JPS62209659A - Correcting device for japanese sentence - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] <Industrial Application Field> The present invention relates to a Japanese text proofreading device that extracts and proofreads notations and expressions that do not meet specific standards from Japanese texts.

<Conventional technology> In Europe and the United States, word processors were developed early, and related technologies have advanced, and devices with spell checking and correcting functions have been put into practical use.

Compared to Western words, which are written with punctuation into each word, Japanese is usually written in kanji, kana, and kanji characters without punctuation, and the orthography is not thorough, so it is difficult to analyze. It is difficult to perform calibration, and no equipment for automating calibration has been put into practical use.

The conventional Japanese proofreading method is to have multiple people work in pairs to read the text together and extract problematic points, or have someone with specialized proofreading knowledge check the original text one by one. It's a method.

For the former, a proofreading device has been proposed that can analyze a text containing kanji and kana, convert it into speech, and read the text aloud with a machine.

However, the latter supports automation of proofreading or specialization of proofreading 8! Since the method requires a higher level of #i technique, only a conceptual proposal has been made, but no specific method for configuring the device or calibration has been reported.

<Problem to be solved by the invention, α> To perform proofreading, specialized knowledge of proofreading is required.
It took time to learn it. In addition, the number of people with specialized proofreading knowledge was small, which was a problem when proofreading a large amount of text or when proofreading needed to be done quickly.

Also, from the perspective of an expert with knowledge of proofreading, it is necessary to pay attention to the details in order to perform proofreading correctly.
He was under a mental burden.

Despite this, there were some errors that were difficult for humans to detect and errors that were easily overlooked, which caused problems in terms of calibration accuracy.

In addition, when ordinary people without specialized knowledge of proofreading create documents using a word processor, etc., reprocessing such as reading, printing, and copying the text file is required to proofread small mistakes. , There was a problem with α in the efficiency of document creation work.

The methods of implementing spell check and correct functions in Europe and the United States were aimed at proofreading notation and terminology, and there was no proofreading equipment that could proofread syntactical mistakes, homophone mistakes, etc.

The present invention analyzes character/symbol strings to be proofread (hereinafter, character/symbol strings may be referred to as character strings without special mention) from Japanese sentences by performing morphological analysis and syntactic analysis. By extracting and presenting the information, it is possible to assist the user of the device in the calibration work, thereby attempting to solve such problems.

Means for Solving the Problems〉 The present invention provides a means for inputting and aggregating Japanese words, a means for storing the inputted Japanese words, and a dictionary consisting of standard notations of Japanese words. vIt is obtained from a means for storing reference Japanese grammar, a means for morphologically analyzing the input Japanese, a means for parsing the Japanese, and a means for displaying the Japanese.

<Operation> By comparing and checking with Japanese-based notation and grammar input into a computer or the like using an input means, some parts are successfully compared and some parts are not. By parsing sentences that are successfully compared and matched using the morphological information obtained as a result of comparison and matching as input, parts with syntactic and semantic problems are extracted as character strings with a high probability of being proofread, and displayed. do. The user of this device uses the displayed character string to perform proofreading using the input editing means.

<Example> The present invention will be described in detail below based on the drawings. FIG. 1 is a block diagram of a Japanese grammar proofing device according to the present invention.

In the figure, 1 is a means for inputting and editing Japanese character strings. 2 is a means for storing the Japanese character string inputted by the input means. A keyboard is usually used as the input means, but even if input is not performed sequentially, for example, 70
It is also possible to use an external storage means for storing the input Japanese character strings, such as a disk or magnetic tape. That is, 1 has been omitted (there may also be 3).

3 is means for storing a dictionary consisting of Japanese character strings that match the criteria. 4 is a means of memorizing Japanese grammar. Because Japanese has no established orthography, writing, grammar, and expressions that meet certain standards are considered correct. Below, when we refer to correct notation, grammar, or expressions, we mean that they meet certain standards.

