JPH053617B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a kana-kanji conversion device that converts a kana input character string into a kana-mixed character string. The present invention relates to a kana-kanji conversion device that can be used to convert kana to kanji.

[Background of the invention] Conventionally, as a method for converting kana to kanji from solid input, there is a method described in "Hitachi Hyoron" May 1980, Vol. 63, No. 5, "HITAC L-320/30H, 50H Document Processing Function"
(hereinafter referred to as Document (1)), a method that combines longest match and back track, ``Showa 53
Proceedings of the 19th National Conference of the Information Processing Society of Japan” 5E−
4. The two-clause longest match method described in ``Kana-Kanji Conversion System for Solid Text'' (hereinafter referred to as Document (2)), ``Information Processing Society of Japan 1981 Computational Linguistics Study Group Materials'' 25-6 ``Table On the phrase structure analysis algorithm using the method and its efficiency'' (Reference (3) below), the method of minimizing the number of phrases is well known.

The method in Reference (1) extracts an independent word from the left end of the input using the longest match, and then repeats the clause conversion process of assigning adjuncts that can be grammatically connected to the independent word until it reaches the right end. If it does not reach that point, it backtracks, tries another phrase conversion, and moves on.

The method in document (2) extracts all possible conversion candidates for two clauses from the left end of the input in a brute force manner, selects the one with the longest sum of the lengths of the two clauses, and converts it to the first clause. This is a method of converting while determining the break for each clause by repeating the same process using that point as the starting point.

The method in document (3) extracts clauses from the left end in a brute-force manner, and selects the combination with the minimum number of clauses from among them as a conversion result.

Above, we have described an example of a method for converting kana to kanji using solid writing input, but in general, with solid writing input, there is room to choose where to place the breaks between bunsetsu, so There are many conversion candidates compared to the case of . For example, the input ``Suugakukai Sekijiyo de'' has many interpretations such as ``Mathematical analysis'', ``Mathematics department transfer'', ``Mathematics banquet'', ``Mathematics or ruins'', etc. conversion) is possible.

One character string from among these conversion candidates is output as a candidate, but as described above, there are many conversion candidates, so it may differ from the operator's request. In such a case, an operation is required to change (correct) the character string to the character string requested by the operator, but conventional operation methods can be divided into the following two methods. 1
The first method is to reconvert the entire input character string by pressing a key that instructs conversion again, and output a candidate that is different from the candidates output previously. (Hereinafter referred to as amendment (1)). The other method is to move the cursor to a conversion string that is different from the string requested by the operator, and perform the necessary operations to change it to the string requested by the operator (hereinafter referred to as correction (2)).
There is a method.

Regarding modification (1), candidates can be changed with a single key operation, but it is necessary to press the key many times until the candidate is the one desired by the operator, and it is not possible to change the candidate to the one desired by the operator. The disadvantage is that the operator does not know whether the

Regarding correction (2), it is necessary to change the correction method depending on the error in the conversion. In other words, if we take the example of "Sugaku Kai Sekiji Yoha" above, "Mathematics Analysis Jojo" divides the input character string into two clauses, "Suugaku" and "Kaiseki Jiyoha", and then divides the input character string into two clauses, "Sugaku Kai Sekiji Yoha", respectively. " is the result of kana-kanji conversion into "On analysis", and "On transfer to the Department of Mathematics", "Suugakuka" and "Isekijiyouha".
This is the result of dividing it into two clauses and converting them into kana-kanji into ``Mathematics Department'' and ``On Transfer.''
Furthermore, ``In mathematics or on the ruins'' is ``Suugaku?''
This is the result of dividing it into two clauses, ``Isekijyouha'' and ``Isekijiyouha,'' and converting them into kana and kanji into ``Mathematics?'' and ``On the register,'' respectively.

Therefore, if you want to change "for mathematical analysis" to "for transfer to the Department of Mathematics,"
and ``Kaisekijiyoha'' need to be re-separated into ``Suugakuka'' and ``Isekijiyoha''. On the other hand, if you try to change "On the transfer to the Department of Mathematics" to "On the mathematics or ruins," you will need to re-separate them because the delimiters are the same for both "Suugakuka" and "Isekijiyouha." There isn't, but
It is necessary to reselect homophones. Therefore, the operation method differs depending on whether or not to change this delimiter. Modification (2) allows the operator to reach the desired candidate faster than modification (1) because the operator performs the modification at the location he or she wants to modify.
Since there is no single method of operation, the burden of correction work on the operator is heavy. Both corrections (1) and (2) took too much time and forced the operator to perform complicated operations.

