Forest-based Tree Sequence to String Translation Model
Zhang, Hui and Zhang, Min and Li, Haizhou and Aw, Aiti and Tan, Chew Lim

Article Structure

Abstract

This paper proposes a forest-based tree sequence to string translation model for syntax-based statistical machine translation, which automatically learns tree sequence to string translation rules from word-aligned source-side-parsed bilingual texts.

Introduction

Recently syntax-based statistical machine translation (SMT) methods have achieved very promising results and attracted more and more interests in the SMT research community.

Related work

As discussed in section 1, two of the major challenges to syntax-based SMT are structure divergence and parse errors.

Forest-based tree sequence to string model

In this section, we first explain what a packed forest is and then define the concept of the tree sequence in the context of forest followed by the discussion on our proposed model.

Training

This section discusses how to extract our translation rules given a triple < F, TS,A > .

Decoding

We benefit from the same strategy as used in our rule extraction algorithm in designing our decoding algorithm, recasting the forest-based tree se-quence-to-string decoding problem as a forest-based tree-to-string decoding problem.

Experiment

6.1 Experimental Settings

Conclusion

In this paper, we propose a forest-based tree-sequence to string translation model to combine the strengths of forest-based methods and tree-sequence based methods.

Topics

parse trees

Appears in 18 sentences as: parse tree (8) parse trees (13)
In Forest-based Tree Sequence to String Translation Model
  1. Therefore, it can not only utilize forest structure that compactly encodes exponential number of parse trees but also capture non-syntactic translation equivalences with linguistically structured information through tree sequence.
    Page 1, “Abstract”
  2. In theory, one may worry about whether the advantage of tree sequence has already been covered by forest because forest encodes implicitly a huge number of parse trees and these parse trees may generate many different phrases and structure segmentations given a source sentence.
    Page 1, “Introduction”
  3. Here, a tree sequence refers to a sequence of consecutive sub-trees that are embedded in a full parse tree .
    Page 2, “Related work”
  4. parse trees .
    Page 2, “Related work”
  5. parse trees ) for a given sentence under a context free grammar (CFG).
    Page 2, “Forest-based tree sequence to string model”
  6. The two parse trees T1 and T2 encoded in Fig.
    Page 3, “Forest-based tree sequence to string model”
  7. Different parse tree represents different derivations and explanations for a given sentence.
    Page 3, “Forest-based tree sequence to string model”
  8. Similar to the definition of tree sequence used in a single parse tree defined in Liu et al.
    Page 3, “Forest-based tree sequence to string model”
  9. them lies in that the sub-trees of a tree sequence in forest may belongs to different single parse trees while, in a single parse tree-based model, all the sub-trees in a tree sequence are committed to the same parse tree .
    Page 3, “Forest-based tree sequence to string model”
  10. The forest-based tree sequence enables our model to have the potential of exploring additional parse trees that may be wrongly pruned out by the parser and thus are not encoded in the forest.
    Page 3, “Forest-based tree sequence to string model”
  11. This is because that a tree sequence in a forest allows its sub-trees coming from different parse trees, where these sub-trees may not be merged finally to form a complete parse tree in the forest.
    Page 3, “Forest-based tree sequence to string model”

See all papers in Proc. ACL 2009 that mention parse trees.

See all papers in Proc. ACL that mention parse trees.

Back to top.

translation model

Appears in 12 sentences as: Translation Model (1) translation model (10) translation models (1)
In Forest-based Tree Sequence to String Translation Model
  1. This paper proposes a forest-based tree sequence to string translation model for syntax-based statistical machine translation, which automatically learns tree sequence to string translation rules from word-aligned source-side-parsed bilingual texts.
    Page 1, “Abstract”
  2. The proposed model leverages on the strengths of both tree sequence-based and forest-based translation models .
    Page 1, “Abstract”
  3. to String Translation Model
    Page 1, “Introduction”
  4. Section 2 describes related work while section 3 defines our translation model .
    Page 2, “Introduction”
  5. Motivated by the fact that non-syntactic phrases make nontrivial contribution to phrase-based SMT, the tree sequence-based translation model is proposed (Liu et al., 2007; Zhang et al., 2008a) that uses tree sequence as the basic translation unit, rather than using single subtree as in the STSG.
    Page 2, “Related work”
  6. (2007) propose the tree sequence concept and design a tree sequence to string translation model .
    Page 2, “Related work”
  7. (2008a) propose a tree sequence-based tree to tree translation model and Zhang et al.
    Page 2, “Related work”
  8. To integrate their strengths, in this paper, we propose a forest-based tree sequence to string translation model .
    Page 2, “Related work”
  9. 3.3 Forest-based tree-sequence to string translation model
    Page 4, “Forest-based tree sequence to string model”
  10. Given a source forest F and target translation T S as well as word alignment A, our translation model is formulated as:
    Page 4, “Forest-based tree sequence to string model”
  11. 4) Decode the translation forest using our translation model and a dynamic search algorithm.
    Page 6, “Decoding”

See all papers in Proc. ACL 2009 that mention translation model.

