JPH02105968A - Automatic test and correction system for japanese sentence error - Google Patents

Abstract

PURPOSE:To speed up processing by extracting only homonyms which are judged to be wrong as a result of decision making, extracting correction candidates for homonym errors automatically by using a homonym table, and outputting only a certain candidate as a correct answer candidate by using information in a homonym concatenation decision table. CONSTITUTION:A table retrieval part 6 tests an error of a homonym included in a noun continuous compound word automatically by using concatenation possible/impossible state information (held in the form of a homonym concatena tion decision table 10) on nouns constituting the noun concatenation compound word to accurately extract only a homonym which is used erroneously. Correc tion candidates are also tested by using the homonym concatenation decision table 10 to output only correction candidates which can be concatenated, so only certain correction candidates can be displayed. Only the candidates which are judged to be wrong as a result of the test and the correction candidates are displayed, so that man-hours are smaller than those by a manual test. Conse quently, only the really homonym error is detected and the correction candidates are displayed.

Description

Detailed Description of the Invention (1) Technical field to which the invention pertains The present invention provides automatic Japanese sentence error verification and correction that automatically detects homonym usage errors in Japanese documents and presents correct answer candidates. It is related to the method.

(2) Conventional technology In the conventional method, homophones that are prone to errors are marked in a dictionary, and all locations in the input document that match the character string of the homophone in the dictionary are candidates for homophone errors. It was extracted as but.

This method had the following problems.

■ All words that are used correctly are also extracted as error candidates.

■ It takes almost the same number of man-hours as manual verification.

■ It is difficult to output correction candidates.

Furthermore, in Japanese Patent Application No. 149448/1983 (Japanese Sentence Error Automatic Verification/Correction Device), errors in the use of homophones can be detected using the meaning categories of words, and candidates can be extracted. There were the following problems.

■ In creating the semantic category linkage determination table that is essential for testing, it is necessary to manually assign a semantic category to each piece of original data, which requires a large amount of man-hours.

■ Because of this, the range of homophones that can be targeted is limited.

■ It is not easy to expand the scope of testing.

(3) Purpose of the invention The purpose of the present invention is to solve the above problems. The object of the present invention is to provide an automatic Japanese sentence error verification/correction method that detects true homophone errors and presents correction candidates.

(4) Structure of the invention (4-1) Features of the invention and differences from the conventional technology The present invention.

Regarding homophone errors in noun continuous compounds caused by errors in Japanese documents input from input devices.

One-noun continuous compound words are divided into word strings by morphological analysis using a Japanese word dictionary and a grammar dictionary.

A first means for obtaining morphological information such as the pronunciation and part of speech of each word.

A homophone table containing a group of homophones, which are words with the same pronunciation, using the pronunciation as a key.

For each homophone recorded in the homophone table, a homophone concatenation determination table holds the possibility of concatenation between the homophone letter and preceding and succeeding character strings using the homophone letter as a key.

The second means searches the homophone table using each word constituting the noun continuous compound divided into word strings as a key, and starts the test when there is a homophone to be tested. With the means of.

Using the font of the homophone to be tested as the key.

The above-mentioned homophone concatenation determination table is searched, and as a result, there is no possibility of concatenation between the homophone and the preceding and following character strings,
Or, when the possibility of this conjunction is not registered in the homophone conjunction determination table.

and a third means for testing the homophone as an error.

A fourth means for obtaining correction candidates for the homophone by searching the homophone table using the pronunciation of the homophone that was determined to be incorrect in the third means as a key.

The above-mentioned homophone concatenation determination table is searched using the font of the correction candidate acquired in the fourth means as a key, and as a result, only correction candidates that have a possibility of concatenation with the character strings before and after 1 are selected for the homophone error. and a fifth means for outputting as a correct answer candidate.

What is conventional technology?

■ Extract only the homonyms determined to be incorrect as a result of the determination.

■ Automatically extract correction candidates for homophone errors using the homophone table, and output only more likely candidates as correct candidates using the information in the homophone conjunction determination table.

