JPS63221475A - Analyzing method for syntax - Google Patents

Abstract

PURPOSE:To improve the average processing efficiency for analysis of a syntax without deteriorating the efficiency of the worst case, by analyzing the syntax after obtaining a part of the syntax of an input sentence by using form information and meaning information in the input sentence as well as the information included in a syntax rule group. CONSTITUTION:An input sentence 1 is successively read into a morpheme analyzing means 3 at and after the head character. The means 3 retrieves a part-of- speech dictionary 2 to confirm character strings that can be used as morphemes. Then the relevant character string is confirmed as a morpheme if the relevant part of speech can be connected to the morpheme confirmed right before that time point. This production is repeated to obtain a morpheme string 4 for the meaning information, the form information added with the part-of-speech information, etc. The strings 4 are read one by one into an incomplete syntax analyzing means 6. The means 6 analyzes the morpheme string 4 and obtains a partial tree string 7 within a buffer area. The string 7 is read by a syntax analyzing means 8 for each partial tree and all possible syntax trees are obtained in the form of an analysis table.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a syntax analysis method for a device that operates according to the sentence structure and meaning of an input sentence.

Conventional technology A case in which a Japanese sentence is targeted as a sentence will be explained below. Computer-applied systems such as machine translation systems and document proofing systems that operate by understanding the sentence structure and meaning of Japanese sentences need to analyze the syntactic structure of Japanese sentences.

In the field of computational linguistics, such sentence analysis methods are being researched, such as ``Lecture on Modern Languages Volume 7: Machine Processing of Language'' (edited by Makoto Nagao, published by Sanseido) and ``Japanese Information In books such as "Processing" (supervised by Makoto Nagao, published by the Institute of Electronics and Communication Engineers),
Representative methods of sentence analysis developed so far are introduced. Among them, the syntactic analysis method based on context-free grammar has declarative and clear syntactic rules, and each rule is independent, so it is easy to develop and maintain the grammar. It also has the characteristic of being able to easily describe the research results of transformative generative grammar, which is one of the linguistic theories.

Furthermore, in the fields of formal language theory and computational theory, efficient analysis methods for analyzing sentences based on context-free grammars are known, and several analysis systems that utilize these methods have been created. "About extended LINGOL, a programming system for natural language processing" (Hozumi Tanaka et al., Transactions of the Institute of Electrical Communication Engineers, 1977, No. 12, 160
1-16o8 pages) is an example. however,
Such parsing systems use parsing techniques that can handle syntactic rules written in the form of general, unrestricted context-free grammars, at the expense of spatial and temporal overrading, e.g. , Early
), if the length of the input sentence is n, then
A storage capacity on the order of n2 and a calculation time on the order of n5 are required. On the other hand, if the syntax rules are limited to a portion of a context-free grammar, faster methods such as those used in translation programs (compilers) for programming languages such as FORTRAN can be used. For example, the syntax rules are called LL(k) grammars. If it belongs to a part of a context-free grammar, a more efficient analysis can be performed using a method called recursive descent. The details are Descent Combi Ring.

Written by N.J.T. Dippy et al. Published by Ellis Harxod Publishing Co., Ltd., 1981 (Descent Compili)
ng.

A, J, T, Davie et al. Ellis Hor
Published by Wood Publishing.

(1981). However, the grammar of many natural languages is ambiguous, and it is impossible to completely analyze natural languages in their original form, since they fall outside the framework of such limited context-free grammars.

Problems to be Solved by the Invention However, in the above method, the syntactic structure of the input sentence is
As long as we try to obtain a complete syntax tree, we have to use an analysis method that can handle even the most complex thing that can be described as a syntactic rule, which poses the problem of considerable spatial and temporal overhead. there were. In addition, in a method that uses limited syntax rules that are different from the syntax rules used in that analysis method before applying a general analysis method, the syntax rules are separated into two groups, so the syntax rules There were problems in that it was difficult to maintain the consistency of the system as a whole, and it was not possible to take advantage of the peculiarities that existed in individual input sentences.

The present invention solves the above-mentioned problems. Before parsing an input sentence using a syntactic analysis means that performs analysis according to general syntactic rules, the present invention analyzes the special characteristics of each input sentence and the given syntax. Considering the characteristics of the rule group, a second parsing method, which is faster and more efficient than the parsing method described above, is used to perform as much analysis as possible, resulting in a single, easy-to-handle parsing method. The present invention provides a syntax analysis method that improves average processing efficiency under a group of syntax rules without reducing efficiency in the worst case.

Means for Solving the Problems The present invention provides a syntax analysis means for analyzing the syntactic structure of an input sentence according to syntactic rules (2), The above object is achieved by an incomplete parsing means for determining a part of the structure.

Operation In the above configuration, various information such as form, syntax, and meaning contained in a human sentence is input to the incomplete syntax analysis means, and is determined by the incomplete syntax analysis means to be applicable among all syntactic rules. only those rules applied to said input sentence,
Fast, but with partial and incomplete parsing. The result is input to a general parser capable of handling all syntax rules, where it is fully parsed.

