JPH0390958A - Sentence check method - Google Patents

Abstract

PURPOSE:To obtain a sentence check method which has a high processing speed and high detecting accuracy by performing a check process to a single expression item of a check subject with a total sentence defined as a check subject range and limiting the subsequent check subject ranges based on the preceding detection areas. CONSTITUTION:An entire sentence to be checked is read by a sentence input means 1 and stored in a sentence storage means 2, and (n) pieces of check means 3(1) - 3(n) are provided for each expression item of the check subject. The means 3(1) which is applied first defines the entire sentence stored in the means 2 as check subject range. Meanwhile the check subject ranges are decided based on the detection result of a check means (i-1) for the check means 3(i) which are subsequently applied (2<=i<=n). A range deciding means 4 decides these check subject ranges, and the check result of the final check means 3(n) is outputted via a result output means 5. Thus it is possible to obtain a sentence check method having high detecting accuracy with no deterioration of the processing speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a text inspection method for detecting inappropriate expressions, expressions requiring attention, etc. from a given text.

C. Prior Art] There is a need for a method for detecting inappropriate expressions, expressions that require caution, etc. from sentences. Expression items to be detected include, for example, the conjunctive particle "ga", which tends to make the connection relation (direct/inverse) ambiguous, and the uke wa body expression, which tends to make the subject ambiguous (uke is the body auxiliary verb). Examples include ``reru'' and ``reru''.

Conventionally, there are the following two methods for this type of text inspection method.

■ Text inspection method using font processing ■ Text inspection method using text analysis processing The first method, the text inspection method using font processing, searches for character strings that meet certain conditions. ``Construction of an algorithm for extracting elaboration information in a tool'' (Suganuma and Hanbo, Proceedings of the 4th Annual Conference of the Japanese Society of Software Science, 1987) and ``A method for extracting passive forms in a Japanese text editing support tool'' (Hanazu and Hyakude, p.
・Takagi, Transactions of the Information Processing Society of Japan, Vol. 28, No. 8, 198
7) and other documents.

For example, when detecting the conjunctive particle ``ga'', search for ``ga'' where the immediately preceding character is one of the characters ``i'', ``da'', ``ta'', and ``n''.

When detecting the passive auxiliary verbs ``re'' and ``rare'', the previous character string is ``ka'', ``sa'', ``ta'', ``ma'', ``ra''.
Search for ``re'' that is not included in ``we'' with ``wa'', ``ga'', ``ba'', ``death'', or ``shina''.

The second method, a text inspection method using text analysis processing, is
After identifying the grammatical units and structures such as sentences, clauses, and words, as well as the part of speech of each word, for all given sentences through sentence analysis using a word dictionary and grammar dictionary, the detection target is identified. It searches for words (or word strings) that correspond to expression items.

This second method is based on the “Japanese document proofreading support system CR”.
ITACJ (Suzuki, Takeda, Fujisaki, Information Processing Society Research Report, 86-JDP-8-5, 1986) and “Japanese writing support system COMET1 (Fukushima, Otake, Inuyama,
Shuto, Institute of Electronics and Communication Engineers Technical Research Report, ○586-21.1
986) and others.

[Problems to be Solved by the Invention] The first conventional method (text inspection method using font processing) described above is a simple processing method and is therefore fast, but it has the problem of low detection accuracy ( For the purpose of inspection, the design is designed to avoid under-detecting the relevant areas, so there is usually an over-detection of more than the relevant areas). For example, in the above-described method for detecting the conjunctive particle "ga" using font processing, "ga" such as "therefore" or "hitotsu ga" is also detected as a conjunctive particle.

On the other hand, the second method (text inspection method using text analysis processing)
This method has high detection accuracy because the word recognition is performed including before and after the relevant part, but compared to the first method, it compares the entire sentence with a word dictionary that generally has tens of thousands of headings. The problem is that it takes much longer to process.

That is, in the conventional method described above, processing speed and detection accuracy are contradictory.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a text inspection method with high processing speed and high detection accuracy.

[Means to solve the problem]

The present invention is a text inspection method for detecting inappropriate expressions, expressions that require attention, etc. from a given text. The inspection target range of subsequent inspection processes is limited based on the detected locations of the previously applied inspection process.

