Edinburgh’s Phrase-based Machine Translation Systems for WMT-14

By Nadir Durrani, Barry Haddow, Philipp Koehn, Kenneth Heafield. 2014

56 citations recorded by the source corpus.

This paper describes the University of Edinburgh’s (UEDIN) phrase-based submissions to the translation and medical translation shared tasks of the 2014 Workshop on Statistical Machine Translation (WMT). We participated in all language pairs. We have improved upon our 2013 system by i) using generalized representations, specifically automatic word clusters for translations out of English, ii) using unsupervised character-based models to translate unknown words in RussianEnglish and Hindi-English pairs, iii) synthesizing Hindi data from closely-related Urdu data, and iv) building huge language on the common crawl corpus.

Subfields: Artificial Intelligence, Molecular Biology, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Molecular Biology

In Natural Language Processing

Citations in the corpus

Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer (2017, neutral): “(Wu et al., 2016) 2.79 39.22 214M 278M 6 days/96 k80s GNMT+RL (Wu et al., 2016) 2.96 39.92 214M 278M 6 days/96 k80s PBMT (Durrani et al., 2014) 37.0 LSTM (6-layer) (Luong et al., 2015b) 31.5 LSTM (6-layer+PosUnk) (Luong et al., 2015b) 33.1 DeepAtt (Zhou et al., 2016) 37.7 DeepAtt+PosUnk (Zhou et…”
Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation (2016, neutral): “Practical implementations of SMT are generally phrase-based systems (PBMT) which translate sequences of words or phrases where the lengths may differ [26].”
Deep Recurrent Models with Fast-Forward Connections for Neural Machine Translation (2016, methodology): “, 2015) while the conventional method of (Durrani et al., 2014) achieves 37.”
Sequence to Sequence Learning with Neural Networks (2014, neutral): “However, if we evaluate the best WMT’14 system [9] (whose predictions can be downloaded from statmt.”
Language Models are Few-Shot Learners (2020, neutral): “prior unsupervised NMT work. SacreBLEUf [Pos18] results reported in AppendixH. Underline indicates an unsupervised or few-shot SOTA, bold indicates supervised SOTA with relative conﬁdence. a[EOAG18] b[DHKH14] c[WXH+18] d[oR16] e[LGG+20] f [SacreBLEU signature: BLEU+case.mixed+numrefs.1+smooth.exp+to”
Findings of the 2014 Workshop on Statistical Machine Translation (2014, methodology): “First, we pit the WMT13 method against two new approaches: that of Hopkins and May (2013, § 3.3.2), and another based on TrueSkill (Sakaguchi et al., 2014, § 3.3.3).”
On Using Very Large Target Vocabulary for Neural Machine Translation (2014, methodology): “, 2015), (•) (Durrani et al., 2014), (∗) Standard Moses Setting (Cho et al.”
A Character-level Decoder without Explicit Segmentation for Neural Machine Translation (2016, neutral): “(3, 5) Durrani et al. (2014).”
Describing Multimedia Content Using Attention-Based Encoder-Decoder Networks (2015, background): “◦: THE STATE-OF-THE-ART PHRASE-BASED STATISTICAL MACHINE TRANSLATION SYSTEM [43].”
EU-BRIDGE MT: Combined Machine Translation (2014, methodology): “UEDIN’s phrase-based systems (Durrani et al., 2014) have been trained using the Moses toolkit (Koehn et al., 2007), replicating the settings described in (Durrani et al., 2013b).”
The Edinburgh/JHU Phrase-based Machine Translation Systems for WMT 2015 (2015, methodology): “In Edinburgh’s submission from the previous year, we used automatically generated word classes in additional language models and in additional operation sequence models (Durrani et al., 2014b).”
Edinburgh SLT and MT system description for the IWSLT 2014 evaluation (2014, methodology): “All available training corpora were utilized, with the exception of the MultiUN corpus and the WMT 109 French-English corpus, which we excluded from both the parallel and the LM training data.”
Vocabulary Manipulation for Neural Machine Translation (2016, background): “But our single system is still about 2 points behind of the best phrase-based system (Durrani et al., 2014).”
Farasa: A Fast and Furious Segmenter for Arabic (2016, methodology): “We used an unsupervised transliteration model (Durrani et al., 2014c) to transliterate the OOV words.”
Understanding and Improving Morphological Learning in the Neural Machine Translation Decoder (2017, methodology): “There have been previous efforts to integrate morphology into MT systems by learning factored models (Koehn and Hoang, 2007; Durrani et al., 2014b) over POS and morphological tags.”
Improving machine translation via triangulation and transliteration (2014, unknown): “…additionally used hierarchical lexicalized reordering (Galley and Manning, 2008), a 5-gram OSM, (Durrani et al., 2013), sparse lexical and domain features, (Hasler et al., 2012), class-based models (Durrani et al., 2014b), a distortion limit of 6, and the no-reordering-over-punctuation heuristic.”