Each point is different.

(4-2) Embodiment FIG. 1 is a basic configuration diagram when configuring the present invention by hardware. A morphological analysis unit that divides words using the grammar dictionary 8 and adds morphological information such as part of speech and pronunciation to each word.

2 determines whether or not a homophone exists in the noun continuous compound word by searching the homophone table 9, and if it exists, extracts the homophone. Part 3 uses information obtained by searching the homophone concatenation determination table IO with the homophone character plane to determine whether the homophone can be concatenated with the character string before or after it. Accordingly, the homophone testing unit 4, which tests the usage error of the homophone,
a homophone candidate extraction unit that searches the homophone table 9 using the pronunciation of the homophone as a key and extracts correction candidates;
5, for the correction candidates extracted by the homophone candidate extraction unit 4, using the homophone connection determination table 10, it is tested whether or not the correction candidates can be connected with a word located before or after the correction candidate, and the connection is determined. Homophone candidate testing unit 6 outputs all acceptable correction candidates to output file 11, and 6 is homophone extraction unit 2. According to the request of the homophone candidate extracting section 4, the homophone table 9 is generated;
Table search unit 7 searches for each 0 and sends necessary information to the request source.

8 is a Japanese word dictionary that describes morphological information such as part-of-speech information and pronunciation of Japanese words; 8 is a grammar dictionary that describes grammatical information such as connections and dependencies between parts of speech of Japanese double-type words; 9 is a homophone table containing a group of homophones, which are words with the same pronunciation, using the pronunciation as a key,
Step 10 calculates the frequency of occurrence of a character string including each homophone face recorded in the homophone table 9 from the existing document, and concatenates the homonym face and the preceding and succeeding character strings based on the frequency of appearance. A homophone conjunctive determination table that holds possibilities using the homophone font as a key.

11 is an output file, and 12 is an automatic Japanese sentence error verification/correction device consisting of an arithmetic unit and memory.

FIG. 2 is a field configuration diagram of the homophone table 9, where 13 is the reading of the homophone which is the key when searching the homophone table 9, and 14 is the plurality of readings of the record. This is the homophone grammatical face of .

FIG. 3 is a field configuration diagram of the homophone conjunctive determination table 10, in which 15 is a homophone character face, which is a key for searching the homophone concatenation determination table 10, and 16 is a field configuration diagram of the homophone concatenation determination table 10.
is character string information regarding character strings that can come before or after a homophone (details are shown in Figure 4). Figure 4 is a configuration diagram of 9 character string information 16, and 17 is a 1 character string face. ,1
8 is.

The forward concatenation possibility 19 describes whether the character string can come before or not, and the backward concatenation possibility 19 describes whether the character string can come after.

FIG. 5 is a schematic flowchart of the operation of the 9-Japanese sentence error automatic verification/correction device 12.

Next, the operation will be explained according to the schematic flow shown in FIG.

(Step Sl) The morphological analysis unit 1 processes a noun continuous compound word that is input to the Japanese sentence error automatic test/correction device 12.

Word candidate extraction using Japanese word dictionary 7 and grammar dictionary 8. Morphological analysis such as part-of-speech contact Vt test was performed.

After dividing the noun continuous compound word into words and adding part-of-speech information, pronunciation, etc. to each word, the analysis result is sent to the homophone extractor 2.

(Step 32) The homophone extraction unit 2 requests the table search unit 6 to search the homophone table 9 using the pronunciation of each word constituting the two-noun continuous compound word as a key. The table search unit 6 searches the homophone table 9, and when the search is successful, sends the record to the homophone extraction unit 2, and when the search is unsuccessful, sends a record to that effect to the homophone extraction unit 2. .

(Step S3) The homophone extraction unit 2 determines whether the search was successful according to the information from the table search unit 6, and if it is successful, the process proceeds to step S4.

If the search fails, this process ends 6 (Step S4) The homophone extraction unit 2 further sets the words that have been successfully retrieved as test target words and sends them to the homophone testing unit 3.

