JPH05233619A - Method for correcting error of japanese language sentence and device therefor - Google Patents

Abstract

PURPOSE:To provide a Japanese language sentence error correcting method which can extract a correction candidate character, and the device for performing its method, even if a character which cannot be analyzed is any kind of character of a KANJI (Chinese character), a KANA (Japanese syllabary), etc., as the result of executing a form analysis of an inputted Japanese sentence. CONSTITUTION:The device is provided with a morpheme analyzing part 3 for parsing a sentence into word units, a word dictionary part 5 for holding a notation and a part of speech used by the morpheme analyzing part 3, or also, by allowing a semantic attribute to correspond, an unanalyzable character detecting part 7 for detecting a character which becomes unanalyzable in the morpheme analyzing part 3 and the corresponding position in the sentence of the unanalyzable character, a dictionary retrieving part 8 for retrieving a word which contains a designated character in the notation from the word dictionary part 5, and a word connection testing part 9 for executing a test of consistency of the word based on the part-of-speech information in a state that a word and an idle character-string retrieved by the dictionary retrieving part 8 and the unanalyzable character in the sentence are replaced, by which the error of a Japanese language sentence is corrected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is mainly used in Japanese word processors, and an error correction method and apparatus for extracting and correcting correction candidate words for errors due to missing characters in a sentence and unnecessary characters erroneously mixed in. Regarding

[0002]

2. Description of the Related Art Japanese sentences created by a Japanese word processor may include errors such as typographical errors and omissions due to conversion errors and editing errors in Kana-Kanji conversion during input.
As a method for correcting such an error in a sentence, for example, Japanese Patent Laid-Open No. 1-281561 discloses a Japanese sentence correction candidate character extraction method. FIG. 7 is a configuration diagram showing the configuration of a Japanese sentence correction device to which the method is applied.

In this figure, in the "unknown word detecting section",
Morphological analysis of Japanese sentences is performed using a Japanese word dictionary and a grammar dictionary, and characters at positions where words are discontinuous in terms of position or grammar are detected as unknown words.
On the other hand, in the “Kanji word table”, words with a high appearance frequency consisting of two Kanji characters are classified as a set of words having the same forward or backward characters, and the occurrence frequency, part of speech, and semantic attribute appear as a pair. Stored in order of frequency. Then, the front one character or the rear one character can be searched as a key.

In addition, the "correction candidate character extraction section" uses the Kanji word table to extract characters that are candidates for correction to the detected unknown word in the form of a plurality of character words, and further selects "correction candidate selection". The “part” selects a correction candidate for a multi-character word including the extracted correction candidate character by using the grammatical connection relationship with the preceding and following words, the semantic acceptance relationship, or the appearance frequency of the word. ing. In addition, a portion surrounded by a broken line in the drawing indicates a processing device including a CPU and a memory.

The flow of operations in this processing apparatus is indicated by arrows in the figure. First, a Japanese sentence input through an "input device" such as a keyboard is processed by an "input processing unit" and then stored in a character code format as an "input Japanese sentence database". Next, in the "unknown word detector", morphological analysis of the Japanese sentence input using the "Japanese word dictionary" and "grammar dictionary" is performed, and the position or grammatical analysis of the word is performed based on the result. Characters at discontinuous points are detected as unknown words.

Further, the "correction candidate character extracting section" extracts a correction candidate character for the unknown word detected using the "kanji word table". Subsequently, the “correction candidate selection unit” selects an appropriate correction candidate character from some correction candidate characters extracted using the “Japanese word dictionary” and the “grammar dictionary”. In this way, the error-relieved “Japanese document database” is created.

[0007]

However, according to the above-described processing apparatus, the correction candidate characters to be extracted are limited to words consisting of two Kanji characters, and Hiragana words and words of two or more characters are corrected. It cannot be extracted as a candidate. Also, it is not possible to delete unnecessary characters that are accidentally mixed in the text. Furthermore, a kanji word table must be prepared in advance for a word consisting of two kanji characters and having a high appearance frequency, and a large storage area is required to hold it.

