JPH0756923A - Japanese sentence generating device capable of eliminating ambiguity - Google Patents

Abstract

PURPOSE:To provide a Japanese sentence generating device generating a Japanese sentence without ambiguity on a computer. CONSTITUTION:A construction check device 3 judges whether a unit sentence including a declinable word is a normal unit sentence which agrees with the regulations of sentence construction information given to the declinable word or not. When it does not agree, correction is urged so that it becomes the normal unit sentence and a sentence connector is detected. It is judged whether two unit sentences connected by the sentence connector have normal structure based on a rule on sentence structure given to the sentence connector or not. When they do not have normal structure, correction is urged so that they have normal structure. A meaning check device 5 uses concept information regulated by a word dictionary storage device 4 and details the modification relation of the respective words in an inputted Japanese original sentence to a range regulated on the computer. A Japanese sentence generation device 6 unifies the processing results of the construction check device and the meaning check device and constitutes the normal Japanese sentence which is uniquely analyzed on the computer.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a Japanese sentence generator for generating a Japanese sentence that can be uniquely interpreted by removing ambiguity from an input Japanese original sentence.

[0002]

2. Description of the Related Art In order to make an electronic computer easier to use, human-machine interfaces (HMI, Hu
It is desired to use natural language for man machine interface). However, when humans describe sentences in natural language, the expressions are often omitted, the expressions are abstracted, and the contents often contain ambiguity. It is not easy to uniquely interpret the included sentence. Therefore, although a method that limits the syntax patterns and terms that can be used in the HMI is partially used in database searches and the like, natural language is generally not used when instructing a processing method to a computer.

[0003]

SUMMARY OF THE INVENTION In view of the above problems, the present invention is particularly directed to Japanese language, and by supplementing omissions in sentences written in Japanese and stricter expressions seen on a computer. We provide a Japanese sentence generator that can remove ambiguity of sentences, and aims to create an environment where end users can easily use a computer with a high degree of directivity using natural language. In particular, it is to enable the sophistication of HMI and natural language-based programming.

[0004]

A Japanese sentence generator according to the present invention is a Japanese sentence generator that removes ambiguity from an input Japanese original sentence and generates a desired Japanese sentence in a conversational manner. Are stored as respective headwords, each headword is provided with a computer attribute of each headword, and conceptual information about other headwords corresponding to a superordinate concept and a subordinate concept for each headword,
Furthermore, when each headword is a syllable, a term dictionary storage unit to which sentence structure information, which is a rule that defines the sentence structure of a unit sentence in which the mnemonic is used, is given, and the input Japanese When the above-mentioned idiom is detected from the original sentence, it is judged whether or not the unit sentence including the above-mentioned idiom is a regular unit sentence that matches the rule of the sentence structure information given to the above-mentioned idiom, and if it does not match. Is displayed on the display unit and prompts the user to correct it so that it becomes a regular unit sentence, and a sentence connector is detected from the input Japanese original sentence, and the sentence connector is combined. It is determined whether or not the two unit sentences that have the regular structure have a regular structure based on a sentence combination rule given in advance for each sentence connector, and if they do not have a regular structure, a display to that effect is provided. The second syntax check that prompts you to correct it by displaying it in the Means and a word in which the other words are not modified in the Japanese original sentence processed by the first and second syntax checking means, the word is used as a headword, and the subordinate concept of the headword is stored in the term dictionary storage unit. A subordinate concept is to read out a word to be represented and, when one of the read out words is selected, hold the selected word and read out a word representing a subordinate concept from the term dictionary storage unit using the selected word as a headword. A first meaning checking unit that repeats until a word having no is selected and a word that is not modified from other words in the Japanese original sentence processed by the first and second syntax checking units finds the word. As a word, a word representing a superordinate concept of the entry word is read from the term dictionary storage unit, and when one of the read words is selected, the selected word is held. Second meaning checking means for repeating the reading of a word representing a superordinate concept from the term dictionary storage unit using the selected word as a headword until a word having no superordinate concept is selected, and input Japanese For the sentence structure of the original sentence,
Words that have been normalized by the second syntax checking means and that have been input by the first and second meaning checking means for each word that does not have a modifier or modified word in the Japanese original text that has been input. And a Japanese sentence composing means for composing a form of a Japanese sentence that can be recognized by a computer from the sentence to which is added.

