Tuning as Ranking

By Mark Hopkins, Jonathan May. Conference on Empirical Methods in Natural Language Processing, 2011

331 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/a13d46125ef505d4e687e25ded74b794efc18323

How researchers see this paper

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

Citations in the corpus

Temporal Attention Model for Neural Machine Translation (2016, methodology): “We tune our SMT system with PRO (Hopkins and May, 2011) to minimize (T ER -”
Coverage Embedding Models for Neural Machine Translation (2016, methodology): “We tune our system with PRO (Hopkins and May, 2011) to minimize (T ER -”
Batch Tuning Strategies for Statistical Machine Translation (2012, methodology): “Introduced by Hopkins and May (2011), Pairwise Ranking Optimization (PRO) aims to handle large feature sets inside the traditional MERT architecture.”
NRC Russian-English Machine Translation System for WMT 2016 (2016, methodology): “The sparse feature model consists of the standard sparse features proposed in Hopkins and May (2011) and sparse hierarchical distortion model features proposed in Cherry (2013).”
BEER: BEtter Evaluation as Ranking (2014, methodology): “In this paper we follow the learning-to-rank approach which produces a local ranking model in a similar way to PRO MT systems tuning (Hopkins and May, 2011).”
Optimizing Statistical Machine Translation for Text Simplification (2016, methodology): “Training algorithms, such as MERT (Och, 2003) or PRO (Hopkins and May, 2011), then directly optimize the model parameters such that the end-to-end simpliﬁ-cation quality is optimal.”
Multi-Source Transformer for Kazakh-Russian-English Neural Machine Translation (2019, methodology): “Each SMT system also includes a hierachical distortion model, a sparse feature model consisting of the standard sparse features proposed in Hopkins and May (2011) and sparse hierarchical distortion model features proposed in Cherry (2013), and a neural network joint model, or NNJM, with 3 words of…”
BRIO: Bringing Order to Abstractive Summarization (2022, methodology): “To coordinate a pre-trained abstractive model, we 1) use it to generate different candidate summaries with various levels of quality,2 then 2) encourage the model to assign higher estimated probabilities to better candidates by fine-tuning the model with a contrastive loss, following the previous wo”
The AFRL-MITLL WMT15 System: There’s More than One Way to Decode It! (2015, methodology): “N-best lists from moses were rescored with 4-way NNJMs, and the system weights were tuned with PRO (Hopkins and May, 2011).”
Fitting Sentence Level Translation Evaluation with Many Dense Features (2014, methodology): “Our model is a linear combination of features trained for ranking similar to RankSVM (Herbrich et al., 1999) or, to readers familiar with SMT system tuning, to PRO tuning (Hopkins and May, 2011): where ~w represents a weight vector and ~x sys a vector of feature values for system output sys .”
Results of the WMT16 Tuning Shared Task (2016, methodology): “ILLC-UVA-BEER (Stanojević and Sima’an, 2015) was tuned using PRO (Hopkins and May, 2011) learning algorithm with new version of BEER evaluation metric.”
A Systematic Comparison of Smoothing Techniques for Sentence-Level BLEU (2014, methodology): “Most importantly, it includes the full set of sparse phrase-pair features used by both Hopkins and May (2011) and Cherry and Foster (2012), which results in nearly 7,000 features.”
Dynamic Data Selection for Neural Machine Translation (2017, methodology): “To create optimal PBMT systems given the available resources, we apply test-set-speciﬁc parameter tuning using PRO (Hopkins and May, 2011).”
Sheffield Systems for the English-Romanian WMT Translation Task (2016, methodology): “Derived from the Pairwise Ranking Optimisation (PRO) algorithm (Hopkins and May, 2011), WRO addresses various limitations of PRO, as we discuss in Section 4.”
QCRI-MES Submission at WMT13: Using Transliteration Mining to Improve Statistical Machine Translation (2013, background): “We call it PROv1 later on.”
Results of the WMT15 Tuning Shared Task (2015, methodology): “…of the available algorithms for direct loss optimization (Och, 2003; Zhao and Chen, 2009), which are usually capable of optimizing only a dozen of features, or one of the optimizers handling also very large sets of features (Cherry and Foster, 2012; Hopkins and May, 2011), or a custom algorithm.”