JPH07175808A - Natural language processor - Google Patents

Abstract

(57) [Summary] [Purpose] The omitted noun phrase is complemented based on the information of the entire preceding sentence. [Structure] Word dictionary 106 and noun thesaurus dictionary 109
And thesaurus dictionary 1 including verb thesaurus dictionary 108
07 and an attribute dictionary 110 that stores attribute information of words. The semantic analysis unit 102 uses the input unit 1
1 based on the dictionary group
The semantic analysis is performed for each sentence, the semantic analysis expressed by the case relation with the 5W1H equivalent phrase having the verb as the core is performed, and the analysis result is stored in the semantic analysis result storage unit 103. The abbreviated information complementing processing unit 104 detects a case element including a noun phrase omitted for each case relation expression with a 5W1H equivalent phrase having a verb as a core, and based on the time-series relation of the verb, a target verb is detected. Verbs having a time series relationship are extracted from the semantic analysis result storage unit 103, and the omitted words and phrases in the semantic analysis result are complemented based on the case elements.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing device, and more particularly to a natural language processing device which can be used in an information retrieval device, a machine translation device, a document summarization support device, etc.

[0002]

2. Description of the Related Art Generally, a technique for processing a sentence composed of a plurality of sentences is called a context processing technique. As one of the techniques, a technique for identifying a referent of a demonstrative pronoun and a context There is a technique for processing the abbreviation of the word by. In particular, Japanese expressions have the characteristic that nouns are often omitted depending on the context, which has been an obstacle to information retrieval, machine translation, document summarization, and the like.

The following methods have been proposed as conventional techniques for processing the omission of words depending on the context. (1) In information retrieval by natural language, an input sentence is analyzed, and a noun phrase that is a sibling to the noun phrase in the preceding context is detected from the input sentence to complement the omitted noun phrase. (Japanese Patent Laid-Open No. 4-220767). (2) In a conversation system, when there are a plurality of instructable objects, the input natural language is converted into a case frame expression, the relative distance between the objects is stored, and the object to be instructed is identified. A "conversation system" for performing relative distance relations between objects (Japanese Patent Laid-Open No. 4-306769).

[0004]

However, in the context processing method described in (1) above, the noun phrase is complemented by using the functional expression table which is the data focusing on the sibling relationship between words. Therefore, the scope of complement is limited. Therefore, if the input noun does not have a sibling relationship with the noun phrase in the preceding context, there is a problem that the omitted noun phrase cannot be complemented for the displayed input character string. .

Further, in the method described in (2) above, the identification of the object to be instructed is performed by using the relative distance of each object, but since it is intended for conversational sentences,
The reference case frame is limited to the latest case frame expression in the conversation history. However, in a real sentence,
Omitting not only the information of the preceding sentence but the entire preceding sentence may occur. Therefore, there is a problem in that the abbreviation complement processing cannot be performed when the input object is pointing to an object in a case frame before the latest.

In view of the above problems, the problem of the present invention is limited to the relationship between siblings of a noun with respect to an abbreviated noun phrase targeted at a sentence that provides a topic consisting of a plurality of sentences. It is an object of the present invention to provide a natural language processing device capable of complementing an instructed noun phrase based on the information of the entire preceding sentence without being limited to the immediately preceding case frame.

[0007]

In order to solve the above problems, the present invention has the following constitution. That is, the present invention is
Input means for inputting a sentence, word information storage means for storing information about words, thesaurus information storage means for storing information about relations between words, including a noun-related word dictionary and a verb-related word dictionary, and word attributes By referring to the word information, thesaurus information, and attribute information, the attribute information storage unit that stores information and a plurality of sentences input from the input unit are analyzed one sentence at a time, and the semantic analysis result is extracted. Semantic analysis means, semantic analysis result storage means for storing the semantic analysis result analyzed by the semantic analysis means, and abbreviation complementation processing for complementing the omitted words in the semantic analysis result based on the thesaurus information. A natural language processing device comprising: means and an abbreviation result output means for displaying a complementation result of the abbreviation expression by the abbreviation complementation processing means.

Further, in the present invention, the semantic analysis means has WHO (subject), WHAT (object), WHEN (time), WHERE (place) and WHY with a verb as a core.
It can be a semantic analysis means for analyzing the meaning expressed in the case relation with the 5W1H equivalent phrase consisting of (reason) and HOW (method).

