JP3508904B2 - Natural language analyzer - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language analysis device for automatically learning and generating abbreviation element complementing rules for complementing abbreviation elements such as words and expressions abbreviated in natural language.

[0002]

2. Description of the Related Art Conventionally, when constructing an abbreviation element complementing rule in a natural language analysis device, it is general to decode an input sentence to be analyzed and manually construct an abbreviation element complementing rule. In addition, the abbreviated element completion rule constructed in this way is the cause of the abbreviated element completion when the abbreviated element is inappropriately complemented based on the analysis result of the natural language analyzer. Experts who have a good understanding of the internal processing of the natural language analysis device have manually added and modified the following rules.

[0003]

As described above, in the method of constructing the omitted element complementing rule in the conventional natural language analysis apparatus, in order to newly create or add or modify a rule, the internal processing of the natural language analysis apparatus is performed. It is necessary to make rules and verify the analysis results by a well-understood expert. Therefore, there is a problem that only an expert can create new rules or make additional corrections. Also,
Since this work is done manually, there is a problem that the time and economic cost for constructing the rules becomes high.

The present invention has been made in view of the above,
The purpose is to extract the abbreviations in the source language and the complements of the abbreviations in the bilingual sentence of the target language,
An object of the present invention is to provide a natural language analysis device that automatically creates and supplements an abbreviation completion rule that complements abbreviations.

[0005]

In order to achieve the above object, the present invention according to claim 1 parses an input sentence of a first natural language which is a source language, and parses an input sentence of the input sentence. A natural language analysis device for generating an input sentence analysis structure as a language analysis structure, and generating and changing an abbreviation element completion rule for complementing an abbreviation element that is a partial structure omitted in the input sentence analysis structure, Bilingual sentence reading means for reading the input sentence and an ideal translated sentence of a second natural language having a parallel translation relationship with the input sentence, and an ideal translated sentence obtained by analyzing the word expression in the input sentence analysis structure having the parallel translation relation and the ideal translated sentence. A translation expression pair extraction means for extracting an input sentence ideal translation word expression pair that is a pair of word expressions in the analysis structure; and in the input sentence analysis structure,
While the input sentence substructure, which is the structure of the portion corresponding to the element for which the omission flag is set, is extracted as the omission element, for the element for which the omission flag is not set, the input sentence ideal translated word expression pair is input. It is determined whether or not an ideal translated sentence partial structure having a bilingual relationship with the sentence partial structure is present. If the result of the determination is that there is an adjacent translated sentence partial structure that is an input sentence partial structure adjacent to the input sentence partial structure. , And an adjacent ideal translation substructure that is an ideal translation substructure that is adjacent to the ideal translation substructure that has a parallel translation relationship with this input sentence substructure, and uses the extracted adjacent input sentence substructure and adjacent ideal translation substructure. , The adjacent input sentence partial structure does not exist, but the adjacent ideal translated sentence partial structure exists at one place, and the input sentence partial structure having a parallel translation relation with the adjacent ideal translated sentence partial structure exists. When not present in the force sentence ideal translation word expression pair, the input sentence omission element extraction means for extracting the adjacent input sentence partial structure as an omission element and the input sentence ideal translation word expression pair extracted by the bilingual expression pair extraction means are provided. From the extracted input sentence ideal translated word expression pairs, the ideal translated sentence substructures in which there is no bilingual input sentence partial structure are sequentially fetched, and the extracted ideal translated sentence substructures are either pronoun expressions or definite article expressions. , The upper ideal translated sentence partial structure directly related to the ideal translated partial structure is extracted, and the input sentence partial structure having a parallel translation relationship with the extracted upper ideal translated partial structure is extracted as the input sentence ideal. Whether or not any of the extracted input sentence partial structures extracted from the translated word expression pair includes the input sentence omitted element extracted by the input sentence omitted element extraction means It makes a determination,
If there is something that includes the input sentence substructure, it is determined whether the structure in the abbreviated element including the input sentence substructure conflicts with the ideal translated sentence substructure that is either a pronoun expression or a definite article expression. As a result, a target language complement element extracting means for extracting an ideal translated sentence partial structure that does not contradict as a target language complement element of an abbreviated element including the input sentence partial structure, and the target language complement element extracted by this target language complement element extracting means A source language complementing element extracting means for converting and extracting a source language complementing element to be complemented with the omitted element, an input sentence omitting element extracting means, the target language complementing element extracting means, and the source language complementing element extracting means, respectively. The extracted abbreviation element,
An abbreviated element completion that adds a new abbreviated element completion rule or changes an existing abbreviated element completion rule based on an abbreviated element completion rule generated from a pair of abbreviated element completion elements consisting of the target language completion element and the source language completion element The gist is to have a rule generation / change means.

According to the first aspect of the present invention, the input sentence which is the first natural language and the ideal translated sentence which is the second natural language are read, and the omitted elements in the input sentence are extracted from the input sentence and the ideal translated sentence. Then, the target language completion element is extracted from the ideal translated sentence that has a bilingual relationship with the input sentence, and the source language completion element to be complemented by converting this target language completion element is extracted. Input by adding a new abbreviation completion rule or changing an existing abbreviation completion rule based on the abbreviation completion rule generated from the abbreviation completion element pair consisting of the target language completion element and the source language completion element It is possible to automatically learn and accumulate abbreviation completion rules suitable for complementing abbreviations in sentences.

The present invention according to claim 2 provides the invention according to claim 1.
The gist of the invention described above is to have an input sentence analysis unit that analyzes the input sentence read by the parallel translation sentence reading unit to generate an input sentence analysis structure.

According to the second aspect of the present invention, the input sentence read by the parallel translation sentence reading means is analyzed to generate the input sentence analysis structure by the input sentence analysis means, and the input sentence analysis structure is used. To extract the omitted element, the target language complement element and / or the source language complement element.

Further, the present invention according to claim 3 provides the invention according to claim 2.
In the invention described above, the gist is that the input sentence analysis means has an input sentence omitted element complementing means for complementing an omitted element in the input sentence.

According to the third aspect of the present invention, the input sentence analysis means complements the omitted elements in the input sentence by the input sentence omitted element complementing means.

The present invention according to claim 4 provides the invention according to claim 1.
The gist of the invention described in any one of (1) to (3) is that it has an ideal translated sentence analysis unit that analyzes the ideal translated sentence read by the parallel translated sentence reading unit to generate an ideal translated sentence analysis structure.

In the present invention according to claim 4, the ideal translated sentence analysis structure is generated by analyzing the ideal translated sentence read by the parallel translated sentence reading means, and the ideal translated sentence analysis structure is utilized. The abbreviated element, the target language complement element and / or the source language complement element are extracted.

The present invention according to claim 5 provides the invention according to claim 4.
The gist of the invention described above is that the ideal translated sentence analysis means has an ideal translated sentence omitted element complementing means for complementing an omitted element in the ideal translated sentence.

According to the fifth aspect of the present invention, in the ideal translated text analysis means, the omitted elements in the ideal translated text are complemented by the ideal translated text omitted element complementing means.

