JPH0577104B2 - - Google Patents

Description

Detailed Description of the Invention "Industrial Application Field" This invention aims to eliminate errors contained in Japanese character strings containing kanji and kana read from an input device in order to create a Japanese document database. This invention relates to a Japanese sentence error testing device that tests the unnaturalness of the pronunciation of words or proper nouns with one kanji character.

"Conventional technology" When creating a Japanese document database by converting a large amount of Japanese documents such as newspaper articles, publication manuscripts, and chemical technology papers into electronic files, it is necessary to avoid misread characters, typos, and omissions that are mixed into the reading results. , The work of verifying and correcting errors such as misplaced insertions is done by manual visual checking.In pairs, one person reads the raw manuscript aloud, and the other person reads the inputted manuscript visually while reading it together. Or, as shown in Japanese Patent Application No. 60-45928 and No. 60-73844, a proofreader sequentially gives proofreading instructions while reading aloud with a synthesized voice after performing morphological analysis. Then, after the reading is finished, the relevant parts are proofread.

However, when processing a large amount of Japanese documents,
In the former manual-requiring work, (1) continuous visual inspection and reading cannot be done for a long time, and the accuracy varies depending on the ability of the proofreader.

(2) Reading errors such as misreading and skipping mistakes occur in the reading process.

In the latter mechanical process of reading aloud with synthesized voice,
Since the reading of a word is given a reading with high frequency in the dictionary, it is especially difficult to find errors in kanji compound words depending on the reading. For example, "work efficiency" Sagi Yoko is confused with "work pit" and is difficult to detect, so it has the disadvantage that it cannot be corrected to the correct answer "work efficiency."

Furthermore, as shown in Japanese Patent Application No. 61-228268, word strings that are prone to errors are registered in advance in a dictionary with incorrect notation, and after morphological analysis, matching with the incorrect notation is performed. There are methods to detect errors. However, even with this method, (1) it is necessary to register a large number of error-prone word strings in the dictionary in advance;

(2) Errors such as misspellings, omissions, and insertions in compound words occur irregularly, and it is practically difficult to register them as erroneous word strings in advance.

This has the disadvantage that detection accuracy is not improved and work efficiency is poor.

"Means for Solving the Problem" According to the present invention, for every single kanji character that has an on-yomi or kun-yomi, an on-yomi or kun-yomi is used as an on-yomi or a kun-yomi to indicate whether the on-yomi or kun-yomi is an on-yomi or a kun-yomi. A single-character kanji reading dictionary is created in advance in which words with information flags called flags are registered, and morphological analysis using a data processing device, a word dictionary, and a grammar dictionary is used to identify single-character kanji independent words within kanji compound words. Alternatively, if a one-letter proper noun for a kanji is extracted, a correct reading is given to that one kanji character using the one-letter kanji guessing dictionary, and then the proofreader reads the speech uttered by the speech synthesizer. The error is tested using the unnaturalness of the reading of the kanji in the Japanese sentence.

As a result, it is possible to work continuously for a relatively long time, and there is little variation in error detection accuracy depending on the proofreader's ability.The proofreader can perform verification work using only the original manuscript and audio, resulting in high work efficiency. .

The conventional technology does not give the reading of the word at the time of recognition to the reading of a single-character kanji independent word or a kanji-character proper noun recognized within a kanji compound word through word recognition using morphological analysis, but instead uses the reading of a single kanji character. The difference is that the reading is given by adding the guess reading of the kun reading registered in the dictionary, and the proofreader tests for errors based on the unnaturalness of the uttered voice.

