LR Parsing Research Articles

Boundelly LR(k)-conflictable grammars

We present a new class of context-free grammars whose sentences are parsable in linear time and space. The class calledboundedly LR(k)-conflictable (BLRC(k)) grammars includes all LR(k) grammars, some non-LR unambiguous grammars and some boundedly ambiguous grammars. A context-free grammar is said to beBLRC(k) if the number of conflict occurences during LR(k) parsing for every sentence of the grammar is inherently bounded. A BLRC(k) grammar can be considered as a natural extension of an LR(k) grammar whose sentences can be parsed by an LR(k) manner with multiple stacks. We show that it is a decidable problem whether a context-free grammar is BLRC(k) for a givenk, whereas it is undecidable for arbitraryk. The result is derived from anLR(k) machine description grammar which describes the behavior of a give LR(k) parser in terms of the grammar symbols. The relationship between the class of BLRC(k) grammars (and languages) and those of other associated grammars (and languages), is also discussed.

A very large vocabulary continuous speech recognition algorithm for telephone directory assistance

This paper proposes a speech recognition algorithm for large vocabulary continuous speech. The proposed algorithm is based on the hidden Markov model (HMM)-LR algorithm using a generalized predictive LR parser and phoneme HMMs. The following three techniques are applied to improve recognition performance and reduce processing time. 1 The forward and the backward likelihood are used to accurately determine the likelihood in the beam search. 2 To reduce the trellis computation in HMM speech recognition and for efficient search, only the speech frames in which the predicted phoneme seems to exist are used by the window for phoneme matching. 3 For efficient search, adjusting identical phoneme sequences are merged by checking the stack and the state of the LR parser. The algorithm was applied to a telephone directory assistance task involving more than 70, 000 subscribers. A recognition experiment for continuous word utterance was done. The sentence recognition rate was 85 percent for speaker-dependent speech recognition; the sentence recognition rate was 71 percent for speaker-independent speech recognition. The sentence understanding rate was 59 percent for speaker-dependent speech recognition with spontaneous utterances.

Realization of realtime large vocabulary continuous speech recognition integrating hidden Markov models and generalized LR parser

Realization of realtime large vocabulary continuous speech recognition integrating hidden Markov models and generalized LR parser

On the reduction of LR( k) parsers

The problem of reducing the number of states in a given LR( k) parser is treated from the standpoint of static merging, introducing a well-defined reduction of the parser. In addition, a locally optimal reduction is presented as a method for reducing the number of states.

An improved LALR( k) parser generation for regular right part grammars

A regular right part grammar (RRPG) is a context-free grammar, in which right parts of productions are finite automata to extend the descriptive power of context-free grammar by including notations for describing repetitions and alternations [6,8]. On LR parsing of RRPGs, extra work is required to identify the left end of a handle at reduction time because a nonterminal can derive potentially infinite number of strings via a single production. For parsing RRPGs, some methods such as grammar transformation from RRPGs to LR(k) context-flee grammars [5,8], augmenting LR(0) automaton with readback machines to recognize the reverse of state sequences leading to a reduction [2,6,7], and stacking an LR state only when the symbol being processed indicates the beginning of a new right part [11], are suggested. In stacking method, the parser is efficient because exactly one state entry is popped from the stack when the right end of a production is found. However, if stacking conflicts occur, the transfor-

Generalized probabilistic LR parsing of natural language (Corpora) with unification-based grammars

We describe work toward the construction of a very wide-coverage probabilistic parsing system for natural language (NL), based on LR parsing techniques. The system is intended to rank the large num...

Generalized LR Parsing

Generalized LR Parsing

Generalized LR parsing

Generalized LR parsing

ADVANCED PARSER-GENERATION TECHNOLOGY FOR AI LANGUAGE APPLICATIONS

A side effect of AI research is the development of programming languages as vehicles for experimentation and demonstration of concepts. These AI languages generally require some form of parser front end, which can be nontrivial to build. Parser generators can ease the task of language development, but commonly available generators use parsing technologies that severely constrain the level of syntactic sophistication, such as allowing at most one symbol of look ahead. Further, these generators are most often targeted for C and Ada applications; parser generators for LISP applications are not widely available. The RAND Advanced Compiler Kit (RACK) is a parser generator for generalized LR parsing and is suitable for AI applications. RACK parsers are unique in their ability to recognize non-LR(k) languages, as well as LR(k) languages for k > 1. RACK is implemented in C and is upwardly compatible with YACC, a widely used parser generator for C applications. RACK generates parsers that interface with C or Common Lisp. RACK also includes features such as arbitrary look-ahead, multiple start symbols, a scanner generator, and a grammar interpreter. In this paper, I describe RACK, its parsing technology, and significant features; I also report performance results comparing RACK to YACC.

Recursive ascent parsing: from Earley to Marcus

An overview is given of recursive ascent parsing, a new functional implementation technique for parsers for context-free grammars. The theory behind it unifies the treatment of hitherto virtually unrelated parsing methods, such as the Earley algorithm and LR parsing. This is effected by banning the distinguishing factors, such as stacks and parsing tables: stacks are replaced by recursive functions, parsing tables by function memoization. In addition to this unification, the theory provides a high-level view on parsing compared to the standard theory. Nevertheless, the functional implementations are as efficient as conventional ones, especially if the functions that constitute the parsers are formulated in low-level imperative languages with efficient function calls, like C. This means that they are important in practice, for parsing both artificial and natural languages. The recursive ascent treatment of LR(0) parsing has a natural generalization that leads to a new family of look-ahead parsers. This family is akin to the one proposed by Marcus a decade ago to model the processing of a natural language in the human mind.