The present invention has been made in view of the above circumstances, and even if the characters that cannot be analyzed as a result of morphological analysis of the input Japanese sentence are kanji, hiragana, etc. An object of the present invention is to provide a Japanese sentence error correction method that enables extraction of the correction candidate character and an apparatus for implementing the method.

[0009]

In order to solve the above problems, the present invention provides a morphological analysis unit for decomposing a sentence into words, a word notation and a part of speech used in the morphological analysis unit, or further a semantic attribute. A word dictionary section that holds the corresponding words, an unanalyzable character detection section that detects a character that cannot be analyzed by the morpheme analysis section and a corresponding position in the sentence of the unanalyzable character, and a designated character in the notation. Based on the part-of-speech information of a word in a state in which the dictionary search unit that searches the included word from the word dictionary unit and the word and the empty character string searched by the dictionary search unit and the unanalyzable character in the sentence are replaced. (EN) A Japanese sentence error correction device is provided with a word connection verification unit that tests whether or not a grammatical connection is possible or further tests the consistency of words based on semantic attributes.

[0010]

According to the present invention, the morpheme analysis unit having the above-described configuration decomposes a Japanese sentence into words using the word dictionary unit that holds the word notation and the part of speech or the semantic attribute in association with each other. Then, the unanalyzable character detection unit detects the unanalyzable character and the corresponding position in the sentence of the unanalyzable character from the result of the morphological analysis.

Further, the dictionary search unit searches the word dictionary unit for a word including the unanalyzable character detected by the unanalyzable character detection unit in the notation. Furthermore, the word connection verification unit is a test of whether or not grammatical connection is possible based on the part-of-speech information of the word in a state where the word and the empty character string searched by the dictionary search unit and the unparsable character in the sentence are replaced, Alternatively, a word consistency test is performed based on the semantic attribute.

As described above, a correction that includes an unparseable character and is grammatically connectable or semantically consistent with the words before and after the unparseable character with respect to the unparseable character in the Japanese sentence. Word candidates are extracted. Then, the unanalyzable character is replaced with the correction candidate word. Alternatively, the unparsable character is replaced with an empty character string and removed from the sentence.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT An embodiment of the present invention will now be described with reference to the drawings.
To explain. FIG. 1 is a Japanese sentence error correction according to the present invention.
It is a block diagram which shows the structure of a primary device. This Japanese sentence error
The correction device is used to input instructions and Japanese sentences to the device.
Nau input unit 1 and the function according to the input from the input unit 1.
Control unit 2 that activates and decomposes Japanese sentences into words
Morphological analysis unit 3 and a text record that stores Japanese sentences
The storage unit 4 and the morphological analysis unit 3 perform morphological analysis on Japanese sentences.
Correspond and hold the word notation and part of speech used when
As a result of morphological analysis, the word dictionary unit 5 is decomposed into word units.
Information about the Japanese text that was
Morphological analysis result storage unit 6 for storing information and morphological analysis results
Characters that cannot be parsed from the result storage unit 6 and the unparseable characters
Unparsable Character Detection to Detect Corresponding Positions in Text
Words that include part 7 and specified characters
The dictionary search unit 8 that searches from the writing unit 5 and the analysis
With the Nohji and the word searched by the dictionary search unit 8 replaced
A simple test for grammatical connection based on word part-of-speech information
Of the words searched by the word connection verification unit 9 and the dictionary search unit 8
It was determined by the word connection verification unit 9 that grammatical connection was possible.
Correction candidate storage unit 10 for storing words and correction candidate storage unit
Correspond the word stored in 10 with the unparseable character in the sentence
Correction candidate display unit 11 to be attached and displayed, and correction candidate display unit
Select one of the words displayed by 11 according to the operator's instruction.
Correction candidate selection unit 12 to select, and the unparsable character in the sentence
Is replaced with the correction candidate word selected by the correction candidate selecting unit 12.
Character string replacing unit 13, the input information from the input unit 1,
Japanese sentences and correction candidate table stored in the strike storage unit 4
From the output unit 14 that displays the information of the processing result of the display unit 11
It is configured.

The operation of the Japanese sentence error correction device configured as described above will be described below. First, when the operator inputs an instruction to perform error correction on a Japanese sentence stored in the text storage unit 4 from the input unit 1, the control unit 2 activates the morphological analysis unit 3. Then, the morphological analysis unit 3
Uses the word dictionary unit 5 to decompose the Japanese sentence stored in the text storage unit 4 into word units, and stores the information in the morphological analysis result storage unit 6.

