Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
Deng, Yonggang and Xu, Jia and Gao, Yuqing

Article Structure

Abstract

In this work, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log—linear model aiming for a balanced precision and recall.

Introduction

Phrase has become the standard basic translation unit in Statistical Machine Translation (SMT) since it naturally captures context dependency and models internal word reordering.

A Generic Phrase Training Procedure

Let e = 6{ denote an English sentence and let f = fi] denote its translation in a foreign language, say Chinese.

Features

Now we present several feature functions that we investigated to help extracting correct phrase translations.

Experimental Results

We evaluate the effect of the proposed phrase extraction algorithm with translation performance.

Discussions

The generic phrase training algorithm follows an information retrieval perspective as in (Venugopal et al., 2003) but aims to improve both precision and recall with the trainable log-linear model.

Conclusions

In this paper, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log-linear model aiming for a balanced precision and recall.

Topics

phrase pair

Appears in 51 sentences as: Phrase Pair (2) Phrase pair (2) phrase pair (34) phrase pairs (20)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Multiple data-driven feature functions are proposed to capture the quality and confidence of phrases and phrase pairs .
    Page 1, “Abstract”
  2. The first problem is referred to as phrase pair extraction, which identifies phrase pairs that are supposed to be translations of each other.
    Page 1, “Introduction”
  3. The most widely used approach derives phrase pairs from word alignment matrix (Och and Ney, 2003; Koehn et al., 2003).
    Page 1, “Introduction”
  4. Other methods do not depend on word alignments only, such as directly modeling phrase alignment in a joint generative way (Marcu and Wong, 2002), pursuing information extraction perspective (Venugopal et al., 2003), or augmenting with model-based phrase pair posterior (Deng and Byrne, 2005).
    Page 1, “Introduction”
  5. Using relative frequency as translation probability is a common practice to measure goodness of a phrase pair .
    Page 1, “Introduction”
  6. Since most phrases appear only a few times in training data, a phrase pair translation is also evaluated by lexical weights (Koehn et al., 2003) or term weighting (Zhao et al., 2004) as additional features to avoid overestimation.
    Page 1, “Introduction”
  7. The focus of this paper is the phrase pair extraction problem.
    Page 1, “Introduction”
  8. High precision requires that identified translation candidates are accurate, while high recall wants as much valid phrase pairs as possible to be extracted, which is important and necessary for online translation that requires coverage.
    Page 1, “Introduction”
  9. We would like to improve phrase translation accuracy and at the same time extract as many as possible valid phrase pairs that are missed due to incorrect word alignments.
    Page 2, “Introduction”
  10. In this work, we present a generic discriminative phrase pair extraction framework that can integrate multiple features aiming to identify correct phrase translation candidates.
    Page 2, “Introduction”
  11. We first train word alignment models and will use them to evaluate the goodness of a phrase and a phrase pair .
    Page 2, “A Generic Phrase Training Procedure”

See all papers in Proc. ACL 2008 that mention phrase pair.

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

Back to top.

word alignment

Appears in 44 sentences as: word alignment (35) Word Alignments (1) word alignments (12)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Experimental results demonstrate consistent and significant improvement over the widely used method that is based on word alignment matrix only.
    Page 1, “Abstract”
  2. The most widely used approach derives phrase pairs from word alignment matrix (Och and Ney, 2003; Koehn et al., 2003).
    Page 1, “Introduction”
  3. Other methods do not depend on word alignments only, such as directly modeling phrase alignment in a joint generative way (Marcu and Wong, 2002), pursuing information extraction perspective (Venugopal et al., 2003), or augmenting with model-based phrase pair posterior (Deng and Byrne, 2005).
    Page 1, “Introduction”
  4. On the other hand, there are valid translation pairs in the training corpus that are not learned due to word alignment errors as shown in Deng and Byrne (2005).
    Page 1, “Introduction”
  5. We would like to improve phrase translation accuracy and at the same time extract as many as possible valid phrase pairs that are missed due to incorrect word alignments .
    Page 2, “Introduction”
  6. One approach is to leverage underlying word alignment quality such as in Ayan and Dorr (2006).
    Page 2, “Introduction”
  7. We employ features based on word alignment models and alignment matrix.
    Page 2, “Introduction”
  8. We first train word alignment models and will use them to evaluate the goodness of a phrase and a phrase pair.
    Page 2, “A Generic Phrase Training Procedure”
  9. Beginning with a flat lexicon, we train IBM Model-l word alignment model with 10 iterations for each translation direction.
    Page 2, “A Generic Phrase Training Procedure”
  10. We then train HMM word alignment models (Vogel et al., 1996) in two directions simultaneously by merging statistics collected in the
    Page 2, “A Generic Phrase Training Procedure”
  11. 1: Train Model-1 and HMM word alignment models 2: for all sentence pair (e, f) do
    Page 2, “A Generic Phrase Training Procedure”

See all papers in Proc. ACL 2008 that mention word alignment.

