JP2775655B2 - Abbreviation word completion device and Japanese analysis system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an abbreviation completion device for mechanically supplementing abbreviations of an input Japanese sentence and a Japanese analysis system provided with such an abbreviation completion device.

[0002]

2. Description of the Related Art In an information processing system such as a computer-assisted Japanese analysis or a machine translation using such a computer, it is often important to understand what abbreviation in an input Japanese sentence corresponds to. Become. This is because Japanese sentences tend to omit elements that can be reproduced by the listener or the reader, and the technique of complementing the abbreviations is extremely important for the information processing system described above.

Conventionally, a technique for complementing this abbreviation is to search for a word having the same case as the omitted case component from a newer one of the previously input sentence or clause, irrespective of the above semantic constraints. There was something to complement.

Further, as an advanced abbreviation complementing technique, there has been a technique using conceptual information such as a semantic feature of a word. In other words, for each predicate such as a verb, the semantic constraint of the word that can be the case component is defined in advance, and the noun in the previously input sentence or clause that matches the constraint from the newest noun Search for and complement the words it finds. For example,
"Weird food was served in front of Taro.
Eat (φ) and (φ) ... ("Φ" in the sentence means omitted).

[0005] If it is defined that the case where the particle of "eating" can be displayed as an animal and that the case where the particle of "o" can be displayed is food, the noun in the first sentence is new. Look up the semantic information in the order of "food", "before", "eyes" and "Taro", and search for animals and food. As a result, "Taro" included in the animal is displayed as the case where "ga" of "eat" is displayed, and food "food" is complemented as the case where "is displayed".

[0006]

However, in the abbreviation complementing technique, it is a minimum necessary condition in a normal sentence that it cannot be complemented unless it meets semantic constraints. Therefore, regardless of this, it was not possible to expect accurate completion when selecting a word of the same case component as the previous word as a complement.

[0007] In addition, the abbreviation complementing technique using conceptual information such as semantic features presupposes that there is a word to be complemented near a place where the abbreviation is omitted.
The probability is not always high. For this reason, even in the related art, very good accuracy was not obtained in complementing the abbreviations. If all of the words abbreviated by these conventional techniques are to be supplemented, the absolute number of errors will increase, and post-editing such as machine translation will be burdensome. In some cases, it was limited to complementing the case components or the subject displayed by the particle "ga". In addition, when applying the abbreviation completion technology to machine translation, for example, in order to select either flower or blossom as a translation of “flower”, it is necessary to supplement a continuous modification component such as “cherry blossom, rose”. . However, such completion was almost impossible with conventional abbreviation completion technology.

[0008] The present invention has been made to solve the above-mentioned conventional problems, and abbreviations are complemented by using information on context including the subject of a previous sentence or section.
An object of the present invention is to realize highly accurate abbreviation complementation and also to complement a required union modification component.

[0009]

In order to achieve the above object, the present invention provides a morphological analysis process for dividing a Japanese sentence into words and adding syntax information and semantic information of the words to the words. A sentence clause that has been subjected to a syntax analysis process that analyzes sentence structures such as dependency relationships and phrase structures between words divided by the analysis process according to the syntax information and semantic information and a predetermined analysis rule regarding the sentence structure. In the abbreviation completion device which inputs and abbreviates the omitted word in the section of the input sentence, the abbreviation completion device determines a word to be a case component among the abbreviations in the input section in a predetermined manner. A case component elimination complementer that complements according to the abbreviation completion rule, a subject determination unit that determines the subject of the section in which the word to be the case component is complemented according to a predetermined theme determination rule, and a case component elimination complementer. A context information holding unit that stores the clause in which the word to be complemented and the subject determined by the subject determining unit as context-related information, and the case component elimination complement unit includes a predicate in the input clause. On the other hand, based on the syntax information and the semantic information added by the morphological analysis process, an abbreviated case detection unit that detects a case in which a component is omitted in the input clause and a condition that a noun phrase of the case component should satisfy, Based on the syntactic information and semantic information added by the morphological analysis processing, a word that satisfies the condition among the previous sentences or clauses stored in the context information holding unit is replaced by a case component in which the word is omitted. A supplementary candidate selecting means for selecting as a possible word, and for a word selected by the complementary candidate selecting means, a predetermined abbreviated complementing rule for the probability that the case component is omitted. Priority determining means for determining priorities in order from the words determined to be highly probable, and the section in which the highest word of the priority is input as the word to be the case component The subject determining unit selects a word that can be a subject of the input clause from the input sentence or a previous sentence or clause stored in the context information holding unit. Candidate selecting means, for the word selected by the subject candidate selecting means, determine the probability of the subject of the input clause by applying a predetermined subject determination rule and referring to predetermined vocabulary relation information, Subject determining means for determining a word determined to have the highest probability as the subject of the input section.

