JPH01287776A - Syntax analyzing method - Google Patents

Abstract

PURPOSE:To reduce the output of an unintentional interpretation by deforming morpheme analytic results into a group of morphemes which have plural start positions and end positions and analyzing the group, and obtaining a syntax analytic result. CONSTITUTION:Respective morphemes in plural morpheme sequences generated as a result of the morpheme analysis of an input sentence 1 are given virtual positions representing the conjunction relations among the morphemes and morphemes which share the virtual positions and 'part of speech' used for syntax analysis are put in one group. A set 5 of groups of morphemes obtained as mentioned above has its syntax analyzed by a parallel type syntax analyzing means 6 to obtain syntax analysis results. Consequently, the order of plural interpretations is easily determined and the output of an unintentional interpretation as a 1st interpretation is reduced.

Description

[Detailed description of the invention] Industrial applications The present invention is used in word processors and the like.

It relates to a syntactic analysis method that analyzes the sentence structure and meaning of input sentences.

2. Description of the Prior Art The following describes an example using Japanese as one sentence.

Computer-applied systems such as machine translation systems and document composition systems that operate by understanding the sentence structure and meaning of a single Japanese sentence need to analyze the syntactic structure of the Japanese sentence.

Conventionally, in the field of computational linguistics, such sentence analysis methods have been researched. Literature such as ``Processing'' (edited by Makoto Nagao, published by the Institute of Electronics and Communication Engineers) introduces representative conventional sentence analysis methods developed to date. However, all of the syntactic analysis methods described in these methods receive a single morphological analysis result and analyze its syntactic structure.

The conventional syntax analysis method will be described below with reference to FIG.

FIG. 2 is a block diagram showing one embodiment of a conventional syntax analysis method. In Figure 2, 1 is an input sentence written in solid kanji/kana mixed notation, 2 is an electronic dictionary used for morpheme recognition, and 3 is a search for the electronic dictionary 2 to find the given kanji/kana mixed sentence. Dictionary search means 4 identifies morpheme candidates from a character string; 4 receives input sentence 1; uses dictionary search means 3 to find and output possible morpheme sequences one by one; serial morphological analysis; 5 is a serial morphological analysis means 4
The individual morpheme sequences outputted by , 6 is a parallel syntactic analysis means that receives one morpheme sequence 5 and calculates all possible syntactic structures, and 7 is the parallel syntax analysis means 6 that calculates all possible syntactic structures for one morpheme sequence 5 In addition, column 8 of all syntax trees is used to obtain the following interpretation.

This is a next interpretation presentation command that instructs the serial morphological analysis means 4 to obtain and output the next morphological analysis result of the input sentence 1.

In a configuration like the one on μ, the operation of lying down will be explained.

An input sentence 1 written in solid letters with a mixture of kanji and kana is read into the morphological analysis means 4. Serial type morphological analysis means 4
The dictionary search means 3 searches for a morpheme whose notation matches the input sentence 10 partial string, determines whether adjacent morphemes can be grammatically connected to each other, and systematically selects morphemes that are considered to be appropriate as the morpheme string 5. When one solution is found,
The result is output as morpheme sequence 5 and the process is stopped.

When the morpheme sequence 5 is output, the parallel parsing means 6 is activated. The parallel syntactic analysis means 6 reads the morpheme sequence 5 and finds all possible syntactic structures of the morpheme sequence 5 according to syntactic rules.

If the number of syntactic structures is 0, that is, the morpheme sequence 5 is
If it cannot be interpreted as a sentence due to syntactic rules, the parallel syntactic analysis means 6 issues a command to the serial morphological analysis means 4 to output the next morphological analysis result. Upon receiving the command, the serial morphological analysis means 4 resumes the suspended process, obtains the next result, and outputs it as a morpheme sequence 5. As long as the morphological analysis results remain and syntactic analysis fails, take out the morphological analysis results one by one as described above.
Attempt to parse. Repeat this action.

In the process, if one or more syntactic structures are found, the parallel syntactic analysis means 6 outputs all the found syntactic structures as a sequence 7 of the syntax tree and stops the operation. on the other hand,
In the process of repeating the above, if there are no more morphological analysis results on this μ (all interpretations have been exhausted), notify the outside that there is no other syntactic structure and terminate the process. . In addition, a request to present another interpretation from an external device that utilizes this parsing result is issued by the Next Interpretation Presentation Command 8.
, a command is issued to the serial morphological analysis means 4 to output the result of the first-order morphological analysis, and the above-described repetitive operation is performed to obtain the sequence 7 of the first-order syntax tree. Therefore, all interpretations of the input sentence 1 are issued by issuing the primary interpretation presentation command 8 several times.

You can ask for it.