Next, the control unit 2 activates the unanalyzable character detection unit 7. Then, the unanalyzable character detection unit 7 refers to the information stored in the morpheme analysis result storage unit 6, detects the unanalyzable character, and at the same time detects the information of the corresponding position in the sentence of the unanalyzed character. The information is returned to the control unit 2. Further, the control unit 2 sets the unanalyzable character returned from the unanalyzable character detection unit 7 as the designated character, and the dictionary search unit 8
To start. The dictionary search unit 8 then uses the word dictionary unit 5
For all the words registered in, it is determined whether or not the specified character is included in the notation by verifying the notation for each word, and only the included words are included in the correction candidate storage unit 10. To store.

Subsequently, the control unit 2 activates the word connection verification unit 9. Then, the word connection verification unit 9 refers to the information stored in the morphological analysis result storage unit 6 after adding the word of the empty character string to the correction candidate storage unit 10, and is stored in the correction candidate storage unit 10. The grammatically inaccessible candidates are deleted from the correction candidate storage unit 10 with the words replaced with the unparseable characters.

Subsequently, the control unit 2 causes the correction candidate display unit 11 to
Is started, and at the same time, information on the position of the unanalyzable character detected by the unanalyzable character detection unit 7 is passed to the correction candidate display unit 11. Then, the correction candidate display unit 11 refers to the text storage unit 4 and the correction candidate storage unit 10, displays the position of the unanalyzable character in the sentence and the correction candidate word in association with each other on the output unit 14, and simultaneously displays the correction candidate selection unit. 12 is activated.

Here, the operator inputs, via the input unit 1, an instruction to select one of the correction candidate words displayed by the correction candidate display unit 11. Then, the correction candidate selection unit 1
When 2 receives this input, it activates the character string replacing unit 13, and at the same time, passes the information of the selected correction candidate word to the character string replacing unit 13. Therefore, the character string replacing unit 13 replaces the uncorrectable character with the selected correction candidate word.

FIG. 2 is a schematic diagram showing an example of a morphological analysis result of a Japanese sentence containing unanalyzable characters. That is, when the created Japanese sentence is stored in the text storage unit 4 and the operator inputs an instruction to correct the error from the input unit 1, the morphological analysis of the Japanese sentence performed by the morphological analysis unit 3 is performed. An example of the results is shown. The analysis result is stored in the morphological analysis result storage unit 6.

More specifically, as shown in the figure, the Japanese sentence "I am running" is stored as information decomposed into word units at the position indicated by diagonal lines. Regarding the content of the decomposed words, "I" is a noun, "ha" is a particle, "run" is a verb "run", or a noun, and the next "ma" is a character that cannot be analyzed in this sentence. , The last "ga" is a case particle.

FIG. 3 is a schematic diagram showing a list of correction candidate words in which an empty character string (displayed as NULL) is added to the words including unparsable characters in the notation. This correction candidate word list is created as follows. That is, the unanalyzable character detection unit 7 refers to the result of the above-mentioned morphological analysis to detect the unanalyzable character "ma", and at the same time, corresponds to the unanalyzable character "ma" in the sentence. The position information is returned to the control unit 2. Therefore, the control unit 2 sets the unanalyzable character "MA" as the designated character and activates the dictionary search unit 8. Here, the dictionary retrieval unit 8 verifies the notation of all the words registered in the word dictionary unit 5, thereby retrieving the word including the designated character "MA" in the notation. The retrieved correction candidate word is stored in the correction candidate storage unit 10.

Correction candidate word list searched and stored
As shown in the figure,Well
(Indicated by underlining)
Well"", "Akara"Well, ...Well","Well
"", "Well","Well","Well","Well"
"WellIt is listed as "...". Figure 4
Replaced the unparsable character "ma" with the above correction candidate word
Replacement text string indicating status and its grammatical connectability
It is a schematic diagram which shows the determination result of. For replacement text string
Is created as follows. That is, the control unit 2
When the connection verification unit 9 is activated, the word connection verification unit 9 will be displayed in FIG.
As shown in FIG.
Add (NULL). Furthermore, the word connection verification unit 9
Correction candidate storage while referring to the morphological analysis result storage unit 6
Each word of the correction candidate and the unanalyzable sentence stored in the section 10
Replace the letter "ma".