According to a fourth aspect of the present invention, there is provided a morphological analysis process for dividing a Japanese sentence into words, and adding syntax information and semantic information of the words to the words. In a Japanese parsing system that performs a parsing process for parsing a sentence structure such as a dependency relationship or a phrase structure according to the above syntax information and semantic information and a predetermined parsing rule regarding the above sentence structure, the omission in the section of the input sentence is omitted. An abbreviation completion unit that complements the selected word, and the abbreviation completion unit complements a word to be a case component among the abbreviations in the input section according to a predetermined abbreviation completion rule. And a subject determination unit that determines the subject of the section in which the word to be the case component is complemented according to a predetermined subject determination rule, and completion of the word to be the case component by the case component elimination complement unit. And a context information holding unit that stores the subject determined by the subject determining unit as context-related information, and the case component omission complementing unit performs the morphological analysis process on the predicate in the input clause. Based on the added syntax information and semantic information, an abbreviation case detecting means for detecting a case in which a component is omitted in the input clause and a condition that a noun phrase of the component of the case should satisfy, and Based on the syntactic information and the semantic information, and selects a word that satisfies the above condition from the previous sentence or clause stored in the context information holding unit as a word that can be a case component in which the word is omitted. By applying a predetermined abbreviated complementing rule to the selecting means and the word selected by the complementing candidate selecting means, the probability that the case component is abbreviated is applied to the word. Priority order determining means for determining the order of priority in the order of the words determined to be highly probable, and abbreviations to be applied to the input section as the word to be the case component with the highest word in the priority order Subject means selecting means for selecting from the previous sentence or section stored in the input section or the context information holding section, the subject determining section, the subject determining section comprising: The probability that is the subject of the input section is determined for the word selected by the subject candidate selecting means by applying a predetermined subject determination rule and referring to predetermined abandonment-related information, and determining that the probability is the highest. Subject determining means for determining the input word as the subject of the input section.

According to a second aspect of the present invention, the abbreviation case detecting means provided in the case component elimination complementing unit has a case in which at least a particle "ga", "wo" or "ni" is displayed. It is characterized by detecting cases in which components are omitted for cases other than temporal cases and place cases and conditions to be satisfied by noun phrases of the components of the cases.

According to the third and sixth aspects of the present invention, words to be used as adjunct modification components among the abbreviations in the above section are complemented on the basis of the relationship between the nouns used as the basis for the subject determination of the section by the subject determining section. It is characterized by having a continuous modification component complementing section.

[0013]

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an abbreviation completion device according to an embodiment of the present invention.

As shown in the figure, an abbreviation complementing device 1 of the present embodiment includes a case component abbreviation complementing unit 2 for complementing a word to be a case component among abbreviations in a section of an input Japanese sentence. A subject determining unit 3 for determining the subject of a section in which a word to be a case component has been complemented, and a continuous modifier abbreviation complementing unit 4 for complementing a word to be a continuous modification component among abbreviations in such sentences or clauses
And a context information holding unit 5 for storing the analysis result of the section and its subject as information related to context (hereinafter, referred to as context information). Further, in this embodiment, a collection of abbreviation completion rules 6 for complementing abbreviations, a collection of subject determination rules 7 for determining a subject, and a vocabulary relation database 8
Is connected to an external memory.

