JPH10269208A - Japanese syllabary-chiness character converting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform KANA (Japanese syllabry)-KANJI (Chinese character) conversion that is high in correct answer rate, by evaluating the priority by making use of field information added to KANJI when there are more than one candidates. SOLUTION: The KANA-KANJI mixture converting means 11 performs retrieval from a word dictionary memory 12 according to a character string stored in an input memory 10 and stores retrieved candidates in a conversion result memory 13. A candidate pointer indicates the storage destination addressees of the conversion result character strings stored in a conversion result memory 13. Then field information on a character that the character pointer indicates is retrieved form KANJI field information memory part 15. On the basis of the retrieved field information, the similarity of the field is calculated and the evaluation result is stored in an evaluation result memory 16. After the final candidate is handled, retrieval from the evaluation result memory 16 is referenced to white the candidates out to an output money 23 in the decreasing order of similarity, and the process is finished. Even when there is made than one candidate, the load of a candidate choice is reduced by the priority evaluation based upon field information on KANJI.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device such as an electronic organizer, a portable information terminal, a word processor, a personal computer, etc., having a function of converting kanji kana.

[0002]

2. Description of the Related Art Conventionally, kanji-kana conversion has been realized in which a character string in which kanji and kana are mixed is converted to a kana-kanji-kanji character string. This is a technique of inputting “meeting” and converting it to “meeting”, for example.

This kanji-kana conversion, which is different from the conversion of inputting common kana characters to kana-kanji kanji characters, is capable of inputting kanji-kana kanji characters. Is effective for inputting characters that are difficult to input, such as a large number of strokes for input or difficulty in recognition.

An advantage of the kanji-kana conversion is that kanji-crossing reduces the number of conversion candidates and improves conversion efficiency as compared to kana-kanji conversion. Furthermore, as disclosed in Japanese Patent Application Laid-Open No. Hei 7-334497, an apparatus having a higher conversion rate is realized by calculating the ratio of handwritten input of Chinese characters and evaluating the priority of the next candidate. This means that if “creation” and “operation” are obtained for “sakusaku”, for example, “saku” is input in kanji, and the ratio is higher when reading “saku” is higher. This is a method of giving priority to "creation" based on this data.

[0005]

As described above, in the kanji-kana mixed conversion, the number of candidates is considerably reduced as compared with the kana-kanji conversion, but the next candidate still exists. Further, as a next candidate peculiar to kanji kana conversion, for example, "Shoko" (to be converted to "Descent") is input to "Keko".
The result is converted into “down” and a candidate such as “go home”, which is difficult for the user to predict, appears, giving a sense of discomfort.
This occurs when the system expects the user to enter and read differently than the reading.

[0006] Further, it is difficult to collect the data of the ratio of the kanji-to-kanji spelling found in the prior art,
In addition, the amount of collected data becomes large. Alternatively, the ratio of cross-writing can be updated and learned at any time,
There is a problem that it is not possible to provide a detailed response according to the content of the input text only by the ratio of each reading.

Therefore, according to the present invention, in a kanji kana conversion apparatus for converting a kana in a kanji kana mixed character string into a kanji, when there are a plurality of candidates, the priority is evaluated using field information added to the kanji. By doing so, kanji kana conversion with a high correct answer rate is performed.

Thus, even when there are many candidates, the priority evaluation is performed based on the kanji field information, so that the burden of candidate selection is reduced. By subdividing the kanji field information, it is possible to give priority to sentences with a wide range of contents, and at the same time, the kanji field learning is performed in accordance with the input sentence, so it can respond to changes in sentence contents. It is possible, and it is possible to prioritize candidates regardless of the content of the sentence.

Further, since the kanji field information is information given to a single kanji, no explosive combination of data occurs, and the absolute amount of required data is small. As a result, the size and cost of the device can be reduced, and it can be realized even on a device with a low processing capacity.