Embodiment Hereinafter, a syntactic analysis method according to the present invention, in which a Japanese sentence written in solid letters with mixed Kanji characters is analyzed based on a context-free grammar and outputted as a syntax tree, will be explained with reference to the drawings. The figure is a block diagram showing the overall configuration for implementing the syntax analysis method according to the present invention. In the figure,
1 is an input sentence to be analyzed, 2 is a part-of-speech dictionary containing morphemes and their parts of speech, and 3 is a part-of-speech dictionary containing morphemes and their parts of speech. Column 4
5 is a morphological analysis means for converting into morphological analysis means, and 5 is a syntactic rule written in a context-free grammar formula, such as "unit sentence - case element/unit sentence", "unit sentence - rendition", "case element - noun phrase/case particle" A syntactic rule group containing rules such as 6 is a morpheme sequence 4
A high-speed incomplete syntactic analysis means based on the LR (1) method that partially parses the input sentence 1 from the part of speech information and the syntactic rule group 5; 7 is output by the incomplete syntactic analysis means 6. Partial Hole sequence 8 sets partial Hole sequence 7 to Early (Earley).
) is a syntactic analysis method that analyzes based on the method (:) and obtains the syntax 9.

The input sentence 1 is read into the morphological analysis means 3 in order from the first character of the sentence and processed. Morphological analysis means 3 (2, therefore,
The part of speech dictionary 2 is searched to identify a character string that can be a morpheme, and if that part of speech can be connected to the immediately preceding (2 recognized morpheme), the character string is recognized as a morpheme.By repeating this operation, , Finally, a morpheme string 4 containing morphological information, semantic information, etc. to which part-of-speech information is added is obtained.The morpheme string 4 is read into the incomplete syntax analysis means 6 one by one starting from the morpheme at the beginning of the sentence and processed. The incomplete syntax analysis means 6 consists of a buffer area for storing intermediate and final results of analysis, and a finite state automaton created from the syntax rule group 5, and the incomplete syntax analysis means 6 consists of a buffer area for storing intermediate and final results of analysis, and a finite state automaton created from the syntax rule group 5.
and the current state of the finite state automaton to determine the action to be taken. However, if there are multiple actions to be performed and it is not decided at will, in other words, LR (
1) When an action is required that goes beyond the scope of the grammar, no syntactic rules are applied, and the shift action (
The read morpheme is simply added to the buffer area. In this way, the incomplete syntactic analysis means 6 analyzes the morpheme sequence 4, and finally obtains the partial h-sequence 7 in the buffer area, but this partial h-sequence 7 is the syntactic rule group 5
Even if they are the same, there are various cases depending on the content of input sentence 1, ranging from the same as morpheme sequence 4 (no full thermal analysis is performed) to a complete syntax (all parsing is performed). It is possible, but generally some morphemes are bottom
It is a sequence of partial hos (final (= part of the syntactic tree obtained) created by grouping up, and the number of elements in that sequence is 1 or more and less than or equal to the number of morphemes in morpheme sequence 4. At the stage of complete syntactic analysis, it is also possible to extract various types of information useful for later syntactic and semantic analysis.
The partial string 7 created by
Each partial water is read and all possible syntax trees are determined in the form of an analysis table.

In the above example, the amount of calculation of the LR(1) method is on the order of nl, both temporally and spatially, where n is the number of morphemes in the morpheme sequence 4, and
The amount of calculation required is smaller than that of the analysis method of (rley).

Moreover, even in the worst case, the amount of calculation is equivalent to the case without the incomplete syntax analysis means 6. In addition, the incomplete syntax analysis means 6 can also extract auxiliary information for the syntax analysis means 8 and pass it to the syntax analysis means 8. In that case, the processing efficiency of the syntax analysis means 8 can be improved. You can improve that.

Note that although the above explanation has been given for the case where Japanese sentences are used as sentences, the invention can also be applied to sentences in various other languages such as English and German.

Effects of the Invention As explained above, according to the present invention, compared to conventional syntax analysis methods, syntax rules that can be applied efficiently while keeping a single syntax rule group can be applied to each individual input. It is possible to automatically judge and apply the processing according to the sentence (2), and the average processing efficiency can be improved without reducing efficiency in the worst case, which has a large effect.

[Brief explanation of drawings]

The figure is a block diagram showing the overall configuration of an embodiment of the syntactic analysis method according to the present invention (2). 1. Input sentence, 2. Part-of-speech dictionary, 3. Morphological analysis means, 4. Morpheme sequence, 5... Syntactic rule group, 6.
... Incomplete syntax analysis means, 7... Partial string, 8...
Syntax analysis means, 9... syntax tree.

Claims (3)

[Claims] (1) When analyzing the syntactic structure of each input sentence according to a group of syntactic rules that define the syntactic structure of a language, the morphological information and semantic information in the input sentence and the information in the group of syntactic rules are used to analyze the syntactic structure of each input sentence. A syntactic analysis method characterized by determining a part of the syntactic structure of an input sentence and then performing syntactic analysis. (2) The syntax analysis method according to claim 1, wherein the syntax rule sentence is written in either a context-free grammar or an extended context-free grammar. (3) The syntactic analysis method according to claim 1, wherein the input sentence is a sequence of Japanese morphemes to which morphological information, semantic information, etc. are added.