Further, the present invention according to claim 6 is the same as claim 4
Alternatively, in the invention described in 5, the ideal translated sentence analyzing means has an ideal translated sentence anaphora analyzing means for determining an instruction target of an anaphoric expression in the ideal translated sentence.

According to the sixth aspect of the present invention, the ideal translated sentence analysis means determines the anaphoric expression instructing target such as a pronoun in the ideal translated sentence.

The present invention according to claim 7 is based on claims 1 to 6.
In any one of the inventions, the abbreviated element complementary rule generation / change means has an abbreviated complementary element pair storage means for recording the abbreviated complementary element pair, and the input sentence and the ideal which are in a bilingual relationship to be learned. The gist is to generate and change the abbreviated element complementing rule by utilizing the set of the abbreviated complementary element pairs stored in the abbreviated complementary element pair storage means accumulated based on the set of translated sentence pairs.

According to another aspect of the present invention, the abbreviated element complementing rule generating / modifying means records the abbreviated complementary element pair in the abbreviated complementary element pair storing means, and ideally corresponds to an input sentence in a bilingual relationship to be learned. The abbreviated element completion rule is generated and changed by using the set of abbreviated complementary element pairs stored in the abbreviated complementary element pair storage means accumulated based on the set of translated sentence pairs.

The present invention according to claim 8 is the same as claim 3
The invention according to any one of claims 1 to 7, wherein the input sentence omitted element complementing means complements the omitted element in the input sentence by using the omitted element completion rule, and repeatedly learns the omitted element completion rule. To do.

According to the present invention, the input sentence omitted element complementing means complements the omitted element in the input sentence by using the omitted element complementing rule, and the learning of the omitted element complementing rule is repeated.

Furthermore, the present invention according to claim 9 provides the invention according to claim 8.
The gist of the invention described above is that the abbreviated element complementation rule generation / change means has a learning end determination means for determining the end of learning.

In the ninth aspect of the present invention, the learning completion determination means determines the end of learning in the omitted element complementation rule generation / change means.

According to a tenth aspect of the present invention, in the invention according to the ninth aspect, a source language analysis structure which is a history of the source language analysis structure which is a result of analyzing the input sentence using the omitted element complementation rule. The gist of the present invention is to have a source language analysis structure history storage means for recording a history, and the learning end determination means terminates learning when the recorded contents of the source language analysis structure history storage means reach a steady state.

According to the tenth aspect of the present invention, the source language analysis structure history is a source language analysis structure history which is a history of the source language analysis structure which is a result of analyzing the input sentence using the omitted element complement rule. And the learning is ended when the recorded contents of the source language analysis structure history storage means reach a steady state.

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the arrangement of a natural language analysis apparatus according to an embodiment of the present invention. The natural language analysis device shown in the figure is an abbreviated element complement for a natural language analysis system that automatically learns and generates abbreviated element completion rules for complementing abbreviated elements such as words and expressions omitted in natural language. To implement the rule learning method,
The first natural language is Japanese and the second natural language is English.

The natural language analysis apparatus shown in FIG. 1 includes a parallel translation reading unit 1, a natural language analysis system unit 2, a rule 3, an ideal translation analysis unit 4, an abbreviated complementary element pair extraction unit 5, and an abbreviated element complementary rule generation / change unit. It is composed of 6.

The bilingual sentence reading unit 1 inputs a pair (bilingual sentence) of a first natural language sentence (input sentence) and a second natural language sentence (ideal translated sentence) which are in a bilingual relationship.

The natural language analysis system unit 2 analyzes the input sentence read by the bilingual sentence reading unit 1 using rule 3 to generate an input sentence analysis structure. Natural language analysis system 2
Is composed of an input sentence syntactic and semantic analysis unit 21 and an input sentence omission element complementing unit 22. Rule 3 is syntactic and semantic analysis rule 3
1 and the omitted element complementing rule 32.

The input sentence syntactic / semantic analysis unit 21 of the natural language analysis system unit 2 performs syntactic / semantic analysis on the input sentence read by the bilingual sentence reading unit 1 using the syntactic / semantic analysis rule 31, and the input sentence syntactic / semantic structure. To generate. The input sentence omitted element complementing unit 22 complements the omitted elements in the input sentence based on the input sentence syntactic and semantic structure generated by the input sentence syntactic and semantic analyzing unit 21 using the omitted element complementing rule 32.

The ideal translation analysis unit 4 is a parallel translation reading unit 1.
The ideal translated sentence read in is analyzed to generate an ideal translated sentence analysis structure. The ideal translated sentence analysis unit 4 includes an ideal translated sentence syntax / means analyzing unit 41, an ideal translated sentence omitted element complementing unit 42, and an ideal translated sentence anaphora analyzing unit 43.

The ideal translated sentence syntax / means analysis unit 41 of the ideal translated sentence analysis unit 4 analyzes the syntactic and meaning of the ideal translated sentence read by the parallel translated sentence reading unit 1 to generate an ideal translated sentence syntactic / semantic structure. The ideal translated sentence omitted element complementing unit 42 complements the omitted elements in the input sentence based on the input sentence syntactic and semantic structure generated by the ideal translated sentence syntactic and semantic analyzing unit 41. Ideal translated sentence anaphora analysis unit 43
Determines an anaphoric expression instruction target such as a pronoun in the ideal translated text based on the ideal translated text syntactic and semantic structure generated by the ideal translated text syntactic and semantic analysis unit 41.

The abbreviated complement element pair extraction unit 5 extracts the abbreviated elements from the input sentence for the input sentence and the ideal translated sentence that are in the bilingual relationship read by the bilingual sentence reading unit 1 and converts them into the abbreviated elements from the ideal translated sentence. The target language completion element to be complemented is extracted, and based on the extracted abbreviation element and target language completion element, the source language completion element that is in parallel with the target language completion element is extracted, and the extracted abbreviation element and target language completion element are extracted. An abbreviation complement element pair which is a pair of an element and a source language complement element is extracted. The abbreviated complement element pair extraction unit 5 includes a bilingual expression pair extraction unit 51, an input sentence omitted element extraction unit 52, a target language complement element extraction unit 53, and a source language complement element extraction unit 54.

The bilingual expression pair extraction unit 51 of the abbreviated complementary element pair extraction unit 5 detects the input word expression in the input sentence from the input sentence and the ideal translated sentence which are read by the bilingual sentence reading unit 1 and which are in parallel with each other, and the realization of this input word. An input sentence ideal translation word expression pair, which is a pair of ideal translation word expressions in an ideal translation sentence having a bilingual relationship, is extracted. The input sentence omission element extraction unit 52 is extracted from the input sentence by the input sentence analysis structure generated by the natural language analysis system unit 2 and the ideal translated sentence analysis structure generated by the ideal translated sentence analysis unit 4 and the bilingual expression pair extraction unit 51. The abbreviated element which is an abbreviated element is extracted using the input sentence ideal translation word expression pair. The target language complementing element extraction unit 53 is extracted from the ideal translation by the input sentence analysis structure generated by the natural language analysis system 2 and the ideal translation analysis structure and the parallel expression pair extraction unit 51 generated by the ideal translation analysis unit 4. A target language complement element to be complemented by an abbreviation element is extracted using an ideal translation word expression pair of an input sentence. The source language complement element extraction unit 54 extracts a source language complement element based on the target language complement element extracted by the target language complement element extraction unit 53.