Embodiment FIG. 1 shows an example of the basic configuration of the present invention. Input device 1 is Kanji OCR (optical reader), Pentouch,
The input device 1 is input or read by the input processing unit 2, and the Japanese sentences read by the input processing unit 2 are stored in a magnetic device in the input Japanese text database 3 with character codes. recorded in the format. The Japanese sentences in the input Japanese sentence database 3 are the word dictionary 4 and the grammar dictionary 5.
Word recognition is performed in the morphological analysis unit 6 using .
From the certified word, 1 kanji in the kanji compound word
A character-independent word or a one-character proper noun in Chinese characters is extracted by a typographical candidate extraction unit 7. Using a kanji one-character guessing dictionary 8, a kanji one-letter guessing adding section 9 adds a guessing to a one-letter kanji independent word or one-letter proper noun extracted by a misspelling candidate extracting section 7. The Japanese pronunciation assigning unit 10 assigns pronunciations to words that are not targeted by the misspelling candidate extracting unit 7. The output of the pronunciations generated by the kanji pronunciation adding section 9 and the Japanese pronunciation adding section 10 is synthesized into speech by the speech output section 11 of the speech synthesis device, and the resultant sound is emitted from the speaker 12.
A correction/repair unit 13 corrects the characters detected as errors based on the voice. The correction operation is performed on the correction terminal 14.
This is done by The input processing unit 2, the word dictionary 4, the grammar dictionary 5, the morphological analysis unit 6, the misspelling candidate extraction unit 7, the expected kanji reading assignment unit 9, the Japanese sentence assignment unit 10, the audio output unit 11, and the correction processing unit 13 are CPUs, It constitutes a processing device 15 consisting of a memory.

All kanji that have on-yomi or kun-yomi in advance 1
A kanji single-character guess-reading dictionary 8 is created in which words with information flags called on-kun flags that indicate whether the on-yomi or kun-yomi reading is the on-yomi or kun-yomi are registered as the on-yomi or kun-yomi for each character. The input Japanese sentence database 3, which is the reading result read by the device 1, is subjected to morphological analysis such as word candidate extraction and part-of-speech connection test.
If this word recognition extracts a one-character kanji independent word or a one-character proper noun in a kanji compound word,
A kanji character reading dictionary 8 is used for each kanji character, and the reading of the kun reading is indexed and assigned. Furthermore, the Japanese pronunciation assigning unit 10 synthesizes the output with pronunciations assigned to words other than the target words of the misspelling candidate extracting unit 7, the above-mentioned kanji guess output, and eo-speech, and the speaker 12 outputs the result as a voice. The system uses the unnatural reading of kanji in Japanese to detect typos. Thereafter, the proofreader uses the correction terminal 14 to correct the errors.

FIG. 2 shows an example of a case where verification is performed using the apparatus of the present invention having the basic configuration example shown in FIG. Here, the second
Fig.-1, Fig. 2-2, and Fig. 2-3 are examples in which a one-character Kanji character independent word or a one-character Kanji character proper noun are generated by typographical errors, misinsertions, and omissions, respectively.
In these examples, 16 is the original text string containing an error,
17 is the kanji string to be tested, 18 is the erroneous character such as a typo, insertion, or omission, 19 is the correct character, 20 is the part of speech in the word recognition result, 21 is the reading in the word recognition result, and 22 is the kanji that has been recognized as a word. It is a one-letter independent word or a one-letter kanji proper noun (kun-yomi).

In the example shown in Figure 2-1, ``Yu'' is mixed in as a misspelling of ``Yang'' and is recognized as an independent word with one kanji character, and with this process, ``Sat'' is changed to the reading of ``Yang'' with one kanji character. Kun-yomi (Tsuchi) and (Yuzuri) are indexed and added to the guess-yomi from the character guess-yomi dictionary, giving the reading an unnatural character.

In the example shown in Figure 2-2, when "jou" is inserted as a mistake and is recognized as a word as an independent word with one kanji character, the processing by this inventive device allows the pronunciation of "jou" to be added to the kanji dictionary. Kun-yomi (Nasake) is added as an index to the guess reading, giving the reading an unnatural feel.

In the example shown in Figure 2-3, ``Sei'' is omitted, and as a result, ``Bi'' is recognized as an independent word with one kanji character, and the processing by this inventive device allows one kanji character to be added to the reading of ``bi''. Kun-yomi (sonae) is indexed and added to the at-yomi from the at-yomi dictionary to give it an unnatural reading.

