JPH0388066A - Automatic detecting/correcting device for error in japanese document - Google Patents

Abstract

PURPOSE:To detect a homonym error in an independent word by extracting an inspection case indicated in an inspection case determining rule out of a clause modifying a word to be inspected and inspecting meaning connecting relation between the meaning category of the word included in the inspection case and the word to be inspected. CONSTITUTION:When a homonym to be inspected exists in an inputted Japanese sentence, the meaning connecting relation between the appearance of the word to be inspected and the meaning category of the word included in the case indicated by the inspection case determining rule 12 out of the clause modifying the word to be inspected is obtained by retrieving meaning connection dictionaries 13, 14 by means of the appearance of the word to be inspected. When the meaning connecting relation does not exist, the existence of a homonym error is decided. A word having the same reading and the same part of speech for the homonym error is extracted as a correction candidate, the correction candidate is substituted with the word inspected as the homonym error, the substituted candidate is inspected by the dictionaries 13, 14 and the correction candidate to be connected is decided as the correct candidate of the homonym error.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a Japanese document processing device, and in particular, to a device for automatically detecting errors in the use of homophones in an input Japanese document and detecting the errors in the use of homophones. This invention relates to an automatic Japanese sentence error detection/correction device that presents correction candidates.

[Conventional technology]

In general, Japanese word processors input a kana matrix or a Roman character string and perform kana-kanji conversion on a word or phrase basis, so homonym selection errors may occur due to word conversion errors or conversion range errors. This type of homophone error requires a thorough understanding of the semantic usage of homophones, so it is not only due to a kana-kanji conversion error, but also due to the creator's assumptions or misconceptions during the process of creating the original document. Mistakes also occur frequently.

Conventionally, automatic detection of homophone errors that occur due to input errors in input Japanese sentences or word selection errors in kana-kanji conversion has been done by marking error-prone homophones in a dictionary in advance, and checking the dictionary in the input document. All locations that matched the character string of the homophone were extracted as homophone errors.

[Problem to be solved by the invention]

The above-mentioned conventional technology has the following problems because all parts of the input Japanese sentence that match the character strings of homophones marked in advance in the dictionary are extracted as homophone errors.

■ All words that are used correctly are also extracted as homonym errors.

■ It is difficult to extract correction candidates.

■ Cannot detect homophone errors for single words. A homophone error in a single word is one that is not a homophone error in a compound word. For example, "returning exhaust gas"
``Go home'' is a single word homophone error.

An example of an automatic detection/correction method and device for homophone errors contained in compound words is the patent application published in 1983 by the present applicant.
-149448 and Japanese Patent Application No. 1-115576, but these target only homophone errors contained in compound words.

The purpose of the present invention is to solve the above-mentioned problems and provide an automatic Japanese sentence error verification/correction device that accurately detects homophone errors such as single words in input Japanese sentences and presents correction candidates. It is about providing.

[Means to solve the problem]

The automatic Japanese sentence error detection/correction device of the present invention includes a semantic concatenation dictionary that describes the concatenation information between the font of a homophone and the meaning category of a word that has a case element shown in the test case determination rule; For each case, the semantic conjunction information in the semantic conjunction dictionary describes the conjunction information with which case element. Furthermore, it performs morphological analysis of the input Japanese sentence and converts the input Japanese sentence into words. A first means of dividing the input Japanese sentence into units, a second means of organizing the input Japanese sentence divided into words into phrases, and a third means of structuring the input sentence by analyzing the dependency relationships between phrases. a fourth means for determining whether or not an input Japanese sentence contains a homophone and, if the input Japanese sentence contains a homonym, and for making the word a test subject;
For each test target extracted by the method, by searching for the test case determination rule using the font of the test target, the case element to be tested for semantic connection with the test target ( a fifth means for determining the test path (referred to as the test path);
The semantic connection between the word semantic category of the test path determined by the fifth means and the test target is tested by referring to the semantic connection dictionary, and if there is no semantic connection between the two, only.