Further, in the present invention, the verb-related word dictionary included in the thesaurus information storage means is a verb-related word dictionary for storing time series relationships of verbs, and the abbreviation complement processing means is the semantic analysis result. 5W1 with the verb as the core, reading the semantic analysis results to be processed
A case element including an abbreviated noun phrase is detected for each case relation expression with the H equivalent phrase, and the verb of the processing target sentence and the time are detected based on the time series relation of the verb read from the verb-related word dictionary. Abbreviation completion that extracts a verb having a series relation from the result of semantic analysis of a preceding sentence, identifies an abbreviated noun phrase based on the case element of the extracted verb, and complements the abbreviation with the noun phrase It can be a processing means.

[0010]

In the natural language processing apparatus of the present invention, the word information storage means stores information regarding words used in the semantic analysis processing, and the thesaurus information storage means stores the meaning between the words stored in the word information storage means. Stores information about physical relationships. The thesaurus information storage means is composed of a noun thesaurus dictionary and a verb thesaurus, the noun thesaurus dictionary stores information about the upper-lower relation between nouns, and the verb thesaurus dictionary stores information about the time-series relation of verbs. Further, the attribute information storage means stores information regarding the attribute relationship of the words stored in the word information storage means.

The input means inputs a plurality of sentences one by one.
Semantic analysis means, a plurality of sentences input from the input means,
Morphological analysis is performed sentence by sentence using the grammatical information stored in the word information storage means. Then, the meaning analysis is performed using the synonym pattern information of the word information storage means, the thesaurus information storage means, and the attribute information storage means, and the meaning analysis result expressed in the case relation with the 5W1H equivalent words and phrases with the verb at the core is defined. Stored in the analysis result storage means.

The omitted information complementing processing means reads the target semantic analysis result from the semantic analysis results one sentence at a time.
Then, the case relation in which the noun phrase is omitted in the case relation expression with the 5W1H equivalent phrase centered on the verb is detected. Based on the time-series relationship of the verbs in the verb thesaurus, a verb having a time-series relationship with the target verb is extracted from the semantic analysis result of the sentence preceding the target sentence. Then, the verb's 5W1H (WHO, WHAT, WEN, WHE
(RE, WHY, HOW) The omitted noun phrase is identified based on the case relation expression with the equivalent phrase, and the omitted noun phrase in the target semantic analysis result is embedded. Next, the abbreviated expression complementing result complemented by the abbreviated information complementing process is output from the abbreviated information complementing process result output means.

As described above, in a sentence composed of a plurality of sentences, by using the verb thesaurus dictionary that describes the time-series relations of verbs, all preceding sentences are referred to and the omitted noun phrase Complementary processing can be performed, and the sentence structure can be correctly extracted.

[0014]

Embodiments of the natural language processing apparatus of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of the natural language processing apparatus of the present invention. The natural language processing apparatus of FIG. 1 includes (1) an input unit 101 for inputting a sentence, (2) a semantic analysis unit 102 connected to the input unit, and (3) a sentence connected to the semantic analysis unit 102. A word information storage unit 106 for storing information on words necessary for analysis, (4) Connected to the semantic analysis unit 102 and an abbreviation complementation processing unit 104, which will be described later, and stores information on the relationship between words, which will be described later. Noun thesaurus dictionary 1
08 and a verb thesaurus dictionary 109, (5) a noun thesaurus dictionary 108 that stores information about the relationship between the nouns that make up the thesaurus dictionary 107, and (6) the same chronological relationship between verbs. A verb thesaurus dictionary 109 for storing information about
(7) An attribute dictionary 110 that is connected to the semantic analysis unit 102 and stores predetermined attribute information, and (8) 10 in the semantic analysis unit.
(2) Connected to the semantic analysis result storage unit 103 and (9) connected to the semantic analysis result storage unit 103, and based on the thesaurus dictionary information, an abbreviation complementation process for complementing the omitted words in the semantic content expression. Unit 104, and (10) abbreviated information complementation result output unit 105 for displaying the abbreviated expression complementation result by said abbreviation complementation processing unit 104.