Problems to be Solved by the Invention However, in the method described above, a sequence of syntactic trees is obtained for each individual morphological analysis result, so it is difficult to obtain the best syntactic tree among the individual morphological analysis results. It is easy to do this, but from among multiple morphological analysis results, the best n parse trees (n≧1) are determined by taking syntactic and semantic structures into consideration. This process is difficult to implement in terms of both storage capacity and processing efficiency. Also, in many application systems.

When one column 7 of the syntax tree is found, in that column,
The processing after μ is performed on the assumption that there is an interpretation intended by the author (or the speaker). The prerequisite for Uke is that the first morpheme sequence 5 output after receiving is highly accurate, but at the stage of morphological analysis,
It is difficult to provide a highly accurate ordering for multiple interpretations (for such a highly accurate ordering.

(requires syntactic/semantic analysis processing), and there was a problem in that an interpretation that was not the author's (or speaker's) intention was used in the post-μ processing. These issues are particularly important in languages such as Japanese, which do not have the custom of separating text.

The first to third present inventions solve the above problems,
By transforming multiple morphological analysis results into a set of morpheme pairs with multiple positions and parsing this set using a conventional parallel parsing method or an improved method, multiple interpretations can be obtained. The object of the present invention is to provide a syntax analysis method in which the ordering between the two is easy and an unintended interpretation is less likely to be output as the first interpretation.

Means for Solving the Problems A first aspect of the present invention is to analyze a plurality of morphological analysis results generated as a result of morphological analysis of an input sentence with a plurality of starting positions.

The above object is achieved by transforming the morphemes into a set of morpheme sets having a plurality of end positions, analyzing the set of morpheme sets by a syntactic analysis means, and obtaining a syntactic analysis result for the plurality of morphological analysis results. It is something.

The second invention achieves the above object by each of the morphological analysis results in the first invention being a sequence of Japanese morphemes to which morphological information, syntactic information, and semantic information are added. be.

The third aspect of the present invention is that the morphological analysis result in the first aspect of the present invention is a graph in which morphemes to which morphological information, syntactic information, and semantic information are added are nodes, and connections between the morphemes are arcs. .

This aims to achieve the above objectives.

Effects The first to third aspects of the present invention, in the above configuration, assign a virtual position representing a connection relationship between the morphemes to each morpheme of a plurality of morpheme sequences generated as a result of morphological analysis of an input sentence, for example. , morphemes that share this virtual position and the "part of speech" used in syntactic analysis are grouped together as one morpheme set. The collection (set) of morphemes thus created is parsed by a parallel parsing means, and a plurality of parsing results are obtained. Note that in the process of obtaining these syntactic analysis results, the results can be ordered in consideration of morphemes, syntax, and meanings.

EMBODIMENT MU Below, an embodiment of the present invention will be described with reference to one drawing. As an example, we can morphemely analyze a Japanese sentence written in solid letters with a mixture of kanji and kana using the concept of the minimum number of clauses, find all morpheme sequences with the minimum number of clauses, and transform these morpheme sequences. , a syntactic analysis method for parsing a set of deformed morpheme pairs using syntactic rules in the form of a free grammar and a parallel syntactic analysis means, and outputting a parse in which multiple parse trees are embedded. I will explain it accordingly.

FIG. 1 is a block diagram mainly showing an apparatus for implementing a syntax analysis method according to an embodiment of the present invention.

In the figure, 11 is an input sentence to be analyzed, 12 is an electronic dictionary containing the notation, pronunciation, and part of speech of morphemes, and 13 is an electronic dictionary containing the notation, pronunciation, and part of speech of morphemes. 14 is a dictionary search means for searching for all the ones that match the left subsequence with a single notation; 14 is a morphological analysis means for finding and outputting all interpretations (morpheme sequences) such that the number of phrases included in the morpheme sequence is the minimum; 15 is a morphological analysis means 14
16 is a set of morpheme sequences having information on orthography, pronunciation, and part of speech output by a position information adding means for calculating the end position and adding these two types of position information to each morpheme in the set 15 of morpheme sequences; 17 is a set of morpheme sequences to which position information outputted by the position information adding means 16 has been added; A morpheme grouping means 18 takes all the morphemes that have the same position information and the same part of speech and groups them together; 519 is a set of morpheme pairs, and 519 is syntactically defined by the syntactic rules written in the form of a one-minute free grammar and the priority calculation rules established for each syntactic rule. Parallel parsing means 20 calculates a predetermined number of most probable syntactic structures, and reference numeral 20 denotes a parsing garment in which a plurality of syntactic structures and associated morpheme sequences created and output by the parallel parsing means 19 are stored. It is. The parallel parsing means 19 may be the Garlay method or the Cocke-Kasami-Younger method.
) can be used to handle input that includes multiple positional information, or a parallel parsing method such as the method described in Specially suited proprietary techniques may also be used.

In the configuration shown above, an outline of the analysis operation will be explained below, taking as an example the case where the sentence "Kuruma hoko wo hakobu" is given as the input sentence 11.