In the abbreviation complementing apparatus 1 of the present embodiment, the Japanese sentence is divided into words by a syntactic analysis device or the like, and the sentence is divided by morphological analysis processing in which syntactic information and semantic information are added to each word, and morphological analysis processing. A sentence clause that has been subjected to syntax analysis processing for analyzing the sentence structure such as the dependency relationship between words and the phrase structure according to the syntax information and semantic information and a predetermined analysis rule regarding the sentence structure is input. As means for performing these morphological analysis processing and syntax analysis processing, means conventionally used can be used.

The case component elimination complementing unit 2 detects the case component (hereinafter referred to as the abbreviated case) in which the word is omitted in the input section as shown in FIG.
A complement candidate selection means 22 for selecting a word corresponding to the abbreviation from context information of a previous sentence or section stored in the context information holding unit 5; Priority determining means 23 for determining the order of priority from words having a high likelihood corresponding to the abbreviation by applying the middle rule (abbreviation completion rule), and the position of the abbreviation in the section where the word with the highest priority is input And the abbreviation complementing means 24.

When a clause of a sentence subjected to the morphological analysis processing and the syntax analysis processing is input, the abbreviated case detecting means 21 pays attention to the predicate in the input clause, and applies the syntactic information and the semantic information added thereto. Based on the predicate, a word that is actually related to the predicate is checked to determine whether or not there is an abbreviated case and a condition that the noun phrase of the component of the case should satisfy.

When an abbreviation is detected, the complement candidate selection means 22 determines a word satisfying the condition of the abbreviation from the context information of the context information holding unit 5 based on the syntax information and the semantic information of the predicate to which the abbreviation is related, that is, Select a candidate for completion. Here, the context information holding unit 5 stores, as context information, the analysis result of the section for which the abbreviation completion processing has been performed previously and the subject of the section.

When a complement candidate is selected, the priority order determination means 23 applies the rules of the omitted complement rule set 6 stored in the external memory to each complement candidate, and the probability that each complement candidate corresponds to the abbreviation is applied. To determine how much there is. Then, the priority order is determined in order from the most probable complement candidate.

Here, the abbreviation complementing rule set 6 includes a probability that the abbreviation corresponds to an abbreviation on condition that the complementing candidate has a characteristic (subject, case component, etc.) in a previous sentence or section. A collection of rules that score how much there is. Then, the likelihood that each complementary candidate corresponds to the abbreviation is evaluated based on the total score of the matching rule among the omitted complementary rules.

The abbreviation completion rule is based on the fact that the previous subject is likely to be an abbreviated case component that displays the particles "ga", "wo", and "ni", and the word of the case component of the previous clause or the main clause of the previous sentence. And words that modify them, words that are used for abbreviation completion there, and words that are likely to be the subject of the next section, such as words that are figuratively localized there, are likely to be components of the abbreviated case However, the fact that the word related to the particle "ga" in the subordinate previous clause, such as the clause describing the cause and reason, does not become a word that complements the nominative case, etc. I have.

The following table shows an example of an omission complement rule. The score of each rule in the table is set based on a survey of linguistic phenomena, and the higher the score, the higher the probability of being abbreviated.

[0024]

[Table 1]

When the priority order of the complementing candidates is determined, the abbreviation complementing means 24 complements the complement candidate having the highest priority to the section input as a word corresponding to the abbreviation. Here, if there are a plurality of abbreviations to be complemented and the same word is a candidate for a plurality of abbreviations, the combination of the most probable (highest total score of rules) and abbreviations Prioritize matching. The other abbreviations are complemented by complement candidates with the next highest priority.

When completion of the abbreviation is completed, the abbreviation completion means 24 sends the complemented clause to the subject judging section 3 and the continuous modifier complement section 4.