Specifically, as shown in the figure, the unanalyzable character "ma" is replaced with each correction candidate word shown in FIG.
"Running - love or decoction - is", "running - Akara is or -
Ga ", ..." ran - masu or be - is "," running - or not -
Ga "," running - or cane - is "," running - or to - is ",
"Run - or be - it is", "running - or them - is", "running -
Or decoction - is "..." running - () - there is such as replacement text string of "is created.

Then, for each column of the replacement text created in this way, the word connection test unit 9 precedes each correction candidate word in which the unanalyzable character "ma" has been replaced with the word "run (verb continuous form). , Or a noun) ”and the subsequent word“ ga (case particle) ”are grammatically connectable.
As a result of this determination, the "run - is - or be", "running - () - is" for the two of grammar connectable with the result (indicated by .smallcircle.), The other for the impossible connection Since it exists (indicated by X), it is deleted from the list of correction candidate words.

FIG. 5 is a schematic diagram showing a state in which correction candidate words are displayed corresponding to unanalyzable characters in a Japanese sentence. That is, the control unit 2 activates the correction candidate display unit 11,
At the same time, by passing the information on the position of the unanalyzable character “m” detected by the unanalyzable character detection unit 7, the correction candidate display unit 11 causes the Japanese sentence stored in the text storage unit 4 and the correction candidate storage unit 10 to be displayed. As shown in the figure, the unanalyzable character "ma" is separated by diagonal lines while referring to the correction candidate words stored in the table, and the correction word candidates "1.mas" and "2. "NULL)" is displayed on the output unit 14. The correction candidate display unit 11 also activates the correction candidate selection unit 12 at the same time.

FIG. 6 shows the unparseable characters in Japanese sentences,
It is a schematic diagram which shows the state replaced with the selected error correction candidate word. That is, when the operator inputs the number 1 from the input unit 1 to select the word "masu" as one of the correction candidate words displayed by the correction candidate display unit 11, the correction candidate selection unit 12 Upon receiving such an input, the character replacement unit 13 is activated, and at the same time, the selected correction word candidate number 1 is passed. As a result, as shown in the figure, the character string replacing unit 13 replaces the unparsable character “m” with the selected correction candidate word “masu”, and “I run, but” correct.

As described above, in the present embodiment, the word dictionary used in the morphological analysis is searched for words that include unanalyzable characters in Japanese sentences in the notation. Furthermore, an empty character string (NUL
L) is added to replace this with the unparsable character in the sentence. Then, in this replaced state, the correction candidate words that cannot be grammatically connected are deleted based on the part-of-speech information, and the unanalyzable characters in the sentence and the correction word candidates are displayed in association with each other. Therefore, if the character string is replaced with the correction word candidate in accordance with the instruction of the operator, an error due to a missing character in the sentence or an unnecessary character mixed by mistake will be corrected.

In this embodiment, the unanalyzable character detector 7 is used after the morphological analysis of the entire Japanese sentence is performed.
However, it is also possible to activate the unanalyzable character detection unit 7 when the recognition of the word immediately after the unanalyzable character is completed in the morphological analysis. Further, the dictionary search unit 8 verifies the notation for each word with respect to all the words registered in the word dictionary unit 5 and determines whether or not the designated character is included in the notation, Although only the included words are detected, it is also possible to detect them by other means. Further, the correction candidate words are displayed on the output unit 14 in the format as shown in FIG. 5, but of course they may be displayed in other formats.
Further, when the correction candidate word is selected, the selection instruction is given by the number attached to the candidate. However, a means for giving an instruction for selection by an instruction device such as a mouse may be used.

