side PASs until the whole decoding procedure is completed (Wu and Fung, 2009b), or directly project semantic roles from the source side to the target side through word alignments during decoding (Liu and Gildea, 2010).
    Page 2, “Related Work”
  2. We maintain word alignments for each phrase pair in the phrase table.
    Page 5, “Integrating the Two Models into SMT”
  3. Whenever a hypothesis covers a new verbal predicate v, we find the target translation 6 for 7} through word alignments and then calculate its translation probability pt(e|C according to Eq.
    Page 5, “Integrating the Two Models into SMT”
  4. We ran GIZA++ on these corpora in both directions and then applied the “grow-diag-final” refinement rule to obtain word alignments .
    Page 6, “Experiments”

See all papers in Proc. ACL 2012 that mention word alignments.

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

Back to top.

significantly improve

Appears in 3 sentences as: significant improvements (1) significantly improve (2)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Experimental results demonstrate that the two models significantly improve translation accuracy.
    Page 1, “Abstract”
  2. Experimental results on large-scale Chinese-to-English translation show that both models are able to obtain significant improvements over the baseline.
    Page 1, “Introduction”
  3. EXperimental results show that both models are able to significantly improve translation accuracy in terms of BLEU score.
    Page 8, “Conclusions and Future Work”

See all papers in Proc. ACL 2012 that mention significantly improve.

See all papers in Proc. ACL that mention significantly improve.

Back to top.

statistical machine translation

Appears in 3 sentences as: statistical machine translation (3)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. Predicate-argument structure contains rich semantic information of which statistical machine translation hasn’t taken full advantage.
    Page 1, “Abstract”
  2. In this paper, we propose two discriminative, feature-based models to exploit predicate-argument structures for statistical machine translation : 1) a predicate translation model and 2) an argument reordering model.
    Page 1, “Abstract”
  3. Recent years have witnessed increasing efforts towards integrating predicate-argument structures into statistical machine translation (SMT) (Wu and Fung, 2009b; Liu and Gildea, 2010).
    Page 1, “Introduction”

See all papers in Proc. ACL 2012 that mention statistical machine translation.

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

Back to top.

translation tasks

Appears in 3 sentences as: translation tasks (3)
In Modeling the Translation of Predicate-Argument Structure for SMT
  1. The two models are integrated into a state-of-the-art phrase-based machine translation system and evaluated on Chinese-to-English translation tasks with large-scale training data.
    Page 1, “Abstract”
  2. In this section, we present our eXperiments on Chinese-to-English translation tasks , which are trained with large-scale data.
    Page 6, “Experiments”
  3. The two models have been integrated into a phrase-based SMT system and evaluated on Chinese-to-English translation tasks using large-scale training data.
    Page 8, “Conclusions and Future Work”

See all papers in Proc. ACL 2012 that mention translation tasks.

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

Back to top.

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