Products of Random Latent Variable Grammars

By Slav Petrov. 2010

96 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.

Citations in the corpus

A Primer in BERTology: What We Know About How BERT Works (2020, background): “Initialization can have a dramatic effect on the training process (Petrov, 2010).”
Bayesian Symbol-Refined Tree Substitution Grammars for Syntactic Parsing (2012, methodology): “Recently, (Petrov, 2010) has reported that combining multiple grammars trained independently gives signiﬁcantly improved performance over a single grammar alone.”
Word Segmentation, Unknown-word Resolution, and Morphological Agreement in a Hebrew Parsing System (2013, background): “This reinforces the ﬁndings of Petrov (2010) who also reports that the ensemble creation using random initialization is exceptionally strong and outperforms other methods of ensemble creation.”
Neural CRF Parsing (2015, background): “The conditions these parsers are operating under are slightly different: the former is a reranker on top of the Stanford Parser (Klein and Manning, 2003) and the latter trains on much larger amounts of data parsed by a product of Berkeley parsers (Petrov, 2010).”
Self-Training with Products of Latent Variable Grammars (2010, methodology): “Given a sentence s and a set of grammars G = { G 1 , · · · , G n } , recall that the decoding algorithm of the product model (Petrov, 2010) searches for the best tree T such that the following objective function is maximized: where log p ( r | s, G ) is the log posterior probability of rule r given…”
Parsing as Language Modeling (2016, methodology): “To obtain silver trees, we parse the entire section of the New York Times (NYT) of the ﬁfth Gigaword (Parker et al., 2011) with a product of eight Berkeley parsers (Petrov, 2010) 2 and ZPar (Zhu et al., 2013) and select 24 million trees on which both parsers agree (Li et al., 2014).”
Inducing Tree-Substitution Grammars (2010, methodology): “In doing so, our model learns local structures for latent linguistic phenomena, such as verb subcategorisation and lexical agreement.”
Distilling an Ensemble of Greedy Dependency Parsers into One MST Parser (2016, background): “An alternative that runs in linear time is transition-based parsing, which recasts parsing as a sequence of actions that manipulate auxiliary data structures to incrementally build a parse tree (Nivre, 2003).”