When there is no abbreviation in the input section (when the abbreviation detection means 21 does not detect an abbreviation), the abbreviation is not subjected to the above-mentioned abbreviation completion processing and the relevant section is replaced with the subject determination section 3.
And sent to the union-modified sentence complementing section 4.

As shown in FIG. 3, the subject determining unit 3 inputs a clause in which case components are complemented, and searches for a word which can be the subject of the clause, in the previous sentence or clause of the context information holding unit 5. Subject candidate selection means 31 for selecting from context information
To apply the rule (subject determination rule) in the subject determination rule collection 7 and the vocabulary relation information in the vocabulary relation database 8 to the selected word to determine the word having the highest probability as the subject as the subject. Determining means 32.

When the section whose abbreviation is complemented by the case component abbreviation complementing section 2 is inputted, the subject candidate selecting means 31 outputs a word to be the subject of the inputted section, that is, the section itself or the context in which the subject candidate was inputted. It is selected from the context information of the information holding unit 5.

The subject candidate selecting means 31 selects words specified in the input section or context information and abbreviations complemented by the case component elimination complementing unit 2 as subject candidates. From the input clauses, the subject that will be selected as the subject candidate is the particle “ha”, “kano”, “mo”, “if”, “nara”
Words pertaining to etc., particle "ga""wo"
Words supplemented by abbreviations displayed in "Ni" and "I think"
The subject of the section related to "to" in the section. From the contextual information, the candidates for the subject are the subject of the previous sentence or section, the words of the case components of the previous sentence or section, the words that modify them, and the words that are figuratively localized there. Such as words that are likely to be the subject of the next section.

When the subject candidates are selected, the subject determining means 32 applies the rules of the subject determination rule collection 7 stored in the external memory to each subject candidate, and determines the probability that each subject candidate corresponds to the subject. Determine if there is. Then, the most probable word is determined as the subject of the input section.

Here, when the subject candidate is selected from the input section, the subject determination rule 7 determines the position of the subject candidate in word order in the section. Role of the subject, whether it is the same as the subject of the previous section, or has a conceptual relationship with the subject of the previous section, such as higher or lower, and the subject candidate was selected from the context information If the subject candidate has a conceptual relationship, such as higher or lower, with the noun phrase of the input clause, or if the subject candidate is specified in the case component, etc. Rules that score the degree of the corresponding probability are collected. Then, the likelihood that each subject candidate corresponds to the subject is evaluated based on the total score of the matching rule among the subject determination rules. The following table shows examples of subject determination rules. Note that the score of each rule in the table is set based on a survey of linguistic phenomena, and the higher the score, the higher the probability that the rule corresponds to the subject. In the table, "Noun phrase"
The “second noun phrase” is a number in which the noun phrases in the input clause are numbered in order from the previous one.

[0033]

[Table 2]

The subject determining means 32 refers to the information in the vocabulary relation database 8 when determining the conceptual relationship between words and the words in the context information, such as words and the upper and lower words. The vocabulary relation database 8 stores a large number of information such as which noun has which relation to which noun. The relation between the stored nouns includes upper and lower, set and element, whole and part, entity and attribute, and the like.

When the subject is determined, the subject determining means 32 sends the information to the continuous modification component complementing unit 4. Note that if the input section is of a type that has no subject (for example,
Subordinate clauses and related clauses) send such clauses to the continuous modification component omission complementing unit 4 without performing the above-described subject determination processing.

The above-mentioned continuous modification component complementing unit 4 is provided with a subject determining unit 3
Based on the relationship between the nouns that became one of the grounds for the subject's decision, the abbreviation of the input clause is complemented with the word that should be the noun modification component. Then, the clause and the subject complemented by the continuous modification component are sent to the context information holding unit 5, and the complemented clause is output and sent to a parser or the like. With this complementation, when abbreviations of the next sentence or section are complemented or the subject is determined, the amount of judgment is increased, and the accuracy can be improved. Also, when the present embodiment is applied to a machine translation system, more accurate translation processing can be performed.