The abbreviated element complementary rule generation / change section 6 generates and changes the abbreviated element complementary rule 32 in the rule 3 based on the abbreviated complementary element pair extracted by the abbreviated complementary element pair extraction section 5.

FIG. 2 shows an example of a basic block configuration of a natural language analysis apparatus using the omitted element complementing rule additionally modified in the present invention. The natural language analysis device shown in the figure includes a first natural language input unit 7 for inputting a first natural language sentence to be analyzed, a natural language analysis system unit 2, a rule 3 used in the natural language system unit, and a natural language analysis unit. The first natural language analysis structure output unit 8 outputs the first natural language analysis structure generated by the system unit 2.

The natural language analysis system unit 2 comprises an input sentence syntactic and semantic analysis unit 21 and an input sentence omitted element complementing unit 22. Rule 3 includes a syntactic and semantic analysis rule 31 and an abbreviation element complementing rule 32. The syntactic and semantic analysis rule 31 includes a syntactic and semantic analysis dictionary 311 and a syntactic and semantic analysis rule 312. The input sentence syntactic and semantic analysis unit 21 of the natural language analysis system unit 2 performs syntactic and semantic analysis on the first natural language sentence read by the first natural language input unit 7 using the syntactic and semantic analysis rule 31, and inputs the sentence. Generate sentence syntactic and semantic structure. The input sentence omitted element complementing unit 22 complements the omitted element of the sentence of the first natural language based on the input sentence syntactic and semantic structure generated by the input sentence syntactic and semantic analyzing unit 21 using the omitted element complementing rule 32.

FIG. 3 shows an example of the syntactic and semantic analysis dictionary 311 according to the embodiment of the present invention. The syntactic and semantic analysis dictionary 311 is
For example, there is a word syntactic and semantic dictionary, and the standard form, part of speech, and semantic attribute for each word of the first natural language, Japanese, are entered in the word syntactic and semantic dictionary. For example, the standard form of “car” is “car”, the part of speech is “noun”, and the semantic attribute is “concrete”. Also, for "wo", the standard form is "wo", the part-of-speech is "particle", and the semantic attribute is "objective". For "I want to buy", the standard form is "buy" and "tai".
It is divided into types, the part of speech is "verb" for "buy", the meaning attribute is "purchase", and the part of speech is "modality" for "tai", meaning attribute Is written as "hope".

FIG. 4 shows an example of the syntactic and semantic analysis rule 312 according to the embodiment of the present invention. Syntax Semantic Analysis Rule 312
For example, there is a verb syntactic / semantic analysis rule, and the verb syntactic / semantic analysis rule includes not only each Japanese verb, but
The verb particle is attached as a pattern, that is,
For example, for the verb "I want to buy", the particle pattern "S" (subject, noun) "ga" (particle) "O" (object, noun) "wo" (particle) "is entered. It is written so as to search the syntactic / semantic analysis structure centered on the verb as a key including the verb, and the selection condition indicating the meaning of the case element of the verb is described.
The particle pattern (S is O
As for the meaning of the case element shown in ()), the verb "buy" is accompanied by the appearance of "hope", and the syntactic and semantic analysis shows that S has a nominative relationship and O has an objective case relationship. When generating a structure, it is indicated that there is a constraint that the syntactic and semantic analysis structure is applied when S has the main "meaning" and O has the meaning of "concrete".

FIG. 5 shows an example of the omitted element complementing rule 32 according to the embodiment of the present invention. The abbreviated element complementing rule 32 is filled with, for example, a syntactic and semantic analysis structural condition and a corresponding complementary element, and complements the complementary element when the syntactic and semantic analysis structure matches the syntactic and semantic analysis structural condition. For example, if an optional (*) verb is accompanied by the modality "prohibition", elements in the nominative case are omitted, and elements in the nominative case are omitted, if not omitted. It shows that the nominative complements "you".

Next, a processing procedure of the omitted element complementing rule learning system for the natural language analysis system having the basic block configuration of FIG. 1 will be described. Here, the procedure of this processing will also be described by taking the bilingual sentence shown in FIG. 6 as an example.

First, the bilingual sentence reading unit 1 inputs a pair (translated sentence) of a first natural language sentence and a second natural language sentence that are in a bilingual relationship. This bilingual sentence is used for automatic learning of the abbreviated element complementing rule for the natural language analysis system. By selecting a set of bilingual sentences (translated sentences) for the learning of this abbreviation complementation rule, the abbreviated elements of the natural language analysis system are suitable for similar field sentences in which abbreviated phenomena similar to the learned bilingual sentences frequently occur. It is possible to tune the complementary processing. In the example of FIG. 6, the first natural language Japanese sentence (input Japanese sentence) “I want to buy a car”
Is input as the bilingual sentence, which is the second natural language English sentence (ideal English translation) "I want to buy my car."

Among the input bilingual sentences, the input Japanese sentence which is the first natural language sentence is the input Japanese sentence which is the first natural language analysis structure by the natural language analysis system unit 2 using rule 3. Generate an analytic structure. The processing procedure of this natural language analysis system will be described using the natural language analysis system having the basic block configuration of FIG.

First, an input Japanese sentence which is a first natural language sentence is, for example, a syntactic and semantic analysis dictionary 311 such as the word syntactic and semantic analysis dictionary shown in FIG. 3 or a verb syntactic and semantic analysis rule shown in FIG. Input sentence syntactic / semantic analysis unit 2 using syntactic / semantic analysis dictionary 31 including syntactic / semantic analysis rules 312
1 to generate a Japanese syntactic and semantic analysis structure. The generated Japanese syntactic / semantic analysis structure is a Japanese sentence whose abbreviation element is complemented by the input sentence abbreviation element complementing section 22 using an abbreviation element completion rule 32 such as the nominative abbreviation element completion rule shown in FIG. Generate an analytic structure. The generated analysis structure of the Japanese sentence is output by the first natural language analysis structure output unit 8.

Information about the analysis structure and rules referred / generated at each stage of the natural language analysis system is stored so that they can be referred to by the abbreviated element complementing rule automatic learning process. For example, individual syntactic and semantic analysis structures and selection conditions obtained when searching the syntactic and semantic analysis rule 312 of FIG. 4 are also stored.

In the above natural language analysis system, the syntactic and semantic analysis dictionary 311 shown in FIG. 3, the syntactic and semantic analysis rule 312 shown in FIG. 4, and the omitted element complementing rule 32 shown in FIG. 5 are used as examples. Thus, the input sentence analysis structure shown in FIG. 7 is generated for the input Japanese sentence shown in FIG.
In these input sentence analysis structures, the words in the input sentence are written according to the relationship between the verb and each element and the function. Although the nominative element is omitted in the input Japanese sentence in FIG. 6, the abbreviated element completion rule 32 in FIG. 5 does not have an abbreviated element completion rule that complements the nominative in the input Japanese sentence. The complementing unit 22 does not complement.

Here, the relationship of "nominative" means that the noun phrase functions as the subject of the verb, and the relationship of "objective" means that the noun phrase functions as the object of the verb. Show.

