JPH0275059A - Error correction processor for japanese sentence - Google Patents

Abstract

PURPOSE:To improve the processing efficiency for production of the correction candidates by using a correction character dictionary which stores the pairs of wrong characters and their correct candidates in the order of characters which are used wrong more easily. CONSTITUTION:A processing part 2 includes a morpheme analyzing part 4, an error detecting part 5, and a correction candidate producing part 6. The part 4 uses a word dictionary 7 and a word connection table 8 to process the Japanese word documents received from an input part 1. The part 5 detects the wrong character strings. Then the part 6 performs its process with use of a character connection table 9 and a correction character dictionary 10 in addition to the dictionary 7 and the table 8. In such a constitution, the uniform process is not applied to all characters and the preference is given to those characters which are easily misused. Thus the correction candidates are produced with high efficiency.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a Japanese sentence error correction processing device that detects erroneous character strings in Japanese sentences and presents correct word candidates.

BACKGROUND ART Conventionally, this type of Japanese error correction has been described, for example, in Information Processing Society of Japan Vo Q, 25 No. 2 Mar.

Some examples are shown in the paper ``Automatic detection of 1''J main text errors using a word analysis program and extraction of correction candidates using a quadratic Markov model'' published in 1984. This is, roughly speaking,
The correction candidate is estimated based on the probability of linking the corrected character candidate with the characters before and after the error' part.

Problems to be Solved by the Invention However, in the case of such a conventional method, all characters are treated equally, so it is not possible to reflect the combinations of characters that are likely to be mistaken, the operating conditions when using manual input, etc.

Furthermore, since the search is performed in character units, even characters that are not grammatically connected to the previous word may be extracted as candidates.

Means for Solving the Problem: A morphological analysis unit and an error detection unit using a word dictionary and a word connection table detect incorrect character strings from a Japanese document, and detect the correct word for the erroneous character string for hiragana errors. In this Japanese sentence error correction processing device, a corrected character dictionary is provided which stores pairs of erroneous characters and correct candidate characters in the order of characters in which errors are likely to occur.

When considering errors in Japanese text, there are some that are easy to make due to the layout of the keys on the keyboard, forgetting to use the shift key, etc., and there are also characters that are less likely to make mistakes. The corrected character dictionary takes this point into consideration and stores pairs of error characters and correct candidate characters in the order in which errors are likely to occur, so it is not possible to process all characters equally. First, since priority is given to items that are likely to cause errors, it is possible to create correction candidates efficiently.

Embodiment A first embodiment of the present invention will be described with reference to FIGS. 1 to 7. FIG. 1 shows a block diagram of a Japanese sentence error correction processing device, which roughly consists of a human power section 1, a processing section 2, and an output section 3. Here, the processing section 2 includes a morphological analysis section 4, an error detection section 5, and a correction candidate creation section 6. First, the morphological analysis section 4 uses the word dictionary 7 and the word connection table 8 to process the Japanese document from the input section 1, and the error detection section 5 detects erroneous character strings. On the other hand, the correction candidate creation section 6 performs processing using the word dictionary 7, word connection table 8, character connection table 9, and correction character dictionary 10, which is a feature of this embodiment.

Note that the correction candidate creation section 6 is provided with a dictionary data specifying means 11 for switching the correction character dictionary]O by manually inputting whether the document to be processed from the input section 1 has been input manually in kana or Roman characters. You can.

22. The word dictionary 7, for example, as shown in FIG.
It consists of codes representing parts of speech, word pronunciation, and word notation.

Further, the word connection table 8, as shown in FIG. 4, for example, is a two-dimensional table that represents word connections in terms of backward connections and forward connections.

As shown in the diagram, "O" if connected, "×" if not connected
I remember that. Using this word connection table 8, the morphological analysis section 4 examines connections between words, and the correction candidate creation section 6 narrows down correction candidates.

In addition, the character connection table 9 represents the connections between characters, and is expressed in a two-dimensional table similar to the word connection table. For example, as shown in Figure 5, the character connection table The character indicates whether or not to connect. As shown in the figure, if it is connected, it is stored as "O", and if it is not connected, it is stored as "x". This is mainly used to check consonants and consonants.