The continuous modification component complementing section 4 is not an essential component, and may not be provided if it is not necessary to supplement the abbreviation of the continuous modification component. In this case, the section whose case component has been complemented by the case sentence elimination complementing unit 2 and the subject determination unit 3
Is sent directly to the context information holding unit 5 from.
The case component complementing section 2 directly outputs the clause in which the case component has been complemented, and sends the clause to the parsing device or the like.

The context information holding unit 5 stores, as the context information, a section and its subject in which the input abbreviation is complemented. The stored context information is used for complementing the abbreviation of the section to be input next and determining the subject.

The abbreviation completion apparatus 1 of the present embodiment configured as described above may be used alone or may be used by being incorporated inside a Japanese analysis system, a machine translation system, or the like. FIG. 4 shows a configuration example in the case of being embedded in a Japanese analysis system.

As shown in the figure, when an abbreviation completion device is included in the Japanese parsing system, the syntactic analysis device 9 sends the morphologically analyzed and syntactically analyzed sentence sections to the case component elimination completion unit 2. Then, the analysis result of the clause is returned to the syntax analysis device 9 via the subject determination unit 3 and the continuous modification component complementing unit 4. In addition, a word to be complemented is sent to the syntactic analyzer 9 once, and may be used as a new analysis result in addition to the analysis result of the sentence or clause (such as a parse tree indicating the dependency relationship of the words in the sentence or clause). it can.

Next, according to the Japanese language analysis system shown in FIG. 4, the abbreviation of the sentence "Bio-related strains are attractive, but small-cap stocks are large and it is difficult to recommend to institutional investors. Price movements are more intense than expected." The operation of the present embodiment will be described by taking the case where complementation is performed as an example. First, when the morphological analysis and the syntax analysis have been performed by the syntax analysis device 9 until “the bio-related strain is attractive,” the analysis result is replaced by the case component elimination complementing unit 2.
Sent to

The case component elimination complementing unit 2 pays attention to the predicate "attractive" in the input analysis result and, based on the information extracted from the "attractive" section of the dictionary, converts "attractive" to a particle. The case that does not have the case to display "wo" or "ni" (hereinafter referred to as "wo" case or "ni" case) and the case to display the particle "ga" (hereinafter referred to as "ga" case) However, it is judged that there is no abbreviation because "bio-related strains" are already involved from the analysis results. Therefore, the analysis result is sent to the continuous modification component elimination complementing unit 4 without complementing the abbreviation.

The subject determining unit 3 determines that there is no subject in this section since there are no words that are subject candidates in both the analysis result and the context information. Then, the determination result is sent to the continuous modification component omission complementing unit 4 and the context information holding unit 5. Further, since there is no subject and the existence of conceptual relationships such as upper and lower levels is not used for determining the subject, the continuous modification component omission complementing unit 4 does not supplement the abbreviations, and the analysis result is directly used as the context information holding unit 5 To the parsing device 9.

Next, the parsing device 9 analyzes up to “bio-related strains are attractive, but many small strains”, and the analysis result is sent to the case component elimination complementing section 2.

The case component elimination complementing unit 2 complements the abbreviation by focusing on the predicate “many”. From the dictionary information of “many”, “many” has “ga” case and “ni” case, but “ga” case is related to “small-cap stock”. It is detected that the case is omitted and the noun condition related to the case is detected. Then, the complementing candidate “bio-related strain” is selected by the complementing candidate selecting means 22, the probability corresponding to “Ni” is determined by the priority order determining means 23, and the “bio-related strain” is predicated by the abbreviation complementing means. It is placed on the "ni" of "many". The table below shows an example of applying the abbreviated complementing rule to the complementing candidate "bio-related strain".