Further, it is preferable that the Japanese original sentence processed by the first syntax checking means is input to the second syntax checking means and further processed. Further, the Japanese sentence constructing means is configured. In addition to displaying Japanese sentences,
It presents different candidates for words, normalization and refinement of Japanese sentences, and information about errors that occurred in the construction of Japanese sentences, which is information that cannot be recognized from the displayed Japanese sentences during the process of generating Japanese sentences. Is also preferable.

[0006]

In the input Japanese original sentence, the first syntax check means detects each verb and the unit sentence including the detected verb matches the rule of sentence structure information given to the verb. It is determined whether or not it is a regular unit sentence that exists, and if it does not match, a message to that effect is displayed on the display unit, and correction is urged so that it becomes a regular unit sentence. Further, the second syntax check means detects a sentence connector from the input Japanese original sentence, and two unit sentences connected by the sentence connector relate to a sentence structure given to the sentence connector. It is determined whether or not the regular structure is based on the rules, and if the regular structure is not established, a message to that effect is displayed on the display to prompt correction so that the regular structure is obtained.

The first and second meaning checking means use the concept information defined in the term dictionary storage section to detail the modification relation of each word of the input Japanese original sentence up to the range specified on the computer. Turn into. The Japanese sentence generating means integrates the processing results of the first and second syntax checking means and the first and second semantic checking to form a regular Japanese sentence uniquely interpreted on a computer.

[0008]

Embodiments of the present invention will now be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a disambiguating Japanese sentence generation apparatus of the present invention, and FIG. 2 is a sentence example stored in a data storage device from an input device of the embodiment of FIG.
FIG. 3 is a diagram showing a processing result in the syntax check device and the meaning check device, and FIG. 3 is a diagram showing contents stored in the term dictionary storage device of the embodiment of FIG.

When a Japanese original sentence (written as an original sentence) described by the user is given to the input device 1 as input information,
The input device 1 stores the input information in the data storage device 2. The syntax check device 3 analyzes the original sentence in Japanese and expresses the sentence structure in a syntax tree format, as is apparent from FIG. That is, the syntax tree is
It has information indicating the relationship between unit sentences and the role of a phrase with respect to a idiom. Next, paying attention to the terms in the original sentence, information about the sentence structure centered on the terms defined so that the computer can uniquely recognize the sentence structure is stored in the term dictionary storage device 4.
, And compare it with the sentence structure of the original sentence revealed by the result of Japanese analysis. As a result of the comparison, if there is a missing structure in the original sentence, it is presented on the display unit to prompt the completion. Also, if there is an excessive structure, it is suggested to delete it.

Further, the syntax check device 3 pays attention to sentence connectors that connect sentences, and rules regarding sentence structures centered on sentence connectors specified so that a computer can uniquely recognize sentence structures. Based on, the regular sentence structure is compared with the original sentence structure. As in the case of the idiom, as a result of the comparison, excess or deficiency in the sentence structure of the original sentence is complemented, and the complemented result is stored in the data storage device 2 as shown in FIG.

The meaning check device 5 is a data storage device 2.
The processing result stored in the syntax check device 3 is read as input information, and the conceptual information on each word of the input information is detected by referring to the term dictionary storage device 4. Using the referenced conceptual information, the semantic ambiguity and errors of the input information are detected, and the input information is refined to a level at which it can be directly associated with computer expressions (words of higher and lower concepts of each word). Extraction from the term dictionary storage device 4). The detailing result is stored in the data storage device 2.

The Japanese sentence generation device 6 includes a data storage device 2
A Japanese sentence with a limited expression format that can be uniquely recognized by a computer by reading out the processing result in the meaning checking device 5 stored in (1) as an expression limited sentence.
To generate. In the structure added by the syntax check device 3, when the corresponding word is unknown, the sentence is generated by replacing it with a symbol or the like for convenience. The generated expression limiting sentence is stored in the data storage device 2.