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

Back to top.

significantly outperforms

Appears in 6 sentences as: significantly outperforms (6)
In Forest-based Tree Sequence to String Translation Model
  1. Experimental results on the NIST MT-2003 Chinese-English translation task show that our method statistically significantly outperforms the four baseline systems.
    Page 1, “Abstract”
  2. Experimental results show that our method significantly outperforms the two individual methods and other baseline methods.
    Page 1, “Introduction”
  3. 1) FTS2S significantly outperforms (p<0.05) FT2S.
    Page 7, “Experiment”
  4. 3) Our model statistically significantly outperforms all the baselines system.
    Page 7, “Experiment”
  5. 4) All the four syntax-based systems show better performance than Moses and three of them significantly outperforms (p<0.05) Moses.
    Page 7, “Experiment”
  6. l) FTSZS significantly outperforms (p<0.05) FTZS consistently in all test cases.
    Page 8, “Experiment”

See all papers in Proc. ACL 2009 that mention significantly outperforms.

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

Back to top.

BLEU

Appears in 5 sentences as: BLEU (5)
In Forest-based Tree Sequence to String Translation Model
  1. Model BLEU (%) Moses 25.68 TT2S 26.08 TTS2S 26.95 FT2S 27.66 FTS2S 28.83
    Page 7, “Experiment”
  2. The 9% tree sequence rules contribute 1.17 BLEU score improvement (28.83-27.66 in Table 1) to FTS2S over FT2S.
    Page 8, “Experiment”
  3. BLEU (%) N-best \ model FT2S FTS2S 100 Best 27.40 28.61 500 Best 27.66 28.83 2500 Best 27.66 28.96 5000 Best 27.79 28.89
    Page 8, “Experiment”
  4. Even if in the 5000 Best case, tree sequence is still able to contribute l.l BLEU score improvement (28.89-27.79).
    Page 8, “Experiment”
  5. 2) The BLEU scores are very similar to each other when we increase the forest pruning threshold.
    Page 8, “Experiment”

See all papers in Proc. ACL 2009 that mention BLEU.

See all papers in Proc. ACL that mention BLEU.

Back to top.

baseline systems

Appears in 4 sentences as: baseline systems (3) baselines system (1)
In Forest-based Tree Sequence to String Translation Model
  1. Experimental results on the NIST MT-2003 Chinese-English translation task show that our method statistically significantly outperforms the four baseline systems .
    Page 1, “Abstract”
  2. We use the first three syntax-based systems (TT2S, TTS2S, FT2S) and Moses (Koehn et al., 2007), the state-of-the-art phrase-based system, as our baseline systems .
    Page 7, “Experiment”
  3. 3) Our model statistically significantly outperforms all the baselines system .
    Page 7, “Experiment”
  4. Experimental results show that our model greatly outperforms the four baseline systems .
    Page 8, “Conclusion”

See all papers in Proc. ACL 2009 that mention baseline systems.

See all papers in Proc. ACL that mention baseline systems.

Back to top.

Chinese-English

Appears in 4 sentences as: Chinese-English (4)
In Forest-based Tree Sequence to String Translation Model
  1. Experimental results on the NIST MT-2003 Chinese-English translation task show that our method statistically significantly outperforms the four baseline systems.
    Page 1, “Abstract”
  2. We evaluate our method on the NIST MT-2003 Chinese-English translation tasks.
    Page 1, “Introduction”
  3. We evaluate our method on Chinese-English translation task.
    Page 6, “Experiment”
  4. Finally, we examine our methods on the FBIS corpus and the NIST MT-2003 Chinese-English translation task.
    Page 8, “Conclusion”

See all papers in Proc. ACL 2009 that mention Chinese-English.

See all papers in Proc. ACL that mention Chinese-English.

Back to top.