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

Back to top.

alignment models

Appears in 20 sentences as: alignment model (9) alignment models (11) alignment model’s (1)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. We employ features based on word alignment models and alignment matrix.
    Page 2, “Introduction”
  2. We first train word alignment models and will use them to evaluate the goodness of a phrase and a phrase pair.
    Page 2, “A Generic Phrase Training Procedure”
  3. Beginning with a flat lexicon, we train IBM Model-l word alignment model with 10 iterations for each translation direction.
    Page 2, “A Generic Phrase Training Procedure”
  4. We then train HMM word alignment models (Vogel et al., 1996) in two directions simultaneously by merging statistics collected in the
    Page 2, “A Generic Phrase Training Procedure”
  5. 1: Train Model-1 and HMM word alignment models 2: for all sentence pair (e, f) do
    Page 2, “A Generic Phrase Training Procedure”
  6. Each normalized feature score derived from word alignment models or language models will be log-linearly combined to generate the final score.
    Page 2, “A Generic Phrase Training Procedure”
  7. All these features are data-driven and defined based on models, such as statistical word alignment model or language model.
    Page 3, “Features”
  8. In a statistical generative word alignment model (Brown et al., 1993), it is assumed that (i) a random variable a specifies how each target word fj is generated by (therefore aligned to) a source 1 word eaj; and (ii) the likelihood function f (f, a|e) specifies a generative procedure from the source sentence to the target sentence.
    Page 3, “Features”
  9. This distribution is applicable to all word alignment models that follow assumptions (i) and (ii).
    Page 3, “Features”
  10. One of them is based on HMM word alignment model .
    Page 3, “Features”
  11. We use the geometric mean of posteriors in two translation directions as a symmetric metric for phrase pair quality evaluation function under HMM alignment models .
    Page 3, “Features”

See all papers in Proc. ACL 2008 that mention alignment models.

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

Back to top.

BLEU

Appears in 13 sentences as: BLEU (13)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. lation engine to minimize the final translation errors measured by automatic metrics such as BLEU (Papineni et al., 2002).
    Page 3, “A Generic Phrase Training Procedure”
  2. We measure translation performance by the BLEU (Papineni et al., 2002) and METEOR (Banerjee and Lavie, 2005) scores with multiple translation references.
    Page 5, “Experimental Results”
  3. BLEU Scores
    Page 6, “Experimental Results”
  4. The translation results as measured by BLEU and METEOR scores are presented in Table 3.
    Page 6, “Experimental Results”
  5. We notice that Model-4 based phrase table performs roughly 1% better in terms of both BLEU and METEOR scores than that based on HMM.
    Page 6, “Experimental Results”
  6. Once we have computed all feature values for all phrase pairs in the training corpus, we discriminatively train feature weights Aks and the threshold 7' using the downhill simplex method to maximize the BLEU score on 06dev set.
    Page 6, “Experimental Results”
  7. Roughly, it has 0.5% higher BLEU score on 2006 sets and 1.5% to 3% higher on other sets than Model-4 based ViterbiExtract method.
    Page 6, “Experimental Results”
  8. - + - BLEU mo“ Phrasetable Size
    Page 7, “Discussions”
  9. After reaching its peak, the BLEU score drops as the threshold 7' increases.
    Page 7, “Discussions”
  10. Table 4: Translation Results ( BLEU ) of discriminative phrase training approach using different features
    Page 7, “Discussions”
  11. Table 5: Translation Results ( BLEU ) of Different Phrase Pair Combination
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention BLEU.

See all papers in Proc. ACL that mention BLEU.

