Generalized LR Research Articles

Fast GLR parsers for extended BNF grammars and transition networks

Abstract The Tomita’s Generalized LR ( 1 ) parsing algorithm (GLR), later improved in many ways, runs in a linear time on LR ( 1 ) grammars, and degrades to a polynomial-time bound if the grammar is not deterministic. We address a useful feature not present in the current GLR ( 1 ) methods: the ability to accept grammars of the Extended BNF type (EBNF), the rules of which contain regular expressions. An EBNF grammar is conveniently represented by a collection of finite automata called a Transition Net (TN). We define, analyze and evaluate a new GLR ( 1 ) algorithm, called GELR, that combines the recent LR ( 1 ) parsing algorithm for TNs with the classical GLR data structures: the Graph-Structured Stack representing multiple stacks, and the Shared Packed Parse Forest for multiple syntax trees. The GELR algorithm is proved correct and an efficient implementation incorporating the state-of-the-art Right-Nulled parsing optimization is available. Experimental measures of the GELR parser size, speed and memory footprint are reported for current programming and web languages, and are compared with those of other parsing algorithms. The findings prove that directly parsing EBNF grammars does not penalize speed. Performance comparisons for different computer languages should also be of interest.

Subspace-Based Representation and Learning for Phonotactic Spoken Language Recognition

Phonotactic constraints can be employed to distinguish languages by representing a speech utterance as a multinomial distribution or phone events. In the present study, we propose a new learning mechanism based on subspace-based representation, which can extract concealed phonotactic structures from utterances, for language verification and dialect/accent identification. The framework mainly involves two successive parts. The first part involves subspace construction. Specifically, it decodes each utterance into a sequence of vectors filled with phone-posteriors and transforms the vector sequence into a linear orthogonal subspace based on low-rank matrix factorization or dynamic linear modeling. The second part involves subspace learning based on kernel machines, such as support vector machines and the newly developed subspace-based neural networks (SNNs). The input layer of SNNs is specifically designed for the sample represented by subspaces. The topology ensures that the same output can be derived from identical subspaces by modifying the conventional feed-forward pass to fit the mathematical definition of subspace similarity. Evaluated on the “General LR” test of NIST LRE 2007, the proposed method achieved up to 52%, 46%, 56%, and 27% relative reductions in equal error rates over the sequence-based PPR-LM, PPR-VSM, and PPR-IVEC methods and the lattice-based PPR-LM method, respectively. Furthermore, on the dialect/accent identification task of NIST LRE 2009, the SNN-based system performed better than the aforementioned four baseline methods.

Linear Programming with Triangular Fuzzy Numbers — A Case Study in a Finance and Credit Institute

The objective of this paper is to deal with a kind of fuzzy linear programming problem involving triangular fuzzy numbers. Then some interesting and fundamental results are achieved which in turn lead to a solution of fuzzy linear programming models without converting the problems to the crisp linear programming models. Finally, the theoretical results are also supported by a real case study in a banking system. The same idea is emphasized to be also useful when a general LR fuzzy numbers is given.

Component-based LR parsing

A language implementation with proper compositionality enables a compiler developer to divide-and-conquer the complexity of building a large language by constructing a set of smaller languages. Ideally, these small language implementations should be independent of each other such that they can be designed, implemented and debugged individually, and later be reused in different applications (e.g., building domain-specific languages). However, the language composition offered by several existing parser generators resides at the grammar level, which means all the grammar modules need to be composed together and all corresponding ambiguities have to be resolved before generating a single parser for the language. This produces tight coupling between grammar modules, which harms information hiding and affects independent development of language features. To address this problem, we have developed a novel parsing algorithm that we call Component-based LR (CLR) parsing, which provides code-level compositionality for language development by producing a separate parser for each grammar component. In addition to shift and reduce actions, the algorithm extends general LR parsing by introducing switch and return actions to empower the parsing action to jump from one parser to another. Our experimental evaluation demonstrates that CLR increases the comprehensibility, reusability, changeability and independent development ability of the language implementation. Moreover, the loose coupling among parser components enables CLR to describe grammars that contain LR parsing conflicts or require ambiguous token definitions, such as island grammars and embedded languages.

Optimized GLR Parsing for Programming Languages

The motivation of adopting the Generalized LR (GLR) parsing algorithm to construct parsers for programming languages is presented. A multi-level strategy for optimizing a GLR parser and the necessary runtime control mechanisms are introduced to the GLR parsers for flexible disambiguation and for invoking semantic actions attached to grammar rules while avoiding the delayed semantic action problem. The algorithm has been implemented in VPGE, a visual parser generation environment. Experimental results show that the speed of the generated GLR parsers is comparable to LALR(1) parsers generated by GNU's Bison when parsing deterministic programming languages.

Even faster generalized LR parsing

We prove a property of generalized LR (GLR) parsing – if the grammar is without right and hidden left recursions, then the number of consecutive reductions between the shifts of two adjacent symbols cannot be greater than a constant. Further, we show that this property can be used for constructing an optimized version of our GLR parser. Compared with a standard GLR parser, our optimized parser reads one symbol on every transition and performs significantly fewer stack operations. Our timings show that, especially for highly ambiguous grammars, our parser is significantly faster than a standard GLR parser.

Generation of LR parsers by partial evaluation

The combination of modern programming languages and partial evaluation yields new approaches to old problems. In particular, the combination of functional programming and partial evaluation can turn a general parser into a parser generator. We use an inherently functional approach to implement general LR( k ) parsers and specialize them with respect to the input grammars using offline partial evaluation. The functional specification of LR parsing yields a concise implementation of the algorithms themselves. Furthermore, we demonstrate the elegance of the functional approach by incorporating on-the-fly attribute evaluation for S-attributed grammars and two schemes for error recovery, which lend themselves to natural and elegant implementation. The parser require only minor changes to achieve good specialization results. The generated parsers have production quality and match those produced by traditional parser generators in sped and compactness

Intensive Determinants of Remote Masking

Five experiments were conducted to determine those aspects of a masking noise for which remote masking is invariant. The relative contributions of spectrum level and band width to remote masking can be described by the general equation LR = 2Ls+Lf+C, where LR is remote masking in db; Ls is spectrum level in db, the SPL of a 1-cps band width; and Lf is band-width level in db re a 1-cps band width. The constant is specific to each group of subjects. For remote masking less than 30 db, the formula does not predict remote masking, but does enable us to equate different conditions of spectrum level and band width. This equation does not hold as well when frequencies above 4000 cps are included in the masking noise. The possible causes of this inconsistency are discussed.