NIST

Appears in 4 sentences as: NIST (5)
In Forest-based Tree Sequence to String Translation Model
  1. Experimental results on the NIST MT-2003 Chinese-English translation task show that our method statistically significantly outperforms the four baseline systems.
    Page 1, “Abstract”
  2. We evaluate our method on the NIST MT-2003 Chinese-English translation tasks.
    Page 1, “Introduction”
  3. We use the FBIS corpus as training set, the NIST MT-2002 test set as development (deV) set and the NIST MT-2003 test set as test set.
    Page 6, “Experiment”
  4. Finally, we examine our methods on the FBIS corpus and the NIST MT-2003 Chinese-English translation task.
    Page 8, “Conclusion”

See all papers in Proc. ACL 2009 that mention NIST.

See all papers in Proc. ACL that mention NIST.

Back to top.

proposed model

Appears in 4 sentences as: proposed model (4)
In Forest-based Tree Sequence to String Translation Model
  1. The proposed model leverages on the strengths of both tree sequence-based and forest-based translation models.
    Page 1, “Abstract”
  2. In this section, we first explain what a packed forest is and then define the concept of the tree sequence in the context of forest followed by the discussion on our proposed model .
    Page 2, “Forest-based tree sequence to string model”
  3. This clearly demonstrates the effectiveness of our proposed model for syntax-based SMT.
    Page 7, “Experiment”
  4. This again demonstrates the effectiveness of our proposed model .
    Page 8, “Experiment”

See all papers in Proc. ACL 2009 that mention proposed model.

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

Back to top.

translation task

Appears in 4 sentences as: translation task (3) translation tasks (1)
In Forest-based Tree Sequence to String Translation Model
  1. Experimental results on the NIST MT-2003 Chinese-English translation task show that our method statistically significantly outperforms the four baseline systems.
    Page 1, “Abstract”
  2. We evaluate our method on the NIST MT-2003 Chinese-English translation tasks .
    Page 1, “Introduction”
  3. We evaluate our method on Chinese-English translation task .
    Page 6, “Experiment”
  4. Finally, we examine our methods on the FBIS corpus and the NIST MT-2003 Chinese-English translation task .
    Page 8, “Conclusion”

See all papers in Proc. ACL 2009 that mention translation task.

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

Back to top.

BLEU score

Appears in 3 sentences as: BLEU score (2) BLEU scores (1)
In Forest-based Tree Sequence to String Translation Model
  1. The 9% tree sequence rules contribute 1.17 BLEU score improvement (28.83-27.66 in Table 1) to FTS2S over FT2S.
    Page 8, “Experiment”
  2. Even if in the 5000 Best case, tree sequence is still able to contribute l.l BLEU score improvement (28.89-27.79).
    Page 8, “Experiment”
  3. 2) The BLEU scores are very similar to each other when we increase the forest pruning threshold.
    Page 8, “Experiment”

See all papers in Proc. ACL 2009 that mention BLEU score.

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

Back to top.

phrase-based

Appears in 3 sentences as: phrase-based (3)
In Forest-based Tree Sequence to String Translation Model
  1. Motivated by the fact that non-syntactic phrases make nontrivial contribution to phrase-based SMT, the tree sequence-based translation model is proposed (Liu et al., 2007; Zhang et al., 2008a) that uses tree sequence as the basic translation unit, rather than using single subtree as in the STSG.
    Page 2, “Related work”
  2. We use seven basic features that are analogous to the commonly used features in phrase-based systems (Koehn, 2003): l) bidirectional rule mapping probabilities, 2) bidirectional lexical rule translation probabilities, 3) target language model, 4) number of rules used and 5) number of target words.
    Page 4, “Forest-based tree sequence to string model”
  3. We use the first three syntax-based systems (TT2S, TTS2S, FT2S) and Moses (Koehn et al., 2007), the state-of-the-art phrase-based system, as our baseline systems.
    Page 7, “Experiment”

See all papers in Proc. ACL 2009 that mention phrase-based.

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

Back to top.

word alignment

Appears in 3 sentences as: word alignment (2) word alignments (1)
In Forest-based Tree Sequence to String Translation Model
  1. Given a source forest F and target translation T S as well as word alignment A, our translation model is formulated as:
    Page 4, “Forest-based tree sequence to string model”
  2. GIZA++ (Och and Ney, 2003) and the heuristics “grow-diag-final-and” are used to generate m-to-n word alignments .
    Page 7, “Experiment”
  3. This is mainly because tree sequence rules are only sensitive to word alignment while tree rules, even extracted from a forest (like in FT2S), are also limited by syntax according to grammar parsing rules.
    Page 7, “Experiment”

See all papers in Proc. ACL 2009 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