JPH09212507A - Character processor and analytic method for character string - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce time necessary for analyzing a modification relation between clauses by applying this analytic method only to an inputted character string to which this analytic method is valid with respect to the analytic system of a large arithmetic quantity. SOLUTION: The inputted character string is divided into words by a morpheme analytic part 2 to be held in a holding part 3 and after then a clause is generated from the words by a clause generation part 4 to be held in a holding part 5. Next, the clause held is the holding part 5 is classified by the class by a clause kind judging part 6. Based on the clause class classified by the class, an analytic system judging part 7 judges whether to execute only modification analysis by a modification analytic part 7 or to also execute case relation analysis by a case relation analytic part 9 and parallel structure analysis by a parallel structure analytic part 10 based on the number of declinable words, the number of substantives modifying declinable words and the number of the clauses to specify the analytic system.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character processing device and a character string analyzing method, and more particularly to a character processing device and a character string analyzing method for analyzing a dependency relation between Japanese phrases.

[0002]

2. Description of the Related Art In a character processing device such as a Japanese word processor, a dependency analysis for analyzing a dependency relation between Japanese phrases has been conventionally used for application software such as voice synthesis and machine translation. . That is,
Dependency analysis is used, for example, in speech synthesis to utilize the result of dependency analysis to determine the pose position, and in machine translation to determine the subject and object.

The basic of the dependency analysis is a simple method of "applying the closest phrase to a conjunction-modifying phrase and the closest phrase to an adjunct-modifying phrase". In addition, in the conventional character processing device, it is general to use a combination of analysis methods such as case relation analysis and parallel structure analysis in order to improve the analysis accuracy of the dependency analysis.

[0004]

However, the case relation analysis and the parallel structure analysis have a problem that the amount of calculation is large as compared with the basic dependency analysis. Moreover, in a character processing device that performs a dependency analysis between bunsetsu by using case relation analysis and parallel structure analysis together, in spite of the fact that the case relation analysis and parallel structure analysis are valid and invalid sentences,
Since these analysis methods are always used for analysis, the amount of calculation becomes unnecessarily large. Especially, for a large amount of documents obtained from information communication networks such as the Internet, it is possible to perform speech synthesis or machine translation in real time. When using application software, the amount of calculation has become a serious problem.

The present invention has been made in view of the above circumstances, and for an analysis method with a large calculation amount, it is applied only to an input character string for which the analysis method is effective.
An object is to provide a character processing device and a character string analysis method that can reduce the time required for dependency analysis between phrases.

[0006]

In order to achieve the above object, a character processing apparatus according to the present invention is a morpheme analysis means for morphologically analyzing an input character string and dividing it into words, and an analysis by the morpheme analysis means. In a character processing device comprising a phrase generating means for generating a phrase from a selected word and a plurality of different character string analyzing means for analyzing the input character string, the phrase generated by the phrase generating means is At least one of the plurality of character string analysis means is selected according to the content.
It is characterized in that it has an analysis method specifying means for specifying more than one character string analyzing means.

Preferably, in addition to the character processing device, a partitioning unit for partitioning a large number of phrases into a plurality of different phrase types, and a phrase generated by the phrase generating unit among the plurality of phrase types. It is characterized in that the analysis method specifying means is executed based on the phrase type determined by the phrase type determining means.

Further, the plurality of character string analysis means include at least a dependency analysis means for analyzing a dependency relation between clauses, and a case relation analysis means for analyzing a case relation between clauses.
And a parallel structure analysis means for analyzing and extracting the parallel structure of the entire character string, wherein the analysis method specifying means specifies the analysis method from these character string analysis means.

Further, the dependency analysis means is always identified by the analysis method identification means, and the case relation analysis means is characterized in that the number of syllables in the character string is first.
Is specified by the analysis method specifying means when the number of consecutive modified type words is the second predetermined number or more, the parallel structure analysis means,
When the number of clauses in the character string is equal to or larger than a third predetermined number, the analysis method specifying means specifies the clause.

Further, the case relationship analyzing means executes an analysis of the case relationship based on a command from the dependency analyzing means, and the parallel structure analyzing means receives the command from the dependency analyzing means. It is characterized by executing an analysis of a parallel structure based on it.