Therefore, the corrected character dictionary 10, which is a feature of this embodiment, is as follows:
As shown in Figure 2, it is broadly divided into two categories: kana input and romaji input. Characters are classified according to the position of the key in the word processor and the presence or absence of shifts, and the erroneous characters are sorted in order of their likelihood of error when inputting. This is a pair of characters that have been corrected (erroneous characters) and characters that are candidates for correction (candidate characters). The dictionary consists of input methods (kana manual/romaji input), error characters, and candidate characters. The correction candidate creation unit 6 creates a correction character dictionary 10 in the error character string.
If there are any erroneous characters, they are replaced as candidate characters and correction candidates are created.

In such a configuration, an outline of error correction processing is shown in the flowchart of FIG. First, the dictionary data specifying means 1 manually selects whether a Japanese document to be subjected to error correction processing was input using kana or Roman characters. This is because, in the present embodiment, since the erroneous character is estimated based on the position of the key at the time of input, errors are corrected using different methods for kana input and for manual Romaji input. After selecting an input method, it is checked whether there is an input character string to be processed, and if there is an input character string, the morphological analysis unit 4 performs morphological analysis on the human-written sentence, and the result is The error detection section 5 performs error detection. Then, the correction candidate generation unit 6 performs correction candidate generation processing on the character string in which an error has been detected, and the resulting correction candidates are displayed on the screen.

During such processing, the processing in the correction candidate creation unit 6 using the correction character dictionary 0 will be explained with reference to the flowchart 1- in FIG. First, when all the erroneous characters in the corrected character dictionary 10 have been checked or the number of candidate words has become quite large (
In this embodiment, the number of characters is 10), and the process ends, but if the state is before then, check whether there is an error character in the correction character dictionary 10 in the error part. We will investigate in order. If not, check the next erroneous character in the same way. As a result of such a check, if there is an erroneous character in the corrected literature society A10 in the error part and there is a candidate character, check whether the candidate character connects with the character immediately before the erroneous character in the error part. Check using connection table 9. If connected, the candidate character is replaced with a character that has the same notation as the erroneous character in the error part.

The word generated by the replacement is in the word dictionary 7, and the word connection table 8
Accordingly, if the word is misconnected with the 11 words immediately before the error part, this word is set as a correction candidate word, and processing proceeds to the next candidate word.

In this way, if there is a character that matches the error character from the beginning of the correction character dictionary 10 in the error part, it is replaced to create a correction candidate, and as described above, all the error characters are checked or the candidate When the amount reaches a certain value, the process ends.

Now, as a specific example, we will explain the target sentence ``Procedures for repurchasing a commuter pass.'' This target sentence 1: When morphologically analyzed by the morphological analysis unit 4, it becomes ``Kekkou''.
becomes an unregistered word, and the error detection unit 5 recognizes it as an error part. After this, the correction candidate creation unit 6 as shown in FIG.
The correction candidate extraction process is performed by First, in the progress of the process described above, there is "tsu" in the corrected character dictionary 10 in the error part "kekko", there is a candidate character, and "ha" is found in the corrected character dictionary 10.
and "tsu" are connected, so the "tsu" in "kekko" is replaced with the first "tsu" of the candidate characters. And word dictionary 7
The dictionary 7 is searched for the reading "Hakko". As illustrated in Figure 3, the word dictionary 7 includes ``issuance'' and ``fermentation.''
Examples include ``poor happiness,''``effect,'' and ``luminescence.'' Since the word ``re'' immediately before the incorrect part is connected to a subverb noun, ``susuki'', which is not connected, is deleted. The remaining words are then used as correction candidate words.

Next, the process proceeds to the next candidate character for the error character "tsu" in the corrected character dictionary 10. First of all, candidate characters for ``r'' are ``
Look up the ``chi'' after ``.

Regarding “chi”, “ha” can be connected with letters, and “F”
and "chi" are replaced, and "hetaka" is in the word dictionary 7 for "hachikou", but since it does not connect to "re", it is not selected as a candidate. We will investigate the incorrect character "tsu" in the same way and find the candidate j)
When one word is exhausted, the next erroneous character "yu" in the corrected character dictionary 10 is checked.

Since "yu" is not in the error part "ketsukou", processing proceeds to the next error character "u". ``U'' is in the error part ``Hatsukou'', so check the first ``U'' of the erroneous character ``U''. Since ``ko'' and ``u'' are not connected, examine the next candidate character ``a'' in the same way as in 1).