[Purpose of the invention]

The purpose of the present invention is to divide an input string into multiple clauses and convert each of them into a string containing kanji and kana, and if the converted string is not the candidate intended by the operator, the input string can be separated. To provide a kana-kanji conversion device capable of quickly correcting by the same type of operation whether the character string is divided in the same way or different from the character string intended by an operator.

[Summary of the invention]

The present invention is characterized by a kana-kanji conversion unit that divides an input character string into different divisional conversion units and converts each divisional conversion unit into a kanji-kana-mixed character string; The likelihood is calculated from among the network storage section that stores the input character strings in the form of a network arranged in series and parallel, and the kanji-kana-mixed character strings for each division conversion unit stored in this network storage section. An optimal candidate generation processing unit that extracts the candidate string with the highest value and displays it on the display unit, and a correction instruction input when the character string containing kanji and kana displayed for the input character string is not the candidate intended by the operator. At the location of the division conversion unit that is not the intended candidate, another candidate with the same division conversion unit like a homonym and a candidate with a different division conversion unit are retrieved from the network storage unit and displayed together on the display unit. A candidate modification processing section is provided to display the candidate, and a processing selection determination processing section is configured to select and perform either one of the optimal candidate creation processing section and the candidate modification processing section, and the optimal candidate creation processing section is further provided with a candidate modification processing section. The operator The object of the present invention is to enable the operator to change to the candidate intended by the operator in a single operation at the location that the operator wants to modify, without having to be aware of whether or not to change the division/conversion unit.

[Embodiments of the invention]

An embodiment of the present invention will be described below.

FIG. 2 is a block diagram showing the structure of the kana-kanji conversion apparatus of the present invention, which is composed of an input section 1, a display section 4, a control section 2, and a storage section 3. The input unit 1 is, for example, a keyboard for inputting kana input, conversion, and candidate change instructions, and the display unit 4 is, for example, a CRT display for displaying conversion results and change candidates. The control unit 2 controls execution of kana-kanji conversion and is configured using a CPU. The storage unit 3 is a memory that temporarily holds input kana character strings and data used in conversion.

FIG. 1 is a processing function block diagram of kana-kanji conversion using this kana-kanji conversion device. Input processing 10
inputs an input kana character string from the input section 1 and instructions for conversion. In the conversion process 20, the control unit 2 creates a conversion candidate with a high likelihood of conversion for this kana input, and performs a conversion process to store it in the storage unit 3. next,
In the process selection determination process 30, the control unit 2 selects the optimal candidate creation process 40 or the candidate modification process 50. After the conversion process 20, the optimal candidate creation process 4
0 is selected and a display process 60 is performed in which a candidate string with the highest likelihood is created from among the candidates in the storage section 3 and displayed on the display section 4. If the character string containing kanji and kana displayed on the display section 4 is different from the operator's request, when the operator instructs the corresponding location to be changed as a candidate from the input section 1, the process returns from the candidate change instruction process 70 to the process selection determination process 30, and the candidate is selected in the process. The control is transferred to the modification process 50, and the candidate modification process 50 creates an alternative candidate to be selected corresponding to the candidate at the location instructed by the candidate change instruction process 70, and the display process 60 displays it on the display unit 4. do. When the operator selects and instructs the desired candidate from among the selected candidates, control transfers from the candidate change instruction process 70 to the optimal candidate creation process 40 via the process selection determination process 30, and connects to the candidate selected by the candidate change instruction. An optimal candidate is created, and a display process 60 displays the candidate selected by the candidate change instruction and the results of changes to its subsequent parts. The above display processing 60 and candidate change instruction processing 7
0, process selection determination process 30, candidate correction process 50,
The display process 60, candidate change instruction process 70, process selection determination process 30, and optimal candidate creation process 40 are repeated until the conversion candidate intended by the operator is obtained.

The operations of the conversion process 20, process selection determination process 30, optimal candidate creation process 40, candidate modification process 50, and candidate change instruction process 70 described above will be described in more detail below.

FIG. 3 is a flowchart showing the operation of the conversion process 20 of FIG. 1. FIG. 4 is a diagram showing an example of a data network created by this conversion process 20 and stored on the storage unit 3. This conversion process 20 will be specifically explained using the example shown in FIG.

