Integrating an Unsupervised Transliteration Model into Statistical Machine Translation

By Nadir Durrani, Hassan Sajjad, Hieu D. Hoang, Philipp Koehn. 2014

103 citations recorded by the source corpus.

We investigate three methods for integrating an unsupervised transliteration model into an end-to-end SMT system. We induce a transliteration model from parallel data and use it to translate OOV words. Our approach is fully unsupervised and language independent. In the methods to integrate transliterations, we observed improvements from 0.23-0.75 ( 0.41) BLEU points across 7 language pairs. We also show that our mined transliteration corpora provide better rule coverage and translation quality compared to the gold standard transliteration corpora.

Subfields: Artificial Intelligence, 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 Natural Language Processing

Citations in the corpus

Edinburgh’s Phrase-based Machine Translation Systems for WMT-14 (2014, methodology): “We explored the use of automatic word clusters in phrase-based models (Durrani et al., 2014a).”
What do Neural Machine Translation Models Learn about Morphology? (2017, methodology): “…increasingly popular in neural MT, although they had also been used in phrase-based MT for handling morphologically-rich (Luong et al., 2010) or closely related language pairs (Nakov and Tiedemann, 2012; Durrani and Koehn, 2014) or for handling the problem of OOV words (Durrani et al., 2014).”
How Grammatical is Character-level Neural Machine Translation? Assessing MT Quality with Contrastive Translation Pairs (2016, background): “(specifically, transliteration of names (Durrani et al., 2014)), and ensuring that polarity is maintained (Wetzel and Bond, 2012; Chen and Zhu, 2014; Fancellu and Webber, 2015).”
The AMU-UEDIN Submission to the WMT16 News Translation Task: Attention-based NMT Models as Feature Functions in Phrase-based SMT (2016, background): “Although SMT seems to be better equipped to handle large vocabularies, the case of Russian still poses problems which are usually solved with specialized transliteration mechanisms (Durrani et al., 2014).”
The Edinburgh/JHU Phrase-based Machine Translation Systems for WMT 2015 (2015, methodology): “We built upon Edinburgh’s strong baselines from WMT submissions in previous years (Durrani et al., 2014a) as well as our recent research within the framework of other evaluation campaigns and projects such as IWSLT2 and EU-BRIDGE3 (Birch et al., 2014; Freitag et al., 2014a; Freitag et al., 2014b).”
Edinburgh’s Syntax-Based Systems at WMT 2015 (2014, methodology): “This is the variant referred to as Method 2 in Durrani et al. (2014).”
PJAIT Systems for the WMT 2016 (2016, methodology): “UTM uses a transliteration phrase translation table to evaluate and score multiple possible transliterations (Durrani et al., 2014).”
Improving machine translation via triangulation and transliteration (2014, unknown): “were translated using in-decoding transliteration method (Durrani et al., 2014c).”
One Size Does Not Fit All: Comparing NMT Representations of Different Granularities (2019, methodology): “, 2016), morphological segmentation (Bradbury and Socher, 2016), characters (Durrani et al., 2014; Lee et al., 2017), and hybrid units (Ling et al.”
The AFRL-MITLL WMT16 News-Translation Task Systems (2016, background): “…IWSLT2015 (Kazi et al., 2015) and WMT15 (Gwinnup et al., 2015), the introduction of Neural Machine Translation rescoring, neural-net based recasing, unsupervised transliteration of out-of-vocabulary (OOV) words (Durrani et al., 2014), and an unique selection process for language modelling data.”
Edinburgh’s Syntax-Based Systems at WMT 2014 (2014, methodology): “Assuming a signiﬁcant fraction of OOV words to be named entities and thus amenable to translit-eration, we applied the post-processing transliteration method described in Durrani et al. (2014) and implemented in Moses.”
N-gram Counts and Language Models from the Common Crawl (2014, methodology): “The Hindi-English system uses transliteration for unknown words (Durrani et al., 2014).”
NRC Russian-English Machine Translation System for WMT 2016 (2016, methodology): “Transliterations were encoded as translation rules with multiple scored alternatives, similar to the approach found to be optimal by Durrani et al. (2014).”
Edinburgh SLT and MT system description for the IWSLT 2014 evaluation (2014, methodology): “We built unsupervised transliteration models [40] to translate OOV words.”
Challenging Language-Dependent Segmentation for Arabic: An Application to Machine Translation and Part-of-Speech Tagging (2017, background): “, 2010; Nakov and Tiedemann, 2012) and OOV words (Durrani et al., 2014).”
Farasa: A Fast and Furious Segmenter for Arabic (2016, methodology): “Among other features, we used lexicalized reordering model (Galley and Manning, 2008), a 5-gram Operation Sequence Model (Durrani et al., 2011), Class-based Models (Durrani et al., 2014b)4 and other default parameters.”
On the Linguistic Representational Power of Neural Machine Translation Models (2019, background): “2010; Nakov and Tiedemann 2012) or for transliterating unknown words (Durrani et al. 2014b).”
Orthographic Syllable as basic unit for SMT between Related Languages (2016, methodology): “We used the target language side of the parallel corpora for character, morpheme and OS level LMs. System details: PBSMT systems were trained us-ing the Moses system (Koehn et al., 2007), with the grow-diag-final-and heuristic for extracting phrases, and Batch MIRA (Cherry and Foster, 2012) for tuni”
Investigating the Usefulness of Generalized Word Representations in SMT (2014, methodology): “We used an unsupervised transliteration model (Durrani et al., 2014) to transliterate OOV words when translating into Russian.”