JPH0432958A - Japanese sentence error word detecting device - Google Patents

Abstract

PURPOSE:To point out only a position which is possibly wrong and to reduce the load on a user by detecting partial character strings, which are used only once among partial character strings counted by a partial character string counting means, as a wrong word. CONSTITUTION:A document is divided by a document dividing means 1 at change points between two kinds of characters, i.e. a change point from KATAKANA(square form of Japanese syllabary) to other characters and a change point from other characters to KATAKANA. Then the partial character string counting means 2 counts how many times the same character string is used in the document as to only partial character strings consisting of only KATAKANA. Then it is judged whether or not the counted character strings in KATAKANA are used only once and the partial character strings in KATAKANA which are used only once in the document are recognized as wrong words. When the partial character strings in KATAKANA which are counted are used only once in the document, an error detecting means 3 displays the detected wrong words to the user. Consequently, only positions which are possibly wrong are pointed out and the load on the user is reduced.

Description

[Detailed Description of the Invention] The present invention relates to a Japanese sentence error word detection device that detects incorrectly written words contained in Japanese sentences electronically expressed using character codes. This invention relates to a Japanese sentence error word detection device. In particular, the present invention relates to a Japanese sentence error word detection device that detects words that may be written incorrectly due to typos or kanji conversion errors, particularly for sentences entered using a keyboard such as a Japanese word processor. For example, it can be applied as a proofreading support function in Japanese word processors, etc.

A publicly known document describing an example of the prior art related to the present invention, that is, an error detection method for Japanese sentences, is "Japanese text creation support system COMETJ (3 people outside Fukushima, IEICE technical report 0586) -21.1986).

The error detection method described in the above-mentioned known document will be explained below. According to this error detection method, there is a method in which a portion where morphological analysis fails is recognized as an error. Morphological analysis is a widely known technique that involves dividing a sentence into words and identifying the part of speech. We use a word dictionary that records the notation of hundreds of thousands of words from Machhogo and their parts of speech, and a connection table that records information on whether each part of speech can be grammatically connected to each other.

Using this connection table, we break down a given sentence into a string of words that can be grammatically connected to each other. At this time, the part of speech of the word can also be recognized. Where morphological analysis fails in a sentence, either a word that is not registered in the word dictionary is used, or there is some kind of error. If a sufficient number of words are registered in the word dictionary, a portion where morphological analysis fails may be recognized as an error.

Next, there is a method in which humans can detect errors using KWIC display. What is KWIC?KeyWord In Cont
This is an abbreviation for ext, which displays the words (character strings) used in a sentence along with the character strings before and after them. Character strings are displayed in order of character code, etc., making it easier to find unreadable character strings that contain errors.

However, the above-mentioned method of recognizing a portion where morphological analysis has failed has the following drawbacks. That is, the first drawback is that the error detection ability is insufficient. Even if a sentence contains an error, it is often impossible to detect the error. In particular, it is often impossible to detect places where part of a kanji compound word is an incorrect kanji. For example, even if you mistakenly enter ``Kenmae'' instead of ``Before the lecture,'' the morphological analysis can divide it into ``Kenmae'' (noun) and ``Kenmae'' (suffix), so it will be treated as an error. It cannot be detected.

The second drawback is that errors are detected excessively.

In general, word dictionaries contain more words than Machhogo, but there are many words that are actually used that are not registered.

In particular, words written in katakana are often used to express technical terms or new concepts, so many words are not registered in word dictionaries.

Therefore, this method incorrectly detects many katakana words as errors.

Furthermore, according to the above-mentioned method for humans to find errors using the KsuIC display, in order to find errors from the 1nIc display,
A human must examine all of the enormous amount of display, which places an excessive burden on the user. Therefore, even if such a function is available, it is expected that it will hardly be used. In addition, although in principle it is possible to find all errors, in reality, as long as humans investigate, there will always be oversights. The disadvantage is that the greater the amount of research that needs to be done, the more likely things will be overlooked.