(Step S5) The homophone testing unit 3 first performs the following case classification and determines a character string (hereinafter referred to as a connected character string) to be checked for connection with the test target word.

[1] When the test target word is the final word of a noun continuous compound word, the concatenated character string is the prefix character string of the test target word.

[2] When the test target word is the first word of a noun continuous compound word, the concatenated character string is the subsequent character string of the test target word.

[3] When the test target word is a word located in the middle of a noun continuous compound word, the concatenated character string is both the front character string and the rear character string of the test target word.

(Step S6) Next, the homophone testing section 3 requests the table search section 6 to search the homophone conjunction determination table 10 using the font of the word to be tested as a key. The table search section 6 searches the homonym conjunction determination table 10 and sends the record to the homonym test section 3.

(Step S7) Furthermore, if the character string information 16 in the record sent from the table search unit 6' satisfies the following conditions, the homophone testing unit 3 executes the process in step S8.
If it does not exist, the process advances to step S9.

[In the case of 11, character string information 16 regarding a partial string of the concatenated character string including the final character of the concatenated character string exists.

In the case of [2], there is character string information 16 regarding a partial string of the concatenated character string that includes the first character of the concatenated character string.

[In the case of 3 characters, there is character string information 16 regarding a substring of the concatenated string that includes the last character of the forward concatenated string, or a partial character of the concatenated string that includes the first character of the backward concatenated string There is character string information 16 regarding the column.

(Step S8) Concatenation possibility ([
l], the forward connection is possible or not for 1 day.

In the case of [2], whether or not the rear connection is possible 19. In the case of [3], if the respective connection possibilities) are not possible, the process advances to step S9, and if the connection is possible, the process ends.

(Step S9) The homophone testing section 3 determines that the test target word is incorrect, and sends the test target word to the homophone candidate extraction section 4.

(Step 5IO) The homophone candidate extraction unit 4 searches the homophone table 9 through the table search unit 6 using the pronunciation of the test target word as a key, obtains correction candidates for the test target word, and obtains correction candidates for the test target word. is sent to the homophone candidate testing section 5.

The homophone candidate testing unit 5 searches the homophone conjunction determination table 10 through the table search unit 6 using each character face of the correction candidate as a key.

Check whether each correction candidate can be concatenated with a concatenated character string.

(Step 511) If there is even one correction candidate that can be connected with the concatenated character string, the process advances to step S12, and if there is none, the process advances to step S13.

(Step 512) All correction candidates that can be connected are output to the output file 11 as correct answer candidates for the test word.

(Step 513) All correction candidates acquired in step S10 are output to the output file 11.

Note that in the above explanation, the homophone table 9. The homophone conjunction determination table 10 was searched again, but the search results in step S3, step S
If all the search results in step 6 are carried around, the search in step SIO becomes unnecessary, and the processing speed can be increased.

Next, the operation will be explained using an example. FIG. 6 shows the results of morphological analysis of the noun continuous compound word used in the example.

Note that the morphological analysis method for obtaining this information is not particularly limited here. Fig. 7 shows an example of the contents of the homophone table 9, and Fig. 8 shows an example of the contents of the homophone concatenation determination table 1o, which collects information on whether or not a homophone can be concatenated with one character before or after it. .

[1] When a character string exists before a homophone, “Natural Chemistry” is used as input for the automatic Japanese sentence error verification/correction device 12.
think of. This noun continuous compound is.

It contains an error, and the correct answer is "natural science."

The morphological analysis unit 1 performs morphological analysis of ``Natural Chemistry'' using the Japanese word dictionary 71 and the grammar dictionary 8.

This morphological analysis result (FIG. 6 (1)) is sent to the homophone extraction section 2.

The homophone extraction unit 2 requests the table search unit 6 to search the homophone table 9 using the pronunciations of the nouns ``Shizen'' and ``Kagaku'' constituting "°Natural Chemistry°" as keys.

