Figure 2 - uploaded by Nadir Durrani
Content may be subject to copyright.
Sentence distribution for the English→German newstest2013 test set comparing system combination output against the best individual system.
Source publication
This paper describes one of the col-laborative efforts within EU-BRIDGE to further advance the state of the art in machine translation between two Euro-pean language pairs, German→English and English→German. Three research institutes involved in the EU-BRIDGE project combined their individual machine translation systems and participated with a join...
Context in source publication
Context 1
... for the German→English translation direction, the best performing individual system outputs are also having the highest BLEU scores when evalu- ated against the final system combination output. In Figure 2 system combination output is com- pared to the best single system pbt 2. We distribute the sentence-level BLEU scores of all sentences of newstest2013. Many sentences have been im- proved by system combination. ...
Similar publications
La démonstration de résumé et de traduction automatique de vidéos résulte de nos travaux dans le projet AMIS. L’objectif du projet était d’aider un voyageur à comprendre les nouvelles dans un pays étranger. Pour cela, le projet propose de résumer et traduire automatiquement une vidéo en langue étrangère (ici, l’arabe). Un autre objectif du projet é...
Citations
... In 2014 the EU-BRIDGE project reported that they achieved significantly better translation performance with gains of up to +1.6 points in BLEU and -1.0 points in TER by combining up to nine different machine translation systems for translation between German and English [14]. Recently Freitag et al. [15] presented novel system combination approach that enhance traditional confusion network system combination approach with an additional model trained by a neural network. ...
... The outputs of all these individual engines are combined using the system combination approach as implemented in Jane, RWTH's open source statistical machine translation toolkit (Freitag et al., 2014a). The Jane system combination is a mature implementation which previously has been successfully employed in other collaborative projects and for different language pairs (Peter et al., 2016;Freitag et al., 2013Freitag et al., , 2014b. ...
... Stacked LSTM (Graves, 2013) 1.67 GF-LSTM (Chung et al., 2015) 1.58 Grid-LSTM 1.47 Layer-normalized LSTM (Chung et al., 2016a) 1.46 MI-LSTM (Wu et al., 2016b) 1.44 Recurrent Highway Networks (Srivastava et al., 2015) 1.42 Recurrent Memory Array Structures (Rocki, 2016) 1.40 HM-LSTM (Chung et al., 2016a) 1.40 Layer Norm HyperLSTM (Ha et al., 2016) 1.38 Large Layer Norm HyperLSTM (Ha et al., 2016) 1.34 ByteNet Decoder 1.33 (Zhou et al., 2016) 20.6 RNN Enc-Dec Att + local p + unk replace 20.9 (3) RNN Enc-Dec Att + BPE in + BPE out 19.98 (4) 21.72 (4) RNN Enc-Dec Att + BPE in + char out 21.33 (4) 23.45 (4) GNMT + char in + char out (Wu et al., 2016a) 22.8 ByteNet 18.9 21.7 (Freitag et al., 2014), result (2) is from (Williams et al., 2015), results (3) are from (Luong et al., 2015) and results (4) are from (Chung et al., 2016b) The ByteNet Decoder that we use for the result has 25 residual blocks split into five sets of five blocks each; for the five blocks in each set the dilation rates are, respectively, 1,2,4,8 and 16. The masked kernel has size 3. ...
We present a neural architecture for sequence processing. The ByteNet is a stack of two dilated convolutional neural networks, one to encode the source sequence and one to decode the target sequence, where the target network unfolds dynamically to generate variable length outputs. The ByteNet has two core properties: it runs in time that is linear in the length of the sequences and it preserves the sequences' temporal resolution. The ByteNet decoder attains state-of-the-art performance on character-level language modelling and outperforms the previous best results obtained with recurrent neural networks. The ByteNet also achieves a performance on raw character-level machine translation that approaches that of the best neural translation models that run in quadratic time. The implicit structure learnt by the ByteNet mirrors the expected alignments between the sequences.
... The outputs of all these individual engines are combined using the system combination approach as imple-mented in Jane, RWTH's open source statistical machine translation toolkit (Freitag et al., 2014a). The Jane system combination is a mature implementation which previously has been successfully employed in other collaborative projects and for different language pairs (Freitag et al., 2013;Freitag et al., 2014b;Freitag et al., 2014c). ...
... In 2014 the EU-BRIDGE project reported that they achieved significantly better translation performance with gains of up to +1.6 points in BLEU and -1.0 points in TER by combining up to nine different machine translation systems for translation between German and English [14]. Recently Freitag et al. [15] presented novel system combination approach that enhance traditional confusion network system combination approach with an additional model trained by a neural network. ...