Thereafter, the process proceeds in the same manner, and when the number of candidates becomes greater than 10 or all the erroneous characters in the corrected literary world N10 have been investigated, the process ends for this erroneous part.

According to this embodiment, by displaying the corrected candidate characters obtained in this way, it is possible to correct the correct Japanese text, such as in [procedures for reissuing a commuter pass]. Since we use a single correction character dictionary 12 that stores candidates in the order of characters that are likely to make mistakes, taking into account key layouts, etc., we can create correction candidates more efficiently than with conventional equivalent processing methods for all characters. It can be carried out. Furthermore, according to this embodiment, since the word connections are also examined to narrow down the correction candidates, words that are not grammatically connected are not extracted as candidates, improving accuracy.

Next, a second embodiment of the present invention will be described in FIGS. 8 to 10.
This will be explained using figures. This embodiment adds a frequency information dictionary 12 that stores frequency information about words in a Japanese document to be processed, and uses the frequency information of this frequency information dictionary 12 to give priority to correct candidate words. This is how it was done.

That is, the morphological analysis unit 4 performs processing using the word dictionary 7 and the word connection table 8, and stores the frequencies of the analyzed words in the frequency information dictionary]2. The correction candidate creation unit 6 then uses a word dictionary 7, a word connection table 8, a character connection table 9, and a correction character dictionary 1.
0 and this frequency information dictionary 12 will be used for processing.

Such a frequency information dictionary 12 consists of pairs of notation and frequency, as shown in FIG. 9, for example, and this frequency is increased by one each time a word is determined during morphological analysis.

FIG. 10 is a flowchart showing correction candidate creation processing according to this embodiment in which such a frequency information dictionary 12 is added. Basically, the processing is the same as in the case of the above embodiment, but if the word generated by substitution exists in the word dictionary 7 and there is a word that can be connected to the word immediately before the error part, The words are rearranged in the order of frequency in the frequency information dictionary 12, and these are used as correction candidate words.

For example, as a specific example, consider the case where the target sentence is ``If you lose a commuter pass that has been issued twice. As in the case of the example, if you replace "tsu" in "hatsuko" with "tsu" and search the word dictionary 7, you will get "issuance", "fermentation", "light happiness", "luminescence", and "effectiveness".And, Following the check to see if the product J4 can be connected, the unconnected product J4 will be deleted.

For each of the remaining words, the frequency information dictionary 12 based on the results of morphological analysis is checked, and the words can be suggested in descending order of frequency.

Then, since "issue" has the highest frequency, it is placed at the beginning of the correction candidate 'dP.

Effects of the Invention As described above, the present invention provides a correction character dictionary that stores pairs of error characters and correct candidate characters in the order of the characters that are most likely to cause errors. Priority can be given to characters that are likely to cause errors due to the presence or absence of combined use of the shift key, etc. Therefore, correction candidate creation processing can be performed more efficiently.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a first embodiment of the present invention;
Figure 2 is a configuration diagram of the corrected character dictionary, Figure 3 is a configuration diagram of the word dictionary, Figure 4 is a configuration diagram of the word connection table, Figure 5 is a configuration diagram of the character connection table, and Figure 6 is the overall configuration. Flowchart 1 showing the process, FIG. 7 is a flowchart showing the correction candidate creation process, FIG. 8 is a block configuration diagram showing the second embodiment of the present invention, FIG. 9 is a configuration diagram of the frequency information dictionary, and FIG. FIG. 10 is a flowchart 1 showing correction candidate creation processing. 4... Morphological analysis unit, 5... Error detection unit, 7...
m-word dictionary, 8...word connection table, 10...corrected sentence 7 dictionary Applicant Ricoh Co., Ltd. -47 Figure

Claims (1)

[Claims] Japanese sentence error correction processing that detects erroneous character strings in Japanese documents using a morphological analysis unit and error detection unit using a word dictionary and word connection table, and presents words that are the correct answer to the erroneous character strings for hiragana errors. 1. A Japanese sentence error correction processing device, characterized in that the device is provided with a correction character dictionary storing pairs of error characters and correct candidate characters in the order of characters in which errors are likely to occur.