〔Example〕

Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a flowchart showing the inspection procedure for one expression item in the present invention. Furthermore, FIG. 2 is an example of a block diagram of an inspection device for one expression item that implements the interrogation procedure. The present invention will be explained below with reference to FIGS. 1 and 2.

First, it is assumed that the entire text to be inspected has been read by the text input means 1 and stored in the text storage means 2.

Here, as shown in FIG. 2, there are n inspection means 3 (3(1)) for one expression item to be inspected.

3(2), . . . 3(n)) are provided.

In FIG. 1, 1 is set to ↓ in step 101, and 1 is incremented by 1 in step ↓05.
The processes 2 to 10.4 are repeated, that is, the n inspection means shown in FIG.
), inspection means 3 (2), ... inspection means 3 (n)
will be applied sequentially. In addition, in this case, the inspection means 3 (1) that is applied first sets the entire sentence stored in the sentence storage means 2 as the inspection target range, as shown in step 10.0 of FIG. However, for the inspection means 3(i) (not shown, 2≦1≦n) applied thereafter, as shown in step 104 in FIG.
-1) Based on the detection results, the inspection target range is determined. Range determining means 4 in FIG. 2 determines the range to be inspected.

Once the n test means have been applied sequentially, the last test means 3 (
Output the detection result of n). In FIG. 2, the result output means 5 outputs the detection results.

The above describes the test for one expression item, for example, the detection of the conjunctive particle ``gako,'' but the test can also be performed for other expression items, such as detecting passive auxiliary verbs. The procedure for testing multiple expression items in this way is to repeat the entire procedure shown in the flowchart in Figure 1.In contrast, the procedure for testing multiple expression items (m items) can be repeated. An example of a block diagram of an inspection device is shown in Fig. 3. In Fig. 3, a plurality of inspection means (
1-1) ~ (1-n), (2-1) ~ (2-n),
..., (rrl-1) to (m-nm) are provided.

Although the procedures and configuration of the present invention have been generally described above,
Next, an example specifically showing the contents of the inspection means will be described.

FIG. 4 is a block diagram showing the configuration of an embodiment of a device for testing the conjunction particle "ga" using the sentence testing method of the present invention. The block diagram of FIG. 4 corresponds to the case where the number n of inspection means in the block diagram of FIG. 2 is two.

Each component of the block diagram in FIG. 4 will be explained.

The text input means 1 is a means for inputting Japanese text mixed with kanji and kana, and can be realized by a keyboard, a kana-kanji conversion device, a character recognition device, or a device for reading a document file created by a word processor.

The text storage means is a means for storing the Japanese text input by the text manual means 1, and can be realized by an IC memory, a magnetic disk device, or the like.

The character string checking means 310 corresponds to the checking means 3(1). The character string search means 310 is a means for searching the Japanese texts stored in the text storage means 2 for a character string that satisfies a certain specific condition. Here, the purpose is to detect the conjunctive particle "ga", so the character strings that meet a specific condition are the characters in the first row of the previous character, "i", "da", "ta", etc. It is ``ga'' which is either ``n''. This character string search means 3 (0) is equivalent to the first conventional text inspection method described in the prior art, and can be easily realized by a person skilled in the art. The range of is notified from the range determining means 4, and the position of the character string detected by the character string searching means 310 (the position of the character string satisfying a specific condition) is sent to the range determining means 4.

As mentioned above, the range determining means 4 sets the inspection target range of the inspection means (1) to be the entire text, and the inspection means (i) (2≦1≦
n) The inspection target range is determined based on the detection location of the inspection means (i-1).

In the embodiment of FIG. 4, since the number n of inspection means is 2,
The range determination means 4 first notifies the character string search means 310 of the inspection means 3 (1) to set the range to be inspected as the entire text of the Japanese text stored in the text storage means 2.

Next, upon receiving the detection position information from the character string search means 310, the range determination means 4 notifies the text analysis means 323 of the inspection means (2) of the inspection target range. The method of determining the inspection target range of this text analysis means 323 is, for example,
For each position of the character string detected by the character string search means 310,
There is a way to make it a range of sentences (or a range separated by punctuation marks) that includes that position.

