Evaluation of the impact of controlled language on neural machine translation compared to other MT architectures

Examining the general impact of Controlled Language (CL) rules in the context of Machine Translation (MT) has been an area of research for many years. The present study focuses on the following question: how do CL rules impact MT output individually? By analysing a German corpus-based test suite of technical texts that have been translated into English by different MT systems, this study endeavours to answer this question at different levels: the general impact of CL rules (rule- and system-independent), their impact at rule level (system-independent) as well as at rule and system level. The results of five MT systems are analysed and contrasted: a rule-based system, a statistical system, two differently constructed hybrid systems, and a neural system. For this, a mixed-methods triangulation approach that includes error annotation, human evaluation, and automatic evaluation was applied. The data was analysed both qualitatively and quantitatively in terms of CL influence on the following parameters: number and type of MT errors, style and content quality, and scores of two automatic evaluation metrics. In line with many studies, the results show a general positive impact of the applied CL rules on the MT output. However, at rule level, only four rules proved to have positive effects on the aforementioned parameters; three rules had negative effects on the parameters; and two rules did not show any significant impact. At rule and system level, the rules affected the MT systems differently, as expected. Rules that had a positive impact on earlier MT approaches did not show the same impact on the neural MT approach. Furthermore, neural MT delivered distinctly better results than earlier MT approaches, namely the highest error-free, style and content quality rates both before and after applying the rules, which indicates that neural MT offers a promising solution that no longer requires CL rules for improving the MT output.

The Centers for Medicare & Medicaid Services Incentive Programs promote meaningful use of electronic health records (EHRs), which, among many benefits, allow patients to receive electronic copies of their EHRs and thereby empower them to take a more active role in their health. In the United States, however, 17% population is Hispanic, of which 50% has limited English language skills. To help this population take advantage of their EHRs, we are developing English-Spanish machine translation (MT) systems for EHRs. In this study, we first built an English-Spanish parallel corpus and trained NoteAidSpanish, a statistical MT (SMT) system. Google Translator and Microsoft Bing Translator are two baseline MT systems. In addition, we evaluated hybrid MT systems that first replace medical jargon in EHR notes with lay terms and then translate the notes with SMT systems. Evaluation on a small set of EHR notes, our results show that Google Translator outperformed NoteAidSpanish. The hybrid SMT systems first map medical jargon to lay language. This step improved the translation. A fully implemented hybrid MT system is available at http://www.clinicalnotesaid.org. The English-Spanish parallel-aligned MedlinePlus corpus is available upon request.

This paper proposes a hybrid machine translation (HMT) system that improves the quality of neural machine translation (NMT) by incorporating statistical machine translation (SMT). Therefore, two NMT systems and two SMT systems were built for the Slovenian–English language pair, each for translation in one direction. We used a multilingual language model to embed the source sentence and translations into the same vector space. From each vector, we extracted features based on the distances and similarities calculated between the source sentence and the NMT translation, and between the source sentence and the SMT translation. To select the best possible translation, we used several well-known classifiers to predict which translation system generated a better translation of the source sentence. The proposed method of combining SMT and NMT in the hybrid system is novel. Our framework is language-independent and can be applied to other languages supported by the multilingual language model. Our experiment involved empirical applications. We compared the performance of the classifiers, and the results demonstrate that our proposed HMT system achieved notable improvements in the BLEU score, with an increase of 1.5 points and 10.9 points for both translation directions, respectively.