Next, the ideal translated sentence analysis unit 4 of FIG. 1 will be described. The ideal translated sentence analysis unit 4 automatically generates an ideal translated sentence analysis structure by analyzing the second natural language sentence (ideal English translated sentence) read by the parallel translated sentence reading unit 1. The processing procedure of the ideal translated sentence analysis unit 4 will be described. First, the ideal English translated sentence that is the second natural language sentence is analyzed by the ideal translated sentence syntactic and semantic analysis unit 41 to generate an English syntactic and semantic analysis structure. The generated English syntactic / semantic analysis structure generates an English analysis structure in which the omitted elements are complemented by the ideal translated sentence omitted element complementing unit 42. The English analytic structure in which the omitted elements are complemented generates an English analytic structure to which the information for the anaphoric expression instruction target is added by the ideal translated sentence anaphora analyzing unit 43.

The ideal translated sentence analysis structure shown in FIG. 7 is generated by the ideal translated sentence analysis unit 4 for the ideal English translated sentence of FIG.
Similar to the input sentence analysis structure, these ideal translation analysis structures describe the words in the ideal translation according to the relationship and function between the verb and each element. Since there is no omission in the ideal English translation of Fig. 6,
Although the ideal translation omitted element complementing unit 42 does not complement,
In the ideal translation anaphora analysis unit 43, since "my" is an anaphoric expression, the information of "I" which is the target of this instruction is "m".
It is added in the analysis structure of y ”.

The input sentence analysis structure and the ideal translated sentence analysis structure obtained as described above are the word representation in the input sentence analysis structure and the word representation in the ideal translated sentence analysis structure which are in the parallel translation relation by the parallel translation expression pair extraction unit 51. The input sentence ideal translation word expression pair that is a pair of is extracted. Such an ideal translated word expression pair of an input sentence is extracted by, for example, referring to a Japanese-English bilingual dictionary holding a bilingual element pair of Japanese which is the first natural language and English which is the second natural language. It The bilingual expression pairs of Japanese and English extracted from the input sentence analysis structure and the ideal translated sentence analysis structure shown in FIG. 7 are extracted by the bilingual expression pair extraction unit 51, and an input sentence word expression pair as shown in FIG. 8 is obtained.

In FIG. 8, the analysis structure obtained from the input sentence analysis structure corresponding to the input sentence word expression “want to buy” and the ideal translation analysis structure corresponding to the ideal translated word expression “want to buy” are obtained. Parsing structure word representation pair, parsing structure obtained from the input sentence parsing structure corresponding to the input sentence word representation "car", and parsing structure obtained from the ideal translation parsing structure corresponding to the ideal translated word representation "car" The two types of word expression pairs are extracted by the bilingual expression pair extraction unit 51. In addition,
In the present embodiment, since the natural language analysis system unit 2 that analyzes the input sentence and the ideal translated sentence analysis unit 4 that analyzes the ideal translated sentence are provided, the input sentence ideal translated sentence word expression pair has the analysis structure of the input sentence and the ideal translated sentence. Although extracted, a method of retaining the pair of each word expression itself is also conceivable.

Using the input sentence ideal translated sentence word expression pair, the input sentence analysis structure, and the ideal translated sentence analysis structure obtained as described above, the input sentence omitted element extraction unit 52 extracts omitted elements in the input sentence.

FIG. 9 is a flowchart showing the processing relating to the input sentence omitted element extraction according to the embodiment of the present invention.

In step S5201, the input sentence omitted element extraction unit 52 of the omitted complement element pair extraction unit 5 first extracts the input sentence ideal translated word expression pair from the result of the bilingual expression pair extraction unit 51. In the example of Fig. 8, "want to buy" and "want to
A pair of analytic structures corresponding to "buy" and a pair of analytic structures corresponding to "car" and "car" are extracted.

In the next step S5202, the input sentence analysis structure is extracted from the result of the natural language analysis system unit 2.
The input sentence analysis structure shown in FIG. 7 is extracted for the input sentence shown in FIG.

In step S5203, the ideal translated sentence analysis structure is extracted from the result of the ideal translated sentence analysis unit 4. For the ideal translated text of FIG. 6, the ideal translated text analysis structure shown in FIG. 7 is extracted.

In step S5204, the input sentence partial structure is sequentially extracted from the input sentence analysis structure extracted in step S5202. In the example of the input sentence analysis structure of FIG. 7,
A partial structure corresponding to "want to buy", a partial structure corresponding to an element with the abbreviation flag of the nominative case, and a partial structure corresponding to "car" of the objective case are extracted.

In step S5205, step S52
It is determined whether the input sentence partial structure extracted in 04 has an omission flag. If it is determined that the omission flag is present (YES), the process proceeds to step S5206, and if not (NO), the process proceeds to step S5207. In the above example, since the omission flag is set for the input sentence partial structure corresponding to the nominative element, the determination is YES and step S
The procedure moves to 5206. Further, since neither the input sentence partial structure corresponding to "want to buy" nor the partial structure corresponding to "car" has the omission flag set, the determination is NO,
The process moves to step S5207.

In step S5206, the input sentence partial structure for which the omission flag is set is extracted as an omission element. In the above example, the partial structure corresponding to the element in the nominative omission flag is extracted as the omission element.

In step S5207, the input sentence ideal translated sentence word expression pair is referenced with respect to the input sentence partial structure, and it is determined whether or not there is an ideal translated sentence partial structure having a parallel translation relationship with the input sentence partial structure. If it exists (YES), it moves to step S5208, and if it does not exist (NO), it moves to step S5211. In the above example, the input sentence partial structure corresponding to “buy” and the input sentence partial structure corresponding to “car” are the ideal translated sentence partial structure corresponding to “want to buy” and the ideal equivalent to “car”, respectively. Since the translated sentence substructure exists, both are YES and step S520.
Go to 8. Further, the input sentence partial structure corresponding to the element with the abbreviation flag of the nominative case is NO because there is no ideal translated sentence partial structure having a bilingual relationship, and the determination is NO, and step S52.
Go to 11.

In step S5208, with respect to the input sentence partial structure, an adjacent input sentence partial structure which is an input sentence partial structure adjacent to the input sentence partial structure is extracted, and an ideal translated sentence partial structure having a parallel translation relationship with the input sentence partial structure is extracted. An adjacent ideal translated partial structure that is an adjacent ideal translated partial structure is extracted. Here, a partial structure adjacent to a certain partial structure indicates, for example, a partial structure immediately below which is directly related to the partial structure.
In the above example, the adjacent input sentence partial structure of the input sentence partial structure corresponding to "buy" is equivalent to the input sentence partial structure corresponding to the element with the abbreviation flag of the nominative case and the "car" of the objective case. It is an input sentence partial structure, and there is no adjacent input sentence partial structure of the input sentence partial structure corresponding to “car”. Also,
Adjacent ideal translation substructure corresponding to "want to buy" which has a bilingual relationship with "I want to buy" The ideal translation substructure is an ideal translation substructure corresponding to the nominative "I" and an objective case "car".
It is an ideal translated substructure corresponding to "," and is adjacent to an ideal translated substructure corresponding to "car," which has a parallel translation relationship with "car." The ideal translated substructure is an ideal translated substructure that corresponds to the possessive "my." is there. Although the partial structure directly below the partial structure is referred to as an adjacent partial structure here, a method in which a partial structure other than this definition is also an adjacent partial structure is also conceivable.