Upon receiving the request, the table search unit 6 first processes "natural". If you search the homonym table 9 using the reading ``shizen'' for ``nature'' as a key, the contents of the homonym table 9 will be as shown in Figure 7, so ゛ nature is on the table. Processing is performed for the 0th order "Chemistry" which does not exist in , and the search fails. Homophone table 9 is searched using ``pronunciation of ``°°chemistry'' as a key.

Since the contents of the homophone table 9 are as shown in FIG. 7, the search for "chemistry" is successful. These pieces of information are sent from the table search section 6 to the homophone extraction section 2. The homophone extraction unit 2 sets the word to be tested = "chemistry" according to the sent information. Furthermore, “chemistry”
Since this is the final word of ``Natural Chemistry'', the preceding character string ``Natural °'' is used as a concatenated character string.These results are then sent to the homophone testing section 3.

The homophone test unit 3 requests the table search unit 6 to search the homophone connection determination table 10 using the test target word=“chemistry” as a key.

Upon receiving the request, the table search unit 6 searches the homophone conjunction determination table IO using the test target word "chemistry" as a key, and the homophone conjunction determination table 10 is as shown in FIG. 8. The record of (1) is obtained, and the information of this record is sent to the homophone testing section 3.

The homophone test unit 3 uses the concatenated character string - a partial character string that includes the last character of "natural" (in this example, Figure 8 shows whether or not the homophone can be concatenated with the first or second character). Therefore, it is determined whether the record contains information regarding the single character string "ZEN"). In the record (1) of FIG. 8, there is character string information related to "zen", and its forward concatenation permission/inhibition information is x. Therefore, the test target word "chemistry" is determined to be incorrect and is sent to the homophone candidate extraction unit 4.

The homophone candidate extraction unit 4 searches the homophone table 9 using as a key the pronunciation of the word "chemistry" determined to be "incorrect" and obtains correction candidates. As is clear from FIG. 7, "Science" is obtained as the correction candidate.Then, this correction candidate "Science" is sent to the homophone candidate testing section 5.

The homophone candidate testing unit 5 replaces the word "chemistry" determined to be "incorrect" with the correction candidate "science" and tests the connection between "science" and the connected character string "nature". Namely.

・Test word - “science” ・Concatenated character string = “nature” First, when the homonym conjunctive determination table 10 is searched through the table search unit 6 using the test target word “science” as a key, homophone concatenation is found. The judgment table 10 is the eighth
As shown in the figure, the record (2) in Figure 8 is obtained. 1. In this record, there is character string information related to "zen" that includes the final character of the concatenated character string, and its forward concatenation The availability information is ○. Therefore, the correction candidate “science”
teeth.

It is determined that connection is possible and output to the output file 11 as a correct answer candidate.

As explained above, the correct answer "Science" can be output to the output file as a correction candidate.

[2] When a character string exists after a homophone
think of. This noun continuous compound is.

It contains an error, and the correct answer is "chemical experiment."

In the morphological analysis department, “scientific experiment” is translated into Japanese word dictionary 7.
1. Morphological analysis is performed using the grammar dictionary 8.

This morphological analysis result (FIG. 6 (2)) is sent to the homophone extraction section 2.

The homophone extraction unit 2 requests the table search unit 6 to search the homophone table 9 using the pronunciations of the nouns ``scientific experiment'' as ``kagaku'' and ``shitsuken'' as keys.

Upon receiving the request, the table search unit 6 first processes "science". If you search the homonym table 9 using the reading ``Science'' as the key, the contents of the homonym table 9 will be as shown in Figure 7, so it will be ``Science''.
The search for ``experiment'' is successful.The homonym table 9 is searched using the ``pronouncement'' of ``experiment'' as the key.The contents of the homonym table 9 are as shown in Figure 7. Therefore, the search for "experiment" fails. These pieces of information are sent from the table search section 6 to the homophone extraction section 2. The homophone extraction section 2 follows the sent information.