A functional LR parser

A purely functional implementation of LR(0) parsers is given, together with a simple correctness proof. For non-LR(0) grammars its time complexity is cubic if the functions that constitute the parser are implemented as memo-functions, i.e. functions that memorize the results of previous invocations. For LR(0) grammars, our algorithm is closely related to the recursive ascent parsers recently discovered by Kruseman Aretz (1988), Barnard and Cordy (1988) and Roberts (1988).

A new parsing method for non‐LR(1) grammars

AbstractOne of the difficult problems that faces a compiler writer is to devise a grammar that is suitable for both efficient parsing and semantic attribution. This paper describes a system that resolves conflicts in LR(1) parsing by taking advantage of information in the parse tree. The system, which functions as part of a compiler generator, rewrites the user's grammar to remove parsing conflicts. It then places code into the generated compiler that rewrites the parse tree during parsing so as to produce the tree of the original grammar. The compiler writer can then write the semantic attribution to fit his or her original grammar without any knowledge of the changes made. The method is expected to be efficient in most cases, even in parsing systems that do not explicitly build the entire parse tree. The method complements previous work in its capabilities and advantages. The system has been implemented and integrated into a compiler generator system.

Book review: Current Issues in Parsing Technology Edited by Masaru Tomita (Kluwer Academic Publishers Boston/Dordrecht/London, 1991)

The 18 papers in this book are selected from the 45 presented at the International Workshop on Parsing Technologies (IWPT-89) sponsored by Carnegie-Mellon University. Papers dealing with Tomita's own parsing algorithm are published in a separate book Tomita, Generalized LR Parsing, Kluwer, 1991).

On efficient implementation of LR-attributed grammars

Several methods have been proposed for the evaluation of attributes during LR parsing. Some of these methods are based on the construction of semantic expressions for computing the values of inherited attributes, associated with states of the LR-automaton. In these methods, the semantic expressions are constructed and evaluated separately. This may lead to situations, where some computations are repeated. This paper exposes the redundant computations in these methods, and an optimization is introduced to avoid them.

Fuzzy partition models and their incremental training for continuous speech recognition.

This paper describes the application of Fuzzy Partition Models (FPMs) and their in-cremental training to continuous speech recognition. FPMs are neural networks with multiple input-output units. Since the outputs are non-negative and their sum is one, they can be regarded as the probabilities of recognizing input speech phonemes. Auto-matic incremental training is developed using the Viterbi alignment to adapt FPMs to continuous speech. The FPMs are retrained automatically by using speech data seg-mented by the Viterbi alignment. We combined FPMs with an LR parser (FPM-LR) and carried out experiments in continuous speech recognition. The recognition rate of the FPM-LR was higher than that of a Time-Delay Neural Network-LR (TDNNLR). Automatic incremental training was more effective with FPMs than with TDNNs.

Knowledge-directed interpretation of mechanical engineering drawings

A methodology for the interpretation of images of engineering drawings is presented. The approach is based on the combination of schemata describing prototypical drawing constructs with a library of low-level image analysis routines and a set of explicit control rules applied by an LR(1) parser. The resulting system (Anon) integrates bottom-up and top-down processing strategies within a single, flexible framework modeled on the human perceptual cycle. Anon's structure and operation are described and discussed, and examples of its interpretation of real mechanical drawings are shown.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

SLR( k) covering for LR( k) grammars

We show that every LR( k) context-free grammar can be right covered by an SLR( k) grammar. We give a method of transforming an arbitrary LR( k) grammar into an SLR( k) grammar which right covers it. The transformation is based on the LR( k) parser for the given grammar.

Language processing in connection with speech translation at ATR interpreting telephony research laboratories

The relationship between speech and language processing is an important problem to be solved in order to achieve continuous large-vocabulary speech recognition for a speech translation system or the human interface of a man-machine system. For the recognition of large-vocabulary continuous speech, first the phonemes are recognized by HMM (Hidden Markov Model). A generalized LR parser is introduced to predict next words/phonemes. The Japanese utterance is successfully recognized by the combined HMM-LR parser (HMM-LR). Many phrase candidates are filtered out of the speech recognition system through the use of linguistic information. An experimental system which translates spoken Japanese into English (SL-TRANS) has been implemented. The translation method consists of analysis, transfer and generation processes.A new method incorporated in the system analyzes the expressions of linguistic intention meanings in utterances. In applying this translation method to a goal-oriented dialogue corpus, involving inquiries and explanations regarding an international conference, the experimental system shows advantages for translating Japanese dialogues into English.

A unifying model for lookahead LR parsing

We present a construction method for lookahead LR parsers that unifies several of the construction algorithms available in the literature. The model allows single, multiple and arbitrary symbol lookahead, each only when required. Three user-supplied parameters are used by the construction method; specific settings of these parameters yield well-known grammar classes such as LALR(k), SLR(k), and several subsets of the LR-Regular class, among others. Thus, rather than using several severely incompatible parser generators, or making unnatural changes to the grammar to accommodate the parsing technique, one may manipulate these parameters to obtain a suitable parser. The model captures the essence of the problem of computing lookahead for LR parsers, and provides a better understanding of the relationships among the corresponding classes of context-free grammars.

HMM speech recognition using stochastic language models.

One of the major reasons for using language models in speech recognition is to reduce the search space. Context-free grammars or finite state grammars are suitable for this purpose. However, these models ignore the stochastic characteristics of a language. In this paper, three stochastic language models are investigated. These models are 1) a trigram model of Japanese syllables, 2) a stochastic shift/reduce model in LR parsing, and 3) a trigram model of context-free rewriting rules. These stochastic language models are incorporated into the syntax-directed HMM-based speech recognition system, and tested by phrase recognition experiments. The phrase recognition rate is improved from 88.2% to 93.2%.