Fast LR parsing using rich (Tree Adjoining) Grammars

2002

The development of large coverage, rich uni cation-(constraint-) based grammar resources is very time consuming, expensive and requires lots of linguistic expertise. In this paper we report initial results on a new methodology that attempts to partially automate the development of substantial parts of large coverage, rich uni cation-(constraint-) based grammar resources. The method is based on a treebank resource (in our case Penn-II) and an automatic f-structure annotation algorithm that annotates treebank trees with proto-f-structure information. Based on these, we present two parsing architectures: in our pipeline architecture we rst extract a PCFG from the treebank following the method of , use the PCFG to parse new text, automatically annotate the resulting trees with our f-structure annotation algorithm and generate proto-f-structures. By contrast, in the integrated architecture we rst automatically annotate the treebank trees with f-structure information and then extract an annotated PCFG (A-PCFG) from the treebank. We then use the A-PCFG to parse new text to generate proto-f-structures. Currently our best parsers achieve more than 81% f-score on the 2400 trees in section 23 of the Penn-II treebank and more than 60% f-score on gold-standard proto-f-structures for 105 randomly selected trees from section 23.

Proceedings of the 42nd Annual Meeting on Association for Computational Linguistics - ACL '04, 2004

We discuss existing approaches to train LR parsers, which have been used for statistical resolution of structural ambiguity. These approaches are nonoptimal, in the sense that a collection of probability distributions cannot be obtained. In particular, some probability distributions expressible in terms of a context-free grammar cannot be expressed in terms of the LR parser constructed from that grammar, under the restrictions of the existing approaches to training of LR parsers. We present an alternative way of training that is provably optimal, and that allows all probability distributions expressible in the context-free grammar to be carried over to the LR parser. We also demonstrate empirically that this kind of training can be effectively applied on a large treebank.

2001

Natural language processing technologies offer ease-of-use of computers for average users, and easeof-access to on-line information. Natural language, however, is complex, and the traditional methods of parsing with a single grammar and parser may result in an inefficient and large system that is difficult to maintain and is fragile in dealing with language irregularities. This paper begins by reviewing an alternative effort in grammar decomposition (also known as grammar partitioning) for natural language parsing, which aims to alleviate these problems. We then propose a novel automatic approach for grammar partitioning, in comparison with a random method of partitioning. Our experiments show that syntactic GLR parsing is formidable for the Wall Street Journal corpus in the Penn Treebank when a single grammar is used. This is due to too many grammar rules for parsing table generation. However, grammar partitioning solves the problem and offers a viable alternative. Our results also show that our automatic grammar partitioning method based on the mutual information criterion fares better than a random partitioning method and exhibits efficiency in parsing as well as high parse coverage.

Tal Traitement Automatique Des Langues, 2005

We investigated the efficacy of beam search parsing and deep parsing techniques in probabilistic HPSG parsing. We first tested the beam thresholding and iterative parsing. Next, we tested three techniques originally developed for deep parsing: quick check, large constituent inhibition, and hybrid parsing with a CFG chunk parser. The quick check, iterative parsing and hybrid parsing greatly contributed to total parsing performance. The accuracy and average parsing time for the Penn treebank were 87.2% and 355 ms. Finally, we tested robustness and scalability of HPSG parsing on the MEDLINE corpus consisting of around 1.4 billion words. The entire corpus was parsed in 9 days with 340 CPUs. RÉSUMÉ. Nous avons étudié l'efficacité de l'analyse de beam search et des techniques de l'analyse profonde dans le probabiliste HPSG analyse. D'abord, nous avons examiné le beam thresholding et l'analyse itérative. Ensuite, nous avons examiné trois techniques développées originalement pour l'analyse profonde: quick check, large constituent inhibition, et l'analyse hybride avec la CFG chunk parser. Le quick check, l'analyse itérative et l'analyse hybride contribuaient considérablement à la performance de l'analyse totale. L'exactitude et le temps d'analyse moyen pour le Penn Treebank étaient 87.2% et 355ms. Finalement, nous avons examiné la robustesse et la extensibilité de HPSG analyse sur le corpus de MEDLINE contenant presque 1.4 milliard de mots. Le corpus entier a été analysé en 9 jours avec 340 CPUs.

IEEE Transactions on Learning Technologies, 2002

1998

In this thesis, we introduce probabilistic models to rank the likelihood of resultant parses within the GLR parsing framework. Probabilistic models can also bring about the bene t of reduction of search space, if the models allow pre x probabilities for partial parses. In devising the models, we carefully observe the nature of GLR parsing, one of the most e cient parsing algorithms in existence, and formalize two probabilistic models with the appropriate use of the parsing context. The context in GLR parsing is provided by the constraints a orded by context-free grammars in generating an LR table (global context), and the constraints of adjoining pre-terminal symbols (local n-gram context).

1991

In this paper a bidirectional parser for Lexicalized Tree Adjoining Grammars will be presented. The algorithm takes advantage of a peculiar characteristic of Lexicalized TAGs, i.e. that each elementary tree is associated with a lexical item, called its anchor. The algorithm employs a mixed strategy: it works bottom-up from the lexical anchors and then expands (partial) analyses making top-down predictions. Even if such an algorithm does not improve tim worst-case time bounds of already known TAGs parsing methods, it could be relevant from the perspective of linguistic information processing, because it employs lexical information in a more direct way. -27 -Procedure 3 Move-dot-left Input A slate s=[

2000

1995

We describe an extension of Earley's parser for stochastic context-free grammars that computes the following quantities given a stochastic context-free grammar and an input string: a) probabilities of successive prefixes being generated by the grammar; b) probabilities of substrings being generated by the nonterminals, including the entire string being generated by the grammar; c) most likely (Viterbi) parse of the string; d) posterior expected number of applications of each grammar production, as required for reestimating rule probabilities. Probabilities (a) and (b) are computed incrementally in a single left-to-right pass over the input. Our algorithm compares favorably to standard bottom-up parsing methods for SCFGs in that it works efficiently on sparse grammars by making use of Earley's top-down control structure. It can process any context-free rule format without conversion to some normal form, and combines computations for (a) through (d) in a single algorithm. Finally, the algorithm has simple extensions for processing partially bracketed inputs, and for finding partial parses and their likelihoods on ungrammatical inputs.

2007

We aim to improve the performance of a syntactic parser that uses a part-of-speech (POS) tagger as a preprocessor. Pipelined parsers consisting of POS taggers and syntactic parsers have several advantages, such as the capability of domain adaptation. However the performance of such systems on raw texts tends to be disappointing as they are affected by the errors of automatic POS tagging. We attempt to compensate for the decrease in accuracy caused by automatic taggers by allowing the taggers to output multiple answers when the tags cannot be determined reliably enough. We empirically verify the effectiveness of the method using an HPSG parser trained on the Penn Treebank. Our results show that ambiguous POS tagging improves parsing if outputs of taggers are weighted by probability values, and the results support previous studies with similar intentions. We also examine the effectiveness of our method for adapting the parser to the GENIA corpus and show that the use of ambiguous POS taggers can help development of portable parsers while keeping accuracy high.