Back to top.

phrase table

Appears in 11 sentences as: phrase table (11) phrase tables (1)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Our baseline phrase table training method is the ViterbiExtract algorithm.
    Page 6, “Experimental Results”
  2. We notice that Model-4 based phrase table performs roughly 1% better in terms of both BLEU and METEOR scores than that based on HMM.
    Page 6, “Experimental Results”
  3. Since the translation engine implements a log-linear model, the discriminative training of feature weights in the decoder should be embedded in the whole end-to-end system jointly with the discriminative phrase table training process.
    Page 6, “Experimental Results”
  4. As a compromise, we fix the decoder feature weights and put all efforts on optimizing phrase training parameters to find out the best phrase table .
    Page 6, “Experimental Results”
  5. The translation results with the discriminatively trained phrase table are shown as the row of “New” in Table 3.
    Page 6, “Experimental Results”
  6. The phrase table extracting procedure is trainable and can be optimized jointly with the translation engine.
    Page 6, “Discussions”
  7. As the figure 1 shows, when we increase the threshold by allowing more candidate phrase pair hypothesized as valid translation, we observe the phrase table size increases monotonically.
    Page 6, “Discussions”
  8. Figure 1: Thresholding effects on translation performance and phrase table size
    Page 7, “Discussions”
  9. Now we compare the phrase table using the proposed method to that extracted using the baseline ViterbiExtract method with Model-4 word alignments.
    Page 7, “Discussions”
  10. The Venn diagram in Table 5 shows how the two phrase tables overlap with each other and size of each part.
    Page 7, “Discussions”
  11. Removing PP1 from the baseline phrase table (comparing the first group of scores) or adding PP1 to the new phrase table
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention phrase table.

See all papers in Proc. ACL that mention phrase table.

Back to top.

language model

Appears in 8 sentences as: language model (7) language models (1)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Each normalized feature score derived from word alignment models or language models will be log-linearly combined to generate the final score.
    Page 2, “A Generic Phrase Training Procedure”
  2. All these features are data-driven and defined based on models, such as statistical word alignment model or language model .
    Page 3, “Features”
  3. We apply a language model (LM) to describe the predictive uncertainty (PU) between words in two directions.
    Page 4, “Features”
  4. Given a history 10711—1, a language model specifies a conditional distribution of the future word being predicted to follow the history.
    Page 4, “Features”
  5. We can find the entropy of such pdf: HLM(w7f’_1) = So given a sentence wiv , the PU of the boundary between word w,- and wi+1 is established by two-way entropy sum using a forward and backward language model : = -|- HLMBUUE—l)
    Page 4, “Features”
  6. Like other log-linear model based decoders, active features in our translation engine include translation models in two directions, lexicon weights in two directions, language model , lexicalized distortion models, sentence length penalty and other heuristics.
    Page 5, “Experimental Results”
  7. The language model is a statistical trigram model estimated with Modified Kneser—Ney smoothing (Chen and Goodman, 1996) using only English sentences in the parallel training data.
    Page 5, “Experimental Results”
  8. We propose several information metrics derived from posterior distribution, language model and word alignments as feature functions.
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention language model.

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

Back to top.

log-linear

Appears in 7 sentences as: log-linear (7)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Note that under the log-linear model, applying threshold for filtering is equivalent to comparing the “likelihood” ratio.
    Page 2, “A Generic Phrase Training Procedure”
  2. Our decoder is a phrase-based multi-stack implementation of the log-linear model similar to Pharaoh (Koehn et al., 2003).
    Page 5, “Experimental Results”
  3. Like other log-linear model based decoders, active features in our translation engine include translation models in two directions, lexicon weights in two directions, language model, lexicalized distortion models, sentence length penalty and other heuristics.
    Page 5, “Experimental Results”
  4. Since the translation engine implements a log-linear model, the discriminative training of feature weights in the decoder should be embedded in the whole end-to-end system jointly with the discriminative phrase table training process.
    Page 6, “Experimental Results”
  5. The generic phrase training algorithm follows an information retrieval perspective as in (Venugopal et al., 2003) but aims to improve both precision and recall with the trainable log-linear model.
    Page 6, “Discussions”
  6. Under the general framework, one can put as many features as possible together under the log-linear model to evaluate the quality of a phrase and a phase pair.
    Page 6, “Discussions”
  7. In this paper, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log-linear model aiming for a balanced precision and recall.
    Page 8, “Conclusions”

See all papers in Proc. ACL 2008 that mention log-linear.

See all papers in Proc. ACL that mention log-linear.

Back to top.

log-linear model

Appears in 7 sentences as: log-linear model (7)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Note that under the log-linear model , applying threshold for filtering is equivalent to comparing the “likelihood” ratio.
    Page 2, “A Generic Phrase Training Procedure”
  2. Our decoder is a phrase-based multi-stack implementation of the log-linear model similar to Pharaoh (Koehn et al., 2003).
    Page 5, “Experimental Results”
  3. Like other log-linear model based decoders, active features in our translation engine include translation models in two directions, lexicon weights in two directions, language model, lexicalized distortion models, sentence length penalty and other heuristics.
    Page 5, “Experimental Results”
  4. Since the translation engine implements a log-linear model , the discriminative training of feature weights in the decoder should be embedded in the whole end-to-end system jointly with the discriminative phrase table training process.
    Page 6, “Experimental Results”
  5. The generic phrase training algorithm follows an information retrieval perspective as in (Venugopal et al., 2003) but aims to improve both precision and recall with the trainable log-linear model .
    Page 6, “Discussions”
  6. Under the general framework, one can put as many features as possible together under the log-linear model to evaluate the quality of a phrase and a phase pair.
    Page 6, “Discussions”
  7. In this paper, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log-linear model aiming for a balanced precision and recall.
    Page 8, “Conclusions”

See all papers in Proc. ACL 2008 that mention log-linear model.

