Bayesian Learning of a Tree Substitution Grammar

By Matt Post, Daniel Gildea. 2009

82 citations recorded by the source corpus.

Tree substitution grammars (TSGs) offer many advantages over context-free grammars (CFGs), but are hard to learn. Past approaches have resorted to heuristics. In this paper, we learn a TSG using Gibbs sampling with a nonparametric prior to control subtree size. The learned grammars perform significantly better than heuristically extracted ones on parsing accuracy.

Subfields: Artificial Intelligence, Probabilistic & Bayesian Methods

How researchers see this paper

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

In Artificial Intelligence

In Probabilistic & Bayesian Methods

Citations in the corpus

A Survey of Machine Learning for Big Code and Naturalness (2017, background): “ion tree programs.Similarly, Wang et al.[185], Yin and Neubig [196] use an LSTM over AST nodes to achieve the same goal. Allamanis and Sutton [12] also create a syntactic model learning Bayesian TSGs [43, 156] (see Section 4.3). Semantic Models(graphs). Semantic code-generatingmodelsview codeasagrap”
Mining Semantic Loop Idioms from Big Code (2016, methodology): “This choice was based on previous work in applying pTSGs to natural language (Post and Gildea 2009; Cohn et al. 2010).”
Program Synthesis and Semantic Parsing with Learned Code Idioms (2019, background): “A pTSG is a probabilistic context-free grammar extended with production rules that expand to a whole AST fragment instead of a single level of symbols [8, 29].”
Inducing Tree-Substitution Grammars (2010, methodology): “The most similar work to that presented here is our own previous work (Cohn et al., 2009; Cohn and Blunsom, 2010), in which we introduced a version of the model described here, along with two other papers that independently introduced similar models (Post and Gildea, 2009; O’Donnell et al., 2009).”
Mining idioms from source code (2014, methodology): “. But, as we have seen, simple methods of formalizing this intuition do not work (see section 2), we resort to methods that are not as simple. We apply recent advanced techniques from statistical NLP [9, 40], but we need to explain them in some detail to justify why they are appropriate for this sof”
Productivity and Reuse in Language (2011, unknown): “Two other models, developed simultaneously with the current framework (Cohn et al., 2009; Post & Gildea, 2009), make use of similar Bayesian non– parametrics to define generative models over sets of subtrees.”
Stylometric Analysis of Scientific Articles (2012, methodology): “Efﬁcient Bayesian procedures have recently been developed that enable the training of large-scale probabilistic TSG grammars (Post and Gildea, 2009; Cohn et al., 2010).”
Bayesian Symbol-Refined Tree Substitution Grammars for Syntactic Parsing (2012, methodology): “In recent years, there has been an increasing interest in tree substitution grammar (TSG) as an alternative to CFG for modeling syntax trees (Post and Gildea, 2009; Tenenbaum et al., 2009; Cohn et al., 2010).”