The Japanese sentence correction device 7 is a data storage device 2.
The original sentence is read as the input sentence, the sentence structure, the word structure, the word candidate information, the error information and the like generated by the Japanese sentence generation device 6, and the original sentence and the expression sentence are presented.
It supports the identification of words in the conversational form with the user by presenting candidate words and the like for the word candidate existing position and the symbolic expression position in the expression-limited sentence. Further, regarding the error detected by each of the syntax check device 3, the semantic check device 5, and the Japanese sentence generation device 6, the error information is presented in the form of conversation with the user, and depending on the response, correction work may be performed on the expression-limited sentence. To do.

When the expression limitation sentence is corrected, the correction result is stored in the data storage device 2 and the process is returned to the input device 1. If not corrected, the process shifts to the output device 8, and the output device 8 outputs the expression-limited sentence as a Japanese sentence with ambiguity removed.

The term dictionary storage device 4 will now be described with reference to FIG. The term dictionary storage device 4 stores a headline, an attribute, concept information, and sentence structure information for each word, and further responds to a request for reference of word information from each device, to obtain a “headline” of the word.
This is a device that performs information search using the key as a key and returns the search result to each device.

The attribute of each word represents a property of the computer expression when the word is associated with the computer expression, and is classified into a function, a frame, a slot and the like. Also,
Conceptual information indicates the semantic relationship between words, and pointers are set between upper and lower concepts. Furthermore, the sentence structure information is
As a regular sentence structure of a word, the sentence structure is uniquely defined for each word.

FIG. 4 is a flow chart showing the overall procedure of ambiguity in sentence structure and error detection in the syntax check device 3.

FIG. 5 is a flow chart showing the overall procedure of ambiguity in meaning and error detection in the meaning checking device 5, FIG. 6 uses lower-level concept information on words, and FIG. 7 uses higher-level concept information on words. 9 is a flowchart showing a procedure for refining the contents of the input information until it becomes possible to directly associate the contents of the input information with the expression on the computer.

It is assumed that the Japanese sentence "if the product code is equal to" 3 "," fax "exists""shown in FIG. 2A is input from the input device 1 of the embodiment shown in FIG. The input information is stored in the data storage device 2, and control is transferred to the syntax check device 3. If the term dictionary storage device 4 stores the information as shown in FIG. 3, the syntax check device 3 searches the term dictionary storage device 4 by the word heading and obtains the sentence structure information of the word. Further, the syntax check device 3 uses the referred sentence structure information to perform normalization for changing the content represented by the input information into a sentence structure that can be uniquely recognized by the computer. The normalization is performed according to the procedure shown in FIG.

First, the sentence structure of the original sentence (referred to as an input sentence structure) is clarified by analyzing the original sentence read from the data storage device 2 in Japanese (step S1). The input sentence structure is organized by focusing on the dependency relation between the clauses of the original sentence (see FIG. 2B).

Next, the word which becomes a mnemonic in the input sentence structure is focused on as the word A, and the term dictionary storage device 4 is searched by the heading of the word A to obtain sentence structure information (step S2).

The partial sentence structure in the input sentence structure centering on the word A is compared with the sentence structure information representing the regular sentence structure of the word A to detect a structural error (step S3). As a result of the comparison, it is determined whether or not the structures match (step S
4). If the structures do not match, the missing part is complemented and the surplus part is deleted in accordance with the regular sentence structure, the input sentence structure is changed, and the deleted part is stored in the data storage device 2 as error information. (Step S5). It is judged whether or not the inspection for all the idioms in the input sentence structure has been completed (step S6), and if not completed, the process returns to step S2, and if completed, the sentence connector in the input structure is focused and set as the word B,
The partial sentence structure in the input structure centering on the word B is compared with the sentence structure rule representing the regular sentence structure of the word B defined in advance, and a structural error is detected (step S7). As a result of the comparison, it is determined whether or not the structures match (step S
8). If the structures do not match, the missing part is complemented and the surplus part is deleted in accordance with the regular sentence structure, the input sentence structure is changed, and the deleted part is stored in the data storage device 2 as error information. It is stored (step S9). It is judged whether or not the inspections for all sentence connectors in the input sentence structure have been completed (step S10), and if not completed, step S
Returning to 7, if it is finished, the syntax check process is finished.