Because of this structure and action,
Compared to conventional technology, word recognition using morphological analysis enables the recognition of single kanji characters without giving the reading of the word at the time of recognition to the reading of a kanji 1-character independent word or kanji 1-character proper noun recognized within a kanji compound word. By adding guess readings of Kunyomi registered in the guess reading dictionary, unnatural pronunciations can be given, which makes it easier for proofreaders to check errors in input Japanese sentences. It was hot.

In addition, in the above, when a single-character kanji independent word is detected in a kanji compound word, a command to perform the target reading instead of the kun-yomi is issued, but at this time, the flag indicating whether the target-yomi is the kun-yomi or the on-yomi is set. Because there is
Kun-yomi can be added. When a single kanji character is recognized as a proper noun, on-yomi is generally used as the reading.

"Effects of the Invention" As explained above, the present invention provides the on-yomi or kun-yomi reading for every single kanji character that has the on-yomi or kun-yomi as the on-yomi or kun-yomi reading.
A kanji single-character guess dictionary is created in advance in which words with an information flag called an on-yomi flag indicating whether the guess reading is an on-yomi or a kun-yomi is created, and morphological analysis is performed using a data processing device, a word dictionary, and a grammar dictionary. When a 1-character independent word or a proper noun with 1 kanji character is extracted from a kanji compound word, use the above 1-character kanji reading dictionary to give the kanji character a ``kun-yomi'' reading. Effectiveness: sagiyokime (correct answer: work efficiency), and then the proofreader uses the unnaturalness of the reading of the kanji in the Japanese sentence to test the sound uttered by a speech synthesizer. ( 1) The proofreader uses the unnaturalness of the Japanese kanji reading uttered by the speech synthesizer to check for errors, and is able to work continuously for a relatively long period of time, and the proofreader's ability is (2) Error information is not displayed on the screen and the proofreader can perform verification work using only the original manuscript and audio, increasing work efficiency; (3) Normal grammatical It is possible to detect errors such as typos, insertions, and omissions in kanji compound words that are difficult to detect using methods that detect typos based on word connections.

There is an advantage.

[Brief explanation of the drawing]

FIG. 1 is a functional configuration block diagram showing an example of the present invention, and FIG. 2 shows a specific example of performing verification using the configuration shown in FIG. 1.
3 diagrams each have 1 kanji character independent word or 1 kanji character
FIG. 3 is a diagram illustrating an example in which character proper nouns are generated due to typos, misinsertions, or omissions.

Claims (1)

[Scope of Claims] 1. A device for verifying errors such as misread characters, misspellings, omissions, misinsertions, etc. caused by input errors of Japanese sentences in a Japanese document database input from an input device or character recognition errors during input. A morphological analysis means that performs word recognition through morphological analysis using a data processing device, a word dictionary and a grammar dictionary, and a compound word consisting of a plurality of kanji string words is extracted from the word recognition result of the morphological analysis. Kanji independent word or kanji 1 of 1 kanji character from kanji string compound word
A spelling error candidate extracting means for extracting character proper nouns, and an on-yomi or kun-yomi that instructs in advance the on-yomi or kun-yomi reading for every single kanji character that has an on-yomi or kun-yomi, and whether the on-yomi or kun-yomi is the on-yomi or kun-yomi. A one-letter kanji guessing dictionary that registers words with information flags called flags, and a one-letter kanji guessing dictionary are used to assign the one-letter kanji independent words or one-letter proper nouns extracted above. A means for assigning a kanji pronunciation, a means for assigning a pronunciation to a portion of the input Japanese sentence excluding the above-mentioned one-character independent word of kanji or a one-letter proper noun; a speech output section that generates each output of the Japanese reading adding means as speech by a speech synthesis device; a correction processing means for a corrector to correct errors while listening to the speech; and a correction processing means for correcting errors while listening to the speech; When a 1-letter kanji independent word or a 1-letter kanji proper noun is included, the kun reading for this 1-letter kanji independent word or 1-letter proper noun is changed to the above kanji 1.
A Japanese sentence error verification device has means for indexing a character guessing dictionary and determining the reading of a single character of that kanji, and verifies errors using the unnaturalness of the reading of kanji in Japanese sentences.