5 is a control means for controlling input/output, collation, information storage, display, etc. The control means includes means for storing results obtained by the control. 6 is a means for parsing Japanese 8n. Reference numeral 7 denotes display means such as a CRT for displaying input Japanese, intermediate results of verification, character strings to be proofread, and the like. 8 is a means for performing morphological analysis using a dictionary, grammar, etc.

FIG. 2 is an example of a character string input by the input means.

In this case, it shows that "wa" has been inputted extra. ``Ha'' exists both as a word ``ha'' that expresses ``ha'' within the limits of common kanji, such as in ``competing for supremacy,'' and ``ha'' as a modal particle.

FIG. 3 is an example of a dictionary stored in 3. 9 is a heading to be compared with the input character string. 10 is one of the morpheme information of the heading.

Figure f54 is an example of the grammar information stored in 4.

For ease of viewing, the headings shown in the m3 diagram are also shown. 11 is the heading or morpheme information connection &1 in Figure 3.
- Information that defines I4C4. For example, the pronoun ``he'' can be connected to case particles and modal particles. 1
2 is information specifying whether each heading can satisfy the clause end condition, O indicates that the clause end condition is met, and X indicates that it does not.

FIG. 5 is a correspondence table between morpheme information and grammatical category names.

13 is an example of a grammar category name. 14 is morpheme information corresponding to 13. A grammar category name in which 14 is blank indicates that it is not a terminator.

FIG. 6 is an example of syntax rules. 15 is a syntax production rule. 16 represents the constraint conditions when applying 15.

Figure 17 is a schematic 70-view of the present invention. 17 stores the input string on the stack inside 5. Step 18 temporarily divides the stored character symbol string using kanji codes or the like. 19 reads each tentatively divided input character string. 20 morphologically analyzes the input character string read in 19 using a dictionary and grammar. Depending on the result, the process branches depending on whether the morphological analysis is successful or unsuccessful. 21 stores morphological information that has been successfully analyzed.

22 uses the morphological information obtained in 21 as input and applies syntactic rules to generate a parse tree. 23 is 16 for a set of syntax rules with a parse tree that has grown to 5ENT.
to determine syntactic validity. 24 extracts syntactic constituent elements with low syntactic validity as character strings to be proofread, and displays them separately from others. 25 performs proofreading based on the information in 24 using selection means or Japanese input/editing means. 26 examines the next input string to be matched.

27 performs processing when either dictionary or grammar matching fails. Since the processing in No. 27 is outside the scope of this patent, it will not be described in detail.

The actual processing flow will be described below using the example sentence shown in FIG. 2 and the dictionary, grammar, and syntax rules shown in FIGS. 3, 4, 5, and 6.

First, the input character string "He goes to school." is stacked at 5. Next, the input string is temporarily divided using the code of the stacked string. For example, split before a character string that changes from hiragana to another character-like code.

As a result, "He is going to school." J is divided into "He is going to school,""He is going to school," and "He is going to school." First of all, “Is it him?”
We will explain the processing of morphemes.

``He is ka wa'' is read into the control device 5 and compared with the dictionary headings shown in FIGS. 3 and 4. In this case, "he" in the dictionary
Stack the information that matches the part in 5 and compare the connection condition of the corresponding heading with the next input character string. In this case, both are "ga" and match. Next, in Figure 3, “
The phrase end condition for the item with the heading ``ga'' is shown in 12 in Figure 4.
Check it out. 12, it can be seen that the clause end condition is satisfied in this case, and "he" is stacked at 5 as a normal clause.

Next, we process the morpheme of ``ha'' in the same way as for ``he'', and as a result, ``he'' and ``ha'' are combined into a -clause and cannot be made into ``he ka wa.'' “he” and “ha”
If it is a bifurcated clause, it is derived that the morphological condition is satisfied.

In the same way, morpheme processing for ``to school'' and ``go.'' is performed.