Further, since the kanji field information can be basically created by extracting the appearance frequency of kanji, unlike the ratio table which requires kanji reading information, processing such as word division by morphological analysis is required. And it is easy to create data. In addition, even in the case of reading a sentence created by another device, even in the pre-processing for obtaining the field information of the input sentence in a batch, the matching of kanji is lighter, faster and easier than word matching. Can be realized. further,
For users who frequently input texts of the same kind, by setting field information individually, field learning can be performed more accurately, and conversion accuracy is improved. The same applies to the case where the field of the input text is known.

[0011]

According to the first aspect of the present invention, in a kana-kanji conversion apparatus for converting an input character string into a kana-kanji mixed character string, at least the notation information of a word and the reading for each notation are stored. A word dictionary, a field information dictionary that stores field information added to each kanji, and learning means for acquiring field information from kanji in a previously input character string based on the field information dictionary; Word search means for searching word candidates from the word dictionary, based on the obtained character strings, and evaluation for prioritizing word candidates searched for by the word search means based on field information obtained by the learning means. The above object is achieved by providing means.

According to the second aspect of the present invention, the field information is composed of n field labels, and each label stores a digitized value of the degree of relevance to the field.
The above problem is solved by configuring the field information as dimensional vector information.

According to the third aspect of the present invention, the evaluation means, when prioritizing the candidates, sets the kanji field information of the n-dimensional vector obtained by the learning means and the n of each word candidate.
The above problem is solved by calculating the inner product of the dimensional vectorized kanji field information and giving priority to the candidate having the larger value.

According to the fourth aspect of the present invention, the field information is composed of n field labels, and each label stores a numerical value of the degree of relevance to the field.
Field information is configured as coordinate information of a three-dimensional space. The evaluation means, in prioritizing candidates, the kanji field information converted into n-dimensional coordinates obtained by the learning means and the n-dimensional coordinate conversion of each candidate. 2. The kana-kanji conversion apparatus according to claim 1, wherein a distance of the obtained kanji field information is calculated, and a candidate having a small value is prioritized.

According to claim 5 of the present invention, the word dictionary is composed of at least notation information, readings for each notation, and part of speech information, and a word is cut out from a character string based on the word dictionary to specify a part of speech. Analysis means is provided, and the learning means solves the above-described problem by dividing a character string input in the past into words by the morphological analysis means and obtaining field information only for predetermined words.

According to the sixth aspect of the present invention, the learning means divides a character string input in the past into words by the morphological analysis means, and does not acquire field information for proper nouns, thereby solving the above problem. Resolve.

According to a seventh aspect of the present invention, there is provided a field setting means for presetting field information corresponding to a sentence inputted by an operator among field labels constituting the kanji field information,
The evaluation means solves the above-mentioned problem by determining priority candidates based on the field label of the field set by the precursor field setting means.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the drawings. FIG. 1 is a basic configuration diagram of an apparatus according to an embodiment of the present invention. Reference numeral 1 denotes input means for inputting various signals such as input of handwritten characters and selection instructions, and may be integrated with display means such as a pen input tablet.

Reference numeral 2 denotes a field setting means for setting in advance the field of the text to be input, which is provided in the input means. Numeral 3 is a field selecting means for selecting necessary field information from the field information, and is provided in the input means. Reference numeral 4 denotes a display unit for displaying the input character, the result of the kanji kana mixed conversion, and the like. Reference numeral 5 denotes input / output control means for controlling each signal of the input means and the output means.

Reference numeral 6 denotes a central processing unit for controlling the entire apparatus. Reference numeral 7 denotes character recognition means for recognizing handwritten input characters from the input means.
y Memory) or a program code stored in an HD (Hard Disk). Reference numeral 8 denotes a memory for storing a character recognition dictionary referred to by the character recognition means.
Is the data stored in.

Reference numeral 9 denotes a memory for storing a character recognition result, which is provided on a RAM (Random Access Memory). Reference numeral 10 denotes a memory for storing an input character string, which is provided on a RAM. Numeral 11 denotes a kanji / kana kana conversion unit for converting the input kanji / kana kana character string, which is a program code stored in the ROM or the HD. Reference numeral 12 denotes a memory for storing a word dictionary referred to by the kanji-kana conversion unit and the morphological analysis unit, and is data stored in the ROM or the HD.
Reference numeral 13 denotes a memory for storing a conversion result of the kanji / kana mixed conversion, which is provided on the RAM. 14 is
RO is an evaluation means for evaluating the priority of conversion result candidates based on kanji field information and field learning data.
It is a program code stored in M or HD.

