Practical Very Large Scale CRFs

By Thomas Lavergne, Olivier Capp'e, François Yvon. Annual Meeting of the Association for Computational Linguistics, 2010

437 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/f22485b565fac413856133fd07efa0b1ae2a1adf

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

Using subcategorization knowledge to improve case prediction for translation to German (2013, methodology): “For predicting the grammatical features, we used the Wapiti Toolkit (Lavergne et al., 2010).”
LIG System for WMT13 QE Task: Investigating the Usefulness of Features in Word Confidence Estimation for MT (2013, methodology): “, 2001) and the corresponding WAPITI toolkit (Lavergne et al., 2010) to train our classifier.”
LIMSI@WMT’15 : Translation Task (2015, methodology): “The ﬁrst two models use standard features for POS tagging as described in (Lavergne et al., 2010).”
Structured Sparsity in Structured Prediction (2011, methodology): “…and scalable algorithms having been proposed for training L 1 -regularized CRFs, based on quasi-Newton optimization (Andrew and Gao, 2007), coordinate descent (Sokolovska et al., 2010; Lavergne et al., 2010), and stochastic gradients (Carpenter, 2008; Langford et al., 2009; Tsuruoka et al., 2009).”
CimS - The CIS and IMS Joint Submission to WMT 2015 addressing morphological and syntactic differences in English to German SMT (2015, methodology): “For each of the four morphological features, we use a linear chain CRF (Lafferty et al. (2001)) trained on stems/lemmas and the respective feature, using the Wapiti toolkit (Lavergne et al., 2010).”
Can We Generate Emotional Pronunciations for Expressive Speech Synthesis? (2020, unknown): “CRFs are trained under cross-validation conditions (with 5 folds) with the Wapiti toolkit [47] and its BFGS algorithm2.”
Encoding linear models as weighted finite-state transducers (2014, methodology): “We then trained a linear tagging model using the Wapiti toolkit [23].”
An Introduction to Conditional Random Fields (2010, methodology): “In practice, models trained with the L1 regularizer tend to be sparser but have roughly the same accuracy as models training using the L2 regularizer [56].”
LIG System for Word Level QE task at WMT14 (2014, methodology): “Motivated by the idea of addressing Word Confidence Estimation (WCE) problem as a sequence labeling process, we employ the Conditional Random Fields (CRFs) for our model training, with WAPITI toolkit (Lavergne et al., 2010).”
Efficient Higher-Order CRFs for Morphological Tagging (2013, methodology): “Lavergne et al. (2010) make use of feature sparsity to signiﬁcantly speed up training for moderate tagset sizes ( < 100 ) and huge feature spaces.”
Non-Uniform Stochastic Average Gradient Method for Training Conditional Random Fields (2015, methodology): “For example, we have focused on the case of (cid:96) 2 -regularization but for large-scale problems there is substantial interest in using (cid:96) 1 -regularization CRFs [Tsuruoka et al., 2009, Lavergne et al., 2010, Zhou et al., 2011].”
Normalizing tweets with edit scripts and recurrent neural embeddings (2014, methodology): “As a sequence labeler we use Conditional Random Fields (Lafferty et al., 2001).”
LIMSI @ WMT’14 Medical Translation Task (2014, methodology): “We use Wapiti (Lavergne et al., 2010), a state-of-the-art CRF implementation, with a standard feature set.”