[0046]

[Table 3]

The subject determining unit 3 determines the subject of this section based on the complemented analysis result and the context information of the context information holding unit 5. Since the abbreviation-completed word becomes a subject candidate, the subject candidate selecting means 31 selects “bio-related strain”.
Is selected as a subject candidate. In addition, the case component "bio-related strain" of the "ga" case in the previous section and "this" showing the entire previous section
Is a subject candidate, and the subject determining means 32 determines the probability that “bio-related strains” will be the subject of this section,
The “bio-related strain” is determined as the subject of this section, and sent to the noun modification component complementing unit 4 and the context information holding unit 5. In addition, when the same word is a subject candidate in a clause and a subject candidate outside a clause at the same time, it is treated as a subject candidate in a clause only and the subject determination rule is applied. Table 4 shows an example of applying the subject determination rule to each subject candidate.

[0048]

[Table 4]

The noun modification component complementing unit 4 complements the noun modification component because the same word as the subject “bio-related strain” is contained in the context information, and the conceptual relationship such as upper and lower is not used as the basis for determining the subject. The analysis result that has been complemented is sent to the syntax analysis device 9 and the context information holding unit 5 without performing.

The syntax analyzer 9 continues the analysis of the next sentence or clause based on the above complementation. Therefore, the context information holding unit 5 stores the analysis result that “bio-related strains are attractive, but (1) bio-related strains are often small-sized strains” and the subject “bio-related strains” in this section.

Next, the parsing device 9 analyzes "bio-related strains are attractive, but {bio-related stocks are small, many are small and difficult to recommend to institutional investors"". Sent to unit 2.

The case component elimination complementing unit 2 complements the abbreviation by focusing on the predicate "recommend". First, the abbreviation case detection means 21 detects that the "ga" case and the "wo" case have been omitted, and detects the noun condition associated with each case. Then, the complement candidate of each case is selected from the context information by the complement candidate selecting means 22. Here, there is no word corresponding to the “ga” case, and only “bio” -related “bio-related stock”, “small-cap stock” and “this (representing the entire previous section)” are selected as complementary candidates.

When the complementing candidates are selected, the priority determining means 23 determines the probability that each of the complementing candidates corresponds to the “を” case, and the priority order is “bio-related stock”, “small-cap stock”, and “this”. Is determined. Table 5 shows an example of applying the abbreviated complement rule to each complement candidate.

[0054]

[Table 5]

When the priority of the complementing candidate is determined, the abbreviation complementing means 24 ranks the "bio-related strain" with the highest priority in the predicate "recommend".

The subject determining unit 3 determines the subject of this section from the complemented analysis result. First, the “bio-related strain” supplemented abbreviated by the subject candidate selection means 31, the subject “bio-related strain” of the previous section and the case component “small strain” of the “ga” case of the previous section are included in the section. Selected as a subject candidate.

When the subject candidates are selected, the subject determining means 3
2, the probability that each subject candidate becomes the subject of this section is determined, and the “bio-related strain” having the highest probability is determined as the subject in the section.

The noun modification component complementing section 4 does not perform complementation because the same word as the subject “bio-related strain” is in the context information and does not use a conceptual upper-lower relationship as a basis for subject determination. The analysis result and the like are sent to the syntax analysis device 9 and the context information holding unit 5. Therefore, the context information storage unit 5 analyzes the results that "bio-related strains are attractive, but (1) bio-related strains are (1) many small-cap stocks, (2) bio-related strains are difficult to recommend to institutional investors," and the subject of this section. "Bio-related strains"
Are stored.

Next, the syntax analyzer 9 analyzes the sentence "The price change is more intense than expected." The analysis result is sent to the case component elimination complementing unit 2.

The case component elimination complementing unit 2 pays attention to the predicate "intense", judges that there is no omission of the case component in any of the "ga" case, the "wo" case, and the "ni" case. It is sent to the judgment unit 3.

The subject determining unit 3 determines the subject of this section based on the analysis result. First, “bio-related stock”, “institutional investor”, and “this (representing the entire previous section)” are selected as the subject candidates by the subject candidate selecting means 31.

When the subject candidates are selected, the subject determining means 3
2, the probability that each subject candidate becomes the subject of this section is determined, and the “bio-related strain” having the highest probability is determined as the subject.