Although the operation of the syntax check device 3 has been clarified from the above description, it will be described more specifically based on the Japanese sentence in FIG. 2 (a). In the syntax check device 3, first, the Japanese sentence that is the input information is analyzed into Japanese,
A dependency structure as shown in FIG. 2B is generated as the input sentence structure.

When "equal" is selected as the above-mentioned word A from the words in the input sentence structure, the term dictionary storage device 4 is referred to obtain the sentence structure information of "equal". As shown in FIG. 3, “equal” has a nominative case and a subject case as a regular sentence structure, the display order in each sentence is unspecified, and both the nominative case and the subject case have an attribute of data as a word corresponding to the case. You know that you want things.

Next, the partial sentence structure centered on "equal" in the input sentence structure is compared with the regular sentence structure referred from the term dictionary storage device 4. In the example, the input sentence structure is the nominative case, the target case. It can be seen that it matches with the regular sentence structure.

Further, similar processing is performed for "exists" as the word A which is a term. Since “exists” has a nominative case as an input sentence structure (see FIG. 2B), and requires a nominative case and a place case as a regular sentence structure (see FIG. 3), there is no place in the input sentence structure. You can see that the case is missing. For the sake of convenience, the word corresponding to the place is the symbol "XX
The input sentence structure is changed to a regular sentence structure by adding the place structure to the input sentence structure.

After normalizing the sentence structure centered on terms, the sentence structure centered on the sentence connector is checked. The following sentence structure is defined as a regular sentence structure for the sentence connector “Nara”.

Regular sentence structure rule: "A sentence connector" if "is subordinate to two sentences representing a condition and a state" (Fig. 2).
(B) Obvious from the syntax tree structure) When the input sentence structure and the above regular sentence structure are compared, FIG.
From (b), it can be seen that the input sentence structure matches the regular sentence structure.

Since there is no sentence connector other than "if" in the input sentence structure, the iteration is terminated, the normalized sentence structure is stored in the data storage device 2, the syntax check process is terminated, and the control is executed. Move to check device 5 (“if”)
Other examples of sentence connectors include "and" or "or" that specify parallelism, and "and" that specify coordination.

Since the term dictionary storage device 4 stores the information as shown in FIG. 3, the meaning checking device 5 searches the term dictionary storage device 4 by the word heading to obtain the conceptual information of each word. Further, the meaning check device 5 uses the referred conceptual information to refine the content represented by the input information to the extent that the expression on the computer can be dealt with, and associates it with a concrete expression. The detailing is performed according to the procedure shown in FIG. 5 and FIGS.

First, paying attention to the modification relation of the syntactic tree serving as a noun phrase, the word A is defined as a noun that does not modify other words. Based on the subordinate word information of word A, the knowledge expression is refined to a level corresponding to the expression on the computer (step S11).
This detailing is performed by the procedure shown in FIG. No error is detected in the input information, and it is determined whether the refinement has been completed normally (step S12). If the refinement has ended abnormally, the knowledge representation is refined (see FIG. 7) until the meaning is uniquely limited based on the superordinate word information of the word A (step S16). On the other hand, when the refinement has been completed normally, the following processing is performed. It is confirmed that the word A is refined until the word A in the sentence structure matches the attribute required for the corresponding case (step S1).
3). As a result of the confirmation, it is judged whether or not the attributes match (step S14), and if they match, step S16.
If there is no match, the word A, the requested attribute, etc. are stored as error information (step S15),
Control goes to step S16. It is determined whether or not the inspection of all the words in the input information is completed, and if not completed, the process returns to step S11, and if completed, the detailed input information is stored in the data storage device 2 and the process is completed.