Test target word = “Science”, and “Science” is “
Since it is the first word of ``scientific experiment,'' the backward character string ``experiment'' is used as a concatenated string.And.

These results are sent to the homophone testing section 3.

The homophone test unit 3 requests the table search unit 6 to search the homophone concatenation determination table 10 using the test target word=“science” as a key.

Upon receiving the request, the table search unit 6 searches the homophone conjunctive determination table 10 using the test word "science" as a key, and the homophone conjunctive determination table 10 is as shown in FIG. 8. The record of (2) is obtained, and the information of this record is sent to the homophone testing section 3.

In the homophone test section 3, the concatenated character string = partial character string that includes the first character of "experiment" (in the current example, Figure 8 shows whether or not it can be concatenated with the first or last character of the homophone). Therefore, it is determined whether the record contains information regarding one character string ``actual''.
) records.

There is character string information related to “actual”, and its backward linkability information is ×. Therefore, the test target word "science" is determined to be incorrect and is sent to the homophone candidate extraction unit 4.

The homophone candidate extraction unit 4 searches the homophone table 9 using the pronunciation of the word "Science" determined to be "incorrect", "kagaku", as a key, and obtains correction candidates. As is clear from FIG. 7, "chemistry" is obtained as a correction candidate. Then, this correction candidate "chemistry" is sent to the homophone candidate testing section 5.

The homophone candidate testing unit 5 replaces the word "science" determined to be "incorrect" with the correction candidate "chemistry" and tests the concatenation of "chemistry" and the connected character string "experiment." That is.

・Test target word = ``chemistry'' ・Concatenated character string - ``experiment'' First, when the homonym conjunctive determination table 10 is searched through the table search unit 6 using the test target word ``chemistry'' as a key, homophone concatenation is found. The judgment table 10 is the eighth
As shown in the figure, the record (1) in FIG. 8 is obtained. In this record, there is character string information regarding "actual" including the first character of the concatenated character string, and the backward concatenation possibility information is O. Therefore, the correction candidate “chemistry”
teeth.

It is determined that connection is possible and output to the output file 11 as a correct answer candidate.

As explained above, the correct answer "Chemistry" can be output to the output file as a correction candidate.

[3] When character strings exist both before and after a homophone, consider "complete elimination of homecoming" as input to the automatic Japanese sentence error verification/correction device 12. This noun continuous compound word contains an error, and the correct answer is "total abolition of regulations."

The morphological analysis section 1 performs morphological analysis of ``total abolition of homecoming'' using the Japanese word dictionary 79 and the grammar dictionary 8. This morphological analysis result (FIG. 6 (3)) is sent to the homophone extraction section 2.

Homophone extraction unit 2 uses the pronunciations of the nouns that make up the phrase ``°°Comprehensive abolition of homecoming'' as ``zenmen'', ``kisei'', and ``tepishi''.
Homophone table 9 is created in table search section 6 using ``key'' as key.
request to search for.

In the table search unit 6 that received the request, first.

Processing is performed for ``Full surface.'' When the homonym table 9 is searched using the reading ``Zenmen°'' of ``Full surface'' as a key, the contents of the homonym table 9 are as follows.

As shown in Figure 7, the search for “full page” fails.
If the homonym table 9 is searched using ``pronounced ``sweet'' as a key, the contents of the homonym table 9 are as shown in FIG. 7, so the search for ``homecoming'' is successful. Finally, process “°abolition”, “°abolition”
Homophone table 9 with “tekishi” as the key
When you search for.

Since the contents of the homophone table 9 are as shown in FIG. 7, the search for "abolition" will fail.

These pieces of information are sent from the table search section 6 to the homophone extraction section 2. In the homophone extraction unit 2, according to the sent information, the word to be tested is set as ``homecoming''.Furthermore, since ``homecoming'' is an intermediate word in ``total elimination of homecoming'', it is a prefix character string. ``Full surface'' and the backward character string ``elimination'' are respectively assumed to be forward concatenated character strings and backward concatenated character strings.Then, these results are sent to the homophone testing unit 3.