Now, assume that the input character string is "Suugakukai Sekijiyouha". In FIGS. 3 and 4, n is a pointer indicating the position of a character counted from the beginning of the input character string. At step 201 in FIG. 3, n is first set to 1 and positioned at the beginning of the input character string. In step 202, it is checked whether the position n is at the end of a clause (referred to as a clause end).
The beginning of a string is specially treated as the end of a clause. step
Moving on to 203, phrase conversion is performed. In phrase conversion,
Conversion for one bunsetsu is performed with position n at the beginning of the bunsetsu, and possible conversion candidates are extracted. Here, the phrase is
An independent word followed by an optional prefix, an independent word followed by an optional suffix, and an optional adjunct. Compound words are considered different clauses. Furthermore, the division conversion unit is each unit in which a clause is divided into a prefix, an independent word, a suffix, and an attached word. In the example in Figure 4, the independent word “mathematics”
and ``mathematics society'', ``mathematics world'' connected with suffixes,
``Mathematics floor'', ``mathematics department'', ``mathematization'', and ``mathematics ka'' connected with adjuncts, as well as the independent word ``kazu'', ``suck'' and ``kazu ga'', ``suck'' connected with adjuncts ga'' is a clause extracted from the beginning as a conversion candidate, and the independent words ``mathematics'', ``number'', ``suck'', and the suffixes ``kai'', ``kai'', ``ki'', ``ka'', `` The adjuncts ``ka'' and ``ga'' are respectively division conversion units. In this case, NHK
The method shown in ``Kana-Kanji Conversion by Computer'', Vol. 25, No. 5, May 1970, Technical Research is well known. Next, in step 204, the likelihood of the conversion candidate is determined, and it is determined whether it should be retained in the storage unit 3 or discarded (referred to as pruning). The likelihood is determined so that the likelihood of dividing the input character string into the minimum number of clauses is high. However, in the same clause, adnominals such as ``kono'' and ``so'' and formal nouns such as ``koto'' and ``mono'' are often used attached to other clauses, so they can be used as nouns or verbs. When counting the number of clauses, instead of treating them as a single independent clause, consider that the likelihood is higher than that of nouns or verbs. For example, nouns, verbs,
Adjectives and adjective verbs are given a weight of 1 clause, formal nouns, auxiliary verbs, adjunctive words, etc. are given a clause count of 0.1, and prefixes and suffixes are treated as semi-independent words and given a clause count of 0.5.
It is assumed that the smaller the number, the higher the likelihood.

Pruning calculates the likelihood from the beginning of the character string to the end of the clause that is currently being judged, and sets that likelihood as the likelihood at the last character position of the clause that is currently being judged. , If the likelihood has already been determined at the same clause end character position, it is compared with that value, and if the difference in value exceeds a predetermined tolerance value, pruning is performed. In this embodiment, this tolerance value is set to the minimum value of the number of clauses at the same character position +1. However, it goes without saying that this is not limited by the method of determining the likelihood and the size of the permissible value for pruning. This example explains how by extracting all clause candidates whose likelihoods are within a certain range and storing them in the storage unit 3, subsequent selection and modification can be done quickly and easily. .

As shown in Figure 4, all phrases extracted from the beginning are stored in the storage unit 3 (step
205). When storing data, it is possible to reduce the storage capacity occupied by storing the data in the form of a network as shown in FIG. For example, "number"
By sharing the right end of the clause with "Suku", it is possible to unify the following clauses. This specific method of data representation is well known as list processing.

In step 206 of FIG. 3, 1 is added to n to determine whether there is any input at the character position of pointer n (step 207), and if so, the process returns to step 202.
When n is 2, there is no right end of the clause at this character position, so the process goes to step 206 and immediately moves the pointer n to the next character position. If n is 3, the phrase ends such as "number" and "suck" are possible, so in step 203, the next phrase is extracted from the main position, and the phrases "amount", "raku", "aboka", " You can get phrases such as ``Hakukai'' and ``Hakuke''. At step 204, "mathematics" has already been obtained as a candidate at the position where n is 4, and the number of clauses is 1.
In the bunsetsu candidates for "frame" and "raku," the number of clauses at the position where n is 4 is 2, which is the sum of 1 for "number" and 1 for "frame" or "raku." Since "Amount" and "Raku" fall within the permissible range from the pruning conditions, they are registered as network data in step 205. As described above, by repeating steps 202 to 207 until n reaches 14, network data as shown in FIG. 4 can be obtained. However, here, ``kai, kai, kana, ka, ka, 上, ba, state'' are used as suffixes.

As described above, the operation of the conversion process 20 in the control section 2 is completed, and then the process moves to the operation of the process selection determination process 30.