In step S5209, step S52
The adjacent input sentence partial structure extracted in step 08, the ideal translated sentence partial structure having a parallel translation relationship with the adjacent input sentence partial structure, and the adjacent ideal translated sentence partial structure do not have the adjacent input sentence partial structure, and the adjacent ideal translated sentence part There is one structure, and it is determined whether or not an input sentence partial structure having a parallel translation relationship with the adjacent ideal translated sentence partial structure does not exist in the input sentence ideal translated word expression pair. If this condition is satisfied (YES), the process proceeds to step S5210, and if not (NO), the process proceeds to step S5211. In the above example, there is no adjacent input sentence partial structure in the input sentence partial structure corresponding to "car", and the adjacent ideal translated sentence portion of the ideal translated sentence partial structure corresponding to "car" that has a bilingual relationship with "car" The structure is possessive "m
There is one ideal translated partial structure corresponding to y ”.
According to the input sentence ideal translated word expression pair of, since there is no input sentence partial structure having a parallel translation relation with “my”, the determination result is YES, and the process proceeds to step S5210. Further, since there are two adjacent input sentence partial structures in the input sentence partial structure corresponding to "buy", the result is NO, and step S521 is performed.
Move to 1.

In step S5210, step S52
The adjacent input sentence partial structure satisfying the condition of 09 is extracted as an omitted element. In the above example, the adjacent input sentence partial structure of the input sentence partial structure of "car" is extracted as an omitted element. In this case, the relationship between the input sentence partial structure of "car" and its adjacent input sentence partial structure is undecided.

In step S5211, it is determined whether or not there is an input sentence partial structure that has not been checked yet.
Finally, when the inspection is completed for all the input sentence partial structures in the input sentence analysis structure generated by the natural language analysis system unit 2, the processing of the input sentence omitted element extracting unit 52 is finished.

With respect to the input sentence analysis structure of FIG. 7, the input sentence omission element extraction unit 52 finally extracts the input sentence omission elements shown in FIG. In the example of FIG. 10, the input sentence omitting element is extracted as a pair of the input sentence partial structure of the input sentence omitting element and the input sentence partial structure immediately above which is directly related.

Next, the processing of the target language complementing element extraction unit 53 will be described. FIG. 11 is a flowchart showing the target language complementing element extraction processing according to the embodiment of the present invention.

In the processing of FIG. 11, first, step S530.
In 1, the target language complement element extraction unit 53 of the omitted complement element pair extraction unit 5 extracts the input sentence ideal translated sentence word expression pair from the result of the bilingual expression pair extraction unit 51. In the example of FIG. 8, a pair of analytic structures corresponding to “buy” and “want to buy” and a pair of analytic structures corresponding to “car” and “car” are extracted.

In step S5302, the input sentence analysis structure is extracted from the result of the natural language analysis system unit 2. Figure 6
For the input sentence of, the input sentence analysis structure shown in FIG. 7 is extracted. In step S5303, the ideal translated sentence analysis structure is extracted from the result of the ideal translated sentence analysis unit 4. For the ideal translated text of FIG. 6, the ideal translated text analysis structure shown in FIG. 7 is extracted. In step S5304, the input sentence omitted element extracting unit 5
The input sentence omission element is extracted from the result of 2. For the input sentence of FIG. 6, the input sentence omitting element of FIG. 10 is extracted.

In step S5305, based on the input sentence analysis structure extracted in step S5302, the ideal translated sentence partial structure in which the input sentence ideal translated sentence word expression pair extracted in step S5301 does not have an input sentence partial structure having a bilingual relationship exists. Take out in order. In the example of the ideal translated sentence analysis structure of FIG. 7, the partial structure corresponding to “I” and the partial structure corresponding to “my” are extracted.

In step S5306, it is determined whether the ideal translated partial structure extracted in step S5305 is a pronoun expression or a definite article expression. If either expression (YES), the process proceeds to step S5307, and if not (NO), the process proceeds to step S5312. In the above example, "I" and "my" are both pronoun expressions, so the determination is yes and the process moves to step S5307. In this embodiment, the pronoun expression and the definite article expression are used as conditions, but other cases may be used as conditions.

In step S5307, the next higher ideal translated sentence partial structure directly related to the ideal translated sentence partial structure which is a pronoun expression or a definite article expression is extracted. In the above example, "I"
The ideal translated partial structure corresponding to "want to buy" is extracted for, and the ideal translated partial structure corresponding to "car" is extracted for "my".

In step S5308, step S53.
An input sentence partial structure having a bilingual relationship with the next higher ideal translated sentence partial structure extracted in 07 is extracted from the input sentence ideal translated sentence word expression pair. In the above example, "I want to buy" is selected from the partial structure corresponding to "want to buy", which is the ideal translated partial structure one level above "I", using the input sentence ideal translated word expression pair in FIG. The input sentence partial structure corresponding to is extracted and "my"
The input sentence partial structure corresponding to "car" is extracted from the partial structure of "car" which is the ideal translated sentence partial structure one above.

In step S5309, step S53.
The input sentence partial structure extracted in 08 is extracted, and it is determined whether or not the input sentence partial structure includes a shorthand expression among the input sentence shorthand expressions extracted in step S5304. If there is an abbreviation that includes the input sentence substructure (YE
S), the process proceeds to step S5310, and if not (NO), the process proceeds to step S5312. In the above example, there is an abbreviation including the input sentence partial structure corresponding to "buy" and an abbreviation including the input sentence partial structure corresponding to "car". Therefore, the determination is YES, and step S53 is performed.
Go to 10.

In step S5310, the structure of the omitted element including the input sentence partial structure extracted in step S5308 determined in step S5309 is the ideal translated sentence part which is the pronoun expression or definite article expression extracted in step S5305. Determine if it conflicts with the structure. If there is a contradiction (YES), the process moves to step S5312,
If there is no contradiction (NO), the process moves to step S5311.

In the above example, the ideal translated partial structure of "I" has the meaning of "subject" and has a nominative relationship with "want to buy", while it has a bilingual relationship with "want to buy". Since there is an abbreviated element that has the meaning of “subject” that has a nominative relationship with “want to buy” in the structure of the abbreviated element that includes the “buy” substructure, the two do not conflict, so N
It is determined to be O, and the process proceeds to step S5311. In addition, the ideal translated partial structure of "my" has the meaning of "subject", its referent is "I", and while it has a relation of possession with "car", it has a bilingual relation with "car". "Cars" in
Since there is an abbreviated element having an undetermined meaning that is in an undecided relationship with the "car" in the structure of the abbreviated elements including the partial structure of No., both are not contradictory, and thus NO is determined and step S5
The process moves to 311.

In step S5311, the ideal translated sentence partial structure determined to be consistent with the structure of the omitted element in step S5310 is extracted as the target language complement element of the omitted element. In the above example, both the ideal translated sentence substructures of "I" and "my" are extracted as the target language complement elements of the abbreviated elements that do not contradict each other.

