Batch Tuning Strategies for Statistical Machine Translation

By Colin Cherry, George F. Foster. 2012

405 citations recorded by the source corpus.

How researchers see this paper

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

In Natural Language Processing

Citations in the corpus

The NiuTrans Machine Translation Systems for WMT19 (2019, methodology): “We rescored 96-best outputs generated by several ensemble systems using a rescoring model consisting of features above by K-batched MIRA (Cherry and Foster, 2012) algorithm which is widely used in Moses4.”
Achieving Human Parity on Automatic Chinese to English News Translation (2018, methodology): “In order to combine the systems described above, we combine n-best hypotheses from all systems and then train a re-ranker using k-best MIRA on the validation set.”
A Challenge Set Approach to Evaluating Machine Translation (2017, methodology): “Tuning was carried out using batch lattice MIRA (Cherry and Foster, 2012).”
Summarizing Source Code using a Neural Attention Model (2016, methodology): “We train a 3-gram language model using KenLM (Heafield, 2011) to use with MOSES, and perform MIRA-based tuning (Cherry and Foster, 2012) of hyper-parameters using DEV.”
The IIT Bombay English-Hindi Parallel Corpus (2017, methodology): “We used the grow-diag-final-and heuristic for extracting phrases, lexicalised reordering and Batch MIRA (Cherry and Foster, 2012) for tuning (default parameters).”
A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU (2014, methodology): “For a sanity check, we compared these results to tuning with our very stable Batch k -best MIRA implementation (Cherry and Foster, 2012), listed as mira , which shows that all of our expected BLEU tuners are behaving reasonably, if not better than expected.”
On the Impact of Various Types of Noise on Neural Machine Translation (2018, methodology): “We optimize feature function weights with k-best MIRA (Cherry and Foster, 2012).”
The Edinburgh/JHU Phrase-based Machine Translation Systems for WMT 2015 (2015, methodology): “In future work we plan to adopt hypergraph MIRA, as well as larger training sets for maximum expected B LEU training.”
Integrating an Unsupervised Transliteration Model into Statistical Machine Translation (2014, methodology): “We tuned with the k-best batch MIRA (Cherry and Foster, 2012).”
Alibaba’s Neural Machine Translation Systems for WMT18 (2018, methodology): “, 2007) of K-batched MIRA algorithm (Cherry and Foster, 2012) to rerank the nbest list.”
Multi-Domain Neural Machine Translation through Unsupervised Adaptation (2017, methodology): “Feature weights were tuned with batch MIRA (Cherry and Foster, 2012) to maximize BLEU on the dev set.”
Edinburgh’s Syntax-Based Systems at WMT 2015 (2014, methodology): “Each system in Tables 5 and 6 was tuned separately with MIRA.”
Edinburgh’s Syntax-Based Systems at WMT 2014 (2014, methodology): “The feature weights were tuned using the Moses implementation of MERT (Och, 2003) for all systems except English-to-German, for which we used k-best MIRA (Cherry and Foster, 2012) due”
Edinburgh SLT and MT system description for the IWSLT 2014 evaluation (2014, methodology): “Model weights for the log-linear 49 Proceedings of the 11th International Workshop on Spoken Language Translation Lake Tahoe, December 4th and 5th, 2014 model combination [28] were optimized with batch k-best MIRA [29] to maximize BLEU [30].”
Farasa: A Fast and Furious Segmenter for Arabic (2016, methodology): “We tuned with the k-best batch MIRA (Cherry and Foster, 2012).”
The Edinburgh/LMU Hierarchical Machine Translation System for WMT 2016 (2016, methodology): “Rather than employing a linearly interpolated LM, we integrate the individual LMs trained over the separate corpora (news2015, Europarl, SETimes2) directly into the log-linear feature combination of the system and let M IRA optimize their weights along with all other features in tuning.”
Revisiting Low-Resource Neural Machine Translation: A Case Study (2019, methodology): “We use MGIZA (Gao and Vo-gel, 2008) to train word alignments, and lmplz (Heaﬁeld et al., 2013) for a 5-gram LM. Feature weights are optimized on the dev set to maximize BLEU with batch MIRA (Cherry and Foster, 2012) – we perform multiple runs where indicated.”