FIG. 5 is a flowchart showing the operation of the process selection determination process 30. In step 301,
It is determined whether the optimal candidate is being created immediately after the conversion process is completed or is being modified. If the optimal candidate is being created immediately after the conversion process is completed, the start position of the modified part of the conversion result is determined in step 302. Initialize the pointer P to 1 to indicate that there is no modified part yet. In the optimal candidate creation process 40, the candidate string with the greatest likelihood from among the candidate strings from point P to the end of the input character string is displayed on the display unit 4 as a conversion result. If there are multiple candidate strings with the same likelihood, refer to the learning table in the storage unit 3 that holds independent words from character strings that have already been determined, and choose the word on this table that has the highest number of words. Select the containing candidate columns as the conversion result. If the candidate string cannot be uniquely determined even after the above, the candidate string with the largest sum of independent word lengths is selected. Other methods for selecting conversion results include methods that use word frequency, and other selection methods can be created by arbitrarily combining the factors described above.

In the case of the example shown in FIG. 4, when P is 1, "in terms of mathematical analysis" is selected as the conversion result and displayed on the display unit 4. Further, in this display processing 60, the conversion result is divided into division units, and the first division unit is highlighted. This highlighted display becomes a cursor used to specify a different conversion location when a conversion different from the candidate requested by the operator is performed.

If the conversion is different from the candidate intended by the operator, the keys necessary to correct the candidate to the candidate intended by the operator are shown in FIG. Here, the arrangement of the keys is not particularly limited.

If the optimal candidate string displayed in the display process 60 is not the candidate intended by the operator, in the candidate change instruction process 70, press the right or left cursor key (hereinafter referred to as the right cursor key or cursor key) shown in FIG. Move the cursor to the correction location,
A correction instruction is given using a correction instruction key (hereinafter referred to as a correction key) shown in FIG. With this instruction,
The control unit 2 modifies the conversion mode, which is the optimal candidate creation mode immediately after the conversion process ends, to a correction mode, and transfers control to the process selection determination process 30 again. In the process selection determination process 30, a mode determination process 301 is performed, and in step 303, the input key is analyzed in the candidate change instruction process 70. For example, if a modification key is input, processing determination processing 304 is performed, and in step 305, it is determined how many characters from the beginning of the input character string the first character of the division conversion unit whose candidates are to be changed is determined. Set. Then, candidate correction processing 5
Transfer control to 0. The candidate correction process 50 displays all candidates starting from point P in the display process 60. In the case of the example shown in Figure 4, when P is 1, "mathematics", "number", and "suck" are selected as candidate groups for selection, and are displayed at the bottom right of the screen in the form of a matrix shown in Figure 7, for example. . In this screen, the bottom 601 is an input line, the top 602 is a conversion result output line, and the bottom right 603 is a selection candidate display area.

Here, when changing the candidate from "mathematics" to "number" or "suck", for example, the numeric keypad and next candidate selection key shown in FIG. 6 are used. (Hereafter, this 2
The street selection method is collectively called the candidate selection key).
When using a numeric keypad, the matrix shown in FIG. 7 and the selection candidates are placed in the same positional relationship, and the operator directly selects the candidate intended by the operator. To change the candidate to "Number", press the numeric keypad "8" in Figure 6.
To change the candidate to "suck", press "9" on the numeric keypad in FIG. 6, and the candidate corresponding to that numeric keypad is set in the display buffer in step 306. The next candidate selection key refers to the storage unit 3 in which candidates are stored in order as shown in FIG.
The display buffer is set at 306. For example, when "Mathematics" is displayed, pressing the next candidate key will retrieve "Number" in FIG.

Key for instructing candidate selection (candidate selection instruction key)
If the character string is inputted, it is passed through the process determination process 304, and in step 307, it is determined which character the next character of the corrected candidate is from the beginning of the input character string, and it is set to P. Then, optimal candidate creation processing 40
transfer control to As described above, the optimal candidate creation process 40 selects the candidate sequence with the highest likelihood from point P, sets it in the display buffer, and displays it in the display process 60. Here, the correction key, the next candidate selection key, and the conversion key for converting the input character string can be processed even if they are the same key by providing the mode determination process 301, key analysis process 303, and process determination process 304 shown in FIG. Therefore, the modification key, next candidate selection key, and conversion key in FIG. 6 can be made into a single key.