Reference numeral 15 denotes a memory for storing kanji field information, which is data stored in a ROM or an HD. 16 is a memory for storing the evaluation result,
It is provided on a RAM. Reference numeral 17 denotes learning means for learning kanji field information from a character string already input, and is a program code stored in a ROM or an HD. Reference numeral 18 denotes a memory for storing the field information learned by the learning means, and is provided on the RAM.

Reference numeral 19 denotes morphological analysis means for morphologically analyzing a character string already input, and is a program code stored in a ROM or an HD. Reference numeral 20 denotes a memory for storing a sentence already input, which is a writable memory such as an HD or FD. Reference numeral 21 denotes a memory for storing the field information selected by the field selection means, which is provided on the RAM. 22 is a memory for storing the field information set by the field setting means,
It is provided on a RAM. 23 is a memory for storing the conversion result character string displayed on the display means;
It is provided on the AM.

Hereinafter, each configuration will be described in order. First, the kanji kana mixing conversion means 11 will be described. Characters input by handwriting after handwritten character recognition are stored in the input memory 10. Then, upon receiving a user's kanji-kana conversion request from the input means, kanji-kana conversion is performed on the character string in the input memory.

In the kanji kana conversion, a dictionary search is performed based on the kanji kana character string, and a part of the hiragana in the character string is converted to kanji. For this reason, the dictionary to be searched needs to have a structure capable of searching from kanji or kana, for example, as shown in FIG.

In the figure, reference numeral 5 denotes a dictionary body portion, where a is a word reading, a is a word notation, and c is reading number information assigned to each character of the notation. For example, "Ogawa / Ogawa / 12 Indicates that the notation "small" is assigned one reading "o", and the notation "kawa" is assigned two readings "gawa". Note that, based on the character number information, it can be determined that the cross-written input “Ogawa” matches the “Ogawa”, and the conversion result “Ogawa” can be obtained. Also, in the kanji kana mixing conversion, various patterns of kanji and kana are written,
For example, a conversion candidate “Ogawa” must be searched from “Ogawa”, “Ogawa”, “Ogawa”, and the like.

For this reason, as a search index,
Index portions such as 1, 2, 3, and 4 in the figure are provided. Reference numerals 1 and 2 are indexes for searching a dictionary based on readings, and reference numerals 3 and 4 are indexes for searching a dictionary based on notations. When shared with a morphological analysis unit described later, grammatical information such as part of speech is also required.

In the kanji-kana conversion process, basically, independent word words to be converted are obtained by searching the word dictionary, and further, a particle part and the like are analyzed to generate candidates in units of phrases. It is common to do. Further, in order to evaluate the priority between these candidates, frequency information, the validity of grammatical connection with preceding and succeeding phrases, and the like are used. Since the conventional technique can be used for the kanji kana conversion processing, detailed description will be omitted. The conversion candidates obtained by the kanji-kana mixing conversion means are stored in the conversion result memory 13.

The kanji field information will be described. The kanji field information defines n classification labels (hereinafter referred to as fields) for each kanji. Examples of disciplines include "politics,""economy,""education," and "arts."
These may be subdivided into definitions as needed.
Then, for each kanji, the degree of relevance to these n fields is quantified and given as kanji field information.

FIG. 7 is a schematic diagram showing the configuration of the kanji field information memory 15. In the figure, a is a storage portion of kanji notation, in which JIS codes are actually stored in ascending order. In the figure, b is a storage part of the field information added to each kanji. The field information can be handled as an n-dimensional vector value (or a coordinate in an n-dimensional space), and is expressed as vec (“school”) = (a1, a2, a3,...) For convenience. In this case, each component of the vector corresponds to an individual field. It should be noted that the field information need not be given to all kanji, and may be limited to necessary kanji.