In this presentation, I would like to introduce the research and products of machine translation in Baidu. As the biggest Chinese search engine, Baidu has released its machine translation system in June, 2011. It now supports translations among 27 languages on multiple platforms, including PC, mobile devices, etc. Hybrid translation approach is important for building an Internet translation system. As we know, the translation demands on the Internet come from various domains, including news wires, patents, poems, idioms, etc. It is difficult for a single translation system to achieve high accuracy on all domains. Therefore, hybrid translation is practically needed. Generally, we build a statistical machine translation (SMT) system, using the training corpora automatically crawled from the web. For the translation of idioms (e.g. “有 志者,事竟成,where there is a will, there is a way”), hot words/expressions (e.g. “一带一路, One Belt and One Road ”), example-based translation methods are used. To improve the translation of date (e.g. “2012年7月6日, July 6, 2012”), numbers (e.g. “三千五百万, thirty-five million), etc, rule-based methods are used as pre-process. To improve translation quality for the resourcepoor language pairs, we used pivot-based methods. Wu and Wang (2007) proposed the triangulation method that combines the source-pivot and the pivot-target phrase tables to induce a sourcetarget phrase table. To fill up the data gap between the source-pivot and pivot-target corpora, Wu and Wang (2009) employed a hybrid method combining RBMT and SMT systems. We also proposed a method to use a Markov random walk to discover implicit relations between phrases in the source and target languages (Zhu et al., 2013), thus to improve the coverage of phrase pairs. We utilized the co-occurrence frequency of source-target phrase pairs to estimate phrase translation probabilities (Zhu et al., 2014). On May 20th this year, we have launched a neural machine translation (NMT) system for Chinese-English translation. The system conducts end-to-end translation with a source language encoder and a target language decoder. Both the encoder and decoder are recurrent neural networks. The strength of NMT lies in that it can learn semantic and structural translation information by taking global contexts into account. We further integrated the SMT and NMT system to improve translation quality. We also released off-line translation packs for NMT system on mobile devices, providing translation services in case that the Internet is unavailable. So far as we know, this is the first NMT system supporting off-line translation on mobile devices. We also investigate the problem of learning a machine translation model that can simultaneously translate sentences from one source language to multiple target languages (Dong et al., 2015). Our solution is inspired by the recently proposed neural machine translation model which generalizes machine translation as a sequence learning problem. We train a unified neural machine translation model under the multi-task learning framework where the encoder is shared across different language pairs and each target language has a separate decoder. This model gets faster and better convergence for both resource-rich and resourcepoor language pairs under the multi-task learning framework. Based on the above techniques, we have released translation products for multiple platforms, including web translation on PC, APP on mobile devices, as well as free API for the thirdparty developers. Our system now support translations among 27 languages, not only including many frequently-used foreign languages, but also

This chapter describes how semi-parallel and parallel data extracted from comparable corpora can be used in enhancing machine translation (MT) systems: what are the methods used for this task in statistical and rule-based machine translation systems; what kinds of showcases exist that illustrate the usage of such enhanced MT systems. The impact of data extracted from comparable corpora on MT quality is evaluated for 17 language pairs, and detailed studies involving human evaluation are carried out for 11 language pairs. At first, baseline statistical machine translation (SMT) systems were built using traditional SMT techniques. Then they were improved by the integration of additional data extracted from the comparable corpora. Comparative evaluation was performed to measure improvements. Comparable corpora were also used to enrich the linguistic knowledge of rule-based machine translation (RBMT) systems by applying terminology extraction technology. Finally, SMT systems were adjusted for a narrow domain and included domain-specific knowledge such as terminology, named entities (NEs), domain-specific language models (LMs), etc.

In this paper, we propose a new method for domain adaptation in Statistical Machine Translation for low-resource domains in English-Vietnamese language. Specifically, our method only uses monolingual data to adapt the translation phrase-table, our system brings improvements over the SMT baseline system. We propose two steps to improve the quality of SMT system: (i) classify phrases on the target side of the translation phrase-table use the probability classifier model, and (ii) adapt to the phrase-table translation by recomputing the direct translation probability of phrases.
 
