Online large-margin training of dependency parsers

McDonald, Ryan; Crammer, Koby; Pereira, Fernando

doi:10.3115/1219840.1219852

Cited by 474 publications

(553 citation statements)

References 20 publications

Supporting

Mentioning

548

Contrasting

Order By: Relevance

“…Related to such online methods is also the MIRA algorithm (Crammer and Singer 2003), which has been used for training structured predictors (e.g. McDonald et al 2005). However, to deal with the exponential size of Y, heuristics have to be used (e.g.…”

Section: Structural Support Vector Machinesmentioning

confidence: 99%

Cutting-plane training of structural SVMs

2009

View full text Add to dashboard Cite

Discriminative training approaches like structural SVMs have shown much promise for building highly complex and accurate models in areas like natural language processing, protein structure prediction, and information retrieval. However, current training algorithms are computationally expensive or intractable on large datasets. To overcome this bottleneck, this paper explores how cutting-plane methods can provide fast training not only for classification SVMs, but also for structural SVMs. We show that for an equivalent "1-slack" reformulation of the linear SVM training problem, our cutting-plane method has time complexity linear in the number of training examples. In particular, the number of iterations does not depend on the number of training examples, and it is linear in the desired precision and the regularization parameter. Furthermore, we present an extensive empirical evaluation of the method applied to binary classification, multi-class classification, HMM sequence tagging, and CFG parsing. The experiments show that the cutting-plane algorithm is broadly applicable and fast in practice. On large datasets, it is typically several orders of magnitude faster than conventional training methods derived from decomposition methods like SVM-light, or conventional cutting-plane methods. Implementations of our methods are available at www.joachims.org.

show abstract

Section: Structural Support Vector Machinesmentioning

confidence: 99%

Cutting-plane training of structural SVMs

2009

View full text Add to dashboard Cite

show abstract

“…Since we only measure passage retrieval and reranking performance, we disabled the answer extraction component. For analyzing parse structures of questions and answers, we integrated the dependency parser MSTParser [3] into the system, and extended OpenEphyra further for our passage retrieval and reranking algorithms.…”

Section: Related Workmentioning

confidence: 99%

Passage Reranking for Question Answering Using Syntactic Structures and Answer Types

Aktolga

Allan

Smith

2011

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Passage Retrieval is a crucial step in question answering systems, one that has been well researched in the past. Due to the vocabulary mismatch problem and independence assumption of bag-of-words retrieval models, correct passages are often ranked lower than other incorrect passages in the retrieved list. Whereas in previous work, passages are reranked only on the basis of syntactic structures of questions and answers, our method achieves a better ranking by aligning the syntactic structures based on the question's answer type and detected named entities in the candidate passage. We compare our technique with strong retrieval and reranking baselines. Experimental results using the TREC QA 1999-2003 datasets show that our method significantly outperforms the baselines over all ranks in terms of the MRR measure.

show abstract

“…• a rule-based Treex tokenizer and detokenizer • a word aligner -GIZA++ (Och and Ney, 2003) • a dependency parser -MST parser for English (McDonald et al, 2005), and its variations for Czech: a version by Novák and Žabokrtský (2007) adapted for Czech in the basic version of Depfix, or MSTperl by Rosa et al (2012a) adapted for SMT outputs in full Depfix • a dependency relations labeller (as the MST parser returns unlabelled parse trees) -a rule-based Treex labeller for English, and a statistical labeller by Rosa and Mareček (2012) for Czech • a named entity recognizer -Stanford NER for English (Finkel et al, 2005), and a simple Treex NER for Czech • a rule-based Treex converter to tectogrammatical (deep syntax) dependency trees There are also other tools that we do not currently use (because they are not part of Treex yet, some of them probably do not even exist yet), but we believe that they would be useful for Depfix as well:…”

Section: Toolsmentioning

confidence: 99%

Depfix, a Tool for Automatic Rule-based Post-editing of SMT

Rosa¹

2014

The Prague Bulletin of Mathematical Linguistics

View full text Add to dashboard Cite

We present Depfix, an open-source system for automatic post-editing of phrase-based machine translation outputs. Depfix employs a range of natural language processing tools to obtain analyses of the input sentences, and uses a set of rules to correct common or serious errors in machine translation outputs. Depfix is currently implemented only for English-to-Czech translation direction, but extending it to other languages is planned.

show abstract

Online large-margin training of dependency parsers

Cited by 474 publications

References 20 publications

Cutting-plane training of structural SVMs

Cutting-plane training of structural SVMs

Passage Reranking for Question Answering Using Syntactic Structures and Answer Types

Depfix, a Tool for Automatic Rule-based Post-editing of SMT

Contact Info

Product

Resources

About