A Simple and Effective Hierarchical Phrase Reordering Model

By Michel Galley, Christopher D. Manning. 2008

380 citations recorded by the source corpus.

While phrase-based statistical machine translation systems currently deliver state-of-the-art performance, they remain weak on word order changes. Current phrase reordering models can properly handle swaps between adjacent phrases, but they typically lack the ability to perform the kind of long-distance re-orderings possible with syntax-based systems. In this paper, we present a novel hierarchical phrase reordering model aimed at improving non-local reorderings, which seamlessly integrates with a standard phrase-based system with little loss of computational efficiency. We show that this model can successfully handle the key examples often used to motivate syntax-based systems, such as the rotation of a prepositional phrase around a noun phrase. We contrast our model with reordering models commonly used in phrase-based systems, and show that our approach provides statistically significant BLEU point gains for two language pairs: Chinese-English (+0.53 on MT05 and +0.71 on MT08) and Arabic-English (+0.55 on MT05).

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): “Table 8: Hierarchical lexicalized reordering model (Galley and Manning, 2008).”
A Challenge Set Approach to Evaluating Machine Translation (2017, methodology): “For each extracted phrase pair, we collected statistics for the hierarchical re-ordering model of Galley and Manning (2008).”
EU-BRIDGE MT: Combined Machine Translation (2014, methodology): “Both systems have been augmented with a hierarchical orientation model (Galley and Manning, 2008; Huck et al., 2013) and a cluster lan-”
A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU (2014, methodology): “We use the hierarchical lexicalized reordering model (RM) (Galley and Manning, 2008), with a distortion limit of 7.”
On the Impact of Various Types of Noise on Neural Machine Translation (2018, methodology): “…of GIZA++ alignments, an interpolated Kneser-Ney smoothed 5-gram language model with KenLM (Heaﬁeld, 2011), hierarchical lexicalized reordering (Galley and Manning, 2008), a lexically-driven 5-gram operation sequence model (OSM) (Durrani et al., 2013), sparse domain indicator, phrase length,…”
The Edinburgh/JHU Phrase-based Machine Translation Systems for WMT 2015 (2015, methodology): “The experiments were conducted on top of the standard hierarchical lexicalized reordering model (Galley and Manning, 2008).”
The Edinburgh/LMU Hierarchical Machine Translation System for WMT 2016 (2016, background): “Similarly to the type of lexicalized reordering models that are in common use in phrase-based systems (Galley and Manning, 2008), our model estimates the probabilities of orientation classes for each phrase (or: rule) from the training data.”
Results of the WMT16 Tuning Shared Task (2016, background): “, 2005) and the other is hierarchical (Galley and Manning, 2008).”
Phrasal: A Toolkit for New Directions in Statistical Machine Translation (2014, methodology): “The standard phrase orientation model (Tillmann, 2004) and the hierarchical model of Galley and Manning (2008) are available.”
Abu-MaTran at WMT 2016 Translation Task: Deep Learning, Morphological Segmentation and Tuning on Character Sequences (2016, methodology): “Model (Durrani et al., 2011) and a Bilingual Neural Language Model (Devlin et al., 2014), as well as three reordering models: word-and phrase-based and hierarchical (Koehn et al., 2005; Galley and Manning, 2008).”
Edinburgh’s Statistical Machine Translation Systems for WMT16 (2016, methodology): “We used the hierarchical lexicalized reordering model (Galley and Manning, 2008) with 4 possible orientations (monotone, swap, discontinuous left and discontinuous right) in both left-to-right and right-to-left direction.”
JU-USAAR: A Domain Adaptive MT System (2016, methodology): “The reordering model was trained with the hierarchical, monotone, swap, left to right bidirectional (hier-mslr-bidirectional) (Galley and Manning, 2008) method and conditioned on both the source”
USAAR: An Operation Sequential Model for Automatic Statistical Post-Editing (2016, methodology): “The reordering model was trained with the hierarchical, monotone, swap, left to right bidirectional (hiermslr-bidirectional) method (Galley and Manning, 2008) and conditioned on both source and target language.”
The RWTH Aachen German-English Machine Translation System for WMT 2014 (2014, methodology): “Additional models used in this evaluation are the hierarchical reordering model (HRM) (Galley and Manning, 2008) and a word class language model (wcLM) (Wuebker et al.”
Improving machine translation via triangulation and transliteration (2014, positive): “Hindi-English: For Hindi-English systems, we additionally used hierarchical lexicalized reordering (Galley and Manning, 2008), a 5-gram OSM, (Durrani et al.”
The RWTH Aachen University English-Romanian Machine Translation System for WMT 2016 (2016, methodology): “Our baseline contains the following models: Phrase translation probabilities and lexical smoothing in both directions, word and phrase penalty, distance-based reordering model, n -gram target language models and enhanced low frequency feature (Chen et al., 2011), a hierarchical reordering model (HRM”
Machine Translation and Monolingual Postediting: The AFRL WMT-14 System (2014, background): “In contrast to our IWSLT 2013 system, all experiments submitted to this year’s WMT evaluation made use of version 2.1 of moses , and incorporated additional decoding features, namely the Operation Sequence Model (Durrani et al., 2011) and Lexicalized Reordering Model (Tillman, 2004; Galley and Manni”
Abu-MaTran at WMT 2015 Translation Task: Morphological Segmentation and Web Crawling (2015, neutral): “…models, namely a phrase-based model with three different orientations (monotone, swap and discontinuous) and a hierarchical model with four orientations (non merged discontinuous left and right orientations), both trained in a bidirectional way (Koehn et al., 2005; Galley and Manning, 2008).”