[0063] adnominal component compensating unit 4, subjected based on the relationship between a noun detected during subject determination (principal phenomena relevant here) the "biotechnology Ltd." to "price movements", the syntax analysis unit 9 and the results It is sent to the context information holding unit 5. Therefore, the context information holding unit 5 stores the analysis result that “｛value movement of bio-related strains is more intense than expected” and the subject “bio-related strains” of this sentence.

The abbreviation complement processing for the sentence "Bio-related stocks are attractive, but small-cap stocks are hard to recommend to institutional investors. Price movements are more intense than expected." The analysis result output from the Japanese analysis system as a result of this processing is as follows: "Bio-related strains are attractive, but (1) Bio-related stocks are (1) many small-cap stocks, and (2) Bio-related stocks are difficult to recommend to institutional investors. The price movement of bio-related stocks is more intense than expected. "

[0065]

As described above, according to the first, second and fourth and fifth aspects of the present invention, an abbreviation complementing apparatus for supplementing abbreviations for Japanese sentences is provided with a case component elimination complementer and a subject. A context information storage unit that stores a context or a sentence whose abbreviation has been previously complemented; and stores the context information with the abbreviation complement or subject of the next sentence or clause. Since it is used at the time of determination, abbreviations with high accuracy can be complemented.

Also, by using the subject of a previous sentence or section as context information, the case component of the case in which the Japanese particle "ga", "wo" or "ni" is closely related to the subject, particularly in Japanese sentences. Has an effect that highly accurate complementation can be performed.

Further, the subject determining section selects subject candidates from the section or the context information in which the subject candidates are being processed, and the subject having the highest probability by judging the probability of being the subject of each selected subject candidate. By providing the subject determining means for determining a candidate as the subject of the section, the subject of the section can be determined accurately, thereby improving the accuracy of the abbreviation completion device using the determined subject. be able to.

According to the third and sixth aspects of the present invention, based on the relationship between nouns which is one of the grounds for determining the subject of a section by the subject determining section, it becomes a continuous modifier component of the abbreviations of the above section. Providing the adjunct modification component complementer that complements a power word can provide more accurate processing when analyzing the next sentence or clause, completing abbreviations, and applying it to a machine translation system. There is.

[0069]

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating an abbreviation completion device according to an embodiment of the present invention.

FIG. 2 is a functional block diagram illustrating functions of a case component elimination complementer of FIG. 1;

FIG. 3 is a functional block diagram illustrating functions of a subject determination unit in FIG. 1;

FIG. 4 is a block diagram showing a Japanese language analysis system according to one embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Abbreviation word completion device 2 Case component omission completion part 3 Subject judgment part 4 Continuous modification component omission completion part 5 Context information storage part 6 Abbreviation completion rule collection 7 Subject judgment rule collection 8 Lexical relation database 9 Syntax analyzer 21 Abbreviation case detecting means 22 Complementary candidate selecting means 23 Priority order determining means 24 Abbreviation word complementing means 31 Subject candidate selecting means 32 Subject determining means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 17/27-17/28

Claims (6)