 Our experiments are conducted with translation direction from English to Vietnamese on two very different domains that are legal domain (out-of-domain) and general domain (in-of-domain). The English-Vietnamese parallel corpus is provided by the IWSLT 2015 organizers and the experimental results showed that our method significantly outperformed the baseline system. Our system improved on the quality of machine translation in the legal domain up to 0.9 BLEU scores over the baseline system,…
 Keywords: 
 Machine Translation, Statistical Machine Translation, Domain Adaptation
 References
 [1] Philipp Koehn, Franz Josef Och, Daniel Marcu, Statistical phrase-based translation, In Proceedings of HLT-NAACL, Edmonton, Canada, 2003, 127-133.
 [2] Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Łukasz Kaiser, Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes and Jeffrey Dean, Google’s neural machine translation system: Bridging the gap between human and machine translation, CoRR, abs/1609.08144, 2016.
 [3] Luisa Bentivogli, Arianna Bisazza, Mauro Cettolo and Marcello Federico, Neural versus phrase-based machine translation quality: A case study, 2016.
 [4] Barry Haddow, Philipp Koehn, Analysing the effect of out-of-domain data on smt systems, In Proceedings of the Seventh Workshop on Statistical Machine Translation, 2012, 422-432.
 [5] Boxing Chen, Roland Kuhn and George Foster, Vector space model for adaptation in statistical machine translation, Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, 2013, pp. 1285-1293.
 [6] Daniel Dahlmeier, Hwee Tou Ng, Siew Mei Wu4, Building a large annotated corpus of learner english: The nus corpus of learner english, In Proceedings of the NAACL Workshop on Innovative Use of NLP for Building Educational Appli-cations, 2013.
 [7] Eva Hasler, Phil Blunsom, Philipp Koehn and Barry Haddow, Dynamic topic adaptation for phrase-based mt, In Proceedings of the 14th Conference of the European Chapter of The Association for Computational Linguistics, 2014, pp. 328-337.
 [8] George Foster, Roland Kuhn, Mixture-model adaptation for smt, Proceedings of the Second Workshop on Statistical Machine Translation, Prague, Association for Computational Linguistics, 2007, pp. 128-135.
 [9] George Foster, Boxing Chen, Roland Kuhn, Simulating discriminative training for linear mixture adaptation in statistical machine translation, Proceedings of the MT Summit, 2013.
 [10] Hoang Cuong, Khalil Sima’an, and Ivan Titov, Adapting to all domains at once: Rewarding domain invariance in smt, Proceedings of the Transactions of the Association for Computational Linguistics (TACL), 2016.
 [11] Ryo Masumura, Taichi Asam, Takanobu Oba, Hirokazu Masataki, Sumitaka Sakauchi, and Akinori Ito, Hierarchical latent words language models for robust modeling to out-of domain tasks, Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, 2015, pp. 1896-1901.
 [12] Chenhui Chu, Raj Dabre, and Sadao Kurohashi. An empirical comparison of simple domain adaptation methods for neural machine translation, 2017.
 [13] Markus Freitag, Yaser Al-Onaizan, Fast domain adaptation for neural machine translation, 2016.
 [14] Jia Xu, Yonggang Deng, Yuqing Gao and Hermann Ney, Domain dependent statistical machine translation, In Proceedings of the MT Summit XI, 2007, pp. 515-520.
 [15] Hua Wu, Haifeng Wang Chengqing Zong, Domain adaptation for statistical machine translation with domain dictionary and monolingual corpora, In Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), Manchester, UK, 2008, pp. 993-1000.
 [16] Adam Berger, Stephen Della Pietra, and Vincent Della Pietra, A maximum entropy approach to natural language processing, Computational Linguistics, 22, 1996.
 [17] 18Santanu Pal, Sudip Naskar, Josef Van Genabith, Uds-sant, English-German hybrid machine translation system, In Proceedings of the Tenth Workshop on Statistical Machine Translation, Lisbon, Portugal, September, Association for Computational Linguistics, 2015, pp. 152-157.
 [18] Louis Onrust, Antal van den Bosch, Hugo Van hamme, Improving cross-domain n-gram language modelling with skipgrams, Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, Berlin, Germany, 2016, pp. 137-142.
 [19] Mark Aronoff, Kirsten Fudeman, What is morphology, V 8. john wiley and sons, 2011.
 [20] Laurence C. Thompson, The problem of the word in vietnamese, In journal of the International Linguistic Association 19(1) (1963) 39-52. https:// doi.org/1080/00437956.1963.11659787.
 [21] Binh N. Ngo, The Vietnamese language learning framework, Journal of Southeast Asian Language Teaching 10 (2001) 1-24.
 [22] Le Hong Phuong, Nguyen Thi Minh Huyen, Azim Roussanaly, Ho Tuong Vinh, A hybrid approach to word segmentation of vietnamese texts, 2008.
 [23] Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondřej Bojar, Alexandra Constantin, Evan Herbst, Moses: Open source toolkit for statistical machine translation, In ACL-2007: Proceedings of demo and poster sessions, Prague, Czech Republic, 2007, pp.177-180.
 [24] Franz Josef Och, Minimum error rate training in statistical machine translation, In Proceedings of ACL, 2003, pp.160-167.
 [25] Andreas Stolcke, Srilm - an extensible language modeling toolkit, in proceedings of international conference on spoken language processing, 2002.
 [26] Papineni, Kishore, Salim Roukos, Todd Ward, WeiJing Zhu, Bleu: A method for automatic evaluation of machine translation, ACL, 2002.
 [27] G. Klein, Y. Kim, Y. Deng, J. Senellart, A.M. Rush, OpenNMT: Open-Source Toolkit for Neural Machine Translation. ArXiv e-prints.
 [28] Pratyush Banerjee, Jinhua Du, Baoli Li, Sudip Kr. Naskar, Andy Way and Josef van Genabith, Combining multi-domain statistical machine translation models using automatic classifiers, In Proceedings of AMTA 2010., 2010.