In step S5312, it is determined whether or not any untranslated ideal translated sentence partial structure remains. Finally, when the inspection is completed for all the ideal translated sentence partial structures in the ideal translated sentence analysis structure generated by the ideal translated sentence analysis unit 4, the processing of the target language complement element extraction unit 53 is finished.

With respect to the input sentence analysis structure of FIG. 7, the target language complementing element extracting unit 53 finally extracts the target language complementing element shown in FIG. It should be noted that the relationship between the omitted element and the “car” that has not been determined as the input sentence omitted element that complements “my” in FIG. 10 is recognized as the possessive case, which is the relationship between “my” and “car”. .

Next, the processing of the source language complementing element extraction unit 54 will be described. The source language complement element extraction unit 54 converts the target language complement element extracted by the target language complement element extraction unit 53 to generate a source language complement element. For the conversion from the target language complement element to the source language complement element, a method using a bilingual dictionary of the source language and the target language can be considered.

For each target language complement element of FIG. 12, the source language complement element shown in FIG. 13 is generated by the source language complement element extraction unit 54.

As described above, the omitted complementary element pair extraction unit 5
The abbreviated complement element generation rule changing unit 6 in FIG. 1 generates an abbreviated element complement rule for the abbreviated complement element pair, which is a pair of the input sentence abbreviated element, the target language complement element, and the source language complement element, and is omitted. A new addition is made to the element completion rule 32, or an abbreviation element completion rule already existing in the abbreviation element completion rule 32 is changed. Regarding the abbreviation complementary element pair in FIG.
Finally, as shown in FIG. 14, a rule entry is added to the omitted element complementing rule of FIG. In the example of FIG. 14, the Japanese verb "buy" is accompanied by the aspect of "hope", and "buy"
Nominative is omitted, and the purpose of “buying” is “car”
Then, if the ownership of "car" is omitted, "buy"
A rule is added to complement "I" to the nominative of "and" to complement "my" to the possessive of "car".

The omitted element complementing rule generation / change unit 6
Input sentence omission elements extracted by the omission complement element pair extraction unit 5,
An abbreviated complement extracted from a set of pairs of input sentences and ideal translated sentences that are the target of learning the abbreviated element completion rule by storing the abbreviated complement element pair that is a pair of target language complement element and source language complement element It is also possible to use a set of element pairs to extract an abbreviation complement rule that is most consistent and effective.

As described above, the one-time process of the omitted element complementing rule learning system for the natural language analysis system having the basic block configuration of FIG. 1 is completed. However, if multiple problems with the same sentence are mixed, not all of them can be extracted in one run, and the dictionaries and rules that have been modified may be worse than before. There is also. Therefore, it is necessary to set the learning end condition and repeat the abbreviated element pair extraction unit 5 and the abbreviated element complementary rule generation / change unit 6 in FIG. 1 until the end condition is satisfied. 1 of learning end judgment
As a proposal, a source language analysis structure history storage unit that holds a history of source language analysis structure generated by the natural language analysis system unit is provided, and it is possible to check whether the source language analysis structure history storage unit is in a steady state.

The procedure of the learning system to be repeatedly executed will be described with reference to the flowchart of FIG. The first time, learning is performed by the same method as shown above. Figure 15 at that point
Step S91 of reading the bilingual sentence of the above and the step S9 of additionally changing the omitted element complementary rule 32 by the omitted complementary element pair extracting unit 5 and the omitted element complementary rule generating unit 6 shown in FIG.
2 is finished. Next, in step S93, the natural language analysis system unit 2 analyzes again using the omitted element complementing rule 32 added and changed in step S92. Next, it is determined whether or not the recorded contents of the analysis structure stored in step S94 as a learning end condition are in a steady state.
At first, it is not the steady state because only the first run is performed. Therefore, the determination is NO, and step S92
Go to. From the second time onward, step S92, step S9
3 is repeated in step S94 until the content retained in the analysis structure reaches a steady state.
Two additional modifications are made. In this way, if the abbreviated complement element pair extraction unit 5 and the abbreviated element complement rule generation / change unit 6 are repeatedly performed, there is no abbreviated element complement rule to be extracted from the above bilingual sentence, and the analysis structure becomes a steady state, so learning is performed at that point. Then, the final abbreviation completion rule is extracted.

The present invention is not limited to the above embodiment, but various modifications and applications are possible within the scope of the claims.

To summarize the above, in the present embodiment,
The first bilingual translation reading unit 1 has a bilingual relationship with each other.
A natural language sentence (input sentence) and a second natural language sentence (ideal translated sentence) are read, and the following processing is performed to automatically create a new abbreviation complement rule and add or correct it. .

That is, the abbreviated complement element pair extraction unit 5 extracts abbreviated elements that are omitted words and / or expressions from the input sentence, and extracts the abbreviated elements and / or expressions that are complemented by the abbreviated elements from the ideal translated sentence. An abbreviation that is a pair of a source language completion element that is a first natural language word and / or expression to be extracted based on an abbreviation element and a target language completion element Extract complementary element pairs. The abbreviated element completion rule generation / change unit 6
The abbreviated complement element rule is generated and changed based on the abbreviated complement element pair extracted by the abbreviated complement element pair extraction unit 5. As a result, it is possible to automatically create and change the abbreviation completion rule for analyzing the abbreviation in the input sentence of the bilingual sentence by the natural language analysis system.

Further, the omitted complement element pair extraction unit 5 includes an input sentence omitted element extraction unit 52 for extracting an omitted element which is an element omitted from the input sentence, and a word and / or expression complemented by the omitted element from the ideal translated sentence. It is possible to provide a target language complement element extraction unit 53 for extracting the target language complement element. As a result, it is possible to identify elements that tend to be translated into target language sentences, such as techniques for extracting omitted elements from source language sentences and fixed expressions such as pronoun expressions from target language sentences. The extraction technique can be used in the omitted complement element pair extraction unit 5.

In addition, the abbreviated complement element pair extraction unit 5 causes the input sentence and the ideal translated sentence to include words and / or expressions in the input word expression and the word in the ideal translated sentence that is in a parallel translation relationship with the input word expression and / or. A bilingual expression pair extraction unit 51 that extracts an input sentence ideal translated word expression pair that is a pair of ideal translated word expressions that are expressions can be provided. In this way, the words and expressions in the ideal translated sentence that have a bilingual relationship with the input sentence can be classified into those that have a bilingual expression in the input sentence and those that do not have a bilingual relation. It is possible to more accurately determine whether the element is a power element.
Further, when there is an element in the input sentence in which the omitted element in the ideal translated sentence has a bilingual relationship with the element to be complemented, it is possible to extract the omitted phenomenon in which the complementary element exists in the same sentence.

Further, the abbreviated complement element pair extraction unit 5 is provided with a source language complement element extraction unit 54 for extracting a complement element in the source language based on the target language complement element extracted by the target language complement element extraction unit 53. be able to. Thus, the position of the omitted element and the element to be supplemented to the omitted element can be extracted only by the first natural language which is the source language.