Next, the details of step S11 will be described with reference to FIG. The above-mentioned word A is a modified word. The existence of the modified word is confirmed (step S2
1), it is determined whether there is a modified word (step S2)
2). If there is a modified word in step S22, the word A
The existence of a subordinate concept word is confirmed (step S23), and it is determined whether or not there is a subordinate concept word (step S24). When there is a subordinate concept word, it is determined whether or not the modified word is a subordinate concept word (step S25). If the modified word is not a subordinate concept word, the conceptual relationship between the subordinate concept word and the modified word is confirmed (step S26), and it is determined whether or not the relation can be associated (step S27). If you can relate it to the input information,
A sentence structure showing a conceptual relationship between the subordinate concept word and the modified word is added (step S28, path formation-described later), the modified word is targeted, the process returns to step S21, and the above-described processing is repeated (step S29).

If there is no subordinate concept word in step S24, and if it is not possible to relate them in step S27, it is assumed that an error is detected in the input information, and the modified word, subordinate concept information of word A, etc. are used as error information. Is stored in the data storage device 2, and the process ends (step S30). When there is no modified word in step S22, the existence of a subordinate concept word of word A is confirmed (step S31), and it is determined whether or not there is a subordinate concept word (step S32). If there is no subordinate concept word, the process ends, and if there is, one determines whether or not there are a plurality of subordinate concept words (step S33). When there are a plurality of subordinate concept words, one proper subordinate concept word is selected as a modified word and added to the input information, and the unselected subordinate concept word is stored in the data storage device 2 as a modified word candidate. Then (step S34), the process ends. When only one subordinate concept word exists, the subordinate concept word is added to the input information as a modified word (step S35), the added modified word is returned to step S21, and the process is repeated (step S36).

Details of step S16 will be described below with reference to FIG. The presence of a modifier that modifies the above-mentioned word A is confirmed (step S41), and it is determined whether there is a modifier (step S42). Step S42
If there is a modifier, the presence of a superordinate concept of word A is confirmed (step S43), and it is determined whether or not there is a superordinate concept (step S44). If there is a superordinate concept word, it is judged whether or not the modified word is a superordinate concept word (step S).
45). If the modified word is not a superordinate concept word, the conceptual relationship between the superordinate concept word and the modified word is confirmed (step S46), and it is determined whether the relation can be established (step S47). If the relation is possible, a sentence structure indicating the conceptual relationship between the superordinate concept word and the modified word is added to the sentence structure which is the input information (step S48, path formation-described later), and the modified word is subjected to the step. It returns to S41 and repeats the above-mentioned processing (step S49).

If there is no superordinate concept word in step S44, and if it is not possible to relate them in step S47, it is assumed that an error is detected in the input information, and the modifier word, superordinate concept information of word A, etc. are used as error information. The data is stored in the data storage device 2, and the process ends (step S50). When there is no modifier in step S42, the existence of a superordinate concept word of word A is confirmed (step S51), and it is determined whether or not there is a superordinate concept word (step S52). If there is no superordinate concept word, the processing ends. If there is, there is determined whether or not there are a plurality of superordinate concept words (step S53). When there are a plurality of superordinate concept words, one proper superordinate concept word is selected as a modifier and added to the input information, and the superordinate conceptual word that is not selected is stored in the data storage device 2 as a modified word candidate. (Step S54), the process ends. If there is only one superordinate concept word, the superordinate concept word is added to the input information as a modifier (step S55), the added modifier is returned to step S41, and the process is repeated (step S5).
6).

Although the operation of each device has been clarified from the above description, it will be described more specifically based on the syntax check result of FIG. It is assumed that the "product code" is selected as the above-mentioned word A from the words in the input information. First, the process of refining the expression is performed by using the subordinate concept information of the "product code" (the refining based on the subordinate concept information is performed according to FIG. 6).

In the case of "product code", it can be seen from the input information that it does not have a modified word. Therefore, the existence of a subordinate term of "product code" is confirmed. It can be seen from FIG. 3 that “product code” has “value”, “column”, and “format” as subordinate terms. In this way, in the process of refinement, when the existence of multiple subordinate terms omitted in the input information is confirmed, one appropriate subordinate term is selected as a modified word, and the remaining modified words are selected. It is stored as a word candidate and the process ends. In the case of "product code", the term "equal" depending on the sentence structure requires "data" as an attribute, etc., so "value" is selected as the modifier.
The “column” and the “form” are stored as the modified word candidates, and the detailed processing based on the subordinate concept information ends.