The character string analysis method according to the present invention is
A morpheme analysis step of morphologically analyzing an input character string into words, a clause generation step of generating a clause from a word analyzed by the morpheme analysis step, and a different step of analyzing the input character string In a method for analyzing a character string including a plurality of character string analyzing steps, at least one or more character string analyzing steps are selected from the plurality of character string analyzing steps according to the content of the clause generated by the clause generating step. It is characterized by having an analysis method specifying step for specifying.

Further, preferably, in addition to the character string analyzing method, a phrase type determining step of determining which of the plurality of phrase types the phrase generated in the phrase generating step belongs to. And the analysis method specifying step is executed based on the phrase type determined by the phrase type determining step.

Further, the plurality of character string analyzing steps include at least a dependency analyzing step of analyzing a dependency relation between clauses, a case relation analyzing step of analyzing a case relation between clauses, and a parallel structure of the entire character string. And a parallel structure analyzing step of extracting and analyzing, and the analyzing method specifying step specifies the analyzing method from these character string analyzing steps.

Further, the analysis method specifying step is characterized in that the dependency analysis step is always specified, and the number of phrases in the character string is equal to or larger than a first predetermined number and the number of continuous modified type phrases is Is a second predetermined number or more, the analysis method identification step identifies the case relationship analysis step, and when the number of clauses in the character string is a third predetermined number or more, the analysis method identification step is performed. The step is characterized by specifying the parallel structure analysis step.

Further, the case relation analysis step is characterized in that case relation analysis is executed based on a command from the dependency analysis step, and the parallel structure analysis step is performed in response to a command from the dependency analysis step. It is characterized by executing an analysis of a parallel structure based on it.

[0016]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a character processing device according to the present invention.

In the figure, 1 is a character input unit such as a keyboard for inputting a character string to be analyzed. The morpheme analysis unit 2 performs morpheme analysis on the character string input by the character input unit 1 and divides the character string into words. The morpheme analysis result holding unit 3 holds the analysis result analyzed by the morpheme analysis unit 2. The clause generation unit 4 uses the morphological analysis result holding unit 3
A bunsetsu is generated based on the held contents held in. The phrase holding unit 5 holds the phrase generated by the phrase generation unit 4.
The bunsetsu type judging unit 6 judges the bunsetsu type for each of the bunsetsu held in the bunsetsu holding unit 5, divides the bunsetsu for each bunsetsu type, and writes the result in the bunsetsu holding unit 5. The analysis method determination unit 7 determines the analysis method of the phrase stored in the phrase storage unit 5 based on the type of phrase obtained by the phrase type determination unit 6. The dependency analysis unit 8 analyzes the dependency of the phrase held in the phrase holding unit 5 based on the analysis method determined by the analysis method determination unit 7. The case relation analysis unit 9 analyzes the case relation between phrases based on the signal from the dependency analysis unit 8. The parallel structure analysis unit 10 analyzes and extracts the parallel structure of the entire sentence based on the signal from the dependency analysis unit 8. The dependency analysis result holding unit 11 retains the analysis result analyzed by the dependency analysis unit 8. Dependency analysis result output unit 12
Outputs the contents held by the dependency analysis result holding unit 11 to an output unit (not shown) such as a display unit or a printing unit.
An analysis dictionary 13 is referred to by the morphological analysis unit 2, the clause type determination unit 6, the dependency analysis unit 8, the case relation analysis unit 9, and the side-by-side structure analysis unit 10 when executing each analysis.

FIG. 2 is a flow chart showing a processing procedure of a character string analyzing method in the character processing apparatus.

In step S1, morphological analysis is performed on the character string input by the morphological analysis unit 2 based on a predetermined part of speech, the character string is divided into words, and the divided words are stored in the morphological analysis result holding unit 3. Hold. As the part of speech used for morphological analysis in the present embodiment, for example, nouns, verbs, adjectives, adjective verbs, adverbs, adnominals, conjunctions,
Nine types of particles and auxiliary verbs are used.

In step S2, the phrase generation unit 4 generates a phrase from the words held in the morphological analysis result holding unit 3 and holds the phrase in the phrase holding unit 5. Here, each bunsetsu consists of one independent word, and an attached word is added as necessary. In this embodiment, seven types of nouns, verbs, adjectives, adjectives, adverbs, adnominals, and conjunctions are used as independent words, and two types of particles, auxiliary verbs, are used as adjuncts.