For the preparation of the kanji field information, the degree of relevance between each field and the kanji may be set one by one. Alternatively, there is a method in which a large amount of text data is classified using a field as a key, and a statistical value of the number of appearances of each kanji in the classified text data is calculated to calculate the degree of association with the field. Alternatively, it is also possible to improve the accuracy of the field information by combining these methods.

Next, the learning means 17 will be described. The learning unit 17 obtains learning information (hereinafter, referred to as a learning field) of a field indicating to which field the entire input sentence is related. The field information of each kanji is used to extract the learning field.

For each kanji in the sentence already input, the kanji field information memory 15 is searched for the respective field information. Then, the retrieved field information is calculated on the basis of a constant equation described later to obtain a learning field. The result of the calculation is stored in the learning memory 18.

The learning field is a vector sum of field information of the appearing kanji calculated based on the following equation. First, the learning field generated up to immediately before is represented by vec (learning) = (k1, k2, k3,...). In this state, when the kanji “school” newly appears, the learning field is updated as vec (learning) ′ = (k1 + a1, k2 + a2, k3 + a3,...). FIG. 8 is a schematic diagram showing the contents of the learning memory.

If the vector sum is simply obtained, each vector component is accumulated, so that the difference between the fields becomes small, and the learning field becomes saturated (all fields have a uniformly high degree of relevance). Would. to solve this problem,
Averaging is performed on each vector component by the number of appearances of kanji. If the value of the learning field, which is a vector sum, is vec (learning) = (K1, K2, K3, ...), and the number of kanji calculated in the learning field is s, the averaged value of the learning field Is as follows: vec (learning) ′ = (K1Ks, K2 ÷ s, K3… s,...) = (K1 ′, K2 ′, K3 ′,...)

Alternatively, there is a method in which a certain limit is set for the component values of the vector, and components (fields) below the limit value are initialized. The limit value may be a fixed value or a function value obtained based on the number of appearances of kanji. For example,
Assuming that the limit value is L and the number of appearances of kanji is s, based on the limit value L calculated by L = α × s α: constant value, vec (learning) ′ = (k1 ′, k2 ′, k3 ′,...) Are initialized to 0 for each component that is equal to or less than the limit value L.

Further, even in the case of kanji to which field information is given, learning may be adversely affected depending on how it is used in a sentence. Specifically, kanji characters such as "Hoshi", "Hikaru", and "Ten" are examples of large values in the field of "space and celestial bodies", but they are part of words such as "Hoshino", "Terio", and "Amakusa". Does not necessarily reflect the field information in the text. Such a tendency is particularly remarkable in the kanji notation of proper noun words. For this reason, it is effective to exclude the kanji in proper nouns in the calculation of the field learning.

The timing of updating the learning data at any time in accordance with the input of the character is performed immediately after the input of the kanji by handwriting input or immediately after the conversion result of the kanji kana mixed with the kana.

Next, the morphological analysis means 19 will be described. As described above, in field learning, it is effective to morphologically analyze sentence data and divide the sentence data into words, and then remove proper nouns. The morphological analysis is to divide a character string mixed with kanji and kana into units called morphemes (independent words, particles, etc.) based on a word dictionary, and lexically divide each character into morphemes.

With regard to morphological analysis, in the case of Japanese, analysis based on a unit of a clause composed of independent words, particles, and the like is generally performed, and the longest match that prioritizes the candidate that matches the dictionary longest from the beginning of the character string is given priority. Or the two-phrase longest match method that gives priority to the longest matching candidate by adding the length of two phrases,
A basic algorithm such as the minimum number of clauses method that gives priority to a candidate that minimizes the number of clauses in the entire sentence has been proposed. Note that the processing content of the morphological analysis may be an existing process, and thus a detailed description is omitted.

As an example of the morphological analysis, if the input character string is “a planetary nebula is a star”, “planet (general noun) +
(Suffix), nebula (general noun) + (particle), star (general noun) + (particle) ”is obtained. Or, if the input character string is "Kyushu Amakusa to be a pilot project of orange fields", "Kyushu (proper noun), Amakusa (proper noun) + (particle), mandarin (general noun) + field (suffix) + (Particle), pilot (general noun), business (general noun) + (particle) ”. By using the result of such morphological analysis, it is possible to limit the use of the kanji "star" in the common noun by the learning means, but not to use the kanji "ten" in the proper noun.