This paper presents a comparison of post-editing (PE) changes performed on English-to-Finnish neural (NMT), rule-based (RBMT) and statistical machine translation (SMT) output, combining a product-based and a process-based approach. A total of 33 translation students acted as participants in a PE experiment providing both post-edited texts and edit process data. Our product-based analysis of the post-edited texts shows statistically significant differences in the distribution of edit types between machine translation systems. Deletions were the most common edit type for the RBMT, insertions for the SMT, and word form changes as well as word substitutions for the NMT system. The results also show significant differences in the correctness and necessity of the edits, particularly in the form of a large number of unnecessary edits in the RBMT output. Problems related to certain verb forms and ambiguity were observed for NMT and SMT, while RBMT was more likely to handle them correctly. Process-based comparison of effort indicators shows a slight increase of keystrokes per word for NMT output, and a slight decrease in average pause length for NMT compared to RBMT and SMT in specific text blocks. A statistically significant difference was observed in the number of visits per sub-segment, which is lower for NMT than for RBMT and SMT. The results suggest that although different types of edits were needed to outputs from NMT, RBMT and SMT systems, the difference is not necessarily reflected in process-based effort indicators.

Corpus-based machine translation (MT) has been the main approach to developing and implementing MT systems in both academia and the industry over the last three decades. In this field, the type and size of the corpus used for training MT engines have presented problems for both statistical MT (SMT) systems as well as neural MT (NMT) systems, being the two dominant corpusbased approaches. Moreover, language pairs such as Turkish-English have been understudied within this framework. This article aims to evaluate the translation quality in Turkish-to-English custom MT systems that have been trained on different corpus sizes and types. Two NMT engines and two SMT engines were trained on the KantanMT platform using two different training corpus types with either only domain-specific cardiology corpus or this corpus plus a mixed-domain corpus. The study conducted both automatic evaluations with metrics including BLEU, F-Measure and TER, as well as a comprehensive human evaluation with metrics including fluency, A/B test, and adequacy. Lastly, the study realized a separate, subjective terminology evaluation in order to investigate how differently MT systems handle terminology, as this is a crucial aspect for specific-domain text types such as cardiology. While the automatic evaluation results suggest the SMT engines to perform better than NMT engines, all human evaluators rated the mixed-domain NMT engine as the highest performing one. However, the terminology evaluation task demonstrated SMT to still be able to perform better and to commit less terminology errors, despite the industry and academia shifting toward NMT engines.