Next, the operation of this embodiment having the above configuration will be described. The input unit 101 reads one sentence of a plurality of sentences, the semantic analysis unit 102 performs morphological analysis using the grammatical information of the word information storage unit 106, and further the word dictionary 106.
The syntactic analysis and the semantic analysis are performed using the semantic information and the thesaurus dictionary 107, and the analysis result is stored in the semantic analysis result storage unit 103. The omitted information complement processing unit 104
The semantic analysis result is read from the semantic analysis result storage unit 103. Then, for each of the synonyms of the semantic analysis result, as shown in the float chart of FIG. 2, the noun phrases omitted from the semantic analysis result of the above sentence are classified by case elements based on the time-series relationship of the verb thesaurus dictionary 109. Embedded in, and complements each noun phrase. The omitted information complementation result output unit 105 outputs the result of the omitted information complementation performed by the omitted information complementation processing unit 104. The above-mentioned processing is performed on the whole sentence.

Next, referring to the float chart of FIG.
The operation of the natural language processing device of FIG. 1 will be described. First, the analysis result of the target sentence is read by the target semantic analysis result reading process (step S1), and it is determined whether the target sentence has the above-mentioned sentence (step S2). For each verb, it is linked with the verb in a case relationship 5
A case in which the word equivalent to W1H is undetermined is detected (step S3). Next, the verb thesaurus information reference processing refers to the time-series relationship of each verb (step S4), and sequentially refers to the verbs of the semantic analysis result of the sentence preceding the target sentence (step S5).
Then, the results are compared (step S6), and if they are the same in the above comparison, the case element of the semantic analysis result including the verb is referred to (step S7).

Next, a comparison process is performed with an undetermined case element in the semantic analysis result of the target sentence (step 8). If the comparison process is the same, the noun constraint condition of both elements is referred to (step S9). By referring to the noun thesaurus dictionary, it is determined whether or not there is an upper-lower relationship between the two noun phrases (step S10), and if there is a higher-lower relationship, the content is stored in the undetermined case element information column. Embedded entry (step S11). If they are not the same in the comparison process of step 7, the verb reference process of the preceding sentence is performed. In this verb reference processing, the sentences preceding the target sentence are referred to in ascending order from the preceding sentence, and the entire sentence preceding the target sentence is the search target.

Next, with reference to FIGS. 3, 4 and 5, the omitted information complementing process in the natural language processing apparatus of the present invention will be described. As shown in FIG. 3, the example sentence 1 is “A company has i486.
We have developed a personal computer equipped with. (Hereinafter referred to as sentence 31), and example sentence 2 indicates “to be released in Japan in March 1992.” (hereinafter referred to as sentence 32). First,
In the preceding sentence 31, it is clearly stated that Company A has developed a personal computer, so in sentence 32, "who (WH
The noun phrase "O)" and the noun phrase "what (WHAT)" are omitted. However, since the chronological relationship of “developing” a “thing” and then “selling” the “thing” is clear from the context, the noun phrase omitted in the sentence 32 can be complemented.

Next, the omission-complementing process in this embodiment will be specifically described. In the present embodiment, the sentence 31 is processed by the semantic analysis unit using the word dictionary and thesaurus, and the semantic analysis result 33 is stored in the semantic analysis result storage unit. Figure 4
An example of the content of the word pattern information of the word dictionary used for the semantic analysis processing is shown in FIG. The verb pattern represents a case relation and a noun constraint condition that can be taken by the verb “development” 35, for example. The shaded case elements in the semantic analysis result 33 of the sentence 31 of FIG. 4 represent the case relation information that the verb “development” can take, although the sentence 31 is not extracted from the semantic analysis result. . For a case element whose information is not extracted from such a sentence, a noun phrase is temporarily stored based on the word pattern information of the word dictionary. Similarly, the statement 32 also has a processing result 34.
Get and store.

Next, the processing result 33 of the sentence 31 is read by the target retrieval result reading processing, and it is processed whether or not there is a preceding sentence to the target sentence. If there is no preceding sentence, there is no information that can be referred to and no omission processing is performed. Similarly, the analysis result 34 of the sentence 32 is read. If there is a preceding sentence,
A case element whose case element is undecided is detected in the case relational expression with the 5W1H equivalent word having the verb as the core in the analysis result. For example, from the semantic analysis result 34, the “who case” in which the case element is undetermined is detected in the case relational expression with the 5W1H equivalent word having the verb “release” 36 as the core. Next, the verb thesaurus dictionary reference processing refers to the time-series relationship of the verb “release” 36, and extracts “commercialization” that has a time-series relationship with “release”.