A sixth means for testing the test subject as having a homophone error; A certain word) is extracted as a correction candidate for the homophone error, and the correction candidate and the word with the homophone error corresponding to the correction candidate are replaced, and the test case for the correction candidate and the correction candidate is calculated. Searches the semantic association dictionary for information on whether or not the word can be linked with the semantic category, and if semantic linkage is possible, outputs the correction candidate as the correct answer candidate, and semantic linkage with all correction candidates is not possible. and a seventh means for outputting all the extracted correction candidates as correct candidates in the case of .

[For production]

When a homophone to be tested (hereinafter referred to as a word to be tested) exists in the input Japanese sentence, the homophone error testing process is started. Then, the semantic category (semantic category is a term used to classify nouns according to their meaning) is based on the font of the word to be tested and the case shown in the test case determination rule among the clauses that modify the word to be tested. (based on a semantic category system prepared in advance) by searching a semantic linkage dictionary that describes such linkability information using the font of the word to be tested, If there is no semantic conjunction, it is determined that it is a homophone error.

For homophone errors, words with the same pronunciation and part of speech are extracted as correction candidates (by searching the meaning conjunction dictionary with the pronunciation of the homophone error), and these correction candidates are extracted as correction candidates for homophone errors. A test is performed in the same manner as above using the semantic concatenation dictionary by replacing the words that have been verified as existing, and correction candidates that can be concatenated are determined as correct candidates for the homophone error.

Thereby, it is possible to test whether a homophone such as a single word appearing in a Japanese sentence is correct or not, and if it is judged to be incorrect, a correction candidate can be extracted.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a basic configuration diagram of an embodiment of the present invention.

16 is the automatic Japanese sentence error detection/correction device, which consists of a CPU, memory, etc. in terms of hardware, but in terms of functionality,
Morphological analysis part 2, clause recognition processing part 2, dependency analysis part 3
, a homophone extraction section 4, a test case extraction section 5, a homophone test processing section 6, a homophone correction extraction section 7, and a meaning conjunction dictionary search section 8.

The morphological analysis unit 1 analyzes the Japanese sentence input to the device 16 by
Word division is performed using the Japanese word dictionary 9 and grammar dictionary 10, and a part of speech, meaning category, etc. is assigned to each word.
The phrase recognition processing unit 2 performs a process of grouping the word-divided input sentence into phrase units based on the parts of speech of the words. The dependency analysis unit 3 determines the dependency relationships between clauses using the dependency rules 11, and structures the input sentence.

The homophone extraction unit 4 determines whether or not a homophone to be tested exists in the input sentence, and if so, extracts the homophone as a word to be tested. Qualification extractor 5
uses the test case determination rule 12 to test the semantic connection of each homonym extracted by the homonym extraction unit 4 with the semantic category of the word which case element has. Decide whether

The homophone test processing unit 6 determines whether each homophone extracted by the homophone extraction unit 4 can be linked with the meaning category of the word that has the case element determined by the test case extraction unit 5. is tested using the semantic concatenation dictionaries 13 and 14 to determine whether the homophone is incorrect.

The homophone correction candidate extraction unit 7 has a semantic linkage dictionary 13.1.
Correction candidates are extracted by searching for 4 using the pronunciation of the word that has been determined to be a homophone error. At this time, only those having the same part of speech as the corresponding homophone error are extracted as correction candidates. Next, the proficiency determination rule 12 is searched using the font of the correction candidate to determine the proficiency for the correction candidate. Furthermore, the homophone correction candidate extracting unit 7 replaces the word with the homophone error and its correction candidate, and determines whether or not the correction candidate can be connected to the semantic category of the word that has the test case in the semantic connection dictionary 13. 14, all correction candidates that can be connected are output as correct candidates to an output file 15, and if all correction candidates cannot be connected, all correction candidates are output to an output file.

The semantic conjunctive dictionary search section 8 searches the homophone conjunctive dictionaries 13 and 14 in response to instructions from the homophone test processing section 6 and the homophone correction candidate extraction section 7.