■--Defeat The present invention was made in view of the above-mentioned circumstances,
We reduced the burden on humans by detecting only the parts where there is a possibility of an error, we also reduced the number of incorrect error detections, and we also made it possible to detect parts that were syntactically correct but were actually incorrect. This was done for the purpose of providing a Japanese sentence error word detection device.

In order to achieve the above-mentioned objects, the present invention provides (1) a Japanese sentence error word detection device that detects error words in Japanese sentences electronically expressed using character codes; A sentence dividing means for dividing into sub-character strings using a species change point as a boundary, and a sub-character string counting means for counting the number of times a katakana sub-character string appears.

and an error detection means for detecting a partial character string in which the number of occurrences of the partial character string counted by the partial character string counting means is one, as an error word.

(2) In a Japanese sentence error word detection device that detects error words in Japanese sentences electronically expressed using character codes,
A sentence dividing means that morphologically analyzes a sentence and divides it into word units; a partial character string counting means that counts the number of occurrences of word strings corresponding to nouns; and occurrences of word strings counted by the department character string counting means. and an error detection means for detecting a partial character string that occurs once as an error word; and (3)
The method is characterized by having an error detection means for detecting a partial character string in which the number of occurrences of the word string counted by the partial character string counting means is one and is not registered in the word dictionary as an error word. be. Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a block diagram for explaining an embodiment of the Japanese sentence error word detection device according to the present invention. In the figure, 1 is a sentence segmentation means, 2 is a substring counting means, and 3 is an error detection means. be.

The sentence dividing means 1 divides a given Japanese sentence into partial character strings corresponding to words. It is assumed that the text is expressed in character codes and in a form that can be processed electronically. The partial character string counting means 2 refers to each character string divided by the sentence dividing means as a partial character string, but here, only partial character strings that meet a certain condition are targeted, and the same partial character string is not included in the sentence. Count how many times it appears. The error detection means 3 basically detects, as an error word, a partial character string that is used only once in the sentence among the partial character strings counted by the partial character string counting means. This is based on the observation that common words are often used multiple times in a sentence, whereas words containing the same error occur only once in a sentence.

FIG. 2 is a flowchart for explaining the operation of the Japanese sentence error word detection device according to the present invention.

Below, each step will be explained in order.

``Shi'' So: Divide a given Japanese sentence at the points where the character type changes. In other words, the text is divided at two types of change points: from katakana to other characters, and from other characters to katakana. There are two types of partial character strings after division: character strings consisting only of katakana characters and character strings consisting only of characters other than katakana characters.

4: Next, count the katakana substrings.

That is, the number of times the same character string is used in a sentence is counted only for partial character strings consisting only of katakana characters.

Shooting: Next, it is determined whether the counted katakana substring is used only once in the sentence. A partial character string of katakana that is used only once in a sentence is recognized as an incorrect word.

If the katakana subcharacter string counted quadratically by μA is used only once in the sentence, the detected error word is presented to the user and corrected if necessary.

Words written in katakana often represent technical terms or new concepts, and are often important words in a text. It is rare for katakana words to appear only once in a sentence, and it is normal for them to be used multiple times.

However, erroneous words such as 'computer', which occur due to typographical errors, rarely appear more than once in a single sentence. Therefore, most of the erroneous words in Katakana can be detected using the above method. The above description corresponds to the description of claim 1.

FIG. 3 is another flowchart for explaining the operation of the Japanese sentence error word detection device according to the present invention. Below, each step will be explained in order.

Bμ So: Divide sentences into word units and identify parts of speech using morphological analysis. In this case, the morphological analysis first identifies clause boundaries using only punctuation marks and character type information, and then determines the part of speech and conjugation of words within the clause. This morphological analysis is known, for example, as described in "Memory of Japanese Dictionaries and Automatic Segmentation of Japanese Sentences" (by 3 authors outside of Nagamaru, Information Processing VO1, 19N (L6, 1978)). It is.