Next, it is judged whether or not the refinement based on the subordinate term has been completed normally. In the case of the above "commodity code", it has been completed normally. Therefore, the attribute of the refined lowest level word "value" and the attribute required by the term for the corresponding case of "commodity code" in the sentence structure. Compare with. "Product code" is the nominative of the term "equal" (Fig. 2 (c)
), The attribute that "equal" requires in the nominative case is "data" (see FIG. 3). On the other hand, since the attribute of “value” is “data” (see FIG. 3), it is confirmed that the attributes match, and as a result, the control shifts to the refinement process using the higher-level concept information (to the higher-level concept information). The refinement based on FIG. 7 is performed).

In the detailing process using the superordinate concept information,
First, the existence of a modifier for the target word is confirmed.
In the case of "product code", there is no modifier in the input information. Therefore, the existence of a broader term of "product code" is confirmed. From FIG. 3, it is confirmed that "product code" has "order slip" and "basic information" as superordinate terms. In this way, in the process of refinement, if the existence of multiple superordinate terms omitted in the input information is confirmed, one proper superordinate term is selected as a modifier, and the rest are qualified candidate words. Then, the process ends.

In the case of the "product code", "order slip" is selected as the modifier, and "basic information" is selected as the modifier candidate.
Is stored in the data storage device 2, and the refinement based on the superordinate concept information ends.

In the refinement of the expression using the concept information, for example, when the modifier is not included in the superordinate concept,
It is assumed that there is an abbreviated word between the superordinate concept word and the modifier, and the concept relation is further traced from the superordinate concept word to the superordinate concept word, and a process to find a path to the modifier is performed. As a result, if it is possible to associate the superordinate concept word and the modifier, the path structure is added to the input information. this is,
Also in the detailed processing using the subordinate concept information, the relationship between the superordinate concept word and the modifier is basically the same except that the relationship between the subordinate concept word and the modified word is replaced.

When an error is detected in the input information in the refinement process using the conceptual information, the word to be erroneous, the conceptual information and the like are stored in the data storage device 2 as error information, and the process relating to the word is completed. Then refine the other words.

When the refinement of all the words in the input information is completed, the information obtained from the term dictionary storage device 4 and the result of the refinement processing are added to the input information (see FIG. 2D), the data storage device. After storing in 2, the control is transferred to the Japanese sentence generator 6.

The Japanese sentence generation device 6 reads the input information (see FIG. 2 (d)) changed by the meaning check device 5 from the data storage device 2 and expresses it in order to make it uniquely recognizable by a computer. Generates an expression-limited sentence that limits The rules for generating the expression-limited sentence are defined by the sentence structure information stored in the term dictionary storage device 4 for the idioms, while the production rules generally used in the Japanese sentence generation device 6 for the nouns and sentence connectors. Is specified.

Regarding the verbs, the syntax check device 3
Since the sentence structure is normalized from the original sentence of Japanese freely described in S., to the expression-limited sentence that can be recognized by the computer, in the Japanese sentence generation device 6, in the sentence structure information of the noun, The noun phrase corresponding to each case is rearranged according to the value of the "order" item (see FIG. 3). If the value of the "order" item is "nul", it means that the order does not matter.

With respect to the nouns and sentence connectors, a noun phrase and a sentence are generated from the detailed sentence structure in the meaning check device 5 based on the generation rule defined by the Japanese sentence generation device 6. An example of production rules is shown below.

Generation Rule Example 1 Sentence structure: "Word A is subordinate to word B in the relation of modification of word B (word A and word B are synonyms)" ↓ Japanese sentence: word B + "no" + word A Generation rule example 2 Sentence structure: “Sentence A is a condition and sentence B is a condition, and two sentences are subordinate to the sentence connector“ if ”. ↓ Japanese sentence: sentence A +“ if ”+ sentence B In the partial sentence structure added by the syntax check device 3, when a symbol is applied because the word cannot be specified, it is specified that the symbol is regarded as a word and a sentence is generated.