In the morphological analysis, it is necessary to analyze an input character string in units of a certain length. Therefore, in handwriting input or the like, it is not possible to learn each input character immediately after input. Therefore, it is necessary to wait for the learning process until there is an input break such as a punctuation mark or a line feed. For example, the learning process is started by delaying the timing by a certain number of characters with respect to the input character.

When the input is performed intermittently or when a sentence created by another device is corrected by the present device, there is no learning data for the already input sentence. For this reason, it is necessary to collectively learn when reading into the sentence memory.
In such a case, it is possible to create learning data by morphological analysis.

Next, the field selecting means 3 and the field setting means 2
Will be described. The field information given to each kanji has a wide range of field settings in order to correspond to various sentence inputs. For this reason, unnecessary field information may have an adverse effect. In such a case, if a sentence is input with contents concentrated on a specific field, adverse effects can be suppressed by limiting necessary field information.

The field selecting means 3 is for the user to temporarily limit the label of the field to be learned by the learning means. In the case where a field is found, the field selection means 3 can select a necessary field or suppress unnecessary fields. The information on the field label selected by the field selection means 3 is stored in the selected field memory 21.
Is stored in

For example, in the calculation of field learning, vec (learning) = (k1, k2, k3, k4, k5, k6,...): Learning field vec (kanji) = (x1, x2, x3, x4, x5, x6,...): When the field labels k2 and k5 are selected as unnecessary labels in a state like an appearance kanji, the component k is calculated in the field learning calculation.
2, neglecting k5, vec (learning) '= (k1 + x1, k3 + x3, k4 + x4, k6 + x6,
…)) (The components corresponding to unnecessary fields may be initial values or indefinite values).

The field setting means 2 is for presetting unnecessary field labels in the field information assigned to each kanji so as not to be used by the learning means or the evaluation means described later. It is the same as the selection means 3, but is effective until it is reset rather than temporarily set. The information on the field label selected by the field setting means 2 is stored in the set field memory 22. For example, assuming that the number of field labels set to be unnecessary by the setting means is u, the field information of kanji is treated as nu-dimensional vector information in all the processes thereafter. The component corresponding to the unnecessary field may be an initial value or undefined.

Next, the evaluation means 14 will be described. The evaluation unit 14 calculates the similarity between the learning field of the already-input text obtained by the learning unit and the field information of the kanji notation part of the conversion candidate obtained by the kanji-kana conversion unit, and gives priority to the candidate. Evaluate the degree.

When the kanji field information is n-dimensional vector information, the similarity is calculated by inner product calculation. For example, the learning field immediately before the conversion is represented by vec (learning) = (k1, k2, k3, k4,...) The field information of the kanji notation part in the candidate is represented by vec (candidate A) = (a1, a2, a3, a4,...), the inner product is calculated as k1 × a1 + k2 × a2 + k3 × a3 + k4 × a4 +..., and a candidate having a larger inner product is determined to have higher similarity.

Alternatively, there is a method in which the field information is regarded as coordinates in an n-dimensional space, and it is determined that the closer the distance between the coordinates is, the higher the similarity is. Distance = sqrt ((k1-a1) ² + (k2-a2) ² + (k3-a3) ² + (k4-a4) ² + ...)
* Sqrt: Calculate the square root, and judge that the candidate with a small distance has higher similarity.

Further, it is possible to limit the components used for these calculations to increase the accuracy of similarity calculation.
In order to avoid an adverse effect of unnecessary fields, there is a method in which only components having a certain value or more are used for calculation for each component in the learning field, or only some higher-order components are used for calculation. The setting of the constant value is experimentally determined according to the setting range of the degree of association with each field in the field information.

Further, when the field information to be used is previously set by the field setting means 2, the similarity calculation is performed using only the components corresponding to the field. Note that, depending on the kanji, field information may not be given, and in that case, similarity calculation is not performed.