4: Next, create a partial string. That is, consecutive prefixes, nouns, and suffixes are combined into one partial character string. In other words, one substring is the longest continuous word string that begins with a prefix or noun, followed by a prefix, noun, or suffix, and ends with a noun or suffix. However, a suffix shall not immediately follow a prefix. Also, words with parts of speech other than prefixes, nouns, and suffixes are not treated as partial character strings.

For example, the sentence ``Finally a new world record,'' is morphologically analyzed as ``6 finally /world/new/record/,'' and the substring ``world record'' is extracted.

The substrings are counted in the second step.

7: Next, it is determined whether the counted partial character string is used only once in the sentence.

That is, a partial character string that is used only once in a sentence is extracted as an error word candidate.

If the partial character string counted in January is used only once in the sentence, it is then determined whether it is registered in the word dictionary.

January rule: If the word is not registered in the word dictionary, it is displayed to the user as an error. That is, an error word candidate is recognized as an error only when it is not registered in the word dictionary used in morphological analysis.

Most nouns that are not compound words are registered in the word dictionary. Therefore, even if the word registered here appears only once in a sentence, it is unlikely to be an error. An unregistered word is either a compound word or an error word.

Similar to katakana, compound words often represent technical terms or new concepts, and are often important words in a sentence. It is rare for a compound word to appear only once in a sentence; it is common for it to appear multiple times.

However, erroneous words that occur due to typos or kanji conversion appear multiple times in a single sentence, such as when you accidentally enter ``world record hand'' instead of ``world record''. Very rarely. Therefore, most errors in compound words can be detected using the above method. The above explanation corresponds to the explanation of claims 2 and 3.

As is clear from the above description, the present invention has the following effects.

(1) Effects corresponding to claim 1: Since only possible mistakes are pointed out, there is less burden on the user, and there is no need to register correct spellings, so there are no unnecessary (mistakes) for katakana ) Fewer error detections. moreover,
Since it does not use grammatical information, it can detect parts that are syntactically correct but are actually incorrect.

(2) Effects corresponding to claim 2: Since only possible errors are pointed out, there is less burden on the user, and since grammatical information is not directly used for error detection, errors may be syntactically correct but actually incorrect. It is also possible to detect locations where Furthermore, since character type information is not used, errors can be detected regardless of character type.

(3) Effect corresponding to title requirement 3: Since it is checked whether the notation is correct, there are fewer unnecessary (erroneous) error detections.

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining an embodiment of the Japanese sentence error word detection device according to the present invention, and FIG. 2 is a flowchart for explaining the operation of the Japanese sentence error word detection device according to the present invention. FIG. 3 is another flowchart for explaining the operation of the Japanese sentence error word detection device. 1... Sentence division means, 2... Partial character string counting means,
3...Error detection means.

Claims (1)

[Scope of Claims] 1. In a Japanese sentence error word detection device that detects error words in Japanese sentences electronically expressed using character codes, a sentence is divided into partial character strings using character type change points as boundaries. a partial character string counting means for counting the number of occurrences of a partial character string in katakana; and an error detection means for detecting an error word in a Japanese sentence. 2. In a Japanese sentence error word detection device that detects incorrect words in Japanese sentences electronically expressed using character codes, there is a sentence dividing means that performs morphological analysis of the sentence and divides it into word units, and words that correspond to nouns. A partial character string counting means for counting the number of occurrences of a string, and an error detection means for detecting a partial character string whose number of occurrences of a word string counted by the partial character string counting means is 1 as an error word. Characteristic Japanese sentence error word detection device. 3. It is characterized by having an error detection means for detecting, as an error word, a partial character string in which the number of occurrences of the word string counted by the partial character string counting means is one and is not registered in the word dictionary. The Japanese sentence error word detection device according to claim 2.