FIG. 5 shows a part of the contents of the verb thesaurus information referred to in this embodiment. Next, the verb of the semantic analysis result 33 of the above sentence is referred to. The verb of the semantic analysis result 33 is not the same as "commercialization", so refer again to the time-series relationship of the verb thesaurus, and refer to "development" that has a time-series relationship with "commercialization". Performs the verb reference process. If the verb of the semantic analysis result 33 is the same as “development”, the case element of the semantic analysis result 33 including the verb “development” 35 is referred to based on the case pattern information of the verb thesaurus. For example, since the case element “who case” of “development” 35 has the same case element as the undetermined case element “who case” in the semantic analysis result 34, the constraint conditions of the noun phrases of both case elements are Perform reference processing. Although details are not specified in the present embodiment, referring to the noun thesaurus information, the noun phrase “A company” in the semantic analysis result 33 and the constraint condition “organization” of the noun in the semantic analysis result 34 show a higher-lower relationship. Since it has it, "A company" is embedded in the information column of the case element "who case" of "release" 36. Similarly, the omission information of the undetermined case element is complemented for every verb, and the processing result 37 is output from the omission information complementation processing result output unit.

[0022]

As described above, the natural language processing apparatus of the present invention includes a verb thesaurus dictionary describing time series relationships of verbs, a noun thesaurus dictionary describing relationships between nouns, an analysis result of language analysis, and Since it is equipped with processing means that complements omitted information by referring to both thesauri described above,
In one sentence composed of a plurality of sentences, it is possible to complement the omitted information by using the sentence of the entire context regardless of the sentence immediately before. As a result, it is possible to correctly extract the structure of a sentence, and it is possible to correctly extract the structure of a sentence composed of a plurality of sentences based on the structure.

[Brief description of drawings]

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

FIG. 2 is a flow chart for explaining an operation of abbreviated information complementing processing by the natural language processing device of FIG.

FIG. 3 is a specific explanatory diagram of omitted information complementing processing.

FIG. 4 is an explanatory diagram showing an example of contents of word pattern information in a word dictionary.

FIG. 5 is an explanatory diagram showing a part of the contents of a verb thesaurus dictionary.

[Explanation of Codes] 101 Input Unit 102 Semantic Analysis Unit 103 Semantic Analysis Result Storage Unit 104 Omitted Information Completion Processing Unit 105 Omitted Information Complementary Result Output Unit 106 Word Dictionary 107 Thesaurus Dictionary 108 Verb Thesaurus Dictionary 109 Noun Thesaurus Dictionary 110 Attribute Dictionary

Claims (3)

[Claims] 1. A thesaurus information storage means for storing a relation between words, including an input means for inputting a sentence, a word information storage means for storing information about words, and a noun-related word dictionary and a verb-related word dictionary. An attribute information storage unit that stores attribute information of a word; and a plurality of sentences input from the input unit, by referring to the word information, thesaurus information, and attribute information,
A semantic analysis unit that analyzes each sentence and extracts the semantic analysis result, a semantic analysis result storage unit that stores the semantic analysis result analyzed by the semantic analysis unit, and a semantic analysis result based on the thesaurus information. 2. A natural language processing apparatus comprising: an abbreviation completion processing means for complementing the omitted words and phrases; and an abbreviation result output means for displaying a result of abbreviation completion by the abbreviation completion processing means. 2. The semantic analysis means according to claim 1, wherein the verb is a core, and WHO (subject), WHAT (object), WWH (time), and WHER are used.
A natural language processing device characterized by being a semantic analysis means for analyzing a meaning expressed by a case relation with a 5W1H equivalent phrase consisting of E (place), WHY (reason) and HOW (method). 3. The verb-related word dictionary included in the thesaurus information storage means according to claim 1 or 2, wherein the abbreviation completion processing means is a verb-related word dictionary for storing time series relationships of verbs. , The semantic analysis result to be processed is read sentence by sentence from the semantic analysis result, and a case element including an abbreviated noun phrase is detected for each case relation expression with a 5W1H equivalent phrase having a verb as a core, and the verb is detected. Based on the time series relationship of verbs read from the related word dictionary, a verb having a time series relationship with the verb of the processing target sentence is extracted from the semantic analysis result of the preceding sentence, A natural language processing device, characterized in that it is an abbreviation complement processing means for identifying an abbreviated noun phrase based on the aforesaid noun phrase and complementing the abbreviated expression with the noun phrase.