The homophone test unit 3 requests the table search unit 6 to search the homophone concatenation determination table 10 using the test target word −°“homecoming°” as a key.

Upon receiving the request, the table search unit 6 searches the test target word °“
When the homonym conjunctive determination table 10 is searched using "homecoming" as the key, the homonym conjunctive determination table 10 is as shown in FIG. 8, so the record (3) in FIG. 8 is obtained, and the information of this record is is sent to the homophone testing section 3.

In the homophone test section 3, the partial character string containing the final character of the concatenated character string = "full face" (in this example, Figure 8 shows whether or not it can be concatenated with one character before or after the homophone). Therefore, it is important to check whether the record contains information about the single character string “face”) and whether the substring containing the first character of the concatenated string = °“elimination゛° (in this example, the 8th
Since the figure shows whether or not a homophone can be concatenated with one character before or after one character, it is determined whether the record contains information regarding one character string "Toru".

In the record (3) in Figure 8, there is character string information related to "Side 1", and its forward linkability information is ×, and character string information related to "Toru" is also present, as well as its backward linkability information. , ×.Therefore, the word to be tested, “homecoming,” is determined to be incorrect, and is sent to the homophone candidate extraction unit 4.

In this example, the connection availability information for both the front and rear is ×.

It is also determined that there is an error if the value is 0.

The homophone candidate extracting unit 4 searches the homophone table 9 using as a key the pronunciation "kisei" of the word °°women' that has been determined to be "incorrect" and obtains correction candidates. 7th
As is clear from the figure, correction candidates include “regulation” and “°
The 3Qi words of "regular" and "existing" are obtained. Then, these correction candidates are sent to the homophone candidate testing section 5.

The homophone candidate testing unit 5 first replaces the word “homecoming” determined to be “incorrect” with the correction candidate “regulation”.
“Regulation” and concatenated string “Full” (forward), “Abolition”
Test the connection with (backward). Namely.

・Words to be tested: “regulation” ・Concatenated string = “Full surface” (front).

"Abolition" (backwards). First, when the homophone conjunctive determination table 10 is searched through the table search unit 6 using the test target word "regulation" as a key, the homophone conjunctive determination table 10 is found in the eighth
As shown in the figure, the record (4) in FIG. 8 is obtained. In this record, there is character string information regarding the "face" that includes the last character of the forward concatenated character string, and the forward concatenation possibility information is O. Furthermore, there is also character string information regarding "Toru" that includes the first character of the backward concatenated character string, and the backward concatenation possibility information is -O.

Therefore, the correction candidate "regulation" is determined to be connectable.

Next, let's check the connection with the correction candidate "Regulation".

From the record (5) in FIG. 8, there is character string information regarding the "face" that includes the final character of the forward concatenated character string, and the forward concatenation permission/inhibition information is x. Further, there is no character string information related to "removal" that includes the first character of the subsequent concatenated character string, and therefore, the correction candidate is determined to be unconcatenable.

Finally, if we check the connection with the correction candidate "existing".

From the record (6) in Figure 8, there is no character string information regarding the "face" that includes the final character of the preceding concatenated character string.
Furthermore, there is no character string information regarding "removal" that includes the first character of the concatenated character string at the end.

Therefore, it is determined that the correction candidate cannot be connected.

From the above, since the only correction candidate that is determined to be concatenable is "Regulation," only "°Regulation" is output to the output file 11 as a correct answer candidate.

As explained above, the correct answer "Regulation" can be output to the output file as a correction candidate.

Because of this structure and operation, compared to conventional methods.

・It is possible to accurately extract homonyms that are incorrectly used.

・Replace the word determined to be "incorrect" with a correction candidate and perform a test using a homophone conjunction determination table,
Since correction candidates are selected, correct candidates can be output.

・Less man-hours than manual verification.

improved in each respect.

In the above example, the possibility of concatenation is expressed as a binary value of Q and Using a ratio, it may be possible to connect when 5 exceeds a certain threshold.