In step S3, the bunsetsu type determining unit 6 classifies the bunsetsu held in the bunsetsu type holding unit 5 by type, and determines which bunsetsu type the bunsetsu belongs to.

Specifically, as shown in FIG. 3, the types of bunsetsu are classified into "synonyms", "hymns", and "others", and these "hymns", "hymns", and "others" are further classified. They are classified into "continuous modification type" and "complex modification type", respectively. Where "definition"
Is the independent part of speech is a verb, an adjective, or an adjective clause
"Symptom" is a phrase whose noun is a part-of-speech of an independent word. Also,
"Others" is a phrase in which the part-of-speech of an independent word is an adverb, a conjunction, or a conjunction. Whether each bunsetsu is the "continuous modification type" or the "adnominal modification type" is the same as the modification type of the last word constituting the bunsetsu. Then, it is registered in the analysis dictionary 13 whether the modification type is the "continuous modification type" or the "adhesion modification type" for each word (the usage form of the word to be used).

In step S4, the analysis method determination unit 7 determines the analysis method based on the type of the phrase determined by the phrase type determination unit 6.

Specifically, as shown in FIG. 4, in the analysis method determination unit 7, whether the dependency analysis only is performed or the case relation analysis and the parallel structure analysis are also performed according to the input character string. Determine whether or not. In the present embodiment, such a criterion is a character string for which case relation analysis and parallel structure analysis are effective, and in this embodiment, dependency analysis is always performed for all clauses, and case relation analysis is performed using the number of terms. "2"
It is determined whether or not the number is more than or equal to 1 and the number of continuous modified type words is "1" or more, and for the parallel structure analysis, the number of clauses is "10" or more.

In step S5, the dependency analysis unit 8 analyzes the character string based on the determination result of the analysis method performed by the above-described analysis method determination unit 7, and holds the analysis result in the dependency analysis result holding unit 11.

In step S6, the dependency analysis result output unit 12 outputs the dependency analysis result to application software such as voice synthesis or machine translation, and the process is terminated.

As described above, in the present character processing apparatus, the dependency analysis is always executed, while the case relation analysis and the parallel structure analysis, which require a large amount of calculation, are carried out depending on the contents of the character string. Since it is executed only for valid character strings, the amount of calculation can be reduced, and the time required for dependency analysis between clauses can be reduced. Thus, even for a large amount of documents obtained from an information communication network such as the Internet, it is possible to avoid the inconvenience at the time of speech synthesis or machine translation due to the increase of the calculation amount.

The present invention is not limited to the above-described embodiment, and in the above-described embodiment, three types of analysis, dependency analysis, case relation analysis, and parallel structure analysis are used. It is also preferable to appropriately add other analysis methods such as response analysis.

In the above embodiment, the case relationship analysis and the parallel structure analysis are executed only when the predetermined requirements are satisfied, but the analysis method to be used may be specified by the user. That is, for example, it is also preferable that the user can specify that the case relationship analysis should always be executed regardless of the determination made by the analysis method determination unit 7 described above.

[0031]

As described above in detail, according to the character processing device and the character string analyzing method of the present invention, at least one or more of a plurality of character string analyzing means or character string analyzing steps are required. Since the character string analysis means of is specified, it is possible to execute only for a predetermined character string input for analysis methods with a large amount of calculation such as case relation analysis and parallel structure analysis, which reduces the amount of calculation. Therefore, it is possible to reduce the time required for dependency analysis between clauses.
Thus, even for a large amount of documents obtained from an information communication network such as the Internet, it is possible to avoid the inconvenience of speech synthesis or machine translation due to an increase in calculation amount as much as possible.

[Brief description of drawings]

FIG. 1 is a block configuration diagram showing an embodiment of a character processing device according to the present invention.

FIG. 2 is a flowchart showing a processing procedure of a character string analysis method according to the present invention.

FIG. 3 is a diagram showing types of clauses.

FIG. 4 is a diagram showing criteria for analysis method determination.