As a result of the above, the morpheme information "he... pronoun", "ga... case particle", "ha... noun", "school... noun", "-... case particle", "go... three-stage verb final shape] is obtained.

Next, by referring to the correspondence table between morphemes and grammatical category names (Fig. 5), we can find "pronoun...N", "case particle..."
Obtain CPJ, "noun...N", etc.

Next, according to Figure 6, the syntax rules [NP, NJ, [NPc-
By applying 4NP and CPJ in the order of sentence order,
Summarize the "he" part with NPCs. In this case, NPCs are grouped together unconditionally by referring to the constraint conditions when generating the syntax.

Next, move on to the ``ha'' and ``school'' parts that contain the problem. In the same way as above, ``ha school'' can be summarized as NP with [NP-N, NJ. In this case, when referring to constraints on syntax generation, "wa" corresponds to the constraints.

By repeating the above-mentioned steps, the Qingbun analysis book shown in FIG. 7 is obtained.

Regarding the growth of a syntax tree, parts that include constraints are more likely to contain syntactic problems than parts that do not.

From the result of syntactic analysis multiplied by M in 5, character strings with restrictions on syntactic growth can be extracted. By displaying this to distinguish it from others, it is possible to proofread the text. The constraint condition can be expressed by the morpheme itself or can be specified by the category name of the morpheme. Further, the criteria for determining the extraction of proofreading character strings from the constraint conditions may be determined by setting the constraint conditions as numerical values or level values, and comparing the constraint conditions with the standard values.

The processing of branches that do not appear in the above explanation will be briefly described. 2
6 is a module that selects a morpheme or a character string to be parsed next. If there is no input string, the series of processing ends. No. 27 is a process that includes a character string that is problematic in terms of morphological analysis, and proofreading can be performed using a method different from that of this patent.

<Effects of the Invention> The effects of the present invention are that, since specialized knowledge of proofreading is internalized, it is not necessary to acquire specialized knowledge, and even an amateur can perform proofreading with high accuracy. Since the main work of calibration can be performed by this device, the user can reduce the time spent on the calibration work.

In addition, it is possible to automatically extract character strings that need proofreading from Japanese texts, and use a proofreading dictionary to present character strings to be replaced and display them separately, making it easier for proofreaders. It is effective in reducing fatigue and mental burden.

Furthermore, it is also possible to extract errors that humans tend to overlook, such as the mistake between ``he'' in katakana and ``he'' in hiragana, which is also effective in improving the accuracy of proofreading. Furthermore, by combining it with a device such as a word processor, it is possible to omit a series of reworks such as printing and copying caused by proofreading, and it is expected that the efficiency of document creation work will be improved.

Furthermore, it is possible to extract syntactic and semantic errors, which can be expected to improve the efficiency of proofreading work.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the apparatus of the present invention, fjS2 is a diagram showing an example of a character string input by the input means, FIG. 3 is a diagram showing an example of a dictionary, and tjS4 is a diagram showing an example of grammatical information. 5 shows an example of a proofreading dictionary, FIG. 6 1 is an input means, 2 is an input character storage means, 3 is a dictionary storage means, 4 is a grammar storage means, 5 is a control means, 6 is a syntax Analysis means, 7 is display means, 8 is morphological analysis means agent Patent attorney Takeshi Sugiyama (and 1 other person) Figure 1 ptS2 Figure 5 Figure 7 He--N-1-NP-. Ga-1---CP--NPC--. Row < ----VP--PRED-j -J Figure 8

Claims (1)

[Claims] A means for inputting and editing Japanese, a means for storing the input Japanese, a means for storing a dictionary consisting of standard notation of Japanese, and a means for storing the grammar of the standard Japanese. , means for morphologically analyzing the input Japanese;
A Japanese grammar proofing device that has means for parsing Japanese words and means for displaying Japanese words extracts candidates for character/symbol strings to be proofread by using the results of morphological analysis as input for parsing, A Japanese text proofreading device characterized by proofreading.