The Japanese word dictionary 9 describes morphological information such as part-of-speech information and semantic categories of Japanese words. The grammar dictionary 10 describes grammatical information such as connectivity between parts of speech of Japanese words. Dependency rule 11 is based on the dependency relationship between clauses (e.g., a clause with the particle "ga" modifies the following pragmatic clause) and the hyuristic rule (e.g., dependency relationships do not intersect, the same case does not intersect). (Multiple clauses with particles do not modify a single phrase). Qualification determination rule 12 has the font of the homophone as a key,
It describes the case elements that have a strong semantic connection with the homophone (for example, wo case, ni case). The forward meaning conjunctive dictionary 13 has two keys, the font of the homophone and its pronunciation, and the meaning category of the word that the homophone and the pronunciation path (the case described in the proficiency determination rule 12) have. It describes information on whether connection is possible between the two.

The backward meaning concatenation dictionary 14 describes information on whether concatenation is possible with a meaning category of a word that can be concatenated after a homophone when the homophone modifies a noun as an adnominal.

The processing flow of the automatic Japanese sentence error detection/correction device is shown in FIG. 2, and the operation of FIG. 1 will be explained below in accordance with this flow.

Step Sl: For the Japanese sentence that is input to the automatic Japanese sentence error detection/correction device main body 16, the morphological analysis unit 1 uses the Japanese word dictionary 9.
, perform morphological analysis such as word candidate extraction and part-of-speech connection test using the grammar dictionary 10, divide the Japanese sentence into word units,
After adding part-of-speech information, meaning categories, etc. to each word, the analysis results are sent to the clause recognition processing section 2.

STEP S2: The clause recognition processing unit 2 groups the input sentence into clause units based on the morphological analysis results, and sends the results to the dependency analysis unit 3.

Step S3: The dependency analysis unit 3 uses the dependency rule 11 to
Determine the dependency relationships between clauses and structure the input sentence. Then, the morphological analysis result and the dependency analysis result are sent to the homophone extraction unit 4.

Step S4: In the homophone extraction unit 4, when a homophone to be tested exists as a single word in the input sentence, it is extracted from the morphological analysis result as a word to be tested. The extraction method is to set a flag in the Japanese word dictionary 9 to indicate that the font of the homophone to be tested is to be tested, or to use the meaning conjunction dictionary 13.14 to check the font of each word in the input sentence. There are several possible ways to search, but this is not particularly limited here.

Step S5 If the word to be tested has been extracted in step S4, the process proceeds to step S6. Otherwise, this process ends.

If the word to be tested is a predicate, the processing is divided depending on whether its inflected form is an adjunctive form or not.

Step S7: When the inflected form of the word to be tested is an adnominal form, the backward semantic conjunctive dictionary 14 is selected as the semantic concatenative dictionary used for the test. In FIG. 2, the above selection is made by assigning the backward semantic linking dictionary to the variable "test dictionary".

Step S8: When the conjugation form of the word to be tested is other than the adnominal form, and when the word to be tested is other than a term, the forward semantic linkage dictionary 13 is selected as the semantic linkage dictionary used for the test. In addition, the second
In the figure, the above selection is made by assigning the forward semantic concatenation dictionary to the variable "r test target dictionary".

Step S9: The test case extracting unit 5 searches the test case determination rule 12 based on the font of the word to be tested, and determines which case element is associated with the meaning category of the word.

This case element is called the test path.

Step SIO: In the homophone test processing unit 6, first, the dependency analysis unit 3
Based on the dependency relationships obtained in , the meaning category of the word that the r test path has among the phrases that modify the word to be tested is extracted.

Step Sll: Next, the semantic linking dictionary 13, 14 is searched for the font of the word to be tested via the semantic linking dictionary search unit 8, and a record regarding the word to be tested is received.

Step S12: Next, step SIO
Obtain the semantic connection information regarding the semantic category number (corresponding to the semantic category and 1:l) that corresponds to the semantic category J of the word in the test path obtained in . If the connection between the meaning category number and the meaning category number corresponding to "semantic category" is possible, step S1
If not, the process proceeds to step S14.

Step S13: This process ends with the word to be tested being OK to be connected.

Step S14: The test target word is determined to be a homophone error, and all information is sent to the homophone correction candidate extraction unit 7.