The input sentence analysis unit analyzes the input sentence to generate an input sentence analysis structure, and uses the input sentence analysis structure to extract omitted elements and / or target language complement elements and / or source language complement elements. be able to. As a result, the abbreviated element that can be judged only by analyzing the input sentence can be extracted, and the target language completion element or source language completion can be more accurately performed by using the constraint condition to the element that can be supplemented to this abbreviated element found by the analysis. Elements can be extracted.

The input sentence analysis unit may include an input sentence omitted element complementing unit 22 for complementing omitted elements in the input sentence. As a result, whether the ellipsis element in each input sentence is a ellipsis element that has not been completed yet and requires a new completion rule, or is an ellipsis element that has already been completed but the wrong element has been completed. , It is possible to distinguish whether or not the correct element has already been complemented and the correct element is the complemented abbreviation element, and it is possible to extract a detailed abbreviation element completion rule.

Further, in the input sentence analysis unit, it is possible to use a natural language analysis system that uses the abbreviated element complementation rule extracted by the present learning method to analyze the input sentence. As a result, the analysis tendency of the natural language analysis system can be directly utilized for extracting the omitted element and / or the target language complement element and / or the source language complement element.

The ideal translated sentence analysis unit 4 analyzes the ideal translated sentence to generate an ideal translated sentence analysis structure, and extracts omitted elements and / or target language complement elements and / or source language complement elements using the ideal translated sentence analysis structure. can do. This makes it possible to more accurately extract the candidate of the target language complement element and the candidate of the source language complement element having the parallel translation relationship from the ideal translation sentence, which are the elements that are likely to be the elements having the translation relationship with the omitted element in the input sentence. The omitted element in the input sentence can be more accurately extracted based on the candidate of.

In the ideal translated sentence analysis unit 4, the ideal translated sentence omitted element complementing unit 42 which complements the omitted elements in the ideal translated sentence.
Can be included. As a result, even if the element to be supplemented to the omitted element in the input sentence is omitted in the ideal translated sentence as well, when the element is complemented by the ideal translated sentence omitted element complementing unit 42, the complemented element is the omitted element of the input sentence. Can be extracted as an element to be complemented with.

Further, the ideal translated sentence analysis unit 4 may include an ideal translated sentence anaphora analyzing unit 43 for determining a target of anaphoric expression such as a pronoun in the ideal translated sentence. As a result, if there is an element in the ideal translated sentence that should be supplemented to the omitted element in the input sentence and it is an anaphoric expression, the ideal translated sentence analysis unit 4
It is also possible to extract the information of the target of the anaphoric expression determined by, and it is possible to extract a more detailed omission element complementing rule. Further, an element to be supplemented to the omitted element in the input sentence exists in the ideal translated sentence, which is the anaphoric expression, and the instruction target in the ideal translated sentence determined by the ideal translated sentence anaphora analysis unit 43 is within the same sentence or the sentence. If an element in the input sentence that is present in another sentence and has a bilingual relationship with the referent in this ideal translation is determined, the element in the input sentence that is bilingual with this referent becomes the omitted element. It can be determined that it is an element to be complemented, and the in-sentence completion rule and inter-sentence completion rule for the omitted element can be extracted.

Further, the omitted element complementation rule generation / change section 6
The abbreviated element completion rule can be generated and modified by using the input sentence analysis structure obtained by the input sentence analysis unit. As a result, it is possible to generate and change a detailed omission element complementing rule according to the analysis structure of the input sentence. Furthermore, by using the natural language analysis system that uses the omitted element completion rules extracted by this learning method for the analysis in this input sentence analysis unit,
The analytic structure output by this natural language analysis system can be directly utilized to efficiently generate the omitted element complement rule.

Further, the abbreviated element complementing rule generation / change section 6 includes an abbreviated element complementary pair recording processing for recording the abbreviated complementary element pair extracted by the abbreviated complementary element pair extraction section 5 in the abbreviated complementary element pair storage section, and learning The abbreviated element completion rule is generated and changed by using the set of abbreviated complementary element pairs stored in the abbreviated complementary element pair storage unit, which is accumulated based on the set of the pair of the input sentence and the ideal translated sentence in the target bilingual relationship. be able to. As a result, the results of various pairs of abbreviated complementary elements extracted from the set of pairs of input translations and ideal translations that are the target of learning are taken into consideration, and the most effective and matching It is possible to extract an abbreviated element complementing rule with good characteristics.

By analyzing the input sentence using the abbreviated element completion rule learning method obtained by this abbreviated element completion rule learning method and repeating the abbreviated element completion rule learning method, the generated and corrected abbreviated element completion rule can be verified. Further, it becomes possible to extract the abbreviated element complementing rule that was not obtained in one learning process.

The end condition determination process automatically determines the learning end condition. As a result, the learning is automatically completed without a person looking at the learning result.

The source language analysis structure recording process records the source language analysis structure history, which is the history of the source language analysis structure, which is the result of the analysis of the input sentence using the above omitted element complement rule, in the source language analysis structure history storage unit. To do. As a result, when the contents recorded in the source language analysis structure history storage unit obtained by the source language analysis structure recording process are in a steady state, learning is automatically terminated and an optimum dictionary or rule is registered.

[0112]

As described above, according to the present invention,
The input sentence and the ideal translated sentence are read, the omitted elements in the input sentence are extracted from the input sentence and the ideal translated sentence, and the target language complement elements that are complemented by the omitted elements are extracted from the ideal translated sentence. The abbreviation complement element pair which is a pair of the abbreviation element and the source language complement element to be complemented to the abbreviation element is extracted, and the abbreviated element complement rule is generated and changed based on the extracted abbreviation complement element pair. As described above, it is possible to automatically accumulate the abbreviation element completion rules suitable for complementing the abbreviation elements in the input sentence without human intervention.

By repeating the abbreviated element complement rule learning method using the abbreviated element complement rule obtained by the abbreviated element complement rule generation / change processing, the registered abbreviated element complement rule can be verified, and further learning can be performed once. It is also possible to extract abbreviation completion rules that were not obtained in the process.

Further, by recording the source language analysis structure obtained by the input sentence analysis processing in the source language analysis structure storage means, the recorded contents in the source language analysis structure storage means obtained by the source language analysis structure recording processing. When is in a steady state, learning is automatically terminated and the optimal omitted element complementing rule can be registered.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a natural language analysis device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a basic configuration of a natural language analysis device that uses an abbreviation element complementation rule that is additionally modified according to an embodiment of the present invention.

FIG. 3 is a diagram showing a structure of a syntactic and semantic analysis dictionary used in the natural language analysis device of FIG.

4 is a diagram showing a configuration of a syntactic and semantic analysis rule used in the natural language analysis device of FIG.

5 is a diagram showing a configuration of an abbreviation element complementing rule used in the natural language analysis device of FIG.

FIG. 6 is a diagram showing an example of a bilingual sentence input to the natural language analysis device of FIG. 1.

7 is a diagram showing a conceptual structure of an input sentence and an ideal translated sentence in the natural language analysis device of FIG.

8 is a diagram showing input sentence ideal translated sentence word expression pairs in the natural language analysis apparatus of FIG. 1. FIG.

9 is a flowchart showing a process of an input sentence omission element extraction unit in the natural language analysis device of FIG.