In addition, in the above example, the homophone conjunctive determination table lO
The character string is one character, but it is easy to expand it to two or more characters.

(5) As described in detail of the invention, (1) according to the present invention, input errors in Japanese documents input from an input device; (2) reading errors due to poor character recognition; or incorrect word selection in kana-kanji conversion; On homophone acidification within noun-sequential compounds that occur.

■ By automatically checking for errors in homophones included in continuous noun compounds using conjunctive information (maintained in the form of a homophone concatenation determination table) of the nouns constituting the continuous noun compound, Only incorrectly used items can be accurately extracted.

■ Since correction candidates are also tested using the homophone conjunction determination table and only correction candidates that can be connected are output, only correction candidates that are likely to be from is can be presented.

■ As a result of the test, only those determined to be incorrect are extracted and correction candidates are presented.

It requires less man-hours than manual verification.

■ When a homophone error is detected, correction candidates are output, making correction work easier.

■ A homophone conjunction determination table can be easily and mechanically created from one document data.

(2) When expanding the test target, it is sufficient to add data only for the expanded portion to the homophone table and the homophone concatenation determination table, resulting in excellent expandability of the test target.

There are advantages.

[Brief explanation of the drawing]

Figure 1 is a basic configuration diagram of the present invention configured by hardware, Figure 2 is a field configuration diagram of a homophone table, Figure 3 is a field configuration diagram of a homophone conjunction determination table, and Figure 4 is a homophone A configuration diagram of character string information in a word concatenation determination table, FIG. 5 is a schematic flow of the operation of the present invention, and FIG. 6 is a morphological analysis result of a noun continuous compound word used to explain the embodiment. FIG. 7 shows an example of the contents of the homophone table, and FIG. 8 shows an example of the contents of the homophone conjunction determination table. 1... Morphological analysis section, 2... Homophone extraction section, 3
... Homonym test unit, 4... Homophone candidate extraction unit. Death: homophone candidate test section, 6: table search section. 7... Japanese word dictionary, 8... Grammar dictionary, 9...
Homophone table, 10... Homophone conjunction judgment table 11... Output file, 12... Automatic Japanese sentence error verification/correction device, 13... Homophone reading,
14... Homophone character surface, 15... Homophone character surface, 16... Character string information, 17... Character string character surface, 1st... Possibility of forward concatenation, 19... Possibility of backward concatenation. . Patent applicant Nippon Telegraph and Telephone Corporation

Claims (1)

[Scope of Claims] Homophone errors in noun continuous compounds caused by errors in Japanese documents input from an input device can be solved by morphological analysis using a Japanese word dictionary and a grammar dictionary. The first method is to divide words into word strings and obtain morphological information such as the pronunciation and part of speech of each word, and a homophone table that stores homophone groups, which are words with the same pronunciation, using the pronunciation as a key. and, for each homophone recorded in the homophone table, a homophone conjunctive determination table that stores the possibility of concatenation between the homophone font and the preceding and succeeding character strings, using the homonym font as a key. and searching the homophone table using each word constituting the noun continuous compound word divided into word strings in the first means as a key, and starting the test if there is a homophone to be tested. Using the second means and the font of the homophone to be tested as a key, the above-mentioned homophone concatenation determination table is searched, and as a result, there is no possibility of concatenation between the homophone and the preceding and succeeding character strings. or a third means for testing the homophone as an error when the possibility of conjunction is not registered in the homophone conjunction determination table; and a third means for testing the homophone as an error in the third means a fourth means for obtaining correction candidates for the homophone by searching the homophone table using the pronunciation of the homophone as a key; and a font of the correction candidate obtained in the fourth means. a fifth means for searching the above-mentioned homophone concatenation determination table using as a key, and outputting only correction candidates that are likely to be concatenated with preceding and succeeding character strings as correct answer candidates for the homophone error; Japanese sentence error automatic test, which is characterized by
Correction method.