[Explanation of symbols]

 2 Morphological analysis unit (morphological analysis unit) 4 Phrase generation unit (Phrase generation unit) 6 Phrase type determination unit (Phrase type determination unit) 7 Analysis method determination unit (Analysis method specifying unit) 8 Dependency analysis unit (Dependency analysis unit) ) 9 Case Relationship Analysis Section (Case Relationship Analysis Means) 10 Parallel Structure Analysis Section (Parallel Structure Analysis Means)

Claims (16)

[Claims] 1. A morpheme analysis unit that morphologically analyzes an input character string and divides it into words, a phrase generation unit that generates a phrase from a word analyzed by the morpheme analysis unit, and a morphological unit of the input character string. In a character processing device including a plurality of different character string analysis means for performing analysis, at least one character is selected from the plurality of character string analysis means according to the content of the clause generated by the clause generation means. A character processing device having an analysis method specifying unit for specifying a column analyzing unit. 2. A partitioning means for partitioning a large number of clauses into a plurality of different clause types, and which of the plurality of clause categories the clause generated by the clause generating means belongs to is determined. The character processing device according to claim 1, further comprising: a phrase type determining unit, wherein the analysis method identifying unit is executed based on the phrase type determined by the phrase type determining unit. 3. The plurality of character string analysis units, at least a dependency analysis unit that analyzes a dependency relation between phrases, a case relation analysis unit that analyzes a case relation between phrases, and a parallel structure of entire character strings. 3. The character according to claim 1 or 2, further comprising a parallel structure analysis means for analyzing and extracting the character, wherein the analysis method specifying means specifies an analysis method from these character string analyzing means. Processing equipment. 4. The character processing device according to claim 3, wherein the dependency analysis unit is always identified by the analysis method identifying unit. 5. The case relation analysis means analyzes the case when the number of phrases in the character string is equal to or larger than a first predetermined number and the number of continuous modified type phrases is equal to or larger than a second predetermined number. 4. The method is specified by the method specifying means.
Alternatively, the character processing device according to claim 4. 6. The parallel structure analysis means is specified by the analysis method specifying means when the number of clauses in the character string is equal to or larger than a third predetermined number.
The character processing device according to claim 5. 7. The character processing device according to claim 3, wherein the case relationship analysis unit executes case relationship analysis based on a command from the dependency analysis unit. . 8. The character processing device according to claim 3, wherein the parallel structure analysis unit executes the parallel structure analysis based on a command from the dependency analysis unit. . 9. A morphological analysis step of morphologically analyzing an input character string into words, a phrase generation step of generating a phrase from the words analyzed by the morphological analysis step, and a morphological analysis of the input character string. In a method of analyzing a character string including a plurality of different character string analyzing steps for performing analysis, at least one or more of the plurality of character string analyzing steps are selected from among the plurality of character string analyzing steps according to the content of the clause generated by the clause generating step. A method for analyzing a character string, comprising an analysis method specifying step for specifying a character string analyzing step. 10. A clause type determining step of determining which of the plurality of clause categories the clause generated by the clause generating step belongs to, and the analysis method specifying step includes a clause type determining step. The character string analysis method according to claim 9, wherein the method is executed based on the phrase type determined in step. 11. The plurality of character string analysis steps include at least a dependency analysis step of analyzing a dependency relation between phrases, a case relation analysis step of analyzing a case relation between phrases, and a parallel structure of entire character strings. 11. The character according to claim 9 or 10, further comprising: a parallel structure analysis step of analyzing and extracting the character, wherein the analysis method specifying step specifies an analysis method from these character string analyzing steps. How to parse the column. 12. The character string analysis method according to claim 11, wherein the analysis method specifying step always specifies the dependency analysis step. 13. The analysis method specifying step, when the number of phrases in the character string is equal to or greater than a first predetermined number and the number of consecutive modified type phrases is equal to or greater than a second predetermined number, the case relationship analysis is performed. 12. A step is specified, and the step is specified.
Alternatively, the character string analysis method according to claim 12. 14. The analysis method specifying step specifies the parallel structure analysis step when the number of clauses in the character string is equal to or larger than a third predetermined number.
A method for analyzing a character string according to claim 13. 15. The character string according to claim 11, wherein the case relationship analysis step executes case relationship analysis based on a command from the dependency analysis step. analysis method. 16. The parallel structure analysis step executes the parallel structure analysis based on a command from the dependency analysis step.
The method of parsing the character string described in any of.