See all papers in Proc. ACL that mention log-linear model.

Back to top.

feature weights

Appears in 6 sentences as: feature weights (6)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. 15: Discriminatively train feature weights N, and threshold 7'
    Page 2, “A Generic Phrase Training Procedure”
  2. These feature weights are tuned on the dev set to achieve optimal translation performance using downhill simplex method.
    Page 5, “Experimental Results”
  3. Once we have computed all feature values for all phrase pairs in the training corpus, we discriminatively train feature weights Aks and the threshold 7' using the downhill simplex method to maximize the BLEU score on 06dev set.
    Page 6, “Experimental Results”
  4. Since the translation engine implements a log-linear model, the discriminative training of feature weights in the decoder should be embedded in the whole end-to-end system jointly with the discriminative phrase table training process.
    Page 6, “Experimental Results”
  5. As a compromise, we fix the decoder feature weights and put all efforts on optimizing phrase training parameters to find out the best phrase table.
    Page 6, “Experimental Results”
  6. The parameter controlling the degree of attenuation in BLT is also optimized together with other feature weights .
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention feature weights.

See all papers in Proc. ACL that mention feature weights.

Back to top.

n-grams

Appears in 5 sentences as: n-grams (5)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Usually all n-grams up to a predefined length limit are considered as candidate phrases.
    Page 2, “A Generic Phrase Training Procedure”
  2. First, due to data sparsity and/or alignment model’s capability, there would exist n-grams that cannot be aligned
    Page 3, “Features”
  3. well, for instance, n-grams that are part of a paraphrase translation or metaphorical expression.
    Page 4, “Features”
  4. Extracting candidate translations for such kind of n-grams for the sake of improving coverage (recall) might hurt translation quality (precision).
    Page 4, “Features”
  5. Second, some n-grams themselves carry no linguistic meaning; their phrase translations can be misleading, for example non-compositional phrases (Lin, 1999).
    Page 4, “Features”

See all papers in Proc. ACL 2008 that mention n-grams.

See all papers in Proc. ACL that mention n-grams.

Back to top.

precision and recall

Appears in 5 sentences as: precision and recall (5)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. In this work, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log—linear model aiming for a balanced precision and recall .
    Page 1, “Abstract”
  2. As in information retrieval, precision and recall issues need to be addressed with a right balance for building a phrase translation table.
    Page 1, “Introduction”
  3. The generic phrase training algorithm follows an information retrieval perspective as in (Venugopal et al., 2003) but aims to improve both precision and recall with the trainable log-linear model.
    Page 6, “Discussions”
  4. It implies a balancing process between precision and recall .
    Page 7, “Discussions”
  5. In this paper, the problem of extracting phrase translation is formulated as an information retrieval process implemented with a log-linear model aiming for a balanced precision and recall .
    Page 8, “Conclusions”

See all papers in Proc. ACL 2008 that mention precision and recall.

See all papers in Proc. ACL that mention precision and recall.

Back to top.

BLEU score

Appears in 5 sentences as: BLEU score (4) BLEU Scores (1)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. BLEU Scores
    Page 6, “Experimental Results”
  2. Once we have computed all feature values for all phrase pairs in the training corpus, we discriminatively train feature weights Aks and the threshold 7' using the downhill simplex method to maximize the BLEU score on 06dev set.
    Page 6, “Experimental Results”
  3. Roughly, it has 0.5% higher BLEU score on 2006 sets and 1.5% to 3% higher on other sets than Model-4 based ViterbiExtract method.
    Page 6, “Experimental Results”
  4. After reaching its peak, the BLEU score drops as the threshold 7' increases.
    Page 7, “Discussions”
  5. On the other hand, adding phrase pairs extracted by the new method only (PP3) can lead to significant BLEU score increases (comparing row 1 vs. 3, and row 2 vs. 4).
    Page 8, “Discussions”

See all papers in Proc. ACL 2008 that mention BLEU score.