The case where the sentence structure shown in FIG. 2D is used as input information will be described below. According to the above rule, the noun phrase corresponding to the nominative "equal" is "the value of the product code of the order form". On the other hand, in the sentence structure information of the term “equal”, the value of the item “order” is “nul
(See FIG. 3), the word order between the noun phrases corresponding to the nominative case and the subject case is arbitrary, so the word order of the input information is followed, and the subordinate unit in the relation of the condition of the sentence connector “Nara” The sentence "the product code value of the order form is equal to" 3 "" is generated.

The same applies to the term "exists". The word order between noun phrases corresponding to the nominative case and the object case is arbitrary, and is represented by the symbol "XXX" between noun phrases corresponding to the place case. , A unit sentence ““ fax ”exists in XXX” that is subordinate to the sentence connector “if” in a state relationship is generated.

Further, by applying the above rule to the sentence connector "if", the expression limitation sentence "if the value of the product code of the order form is equal to" 3 "," fax "exists in XXX". Will be generated.

The generated expression-limited sentence is stored in the data storage device 2.
Then, the processing is transferred to the Japanese sentence correction device 7.

In the Japanese sentence correction device 7, the original sentence and the expression-limited sentence, sentence structure, word candidate information, error information, etc. generated in the Japanese sentence generation device 6 are read as input information, and the original sentence, expression Present the limited text. In the presentation of the expression-limited sentence, the place where the word candidate and the place where the word is represented by a symbol are clearly shown to the user, and the user is prompted to identify an appropriate word. The task of identifying words is
If there are word candidates, present the candidate list to the user,
In the case of replacing a symbol with a word, the corresponding word is collected from the term dictionary storage device 4 based on the information about the case attribute in the sentence structure information and presented to the user as a word candidate, thereby Being supported in a conversational manner. further,
With respect to the error detected in each of the syntax check device 3, the semantic check device 5, the Japanese sentence generation device 6 and the like, by presenting the error information, the error location, the error content, etc. To help.

Below, the expression-limited sentence "if the product code value of the order form is equal to" 3 "," fax "exists in XXX""generated by the Japanese sentence generation device 6 is used as input information. The case will be described.

In the above input information, if the place where the word candidate exists and the place where the symbol is substituted are underlined, the user is informed that "the product code value of the order form is" 3 "".
"Fax" exists in XXX if
Will be presented.

It is assumed that the user selects the correction of the word "order slip". When the other word candidates related to the “order slip” are “basic information”, the Japanese sentence correction device 7 uses the word “order slip”.
The user is informed that there is a candidate "basic information", and the user is prompted to select which of "order slip" and "basic information" is appropriate. When the user selects "basic information", the "order slip" in the expression-limited sentence is replaced with "basic information".

If the user selects the correction of the symbol "XXX", the Japanese sentence correction device 7 first displays "XX".
A word candidate that replaces "X" is searched. Sentence structure "XXX"
Since the attribute required for the location case corresponding to is “data”, the term dictionary storage device 4 is searched for with the attribute “data”,
Collect word candidates such as “value” (see FIG. 3). The Japanese sentence correction device 7 presents the collected word candidates to the user and prompts the user to select a word. If the user selects "value",
"Value" replaces "XXX" in the expression-limited sentence. If there is no suitable word among the word candidates presented by the Japanese sentence correction device 7, the user can freely input the word.

As a result of the above correction work, the expression limitation sentence "If the value of the product code of the order form is equal to" 3 "," Fax "exists in XXX ""is" the product code of the basic information. " If the value of is equal to "3", "fax" is present in the value. "

When the user corrects the expression limitation sentence, the correction result is stored in the data storage device 2 and the process is returned to the input device 1. If not corrected, the processing shifts to the output device 8, and the output device 8 outputs the expression-limited sentence as a Japanese sentence without ambiguity. In the case of this example, since it has been corrected, the correction result is stored in the data storage device 2, and the disambiguation processing of the Japanese sentence as a new original sentence is repeated.