Step S15: The homophone correction candidate extracting unit 7 searches the semantic concatenation dictionary 13. with the reading of the homophone error via the semantic concatenation dictionary search unit 8.
14, and all words with the same pronunciation and part of speech are extracted as correction candidates for the word with the homophone error. Here, let the number be m.

Step 816: Processing proceeds with the i-th correction candidate as the target.

First, let i=1.

Step S17: Using the face of the correction candidate as a key, test path determination rule 12
Search for and decide the test path.

Step 818: Extract case elements having the same particle as the test path obtained in step S17 from among the clauses modifying the word to be tested, and obtain the meaning category of the word possessed by the case element.

In step S19, the meaning category number (semantic category and 1 :1)) is obtained. If the connection between the two is possible, the process advances to step S20; otherwise, the process advances to step S21.

Step S20: Output the correction candidate as a correct answer candidate to the output file 15, and proceed to step S21.

Step S21: In order to process the next correction candidate, i=i +
Set to 1.

Soup S22: Processing is distributed depending on whether processing is completed for all m correction candidates. If the processing has been completed for all correction candidates, the process advances to step S21; otherwise, the process returns to step SF.

Step S23: If even one correct answer candidate is output to the output file 15, this process ends; otherwise, the process proceeds to step S24.

Step S24: Output file 1 with all m correction candidates as correct answer candidates.
Output to 5.

Note that if a plurality of homonyms are extracted in step S4, step 85 is performed for each homonym.
Perform the following processing.

FIG. 3 shows an example of the field configuration of the forward semantic linking dictionary engineer 3, and the same applies to the backward semantic linking dictionary engineer 4. In FIG. 3, numeral 17 indicates the face of the homophone, which is the key when searching the semantic association dictionary, and numeral 18 indicates the face of the homophone, which is the key when searching the semantic association dictionary in the homophone correction candidate extraction unit 7. Reading, engineering 9 is the part of speech of the homophone,
20 indicates whether or not the 17 characters can be connected to the meaning category of the word in the test case, based on the number assigned to each meaning category in the semantic category system (consisting of N meaning categories in Figure 3) prepared in advance. This is a linkability information section corresponding to a category number that describes information. 2nd field is a field indicating whether or not each of the 20 meaning category numbers can be connected, and #
n (1≦n≦N) is "O" if concatenation is possible, "X" if conjunctive is not possible, and "X" if concatenation is unknown (if concatenation is possible for the character face and there is a homophone relationship) If the word can also be connected to the written word), "Δ" is shown. 22 indicates the 3rd record on the forward meaning conjunction dictionary.

FIG. 4 shows an example of the configuration of the qualification determination rule 12.

In FIG. 4, numeral 23 is the face of a homophone that is a key when searching for a test case determination rule, and 24 is the face of a particle indicating the test case for the homophone 23. 25 indicates the l record of the qualification determination rule 12.

Next, a specific example will be explained. An example sentence is, ``We will reduce the amount of exhaust gas to half the current amount.'' (Correct answer: regulate).

5 and 6 show examples of the contents of the forward meaning conjunction dictionary 13 and the qualification determination rule 12, respectively. In the forward meaning conjunction dictionary in Figure 5, rOJ is connectable, “×” is not connectable, and “Δ
” means unknown connection.

Figure 7 shows an example sentence, ``We will reduce the amount of exhaust gas to half of the current amount.

This figure shows the processing process for ``.

First, the morphological analysis unit 1 divides words and assigns readings, parts of speech, meaning categories, etc. to each word. Next, the phrase recognition processing unit 2 groups the phrases into phrases (in the example sentence, reduce the exhaust gas to /the current /half of the amount /go home/), and the dependency analysis unit 3 adds dependency rule 11. The dependency relationship between clauses is recognized using this method, and the input sentence is structured (Step S1-Step S3).

, FIG. 7(a) shows the result of morphological analysis of an example sentence, and FIG. 7(b) shows the result of dependency analysis.

Next, the homophone extraction unit 4 extracts homophones to be the words to be tested. In the example sentence.