(57) [Claims] 1. A morphological analysis process that divides a Japanese sentence into words and adds syntax information and semantic information of the words to the words, and the dependency relationships and phrase structures of the words divided by the morphological analysis process. A sentence section that has been subjected to a syntax analysis process for analyzing the sentence structure according to the syntax information and semantic information and a predetermined analysis rule regarding the sentence structure is input, and the omitted words in the input sentence section are input. In the abbreviation completion device for complementing, the abbreviation completion device includes a case component abbreviation completion unit that complements a word to be a case component among the abbreviations in the input section according to a predetermined abbreviation completion rule; A subject determining unit that determines the subject of the section in which the word to be completed is complemented in accordance with a predetermined subject determining rule; and a section in which the word to be a case component is complemented by the case component elimination complementing unit and the subject determining unit. A context information storage unit for storing the determined subject matter as the information about the context I, the price components omitted complementing unit, the syntax information to the predicate in clauses entered added by the morphological analysis and An abbreviation case detecting means for detecting a case in which a component is omitted in an input clause and a condition to be satisfied by a noun phrase of the case component based on the semantic information, and syntactic information and meaning added by the morphological analysis processing
Based on the information, previously stored in the above context information holding unit
Words that satisfy the above conditions in the sentences or clauses of
Candidate selection means for selecting as a word that can be a case component in which is omitted, and the case formation for the word selected by the complement candidate selection means
Minutes are abbreviated words.
Priority order determining means for judging by applying and determining the order of priority in the order of the words judged to be highly probable, and the section in which the highest word of the priority order is input as the word to be the case component The subject determining unit selects a word that can be the subject of the input clause from the input sentence or the previous sentence or clause stored in the context information holding unit. Candidate inputting means, and inputting the word for the word selected by the subject candidate selecting means.
Apply the prescribed subject judgment rule to the probability that is the subject of the selected section
Judgment by referring to the specified vocabulary-related information.
An abbreviation complementing device comprising: a subject determining means for determining a word determined to have the highest likelihood as the subject of the input section. 2. An abbreviation case detecting means provided in a case component omission complementing unit, wherein at least a case is indicated by a particle "ga", "wo" or "ni", and a component other than a temporal case and a place case is a component. 2. The abbreviation completion device according to claim 1, wherein a condition to be satisfied by the omitted case and the noun phrase of the component of the case is detected. 3. An adjunct modification component complementing unit that supplements a word that is to be a adjunct modification component among the abbreviations in the section based on the relationship between the nouns used as the basis for the subject judgment of the clause by the subject judgment unit. The abbreviation completion device according to claim 1, characterized in that: 4. A morphological analysis process in which a Japanese sentence is divided into words and syntax information and semantic information of the words are added to the words, and a dependency relationship between the words divided by the morphological analysis process, a phrase structure, and the like. A parsing process for parsing the sentence structure of the sentence according to the syntax information and the semantic information according to a predetermined analysis rule regarding the sentence structure. A word complementer, wherein the abbreviation complementer complements a word to be a case component among the abbreviations in the input section according to a predetermined abbreviation completion rule; A subject judging unit for judging the subject of the section in which the power word is complemented in accordance with a predetermined subject judging rule; a clause in which the word to be a case component is complemented by the case component elimination complementing unit; A context information storage unit for storing the subject matter has been determined by the determining unit as the information on the context, the price components omitted complementary portion, syntactic information and meaning the predicate in clauses entered added by the morphological analysis An abbreviation case detecting means for detecting a case in which a component is omitted in an input clause and a condition to be satisfied by a noun phrase of the case component based on the information, and syntactic information and meaning added by the morphological analysis process
Based on the information, previously stored in the above context information holding unit
Words that satisfy the above conditions in the sentences or clauses of
Candidate selection means for selecting as a word that can be a case component in which is omitted, and the case formation for the word selected by the complement candidate selection means
Minutes are abbreviated words.
Priority order determining means for judging by applying and determining the order of priority in the order of the words judged to be highly probable, and the section in which the highest word of the priority order is input as the word to be the case component Subject selection means for selecting a word that can be the subject of the input section from the previous sentence or section stored in the input section or the context information holding section. Means, and inputting the word for the word selected by the subject candidate selecting means.
Applying the prescribed subject judgment rule to the probability
Judge by referring to the predetermined vocabulary information,
A subject determining means for determining a word determined to have the highest probability as the subject of the input section. 5. An abbreviation case detecting means provided in a case component omission complementing unit, wherein at least a case is displayed for a particle "ga", "wo" or "ni", and a component other than a time case and a place case is a component. 5. The Japanese analysis system according to claim 4, wherein a condition to be satisfied by the omitted case and the noun phrase of the component of the case is detected. 6. An adjunct modification component complementing unit that supplements a word that should be a adjunct modification component among the abbreviations in the above clause based on the relationship between the nouns used as the basis for the subject judgment of the clause by the subject judgment unit. The Japanese language analysis system according to claim 4, wherein: