## Error Detection for Statistical Machine Translation Using Linguistic Features

### Cached

### Download Links

Citations: | 4 - 0 self |

### BibTeX

@MISC{Xiong_errordetection,

author = {Deyi Xiong and Min Zhang and Haizhou Li},

title = {Error Detection for Statistical Machine Translation Using Linguistic Features},

year = {}

}

### OpenURL

### Abstract

Automatic error detection is desired in the post-processing to improve machine translation quality. The previous work is largely based on confidence estimation using system-based features, such as word posterior probabilities calculated from N-best lists or word lattices. We propose to incorporate two groups of linguistic features, which convey information from outside machine translation systems, into error detection: lexical and syntactic features. We use a maximum entropy classifier to predict translation errors by integrating word posterior probability feature and linguistic features. The experimental results show that 1) linguistic features alone outperform word posterior probability based confidence estimation in error detection; and 2) linguistic features can further provide complementary information when combined with word confidence scores, which collectively reduce the classification error rate by 18.52 % and improve the F measure by 16.37%. 1

### Citations

1560 | Bleu: a method for automatic evaluation of machine translation
- Papineni, Roukos, et al.
- 2002
(Show Context)
Citation Context ...s the development set for the translation task. In order to calculate word posterior probabilities, we generate 10,000 best lists for NIST MT-02/03/05 respectively. The performance, in terms of BLEU (=-=Papineni et al., 2002-=-) score, is shown in Table 4. 6 Experiments We conducted our experiments at several levels. Starting with MaxEnt models with single linguistic feature or word posterior probability based feature, we i... |

1123 | A maximum entropy approach to natural language processing - Berger, Pietra, et al. - 1996 |

953 | Open source toolkit for statistical machine translation - Koehn, Hoang, et al. - 2007 |

489 | Minimum error rate training in statistical machine translation
- Och
- 2003
(Show Context)
Citation Context ...ation Chinese Treebank Multiple Translation Chinese Table 2: Training corpora for the translation task. on Xinhua section of the English Gigaword corpus (181.1M words). For minimum error rate tuning (=-=Och, 2003-=-), we use NIST MT-02 as the development set for the translation task. In order to calculate word posterior probabilities, we generate 10,000 best lists for NIST MT-02/03/05 respectively. The performan... |

374 | Parsing English with a link grammar
- Sleator, Temperley
- 1991
(Show Context)
Citation Context ...lenge, we select the Link Grammar (LG) parser 3 as our syntactic parser to generate syntactic features. The LG parser produces a set of labeled links which connect pairs of words with a link grammar (=-=Sleator and Temperley, 1993-=-). The main reason why we choose the LG parser is that it provides a robustness feature: null-link scheme. The null-link scheme allows the parser to parse a sentence even when the parser can not fully... |

237 |
Binary Codes Capable of Correcting Deletions
- Levenshtein
- 1966
(Show Context)
Citation Context ...nce f, let {en} N 1 be the Nbest list generated by an SMT system, and let ei n is the i-th word in en. The major work of calculating word posterior probabilities is to find the Levenshtein alignment (=-=Levenshtein, 1966-=-) between the best hypothesis e1 and its competing hypothesis 3 Available at http://www.link.cs.cmu.edu/link/Figure 1: An example of Link Grammar parsing results. en in the N-best list {en} N 1 . We ... |

41 | Multi-engine machine translation guided by explicit word matching
- Jayaraman, Lavie
- 2005
(Show Context)
Citation Context ...sing the probability of correctness calculated for each hypothesis (Zens and Ney, 2006) or by generating new hypotheses using Nbest lists from one SMT system or multiple systems (Akibay et al., 2004; =-=Jayaraman and Lavie, 2005-=-). In this paper we restrict the “parts” to words. That is, we detect errors at the word level for SMT. A common approach to SMT error detection at the word level is calculating the confidence at whic... |

37 | Word-level confidence estimation for machine translation
- Ueffing, Ney
- 2007
(Show Context)
Citation Context ...reference translations) and incorrect parts. Automatically distinguishing incorrect parts from correct parts is therefore very desirable not only for post-editing and interactive machine translation (=-=Ueffing and Ney, 2007-=-) but also for SMT itself: either by rescoring hypotheses in the N-best list using the probability of correctness calculated for each hypothesis (Zens and Ney, 2006) or by generating new hypotheses us... |

23 | N-Gram posterior probabilities for statistical machine translation - Zens, Ney - 2004 |

10 |
Confidence estimation for machine translation. Final report, JHU/CLSP Summer Workshop. http://www.clsp.jhu.edu/ws2003/groups/ estimate
- Blatz, Fitzgerald, et al.
- 2003
(Show Context)
Citation Context ... three steps: 1) Calculate features that express the correctness of words either based on SMT model (e.g. translation/language model) or based on SMT system output (e.g. N-best lists, word lattices) (=-=Blatz et al., 2003-=-; Ueffing and Ney, 2007). 2) Combine these features together with a classification model such as multi-layer perceptron (Blatz et al., 2003), Naive Bayes (Blatz et al., 2003; Sanchis et al., 2007), or... |

6 | sentencelevel confidence measures for machine translation - Word- |

4 | 2003. Confidence estimation for translation prediction - Gandrabur, Foster |

4 | Estimation of confidence measures for machine translation
- Sanchis, Juan, et al.
(Show Context)
Citation Context ...ices) (Blatz et al., 2003; Ueffing and Ney, 2007). 2) Combine these features together with a classification model such as multi-layer perceptron (Blatz et al., 2003), Naive Bayes (Blatz et al., 2003; =-=Sanchis et al., 2007-=-), or loglinear model (Ueffing and Ney, 2007). 3) Divide words into two groups (correct translations and errors) by using a classification threshold optimized on a development set. Sometimes the step ... |

4 | Error Detection Using Linguistic Features
- Shi, Zhou
- 2005
(Show Context)
Citation Context ...and Ney, 2007; Sanchis et al., 2007; Raybaud et al., 2009). Experimental results show that they are useful for error detection. However, it is not adequate to just use these features as discussed in (=-=Shi and Zhou, 2005-=-) because the information that they carry is either from the inner components of SMT systems or from system outputs. To some extent, it has already been considered by SMT systems. Hence finding extern... |

2 | Maximum Entropy Modeling Tooklkit for Python and C++. Available at http://homepages.inf.ed.ac.uk/s0450736 /maxent toolkit.html - Zhang - 2004 |

1 | Eiichiro Sumitay, Hiromi Nakaiway, Seiichi Yamamotoy, and Hiroshi G. Okunoz. 2004. Using a Mixture of N-best Lists from Multiple MT Systems in Rank-sum-based Confidence Measure for MT Outputs - Akibay |

1 | 2006 Corpus-based Machine Translation Evaluation via Automated Error Detection in Output Texts - Elliott |