“Go home” becomes the word to be tested (steps S4, S
5). In this embodiment, a flag indicating that it is a test target is set in the record of the Japanese word dictionary 9 at the time of morphological analysis, but as another method, the homonym extraction unit 4 can key the font of each word. It is also conceivable to extract the words to be tested by searching the semantic linking dictionary 13, 14 as follows. Further, the homophone extraction unit 4 selects either the forward semantic conjunctive dictionary 13 or the backward semantic conjunctive dictionary 14 as the dictionary to be tested, using the part of speech and conjugated form of the word to be tested. In the case of the example sentence, "Go home" is a predicate, but it is a final form, so 3 of the forward meaning conjunctive dictionary is selected as the dictionary to be tested (step 86 to step S8).
).

The qualification extracting unit 5 determines the qualification using the qualification determination rule 12. In the example sentence, the test qualification determination rule 12 is searched for the stem word "homecoming" of the test target word "homecoming".

Since the proficiency determination rule 2 is as shown in FIG. 6, the proficiency case is determined to be the "ni" case (step S9).

Next, the homophone test processing unit 6 acquires the meaning category of the word with the test case (step 510).
). The structure of the example sentence is as shown in Figure 7 (b), so the clause modifying the test word ``homecoming'' is ``
There are two types: "exhaust gas" and "quantity." Among these, the test case "ni" is "r quantity ni", so the semantic category "quantity" (semantic category number=154) of the word "quantity" in this clause is acquired. Next, in the homophone test processing unit 6, the stem word ``Kikaku'' of the test target word ``Khomeishuru'' is used in the test target dictionary (in the example sentence, the forward meaning conjunctive dictionary 1
3) is searched (step 511). Since the forward meaning conjunction dictionary 13 is as shown in FIG. 5, the possibility of concatenation with the word having the meaning category number 154 for "homecoming" is rXJ. Therefore, "Go home" is determined to be a homophone error (Step S12. Step 514).

FIG. 7(c) shows the processing contents of steps 84 to S14.

The homophone correction candidate extraction unit 7 performs a forward semantic conjunction of the homophone error ``Kisei'', which is the stem reading of ``Kisei'', which is the pronunciation of the stem of ``Kisei'', which has the same part of speech as ``Kisei''. It is searched from the dictionary 13 and extracted as a correction candidate. Since the forward meaning conjunction dictionary 13 is as shown in Figure 5, there are two correction candidates: "regulation" and "regulation" (m = 2).
) is obtained (step 515). FIG. 7(d) shows the contents of this process.

Next, the homophone correction candidate extraction unit 7 first processes the correction candidate "regulation" (step 516).
. The qualification determination rule 12 is searched using the word "regulation" of the correction candidate, and the qualification is determined. Since the proficiency determination rule 12 is as shown in Figure 6, the proficiency rating for "Regulation" is "
The example sentence is structured as shown in Figure 7 (b), so the phrase that modifies the test word ``go home'' is ``exhaust gas'' (step 517). There are two. Among these, the valid case for the correction candidate "regulation" is "r exhaust gas", so the meaning category of the word "exhaust gas" in this clause is "
Gas" (semantic category number = 43) is obtained (step 818). From FIG. "Restrictions" is output to the output file 15 as a correct answer candidate for "Homecoming" (step 520).

FIG. 7(e) shows the processing details for this correction candidate "regulation".

Next, processing is performed on the correction candidate "Regulation" (step 521), and in the same way as "r-Regulation", the test case for "Regulation" is also determined to be "O" case. “Regulations” and meaning category number 4
As for whether or not it can be connected with 3, it is "x" which means that it cannot be connected, so "regulation" is not a correct answer candidate.

FIG. 7(f) shows the processing details for this correction candidate "Regulation".

When all correction candidates have been processed (Step 5)
22) Since "regulation" is output as a correct answer candidate, step 824 is skipped and the present process is ended (step 523).

As described above, ``Regulate J'' is obtained in the output file 15 as a correct answer candidate for the homophone error ``Return'' included in ``Reduce the amount of r exhaust gas to half the current amount''.

〔Effect of the invention〕

As is clear from the above description, the automatic Japanese sentence error detection/correction device of the present invention provides the following effects.

- Since it is determined whether or not there is a homophone error by testing the semantic linkage, it is possible to accurately detect only homophone errors.