As an example of the operation of the evaluation means, for example, if kanji kana conversion is performed based on the input character string "public" to obtain "public private", "legal envoy" and "kouji", then the kanji notation of each kanji is described. The field information is searched from the kanji field information memory 15. Then, the field information and the learning field of each kanji are calculated based on the above calculation method. As a result, for example, in the normal frequency information stored in the dictionary, the priority order is "Public private → Public ministry → Public corporation", but the evaluation based on the kanji field information changes to "Minister → Public private → Public corporation". Change priority.

Also, depending on the input characters, the number of kanji characters to be converted may be different. For example, a case where "modernization" or "modernization" is searched for the input character "is close". In this case, if the sum of similarities for each kanji is calculated, a candidate having a larger number of characters is often given priority. Therefore, for a candidate consisting of a plurality of characters, it is necessary to perform processing such as selecting one character having the highest similarity as a representative value of the candidate, or dividing the sum of similarities by the number of characters and averaging. Each candidate whose priority is determined is stored in the output memory 23 in order of priority.

Next, the input means 1 and the character recognition means 7 will be described. It is assumed that the input means of the present apparatus is integrated with the display means and has a handwriting input function such as a pen input tablet. Furthermore, handwriting input requires handwritten character recognition means capable of inputting kanji and kana. Note that the input means and the character recognition means may be existing techniques, and thus detailed description is omitted.

Hereinafter, the flow of the processing will be described with reference to the processing flowcharts of FIGS. The general flow of the processing is shown in FIG.
After performing 1), pre-processing related to learning (step 20)
Perform 2).

Thereafter, the process branches to each process according to the input signal from the input means 1. If the input signal is a request to end the processing (step 203), the apparatus ends the processing. On the other hand, if the instruction is a conversion of kanji to kana conversion (step 204), a kanji to kana conversion process is performed on the character string input so far (step 2).
07), display processing of conversion candidates is performed (step 20).
8).

After that, it waits for a decision instruction by a conversion candidate selection signal or the like (step 209). If the conversion instruction is not a kanji-kana mixed conversion instruction (step 204), a character recognition process is performed by the character recognition means 7 because the input signal is a character (step 205). Next, the recognized result is stored in the recognition result memory 9, and the input character string is displayed in a predetermined display area of the display means 4 (step 206).

FIG. 3 shows the flow of the process of the initial setting (step 201 in FIG. 2). In the device provided with the field selecting means 3,
If there is a field selection request (step 301), the selected field information is set in the selected field memory 21 (step 302).

FIG. 9 is a schematic diagram showing the state of input and display means when selecting a field. Further, in the device provided with the field setting means 2, when there is a field setting request (step 303), the set field information is set in the set field memory 22 (step 304). FIG. 10 is a schematic diagram showing the state of input and display means when selecting a field.

FIG. 4 shows a pre-learning process (step 202 in FIG. 2).
2 shows the flow of processing. The pre-learning process is a process of collectively creating a learning field when a text has been read from an external storage medium or the like by the learning means 17. First, the sentence data is read into the sentence memory 20 (step 40).
1). Next, 0 is set to the sentence pointer provided in the learning memory 21 (step 402).

The sentence pointer is a variable that indicates the address where the character string in the sentence memory is stored, and it is assumed that 0 points to the beginning of the sentence. Then, it is checked whether the sentence pointer is at the end of the sentence (step 403). If the sentence pointer is not at the end of the sentence, the sentence data is cut out from the position indicated by the sentence pointer (step 404). Extraction of text data is performed in units of text breaks such as punctuation and line feed.

The morphological analysis means 19 performs morphological analysis processing on the extracted sentence data (step 4).
05). Next, 0 is set to the character pointer provided in the learning memory 18 (step 406). Note that the character pointer is a variable that indicates the address of the storage destination of the character string to be subjected to morphological analysis, and 0 indicates the head of the character string. Then, it is checked whether the character pointer is at the end of the character string (step 407). If it is not the end of the character string, each character pointed to by the character string pointer is checked.

First, when the character is a kanji (step 4
08), referring to the result of the morphological analysis, it is checked whether the word is a proper noun and a part of the analyzed word (step 409).
If it is a kanji and not a notation in a proper noun, the corresponding field information is searched from the kanji field information 15 (step 410). Learning memory 18 using searched field information
Are updated (step 411) and stored in the learning memory again.