[0059]

As described above, according to the present invention, a Japanese sentence including ambiguity can be associated with a computer expression by the first and second syntax checking means and the first and second semantic checking means. The work of refining to various levels will be performed semi-automatically in a conversational manner, and there is an effect that it is possible to easily understand the meaning of natural language using a computer.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a Japanese sentence generation device capable of disambiguation according to the present invention.

2 (a), (b), (c) and (d) are sentence examples stored in a syntax tree format from the input device of the embodiment of FIG. 1 in a data storage device, a syntax check device, and a semantic check. It is a figure which shows the process result in an apparatus.

FIG. 3 is a diagram showing contents stored in a term dictionary storage device of the embodiment of FIG.

FIG. 4 is a flowchart showing a checking operation of the syntax checking device in the embodiment of FIG.

5 is a flowchart showing a checking operation of the meaning checking device in the embodiment of FIG.

6 is a flowchart showing a detailed operation related to a subordinate concept in the checking operation of the meaning checking device in FIG. 1. FIG.

7 is a flowchart showing a detailed operation related to a superordinate concept in the checking operation of the meaning checking device in FIG. 1. FIG.

[Explanation of symbols]

 1 Input Device 2 Data Storage Device 3 Syntax Check Device 4 Term Dictionary Storage Device 5 Semantic Check Device 6 Japanese Sentence Generation Device 7 Japanese Sentence Correction Device 8 Output Device S1 to S56 Steps

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location 7315-5L G06F 15/20 550 E 7315-5L 554 M

Claims (3)

[Claims] 1. A Japanese sentence generator for removing ambiguity from an input Japanese original sentence and generating a desired Japanese sentence in a conversational manner, wherein each word is stored as an entry word, and each entry word is The computer attribute of each headword and the conceptual information about other headwords corresponding to the superordinate concept and the subordinate concept for each headword are given, and when each headword is a noun, , The term dictionary storage unit to which sentence structure information, which is a rule that defines the sentence structure of the unit sentence in which the above-mentioned verb is used, is provided, and the above-mentioned verb is detected from the input Japanese original sentence, It is determined whether or not the included unit sentence is a regular unit sentence that conforms to the rules of sentence structure information given to the above-mentioned defensive word. To be a unit sentence First sentence checking means for urging positiveness, and a sentence connector detected from the input Japanese original sentence, two unit sentences connected by the sentence connector are given in advance for each sentence connector. Second syntax check means for judging whether or not the regular structure is based on the sentence combination rule, and displaying the fact on the display to prompt the correction so that the regular structure is obtained when the regular structure is not formed. And for a word in which the other words are not modified in the Japanese original sentence processed by the first and second syntax check means, the word dictionary is used as a headword to represent a subordinate concept of the headword from the term dictionary storage unit. When a word is read and one of the read words is selected, the selected word is held and a word representing a subordinate concept is read from the term dictionary storage unit using the selected word as a headword. A first meaning checking unit that repeats until a word not having is selected and a word that is not modified from other words in the Japanese original sentence processed by the first and second syntax checking units finds the word. A word representing a superordinate concept of the headword is read from the term dictionary storage unit as a word, and when one of the read words is selected, the selected word is held and the selected word is further superordinate as a headword. Second meaning checking means for repeating the reading of the word representing from the term dictionary storage unit until a word having no superordinate concept is selected; and the first and the first with respect to the sentence structure of the input Japanese original sentence. Two
Are added to the words that are normalized by the syntax check means and have no modifiers or modifiers of the input Japanese original sentence and that are extracted and retained by the first and second meaning check means. A Japanese sentence generation device having a Japanese sentence composing means for composing a form of a Japanese sentence that can be recognized by a computer from the formed sentence. 2. The Japanese sentence generation device according to claim 1, wherein the Japanese original sentence processed by the first syntax checking means is input to the second syntax checking means and further processed. 3. The Japanese sentence composing means displays the composed Japanese sentence, and presents another candidate of the word, which is information that is generated in the process of generating the Japanese sentence and cannot be recognized from the displayed Japanese sentence, and the Japanese sentence. 3. The Japanese sentence generation device according to claim 1 or 2, which performs normalization, detailing, and presentation of information about an error that has occurred in the construction of a Japanese sentence.