<p><strong> </strong>It is practically impossible for pure machine translation approach to process all of translation problems; however, Rule Based Machine Translation and Statistical Machine translation (RBMT and SMT) use different architectures for performing translation task. Lexical analyser and syntactic analyser are solved by Rule Based and some amount of ambiguity is left to be solved by Expectation–Maximization (EM) algorithm, which is an iterative statistic algorithm for finding maximum likelihood. In this paper we have proposed an integrated Hybrid Machine Translation (HMT) system. The goal is to combine the best properties of each approach. Initially, Arabic text is keyed into RBMT; then the output will be edited by EM algorithm to generate the final translation of English text. As we have seen in previous works, the performance and enhancement of EM algorithm, the key of EM algorithm performance is the ability to accurately transform a frequency from one language to another. Results showing that, as proved by BLEU system, the proposed method can substantially outperform standard Rule Based approach and EM algorithm in terms of frequency and accuracy. The results of this study have been showed that the score of HMT system is higher than SMT system in all cases. When combining two approaches, HMT outperformed SMT in Bleu score.</p>

The present article focuses on improving the performance of a hybrid Machine Translation (MT) system, namely PRESEMT. The PRESEMT methodology is readily portable to new language pairs, and allows the creation of MT systems with minimal reliance on expensive resources. PRESEMT is phrase-based and uses a small parallel corpus from which to extract structural transformations from the source language (SL) to the target language (TL). On the other hand, the TL language model is extracted from large monolingual corpora. This article examines the task of maximising the amount of information extracted from a very limited parallel corpus. Hence, emphasis is placed on the module that learns to segment into phrases arbitrary input text in SL, by extrapolating information from a limited-size parsed TL text, alleviating the need for an SL parser. An established method based on Conditional Random Fields (CRF) is compared here to a much simpler template-matching algorithm to determine the most suitable approach for extracting an accurate model. Experimental results indicate that for a limited-size training set, template-matching generates a superior model leading to higher quality translations.

In this paper, we present our work on a case study between statistical machine translation (SMT) and rule-based machine translation (RBMT) systems on English-Indian language and Indian to Indian language perspective. Main objective of our study is to make a five-way performance comparison: such as, (a) SMT and RBMT; (b) SMT on English–Indian language; (c) RBMT on English–Indian language; (d) SMT on Indian to Indian language perspective; (e) RBMT on Indian to Indian language perspective. Through a detailed analysis, we describe the rule-based and the statistical machine translation system developments and its evaluations. Further, with a detailed error analysis, we point out the relative strengths and weaknesses of both the systems. The observations based on our study are: (a) SMT systems outperform RBMT; (b) In the case of SMT: English to Indian language MT systems perform better than Indian to English language MT systems; (c) In the case of RBMT: English to Indian language MT systems perform better than Indian to English language MT systems; (d) SMT systems perform better for Indian to Indian language MT systems compared to RBMT. Effectively, we shall see that even with a small amount of training corpus SMT system has many advantages for high-quality domain-specific machine translation over that of a rule-based counterpart.

Access to information is increasingly global, which brings with it the growth in a non-English speaking public and, as such, a demand for tools that allow users to access this information. Faced with this scenario, the development of new tools and technologies is essential. This is the case of machine translation (MT), one of the most popular and used tools to overcome linguistic barriers. As a result, linguistic companies are increasingly offering services related to MT. In addition, MT systems can be integrated with other tools and resources such as Translation Memories or multilingual terminology databases in order to improve the efficacy of the translation process. In this chapter an overview of MT development since its early beginnings in the late 1950s until the present day is proposed. The evolution of MT techniques will be reviewed starting from the traditional approaches—statistical and rule-based MT—and finishing with the latest trends, which include hybrid MT systems, adaptive and neural MT.

Machine Translation has been a field of study for over six decades, but it has acquired substantial prominence in the last decade as processing capacity in personal computers has increased. The purpose of this paper is to discuss the usage of Sanskrit as a source, target, or supporting language in various Machine Translation systems. To investigate Machine Translation, researchers use a variety of strategies, including corpus-based, direct, and rule-based approaches. The primary goal of employing Sanskrit in Machine Translation is to evaluate its appropriateness, lexicon, and performance when proper Machine Translation methods are used. The research examines various modelling strategies for developing a machine translation system, specifically Statistical and Neural Machine Translation, in order to bridge the gap between Sanskrit and its current successor, Hindi. Interpretations are formed in Statistical Machine Translation by matching words from the source and target languages with statistical models and bilingual text corpora to learn parameters. Neural Machine Translation, on the other hand, uses an artificial neural network to predict the likelihood of a word sequence, frequently modelling entire phrases within a single integrated model. Neural Machine Translation is implemented using an encoder-decoder architecture with an attention mechanism. One of the most significant contributions of this paper is the use of different data sources, data collecting, and scraping to create a complete dataset. According to the study's findings, Neural Machine Translation outperforms the Statistical Machine Translation modelling technique. Furthermore, the paper examines the distinctive qualities of the Sanskrit language as well as the difficulties encountered by researchers in digesting Sanskrit while constructing the machine translation system. This study investigates the use of Sanskrit in Machine Translation and analyses several modelling methods, such as Statistical and Neural Machine Translation. The paper emphasizes the advantages of Neural Machine Translation and discusses the unique characteristics and challenges of the Sanskrit language in machine translation development.