[0029]

As described above, in the Japanese sentence error correction device of the present invention, a word that includes a character that cannot be analyzed is searched using the notation of the word used in the morphological analysis as a key. A possible word dictionary is searched, and the possibility of grammatical connection is tested with the searched word group replaced with unparsable characters in the sentence. Therefore, it is possible to extract uncorrectable characters of any character type such as kanji and hiragana without limiting the number of characters of the correction candidate word. Further, it is possible to extract correction candidates without fail for any error caused by any missing characters or erroneously mixed unnecessary characters, and correct the error with high accuracy. Furthermore, without using a special table for searching correction candidate words,
Since a large storage area is not required, its practical effect is extremely large.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a configuration of a Japanese sentence error correction device according to the present invention.

FIG. 2 is a schematic diagram showing an example of a morphological analysis result of a Japanese sentence containing unanalyzable characters.

FIG. 3 is a schematic diagram showing a list of correction candidate words in which an empty character string is added to words including unanalyzable characters in the notation.

FIG. 4 is a schematic diagram showing a replacement text string showing a state in which an unparseable character is replaced with a correction candidate word, and a grammatical connectability determination result.

FIG. 5 is a schematic diagram showing a state in which correction candidate words are displayed corresponding to unanalyzable characters in a Japanese sentence.

FIG. 6 is a schematic diagram showing a state in which an unanalyzable character in a Japanese sentence is replaced with a selected error correction candidate word.

FIG. 7 is a configuration diagram showing a configuration of a Japanese sentence correction device of a conventional device.

[Explanation of symbols]

 1 Input Section 2 Control Section 3 Morphological Analysis Section 4 Text Storage Section 5 Word Dictionary Section 6 Morphological Analysis Result Storage Section 7 Unanalyzable Character Detection Section 8 Dictionary Search Section 9 Word Connection Verification Section 10 Correction Candidate Storage Section 11 Correction Candidate Display Section 12 Correction candidate selection unit 13 Character string replacement unit 14 Output unit

Claims (4)

[Claims] 1. A morphological analysis step of decomposing a sentence into words when correcting an error in a Japanese sentence, a character that cannot be analyzed in the morphological analysis step, and a correspondence between the unanalyzable character in the sentence. An unanalyzable character detection step of detecting a position, a dictionary search step of searching a word including a specified character in the notation used in the morpheme analysis step in a word dictionary that holds the notation and the part of speech in association with each other, Execute the word connection verification step that tests whether or not grammatical connection is possible based on the part-of-speech information of the word with the word and empty character string searched in the dictionary search step replaced with the unparsable character in the sentence. Omissions in sentences,
Characterized by extracting a candidate for a correction word that includes the unparsable character with respect to the unparsable character due to an unnecessary character that is mistakenly mixed and is grammatically connectable with the words before and after the unparsable character Japanese sentence error correction method. 2. A morpheme analysis unit that decomposes a sentence into words, a word dictionary unit that holds the notation of a word used in the morpheme analysis unit and a part of speech, and the morpheme analysis unit cannot analyze the words. The unanalyzable character detection unit that detects a corresponding position in the sentence between the character and the unanalyzed character, the dictionary search unit that searches the word dictionary unit for a word that includes the specified character in the notation, and the dictionary search unit. Equipped with a word connection verification unit that tests whether or not grammatical connection is possible based on the part-of-speech information of a word with the searched word and empty character string replaced with the unparsable character in the text A candidate for a correction word that includes the unparsable character with respect to the unparsable character due to an unnecessary character erroneously mixed and is grammatically connectable with the words before and after the unparsable character. Japanese Text error correction device. 3. The morphological analysis step uses a word dictionary that holds notations, parts of speech, and semantic attributes in association with each other, and further, the word connection verification step uses words and empty character strings searched in the dictionary search step. In addition to testing whether or not grammatical connection is possible based on the part-of-speech information of a word with the unparsed characters in the sentence replaced, it is also possible to test the consistency of the word based on the semantic attribute. Claim 1 characterized by
How to correct the written Japanese text error. 4. The word dictionary unit is a word dictionary unit that holds a notation, a part of speech, and a semantic attribute in association with each other, and the word connection verification unit searches words and empty character strings and sentences in the dictionary searched by the dictionary search unit. In addition to testing the grammatical connection based on the part-of-speech information of the word in the state of replacing the unparsable character, the word connection verification unit for testing the consistency of the word based on the semantic attribute 3. The Japanese sentence error correction device according to claim 2, wherein