■ Since only those that satisfy the semantic conjunction are extracted as correct answer candidates for homophone errors, only likely candidates can be presented.

■ The test case shown in the test case determination rule is extracted from the clauses that modify the test target word, and the semantic linkage between the word semantic category of this test case and the test target word is tested. Homophone errors in single words can be detected.

[Brief explanation of drawings]

Fig. 1 is a basic configuration diagram of an embodiment of the present invention, and Fig. 2 is a basic configuration diagram of an embodiment of the present invention.
Figure 3 is a diagram showing an example of the field configuration of a semantic linkage dictionary, Figure 4 is a diagram showing an example of the field configuration of a qualification determination rule, and Figure 5 is a diagram showing the forward direction. Figure 6 is a diagram showing an example of the content of a meaning conjunction dictionary, Figure 6 is a diagram showing an example of the content of a qualification determination rule, and Figure 7 is a diagram showing the processing process for the example sentence ``Reduce the amount of exhaust gas to half the current amount.'' be. l... Morphological analysis unit, 2... Clause recognition processing unit. 3... Dependency analysis section, 4... Homophone extraction section. 5... Test case determination section, 6... Homonym test processing section, 7... Homophone correction candidate extraction section. 8... Semantic conjunction dictionary search unit, 9... Japanese word dictionary, 10... Grammar dictionary, 11... Dependency rule, 12... Certification determination rule, 13... Forward semantic conjunction Dictionary, 14... Backwards meaning conjunction dictionary, 15.
・・Output file, 16・Japanese sentence error automatic detection/correction device main body, 17・font of homophones, 18
...Pronunciation of the homophone, 19...Part of speech of the homophone, 20...Connection availability information for the corresponding semantic category numbers, 21...Field showing connection availability information for each semantic category number, 22 ...Lou code of the semantic conjunction dictionary, 23...Character of homophones, 24...Character of particle indicating certifying case, 25...Lou code of test case determination rule. Condolence ■ Figure Figure 3] Beg for the wrong field ``No. Inner axis row of the dictionary, Figure 6, Seeds Hikibayashi 4 Inutaku Ruri, Naikanakaj Figure 7 (C) rltei4to?”A'iha...1:+:Kuyo Figure 7 ( 4) (e)

Claims (1)

[Claims] (1) In a device that automatically detects and corrects errors in homophones contained in Japanese sentences created by a Japanese input device such as a Japanese word processor, the font and proficiency determination of homophones is performed. A semantic linkage dictionary that describes whether the case element shown in the rule can be linked with the semantic category of the word, and a semantic linkage dictionary that describes the linkage information with which case element for each homophone. a first means for performing morphological analysis of the input Japanese sentence and dividing the input Japanese sentence into word units; and a first means for dividing the input Japanese sentence into word units by morphologically analyzing the input Japanese sentence; A second means of structuring the input sentence by analyzing the dependency relationships between clauses, and a third means of structuring the input sentence by analyzing the dependency relationships between clauses, and determining whether or not the input Japanese sentence contains homonyms, If the word is a test target, a fourth means is used, and for each test target extracted by the fourth means, the above-mentioned qualification determination rule is searched for the character of the test target. A fifth means for determining a case element (referred to as a test case) to be tested for semantic connection with the test target, and a semantic category of the word of the test case determined by the fifth means and the test target. A sixth means for testing the semantic linkage of by referring to the semantic linkage dictionary, and testing the test target as a homophone error only when there is no semantic linkage relationship between the two. and extracting words (words in a homonymous relationship) that have the same pronunciation and part of speech as the word tested as a homophone error in the sixth means as correction candidates for the homophone error; Replace the correction candidate with the homophone error word corresponding to the correction candidate, and search the semantic connection dictionary for information on whether or not the correction candidate can be connected with the semantic category of the word that has the qualification for the correction candidate. , if a semantic link is possible, the correction candidate is output as a correct answer candidate, and if a semantic link with all correction candidates is not possible, all extracted correction candidates are output as correct answer candidates. 7 means, and an automatic Japanese sentence error detection system characterized by:
correction device.