This paper presents a hybrid machine translation (HMT) system that pursues syntactic analysis to acquire phrases of source sentences, translates the phrases using multiple online machine translation (MT) system application program interfaces (APIs) and generates output by combining translated chunks to obtain the best possible translation. The aim of this study is to improve translation quality of English – Latvian texts over each of the individual MT APIs. The selection of the best translation hypothesis is done by calculating the per-plexity for each hypothesis using an n-gram language model. The result is a phrase-based multi-system machine translation system that allows to improve MT output compared to individual online MT systems. The proposed approach show improvement up to +1.48 points in BLEU and-0.015 in TER scores compared to the baselines and related research. 1 Introduction Although machine translation (MT) has been researched for many decades and there are many on-line MT systems available, the output of MT systems in many cases still has low quality. This problem is especially relevant to the less resourced languages with rich morphology and rather free word order (e.g. Latvian, Croatian or Estonian) where linguistic and human resources are limited. As these are rather small languages, they are not the main interest of the global MT developers. Better solutions are provided by local companies for particular domain, e.g. hugo.lv 1 as Latvian e-government solution provides good translation of legal texts [1]. Although such translation systems usually translate better than global multilingual MT systems [2], translation quality for general domain texts is still insufficient. 1 Latvian public administration machine translation service-http://hugo.lv
... Our system is furthermore one component in a combination of systems by members of the HimL project and another EU-funded project, QT21. 3 Measured in BLEU (Papineni et al., 2002), the QT21/HimL submission yields top translation quality amongst the shared task submissions. 4 The QT21/HimL submission highlights the continued success of system combinations based on the Jane machine translation toolkit (Freitag et al., 2014a) in open evaluation campaigns (Freitag et al., 2013;Freitag et al., 2014b;Freitag et al., 2014c). A description of the QT21/HimL combined submission is given by Peter et al. (2016). ...
This paper describes the hierarchical phrase-based machine translation system built jointly by the University of Edinburgh and the University of Munich (LMU) for the shared translation task at the ACL 2016 First Conference on Machine Translation (WMT16). The WMT16 Edinburgh/LMU system was trained for translation of news domain texts from English into Romanian. We participated in the shared task for machine translation of news under “constrained” conditions, i.e. using the provided training data only.
... Combining several single machine translation engines has proven to be highly effective in previous submissions, e.g. (Freitag et al., 2013;Freitag et al., 2014a;Peter et al., 2015). We therefore used a similar approach for this evaluation. ...
This paper describes the statistical machine translation system developed at RWTH Aachen University for the English!Romanian translation task of the ACL 2016 First Conference on Machine Translation (WMT 2016). We combined three different state-ofthe-art systems in a system combination: A phrase-based system, a hierarchical phrase-based system and an attentionbased neural machine translation system. The phrase-based and the hierarchical phrase-based systems make use of a language model trained on all available data, a language model trained on the bilingual data and a word class language model. In addition, we utilized a recurrent neural network language model and a bidirectional recurrent neural network translation model for reranking the output of both systems. The attention-based neural machine translation system was trained using all bilingual data together with the backtranslated data from the News Crawl 2015
... This complexity creates a challenge for the phrase-based SMT (PBSMT) paradigm that only sees a sentence as a sequential structure. To overcome this problem, the output of PB-SMT systems can be combined with the output of "syntax-aware" MT systems (rule-based, syntaxbased, etc. (Freitag et al., 2014b;Avramidis et al., 2015;Li et al., 2015)). ...
... These limitations can be overcome in PBSMT by combining with the outputs of "syntax-aware" MT systems (rule-base MT (RBMT), syntax-based MT (SBMT)) (Costa-Jussà et al., 2012; Avramidis et al., 2015;Li et al., 2015). The combination of system outputs is often performed with the help of Confusion Network (CN) decoding as an effective means to recombine translation alternatives at the word level (Deng et al., 2015;Freitag et al., 2014a;Freitag et al., 2014b;Zhu et al., 2013). Less costly solutions seek to better explore the potential of phrase-based architectures. ...
... A combination of several single machine translation engines has proven to be highly effective on previous joint submission, e.g. [1,2], and a similar approach is used for this task. We train individual systems using state-of-the-art phrase-based and hierarchical phrase-based translation engines. ...
This work describes the statistical machine translation (SMT) systems of RWTH Aachen University developed for the evaluation campaign of the International Workshop on Spoken Language Translation (IWSLT) 2015. We participated in the MT and SLT tracks for the German→English language pair. We employ our state-of-the-art phrase-based and hierarchical phrase-based baseline systems for the MT track. The phrase-based system is augmented with joint translation and reordering model and maximum expected BLEU training for phrasal, lexical and reordering models. Furthermore, we apply feed-forward and recurrent neural language and translation models for reranking. We also train attention-based neural network models and utilize them in reranking the n-best lists for both phrase-based and hierarchical setups. On top of all our systems, we use system combination to enhance the translation quality by combining individually trained systems. In the SLT track, we additionally perform punctuation prediction on the automatic transcriptions employing hierarchical phrase-based translation.
... The Edinburgh/JHU phrase-based translation systems for our participation in the WMT 2015 shared translation task 1 are based on the open source Moses toolkit (Koehn et al., 2007). We built upon Edinburgh's strong baselines from WMT submissions in previous years as well as our recent research within the framework of other evaluation campaigns and projects such as IWSLT 2 and EU-BRIDGE 3 (Birch et al., 2014;Freitag et al., 2014a;Freitag et al., 2014b). ...
This paper describes the submission of the University of Edinburgh and the Johns Hopkins University for the shared translation task of the EMNLP 2015 Tenth Workshop on Statistical Machine Translation (WMT 2015). We set up phrase-based statistical machine translation systems for all ten language pairs of this year’s evaluation campaign, which are English paired with Czech, Finnish, French, German, and Russian in both translation directions. Novel research directions we investigated include: neural network language models and bilingual neural network language models, a comprehensive use of word classes, and sparse lexicalized reordering features.