The inspection means 3 (2) includes a word dictionary 3209 and a grammar dictionary 32.
1. Text analysis means 322. Analysis result storage means 323 and word string search means 324 correspond to this.

In the word dictionary 320, the notation, part of speech, etc. of words are registered. The grammar dictionary 321 describes grammatical conditions regarding word arrangement, sentence structure, etc.

The sentence analysis means 322 includes a word dictionary 320 and a grammar dictionary 32.
With reference to 1, for the given sentence, write sentences, clauses,
It is a means of identifying the grammatical units and structures of words, as well as the part of speech of each word.

The analysis result storage means 323 is a means for storing the analysis results of the text analysis means 322 (IC memory).

(Realized using magnetic disks, etc.)

The word string search means 324 targets the word string as the sentence analysis result stored in the analysis result storage means 324.
Search for words or word strings that meet certain conditions. Here, the word string search means 324 uses the conjunctive particle "ga"
Search for (words whose notation is "ga" and whose part of speech is a conjunctive particle). This word string search means 324 is equivalent to the conventional second text inspection method described above, and can be easily implemented by those skilled in the art. Note that the position of the word (or word string) detected by the word string search means 324 is sent to the result output means 5.

The result output means 5 is a means for outputting the detection result of the last detection means, here the detection result of the word string search means 324, as described above, and can be realized by a display, a printer, or the like. The output format is to display the text stored in the text storage means 2 while enclosing the detected part in parentheses, inverting or
There are methods such as a method of highlighting by blinking or hanging, and a method of displaying the detected location along with the surrounding context in KWIC.

In the method of detecting the conjunctive particle "ga" performed by the character string search means 310, places other than the conjunctive particle "ga" are also detected excessively, such as "ga" in "therefore" and "ga" in "hitotsu ga". However, the conjunctive particle "ga" is not missed.

Therefore, the range to be analyzed by the text analysis unit 323 determined by the range determination unit 4 includes all occurrence positions of the conjunctive particle "ga", and the amount of text is usually smaller than the entire text. That is, the sentence analysis means 322 narrows down the analysis to only parts where the conjunctive particle "ga" is likely to appear.

In addition, the above embodiment is an inspection device that uses the conjunctive particle "ga" as an expression item to be detected, but Uke is an inspection device that detects body expressions, etc., and other expression items are detected. This inspection device can also be realized with the same configuration as shown in FIG.

〔Effect of the invention〕

As explained above, according to the present invention, when there is an inspection means A that has a high processing speed but low detection accuracy and a tendency to over-detect, and an inspection means B that has a high detection accuracy but a slow processing speed, the inspection means A Based on the inspection results of the entire text by
A text inspection method that limits the inspection range of inspection means B can be realized. In that case, if inspection means A does not miss any detections, the final inspection result and detection accuracy will be the same as when inspection means B is applied to the entire text. Moreover, the processing speed is faster than when the entire text is processed by inspection means A.
There is only a slight delay, and it is considerably faster than when the entire text is processed by inspection means B.

That is, compared to the conventional method, a text inspection method with high detection accuracy can be obtained without significantly reducing the processing speed.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing an inspection procedure for one expression item according to the present invention, FIG. 2 is a block diagram showing an example of the configuration of an inspection device that implements the inspection procedure of FIG. FIG. 4 is a block diagram showing an example of the configuration of a testing device for testing expression items. FIG. 4 is a block diagram showing the construction of an embodiment of a testing device for the conjunction particle "ga" using the present invention. 1...Text input means, 2...Text storage means, 3...Inspection means, 4...Range determining means, 5...... Result output means, 310...
... Character string search means, 320 ... Word dictionary, 32
1... Grammar dictionary, 322... - Sentence analysis means, 323... Analysis result storage means, 324...
...Word string search means.

Claims (1)

[Claims] In a text inspection method that detects inappropriate expressions, expressions that require attention, etc. from a given text, an inspection process is performed for one expression item to be detected, with the entire text of the given text as the inspection target range, and subsequent A text inspection method characterized in that an inspection target range of an inspection process is limited based on a detection location of a previously applied inspection process.