FIG. 9 shows an example of a screen displayed on the display unit 4 immediately after input processing 10 is finished, and FIG. 10 shows an example of a screen display in which the candidate string with the highest likelihood for the input character string to be converted is selected and displayed. shows. 602a is a highlighted display (cursor). When the operator requests ``Mathematics Department Transfer Top'' as the conversion result of the input character string ``Suugaku Kai Sekiji Yoha'', the operator uses the right cursor key in Fig. 6 to make corrections, and moves the cursor 602a to the As shown in FIG. 11, move from "Mathematics" to "Analysis" and press the modify key in FIG. 6. The process selection determination process 30 analyzes the correction instruction in step 303, and in step 305 analyzes that the input character string should be corrected starting from the fifth ``ka'', and sets P to 5. Candidate correction process 50
is the value of P, specifically the 5th "ka" in the input string
All the candidates starting from are set in the display buffer from the network data shown in FIG. 4 (see FIG. 17), and a matrix is displayed at the bottom right of the screen as shown in FIG. 12 in display processing 60. In FIG. 6, the operator directly selects ``Department'' with the numeric keypad ``1'' or continues to press the next candidate selection key to repeat corrections until ``Department'' is selected. When "Department" is selected using the next candidate selection key, the processing selection determination process 30 analyzes the modification instruction in step 303, and sets the selection candidate "department" in the display buffer in step 306. In step 307, P is set to 6 because "i" is the sixth input character next to the fifth kana-kanji conversion result "ka" in the input character string. In the optimal candidate creation process 40, the candidate string with the highest likelihood following "family" is selected and set in the display buffer, and the screen display example shown in FIG. 13 is output to the display unit 4. Since the adjacency relationship between the candidates is maintained in this way, if the division conversion unit is changed in the candidate change instruction process 70, the subsequent display can also be corrected at the same time, making it possible to perform subsequent correction operations. can be reduced.

The candidate shown in FIG. 14 and the candidate intended by the operator are different in the "Iseki" part between "Ruins" and "Transfer". Therefore, it can be corrected by performing the same operation as when correcting "analysis" to "department". The flow is shown in FIGS. 14 to 16.
Cursor 602 indicating the correction location in FIG.
If a is moved from "family" to "ruin", "ka"
Candidates starting with are deleted from the screen display example. However, it is also possible to output all candidates starting with "i" when the cursor 602a is moved to "ruins".

The series of processes described above are repeated to correct the candidates to the ones intended by the operator.

〔Effect of the invention〕

According to the present invention, even if an incorrect conversion is performed, corrections can be made in the same division conversion unit at any location in the converted character string, such as correction of homophones (e.g., "analysis").
The same operation can be used to correct corrections (correcting ``Kaiseki'' to ``Kaiseki'') or corrections in different division conversion units (for example, correcting ``Analysis'' to ``Kaiseki''), so there are no errors in the conversion results. Even if a problem occurs, you can immediately change to another candidate and make corrections easily.

[Brief explanation of the drawing]

The drawings show an embodiment of the present invention, and FIG. 1 is a functional block diagram of the kana-kanji conversion device, FIG. 2 is a block diagram of the structure, FIG. The figure is a data network configuration diagram, Figure 5 is a flowchart showing the operation of process selection determination processing, Figure 6 is a key arrangement diagram for modifying candidates, and Figures 7 and 9 to 16 are FIGS. 8 and 17 are explanatory diagrams showing screen display information, and FIGS. 17A and 17B are explanatory diagrams showing examples of correction candidate data. 1...Input section, 2...Control section, 3...Storage section,
10...Input processing, 20...Conversion processing, 30...
Process selection determination process, 40...optimum candidate creation process,
50... Candidate correction processing, 60... Display processing, 70
...Candidate change instruction processing.

Claims (1)

[Claims] 1. A kana-kanji conversion unit that divides an input string into multiple division conversion units and converts each division conversion unit into a kanji-kana-mixed character string, and a kanji-kana-mixed unit that outputs a kanji-kana-mixed character string complement for the input string. In a kana-kanji conversion device comprising a character string output unit, and a display unit that displays the conversion result by the kana-kanji conversion unit and character string candidates output from the kanji-kana mixed character string output unit, each divided conversion unit a network storage unit that stores each character string containing kanji and kana in the form of a network in which the character strings are arranged in series and parallel according to the input character string; An optimal candidate creation processing unit that extracts the candidate string with the highest likelihood from the Kanji/Kana mixed character strings for each conversion unit and displays it on the display unit, and one instruction input that corrects the converted and displayed result. Candidate correction processing in which another candidate having the same division conversion unit, such as a homonym, and a candidate having a different division conversion unit are retrieved from the network storage unit and displayed together on the display unit. and a processing selection determination processing section that selects and performs processing of either the optimal candidate creation processing section or the candidate modification processing section, and the optimal candidate creation processing section is extracted by the candidate modification processing section. the candidate string with the highest likelihood is selected from the network storage section from among the candidate strings following the candidate selected by the selection key from among the candidate groups displayed on the display section and displayed on the display section; A kana-kanji conversion device characterized in that it is configured to