When the field information is not added to the kanji pointed to by the character pointer, the calculation of the learning field is not performed. Then, the character pointer is updated by one character (step 412).
And returns to step 407. If the character is not a kanji in step 408 and if it is a notation in a proper noun in step 409, the character pointer is updated and the process returns to step 407. If the character pointer has reached the end of the character string in step 407, the text pointer is updated to the start address of the next text (step 413).
Then, the process returns to step 403. If the text pointer has reached the end of the text in step 403, the pre-learning process ends.

FIG. 5 shows the flow of the kanji kana conversion process (step 207 in FIG. 2). The kanji kana conversion means 11 searches the word dictionary memory 12 based on the character string stored in the input memory 10 (step 501).
The searched candidates are stored in the conversion result memory 13 (step 502). Next, 0 is set to the candidate pointer provided in the evaluation result memory 16 (503). Note that the candidate pointer is a variable that indicates the storage destination address of the conversion result character string stored in the conversion result memory 13, and 0 indicates the leading candidate.

Next, it is checked whether or not the candidate pointer points after the final candidate (step 504). If it is not next to the final candidate, 0 is set to the character pointer provided in the evaluation result memory 16 (step 505). The character pointer is a variable indicating the storage destination address of the candidate character string currently pointed to by the candidate pointer among the conversion candidates stored in the conversion result memory 13, and 0 indicates the head of the candidate character string. And Then, it is checked whether the character pointer points to the end of the character string (step 506). If not, the field information of the character pointed to by the character pointer is searched from the kanji field information memory 15 (step 50).
7). Next, the similarity of the field is calculated based on the searched field information (step 508), and the evaluation result is stored in the evaluation result memory 16.

If the field information is not searched, the similarity calculation is not performed, and a default (for example, 0) evaluation value is given, and the process is terminated. Then, the character pointer is updated by one character (step 509), and the process returns to step 506. If the character pointer has reached the end of the character string in step 506, the previously calculated similarity is averaged by the number of characters (step 510), and recalculated and averaged similarity is stored in an evaluation result memory. (Step 51)
1). When the number of characters is 1, the averaging process is not performed. Then, the candidate pointer is updated to the next candidate (step 512), and the process returns to step 504. Step 50
When the next candidate is reached after the last candidate, the candidate is written to the output memory 23 in descending order of similarity with reference to the evaluation result memory (step 513), and the process is terminated.

FIGS. 11 and 12 are schematic diagrams showing the state of the input and display means. FIG. 11 shows a case where the character “public” is input by handwriting input, in which a is a text display area for displaying the input text data, b is a handwriting input area for performing handwriting input, and c is a handwriting input area. This is a conversion candidate list display area for displaying a list of conversion candidates for kanji kana conversion, and the same applies to FIG. 12 and subsequent figures.

FIG. 12 shows that the input character string “public”
After performing conversion with Kanji and Kana, evaluation processing is performed based on the field of study of the sentence that was input immediately before, and the first candidate "Minister" is displayed in the sentence display area, and the conversion candidate list display area Shows the case where “candidate, public and private, corporation” candidates are displayed in order of priority.

[0071]

According to the present invention, in a conversion device for converting an input character string to a kana-kanji character string, when there are a plurality of candidates, the priority is evaluated using the field information added to the kanji. Thus, kana-kanji conversion with a high correct answer rate is performed.

Therefore, even when there are many candidates, the burden of selecting candidates can be reduced by the priority evaluation based on the kanji field information. Further, by subdividing the kanji field information, it is possible to give priority to sentences having a wide range of contents.

Since the field learning of kanji is performed in accordance with the input sentence, it is possible to cope with a change in the sentence content, and it is possible to prioritize candidates regardless of the sentence content. Since the kanji field information is information given to kanji alone, there is no explosive combination of data that gives a field for each word.
The absolute amount of data required is small. As a result, the present invention can be realized even in an apparatus having a small processing capacity and a low processing capacity.