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

Back to top.

sentence pair

Appears in 4 sentences as: sentence pair (4)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. 1: Train Model-1 and HMM word alignment models 2: for all sentence pair (e, f) do
    Page 2, “A Generic Phrase Training Procedure”
  2. Phrase pair filtering is simply thresholding on the final score by comparing to the maximum within the sentence pair .
    Page 2, “A Generic Phrase Training Procedure”
  3. Given a phrase pair in a sentence pair , there will be many generative paths that align the source phrase to the target phrase.
    Page 3, “Features”
  4. Given a sentence pair , the basic assumption is that if the HMM word alignment model can align an English phrase well to a foreign phrase, the posterior distribution of the English phrase generating all foreign phrases on the other side is significantly biased.
    Page 4, “Features”

See all papers in Proc. ACL 2008 that mention sentence pair.

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

Back to top.

translation probability

Appears in 4 sentences as: translation probability (4)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Using relative frequency as translation probability is a common practice to measure goodness of a phrase pair.
    Page 1, “Introduction”
  2. The translation probability can also be discriminatively trained such as in Tillmann and Zhang (2006).
    Page 1, “Introduction”
  3. In estimating phrase translation probability , we use accumulated HMM-based phrase pair posteriors
    Page 6, “Experimental Results”
  4. as their ‘soft’ frequencies and then the final translation probability is the relative frequency.
    Page 6, “Experimental Results”

See all papers in Proc. ACL 2008 that mention translation probability.

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

Back to top.

development set

Appears in 3 sentences as: development set (3)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. A significant deviation from most other approaches is that the framework is parameterized and can be optimized jointly with the decoder to maximize translation performance on a development set .
    Page 2, “Introduction”
  2. In the final step 4 (line 15), parameters {Am 7'} are discriminatively trained on a development set using the downhill simplex method (Nelder and Mead, 1965).
    Page 3, “A Generic Phrase Training Procedure”
  3. We use feature functions to decide the order and the threshold 7' to locate the boundary guided with a development set .
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention development set.

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

Back to top.

end-to-end

Appears in 3 sentences as: end-to-end (3)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. We present a generic phrase training algorithm which is parameterized with feature functions and can be optimized jointly with the translation engine to directly maximize the end-to-end system performance.
    Page 1, “Abstract”
  2. Since the translation engine implements a log-linear model, the discriminative training of feature weights in the decoder should be embedded in the whole end-to-end system jointly with the discriminative phrase table training process.
    Page 6, “Experimental Results”
  3. It can be optimized jointly with the translation engine to directly maximize the end-to-end translation performance.
    Page 8, “Conclusions”

See all papers in Proc. ACL 2008 that mention end-to-end.

See all papers in Proc. ACL that mention end-to-end.

Back to top.

n-gram

Appears in 3 sentences as: n-gram (3)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. Trying to find phrase translations for any possible n-gram is not a good idea for two reasons.
    Page 3, “Features”
  2. We will define a confidence metric to estimate how reliably the model can align an n-gram in one side to a phrase on the other side given a parallel sentence.
    Page 4, “Features”
  3. Now we turn to monolingual resources to evaluate the quality of an n-gram being a good phrase.
    Page 4, “Features”

See all papers in Proc. ACL 2008 that mention n-gram.

See all papers in Proc. ACL that mention n-gram.

Back to top.

parallel sentence

Appears in 3 sentences as: parallel sentence (2) parallel sentences (1)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. We will define a confidence metric to estimate how reliably the model can align an n-gram in one side to a phrase on the other side given a parallel sentence .
    Page 4, “Features”
  2. The training corpus consists of 40K Chinese-English parallel sentences in travel domain with to-
    Page 5, “Experimental Results”
  3. Ideally, a perfect combination of feature functions divides the correct and incorrect candidate phrase pairs within a parallel sentence into two ordered separate sets.
    Page 7, “Discussions”

See all papers in Proc. ACL 2008 that mention parallel sentence.

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

Back to top.

Chinese-English

Appears in 3 sentences as: Chinese-English (3)
In Phrase Table Training for Precision and Recall: What Makes a Good Phrase and a Good Phrase Pair?
  1. We do experiments on IWSLT (Paul, 2006) 2006 Chinese-English corpus.
    Page 5, “Experimental Results”
  2. The training corpus consists of 40K Chinese-English parallel sentences in travel domain with to-
    Page 5, “Experimental Results”
  3. Our experimental results on IWSLT Chinese-English corpus have demonstrated consistent and significant improvement over the widely used word alignment matrix based extraction method.
    Page 8, “Conclusions”

See all papers in Proc. ACL 2008 that mention Chinese-English.

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

Back to top.

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