In general, advances in translation technology tools have enhanced translation quality significantly. Unfortunately, however, it seems that this is not the case for all language pairs. A concern arises when the users of translation tools want to work between different language families such as Arabic and English. The main problems facing Arabic<>English translation tools lie in Arabic’s characteristic free word order, richness of word inflection – including orthographic ambiguity – and optionality of diacritics, in addition to a lack of data resources. The aim of this study is to compare the performance of translation memory (TM) and machine translation (MT) systems in translating between Arabic and English.The research evaluates the two systems based on specific criteria relating to needs and expected results. The first part of the thesis evaluates the performance of a set of well-known TM systems when retrieving a segment of text that includes an Arabic linguistic feature. As it is widely known that TM matching metrics are based solely on the use of edit distance string measurements, it was expected that the aforementioned issues would lead to a low match percentage. The second part of the thesis evaluates multiple MT systems that use the mainstream neural machine translation (NMT) approach to translation quality. Due to a lack of training data resources and its rich morphology, it was anticipated that Arabic features would reduce the translation quality of this corpus-based approach. The systems’ output was evaluated using both automatic evaluation metrics including BLEU and hLEPOR, and TAUS human quality ranking criteria for adequacy and fluency.The study employed a black-box testing methodology to experimentally examine the TM systems through a test suite instrument and also to translate Arabic English sentences to collect the MT systems’ output. A translation threshold was used to evaluate the fuzzy matches of TM systems, while an online survey was used to collect participants’ responses to the quality of MT system’s output. The experiments’ input of both systems was extracted from Arabic<>English corpora, which was examined by means of quantitative data analysis. The results show that, when retrieving translations, the current TM matching metrics are unable to recognise Arabic features and score them appropriately. In terms of automatic translation, MT produced good results for adequacy, especially when translating from Arabic to English, but the systems’ output appeared to need post-editing for fluency. Moreover, when retrievingfrom Arabic, it was found that short sentences were handled much better by MT than by TM. The findings may be given as recommendations to software developers.

Machine translation (MT) outputs are widely scored using automatic evaluation metrics and human evaluation scores. The automatic evaluation metrics are expected to be easily computable and a reflection of human evaluation. Traditional string-based metrics such as BLEU, ChrF++ scores, are widely used to evaluate MT systems, but fail to account for synonyms that appear in the state-of-the-art neural machine translation (NMT) systems, owing to their inability to evaluate paraphrases. While similarity-based metrics such as Yisi, BERTScore address this issue, these metrics need to be modified to better evaluate morphologically rich Indian languages such as, Tamil and Hindi. The current work proposes a novel and individual sentence-BERT based similarity (SBSim) metric, that makes use of a paraphrase-BERT model and sentence-level embedding to evaluate NMT outputs. The effectiveness of the BLEU, ChrF++, Yisi, BERTScore, and the proposed SBSim are evaluated on English-to-Tamil and English-to-Hindi NMT outputs. The sentence-level metric correlation of the proposed SBSim metric with respect to human scores is observed to outperform the existing metrics with a correlation of 0.9123 and 0.9052 for English-to-Tamil and English-to-Hindi NMT systems, respectively. Further, the average metric correlation of the SBSim metric is also observed to be the highest with a value of 0.9801 and 0.9836 for these NMT systems, respectively. The proposed metric is also evaluated on WMT2020 dataset and reports the highest correlation of 0.7129 with the human scores.