Further, since the kanji field information can be basically created based on the appearance frequency of the kanji, there is no need to use a ratio table or the like which requires kanji reading information, and data can be easily created. Also, in pre-processing to collectively obtain field information of already input text, such as when reading text created on other devices, Kanji collation has an overwhelming processing load compared to word collation. It is light, fast and easy to implement.

Further, for a user who frequently inputs texts of the same kind, by setting the field information individually, field learning can be performed more accurately, and conversion accuracy can be further improved.

[Brief description of the drawings]

FIG. 1 is a basic configuration diagram of an apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a flow of processing according to the embodiment of the present invention.

FIG. 3 is a flowchart showing a flow of processing according to the embodiment of the present invention.

FIG. 4 is a flowchart showing a flow of processing according to the embodiment of the present invention.

FIG. 5 is a flowchart showing a flow of processing according to the embodiment of the present invention.

FIG. 6 is a schematic diagram showing a structure of a kanji / kana kana conversion dictionary.

FIG. 7 is a schematic diagram showing the contents of kanji field information.

FIG. 8 is a schematic diagram showing the contents of a learning memory.

FIG. 9 is a schematic diagram showing a state of input and display means when a field selection process is activated.

FIG. 10 is a schematic diagram showing a state of input and display means when a field setting process is activated.

FIG. 11 is a schematic diagram showing a state of input and display means.

FIG. 12 is a schematic diagram illustrating a state of input and display means.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 input means 2 field setting means 3 field selecting means 4 display means 5 input / output control means 6 central processing unit 7 character recognition means 8 character recognition dictionary memory 9 recognition result memory 10 input memory 11 kanji kana conversion means 12 word dictionary memory 13 Conversion result memory 14 Evaluation means 15 Kanji field information memory 16 Evaluation result memory 17 Learning means 18 Learning memory 19 Morphological analysis means 20 Sentence memory 21 Selected field memory 22 Setting field memory

Claims (7)

[Claims] 1. A kana-kanji conversion device for converting an input character string into a kana-kanji mixed character string, comprising: a word dictionary storing at least word notation information and readings for each notation; and field information added to each kanji. A field information dictionary that stores and stores, based on the field information dictionary, learning means for acquiring field information from kanji in a previously input character string, and searching for a word candidate from the word dictionary by the input character string A kana-to-kanji conversion device, comprising: a word search unit that performs a search; and an evaluation unit that prioritizes word candidates searched by the word search unit based on field information obtained by the learning unit. 2. The field information is composed of n field labels, each of which stores a numerical value of the degree of relevance to the field, and stores the field information as n-dimensional vector information for each kanji. 2. The kana-kanji conversion device according to claim 1, wherein the kana-kanji conversion device is configured. 3. The evaluation means calculates an inner product of n-dimensional vector kanji field information obtained by the learning means and n-dimensional vectorized kanji field information of each word candidate in prioritization of candidates. 3. The kana-kanji conversion device according to claim 2, wherein a candidate having a large value is prioritized. 4. The field information is composed of n field labels, and each label stores a numerical value of the degree of relevance to the field, and for each kanji, the field information is coordinate information in an n-dimensional space. The evaluation means, when prioritizing the candidates, calculates the distance between the n-dimensional coordinated kanji field information obtained by the learning means and the n-dimensional coordinated kanji field information of each candidate. 2. The kana-kanji conversion device according to claim 1, wherein a candidate having a small value is calculated and given priority. 5. The vocabulary analysis means comprising at least notation information, readings for each notation and part of speech information, extracting words from a character string based on the word dictionary, and specifying a part of speech, and the learning means. 2. The kana-kanji conversion apparatus according to claim 1, wherein the character string input in the past is divided into words by the morphological analysis means, and field information is obtained only for predetermined words. 6. The kana-kanji character according to claim 1, wherein said learning means divides a character string inputted in the past into words by said morphological analysis means, and does not acquire field information for proper nouns. Conversion device. 7. A field setting means for presetting field information corresponding to a sentence inputted by an operator among field labels constituting kanji field information, wherein said evaluation means sets said field information by said precursor field setting means. 5. The kana-kanji conversion apparatus according to claim 2, wherein priority candidates are determined based on a field label of the field.