The input sentence 11 is read into the morphological analysis means 14, and a systematic dictionary search is performed from the beginning of the sentence to the end using the dictionary search half-input 13 to determine whether adjacent morphemes can be grammatically connected. is done using an internal connection table. Among the several morpheme sequences thus determined, those with the smallest number of clauses (formally defined by a segmental free grammar) are all output as a set 15 of morpheme sequences. In this case, the set of morpheme sequences 15 is.

(``car'', ``de'', ``box'', ``wo'', ``carry'').

('car', 'de', 'child', 'wo', 'carry'), ('come'
"until""box""to""carry"),("come""until"
``ha'',``child'',``wo'',``carry''),(``come'',``between'',``de'',``box'',``wo'',``carry'').

(``come'', ``between'', ``then'', ``child'', ``wo'', ``carry'').

(``carry'', ``until'', ``box'', ``wo'', ``carry'').

This is a set of many morpheme sequences such that the number of clauses is 3. The position information adding means 16 uses this morpheme sequence set 1
5, each morpheme sequence is simultaneously scanned from the beginning, and a start position and an end position expressing the connection relationship between the morphemes are given to each morpheme. For example, in the case of the morpheme "kuru", two types of morphemes follow: "to" and "ma".

In the input sentence, there are two other morphemes that share the same substring ``kuru'' with ``kuru'': ``kuru'' and ``keikikyo'', and ``kuru'' has ``ma'' and ``ma''. but.

``Kantei'' is followed by ``to''. Therefore,
The starting position of ``Kuru'' and ``Kuru'' is 0, the ending position is 1 and 2, and the starting position of ``Kakei'' is O1, and the ending position is 2. "
The starting position of "between" is 1. The starting position of "until" is determined to be 2. This position information expresses that ``kuru'' and ``karu'' are followed by ``ma'', but ``kokuden'' is not followed by ``ma''. The set 15 of morpheme sequences assigned one position in this manner is passed to the morpheme grouping means 17. The morpheme grouping means 17 sets the starting position in the set of morpheme sequences taken.

All morphemes that share a final position and a part of speech are grouped together as a morpheme set, and processed as one morpheme in the syntactic/semantic analysis process after μ. For example, in the previous example, "
"come" and "kari" are both used together.

Starting position is O1 Ending position ht1 and 21 Part of speech is "verb"
and [: (0), (1, 2), verb, ('come',
It becomes a set of morphemes like ``kuru'')].

A set containing such pairs as elements is the set 1 of morpheme pairs.
Output as 8. The parallel syntactic analysis means 19 considers this set of morpheme pairs 18 to be a sequence of morphemes that have multiple positions of normal meaning (that is, the character of the input sentence), and uses the built-in form of the free branch grammar. According to the syntactic rules of , the most probable is determined by using the priority calculation rules based on the information grouped together in the set of morphemes (e.g., ``(come'', ``carry'')'' etc.). A predetermined number of syntactic groove structures are determined and outputted as an analysis table 20.

In addition, in the explanation above, we explained the case where Japanese was used, but English, German, Chinese, etc.

It can also be applied to various other languages.

Effects of the Invention As explained above, according to the present invention, compared to conventional syntactic analysis methods, it is easier to compare and order multiple interpretations, taking into account the ambiguity that occurs in morphological analysis, and it is possible to This has the advantage that an interpretation that does not occur is rarely output as the first interpretation. In particular, the input text was written all over the place. The present invention is particularly useful in the case of Japanese written in kana notation, since there are many ambiguities that arise in morphological analysis, and it is often difficult to order the results of morphological analysis at the stage of morphological analysis. .

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration mainly of an apparatus for implementing the syntax analysis method according to the present invention. FIG. 2 is a block diagram showing a configuration centered on a device for implementing a conventional syntax analysis method. DESCRIPTION OF SYMBOLS 11... Input sentence, 12... Electronic dictionary 413... Dictionary search means, 14... Morphological analysis means, 15... Set of morpheme sequences, 16. Position information adding means, 17... Morphological summary Raising means, 18... Set of morpheme pairs, 19
...Parallel parsing means, 20...Analysis clothes. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
figure

Claims (3)

[Claims] (1) Transform a plurality of morphological analysis results resulting from morphological analysis of an input sentence into a set of morpheme pairs having a plurality of start positions and a plurality of end positions, and convert the set of morpheme pairs by a syntactic analysis means. A syntactic analysis method, characterized in that the syntax analysis results are obtained for the plurality of morphological analysis results. (2) The syntactic analysis method according to claim 1, wherein each of the morphological analysis results is a sequence of Japanese morphemes to which morphological information, syntactic information, and semantic information are added. (3) The syntactic analysis method according to claim 1, wherein the morphological analysis result is a graph in which nodes are morphemes to which morphological information, syntactic information, and semantic information have been added, and arcs are connections between morphemes.