10 is a diagram showing input sentence omission elements in the natural language analysis apparatus of FIG.

11 is a flowchart showing the processing of a target language complementing element extraction unit in the natural language analysis apparatus of FIG.

12 is a diagram showing an input sentence omission element and a target language complement element in the natural language analysis apparatus of FIG.

13 is a diagram showing an input sentence omission element, a target language complementing element, and a source language complementing element in the natural language analysis apparatus of FIG.

FIG. 14 is a diagram showing an example of an abbreviated element complement rule obtained as a result of adding and changing processing of an abbreviated element complement rule in the natural language analysis apparatus of FIG. 1;

FIG. 15 is a flowchart showing a learning process according to another embodiment of the present invention.

[Explanation of symbols]

1 Bilingual text reading section 2 Natural Language Analysis System Department 3 rules 4 Ideal translation analysis part 5 Omitted complement element pair extraction unit 6 Omitted element completion rule generation / change section 7 First natural language input section 8 1st natural language analysis structure output section 21 Input sentence syntax and semantic analysis part 22 Input sentence abbreviation element completion section 31 syntactic and semantic analysis rules 32 Omitted element completion rules 41 Ideal Translation Syntax Semantic Analysis 42 Ideal Translation Omission Element Complementation Section 43 Ideal translated sentence anaphora analysis section 51 bilingual expression pair extraction unit 52 Input sentence omitted element extraction unit 53 Target language complementing element extractor 54 Source Language Complementary Element Extraction Unit 311 syntactic and semantic analysis dictionary 312 Syntax Semantic Analysis Rule

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-8-101837 (JP, A) JP-A-7-244664 (JP, A) Mihoko Kitamura, Yuji Matsumoto, Translation knowledge from bilingual bilingual corpus Automatic acquisition, IEICE Technical Report NLCC94-2, Japan, May 13, 1994, Vol. 94, No. 32, p. 9-p. 16 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28

Claims (10)

(57) [Claims] 1. An input sentence of a first natural language which is a source language is analyzed, an input sentence analysis structure as a source language analysis structure which is an analysis structure of the input sentence is generated, and in the input sentence analysis structure, A natural language analysis device for generating and changing an abbreviated element completion rule for complementing an abbreviated element which is an abbreviated partial structure, wherein the input sentence and an ideal translated sentence of a second natural language having a bilingual relationship with the input sentence And an input sentence ideal translated word expression pair that is a pair of a word expression in the input sentence analysis structure and a word expression in the ideal translated sentence analysis structure that have analyzed the ideal translated sentence. Bilingual expression pair extraction means, and in the input sentence analysis structure, the input sentence partial structure, which is the structure of the portion corresponding to the element for which the omission flag is set, is extracted as an omission element while the element for which the omission flag is not set is extracted Against Then, it is determined whether or not the input sentence ideal translated word expression pair has an ideal translated sentence partial structure having a bilingual relationship with the input sentence partial structure, and if there is a result of the determination, the input sentence partial structure is present. The adjacent input sentence substructure, which is the input sentence substructure adjacent to, and the ideal ideal translation partial structure which is the ideal translation partial structure adjacent to the ideal translation partial structure which is in parallel with this input sentence partial structure, are extracted and extracted. By using the adjacent input sentence partial structure and the adjacent ideal translated sentence partial structure, the adjacent input sentence partial structure does not exist, but the adjacent ideal translated sentence partial structure exists in one place and has a parallel translation relationship with the adjacent ideal translated sentence partial structure. An input sentence omitted element extracting means for extracting the adjacent input sentence partial structure as an omitted element when the input sentence partial structure does not exist in the input sentence ideal translated word expression pair; The input sentence ideal translated word expression pair extracted by the extracting means is extracted, and the ideal translated sentence partial structure in which the input sentence ideal translated word expression pair that does not have the input sentence partial structure having the bilingual relationship does not exist is sequentially extracted and extracted. When the ideal translated sentence substructure is either a pronoun expression or a definite article expression, the ideal translated sentence partial structure immediately above which is directly related to the ideal translated sentence partial structure is extracted, and the extracted ideal translated sentence partial structure immediately above is extracted. An input sentence partial structure having a bilingual relation is extracted from the input sentence ideal translated sentence word expression pair, and among the extracted input sentence partial structures, there is one including an input sentence omitting element extracted by the input sentence omitting element extracting means. If there is something that includes the input sentence substructure, the structure in the abbreviated element is either a pronoun expression or a definite article expression. A target language complement element extraction means for determining whether or not the target sentence complement structure is inconsistent, and as a result of this judgment, an ideal translated partial structure that does not conflict is extracted as a target language complement element of an abbreviated element including the input sentence partial structure, A source language complementing element extracting means for converting the target language complementing element extracted by the target language complementing element extracting means to extract a source language complementing element to be complemented with the omitted element; the input sentence omitting element extracting means; A new complement based on an abbreviated element completion rule generated from an abbreviated complement element pair consisting of the language complement element extraction means, the source language complement element extraction means, the target language complement element, and the source language complement element, respectively. A natural language having an abbreviated element completion rule generating and changing means for adding an abbreviated element completion rule or changing an already existing abbreviated element completion rule Analyzer. 2. The natural language analysis apparatus according to claim 1, further comprising an input sentence analysis unit that analyzes the input sentence read by the parallel translation sentence reading unit to generate an input sentence analysis structure. 3. The natural language analysis apparatus according to claim 2, wherein the input sentence analysis means includes input sentence omitted element complementing means for complementing omitted elements in the input sentence. 4. The ideal translated text analysis means for analyzing the ideal translated text read by the parallel translated text reading means to generate an ideal translated text analysis structure, according to any one of claims 1 to 3. Natural language analyzer. 5. The natural language analysis apparatus according to claim 4, wherein the ideal translated sentence analysis means includes ideal translated sentence omitted element complementing means for complementing omitted elements in the ideal translated sentence. 6. The natural language analysis device according to claim 4, wherein the ideal translated sentence analysis unit includes an ideal translated sentence anaphora analyzing unit that determines an instruction target of an anaphoric expression in the ideal translated sentence. 7. The abbreviated element complementing rule generation / change means has an abbreviated complementary element pair storage means for recording the abbreviated complementary element pair, and stores a pair of the input sentence and the ideal translated sentence in a bilingual relationship to be learned. 7. The abbreviation element complementing rule is generated and changed using a set of the abbreviation complementary element pairs stored in the abbreviation complementary element pair storage means based on a set. The described natural language analysis device. 8. The input sentence omitted element complementing means complements an omitted element in the input sentence by using the omitted element completion rule, and repeats learning of the omitted element completion rule. The natural language analyzer according to any one of 1. 9. The natural language analysis apparatus according to claim 8, wherein the omitted element complementation rule generation / change means includes learning end determination means for determining the end of learning. 10. A source language analysis structure history storage unit for recording a source language analysis structure history, which is a history of the source language analysis structure, which is a result of analyzing the input sentence using the omitted element complementing rule, 10. The natural language analysis apparatus according to claim 9, wherein the learning end determination means ends learning when the recorded content of the source language analysis structure history storage means is in a steady state.