JPH11238056A - Predictive japanese input method/system and medium for programming/recording the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the efficiency of document input and to easily input Japanese by precisely giving priority results obtained through automatically executing input/prediction from an arbitrary position during the input of a sentence and showing them to a user. SOLUTION: A segment part 107 takes out a partial character string group from a read character string which is inputted or prediction-supplemented, and a KANJI (Chinese character/KANA (Japanese character) notation where the partial character string taken out is forward-matched with the reading is retrieved at a retrieval part 108 by using a dictionary. Control parts 103, 110 and 111 assigns a priority order to the retrieved KANJI/KANA notation by using one of (1) frequency information on the KANJI/KANA notation character string and frequency information on at least a part of the partial character string of the reading character string, (2) frequency information on the reading character string, (3) likelihood information for respective KANJI/KANA notation characters and (4) likelihood information for respective reading character strings in a dictionary 109, and Japanese is predicted/inputted.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and system for predicting and inputting Japanese language, and a recording medium on which the method is programmed and recorded. The present invention relates to a method and a system for predicting and inputting Japanese words and a recording medium that can be input.

[0002]

2. Description of the Related Art In recent years, it has become common to create a document on a computer, and a document can be easily input by using a dedicated machine such as a word processor or a general-purpose machine such as a personal computer. In particular, in recent years, communication through communication of documents on a computer has become generalized, and opportunities for creating documents on a computer have been further increased.

[0003] On the other hand, various electronic devices such as computers have been downsized due to technological advances, and portable devices have been widely used. In a mobile device or the like, the size of an operation unit such as a keyboard is small, so that the burden of inputting a document is large, and it is desired to reduce the burden related to user input.

[0004] In order to solve such a problem, a document input device for inputting the beginning of the idiom to be input and searching for and presenting the idiom to be input using the input as a key (Japanese Patent Application Laid-open No. Sho 60/1985).
(Japanese Patent Application Laid-Open No. 1-26).
[0568] A kana-kanji conversion system (Japanese Patent Laid-Open No. 8-147289) is known which inputs the beginning of a reading to be input, and searches for and presents the reading to be input using the key as a key. In Japanese Patent Application Laid-Open No. 60-105029, information on the frequency of reading idioms is stored, and the priorities of candidates are determined based on the information. In Japanese Patent Application Laid-Open No. 1-268568, priorities are assigned to candidates based on past input histories. In Japanese Patent Application Laid-Open No. 8-147289, frequency information of conversion candidate words is stored, and priorities are assigned to the candidates based on the frequency information.

[0005]

However, according to the above-mentioned conventional methods, it is assumed that the head of a character string to be input is input, and the prediction is performed by using the key as a key to search the dictionary forward. It has a mechanism. Therefore, since there is no mechanism for automatically predicting the subsequent input at an arbitrary position during input of a sentence, when using it during input of continuous phrase input, which has become common in recent years, conversion After interrupting the input operation, such as pressing the conversion key-times in the middle of, the key operation such as inputting the beginning of the character string to be input again or instructing a prediction search at the position where prediction is desired is required, The operation became complicated, and the operator was burdened.

On the other hand, in order to perform prediction from an arbitrary position during input, it is difficult to accurately perform prioritization between prediction candidates using only conventional words and their reading frequency information. It was difficult in practice to make automatic predictions. In particular, it is not suitable for control of displaying only the top several prediction candidates, and it is necessary to develop a new method for prioritizing prediction candidates.

The present invention has been made in view of the above circumstances, and performs highly accurate white dynamic prediction during user input and at any point after prediction complementation, and presents it to the user in an easy-to-understand manner. It is an object of the present invention to provide a method and system for inputting Japanese prediction and a recording medium on which the method is programmed and recorded.

[0008]

According to the present invention, there is provided a method for inputting Japanese predictive characters, comprising inputting a reading character string, extracting at least a partial character string group from the input or predicted and complemented reading character string. A dictionary is used to search for a kanji kana notation whose pronunciation is in front of the partial character string, and (1) frequency information for the kanji kana notation character string in the dictionary and at least a part of the partial character string of the reading character string. Using frequency information, or (2) prioritizing kanji kana notation searched using frequency information for at least a part of a partial character string of a read character string in the dictionary and performing Japanese prediction input. Features. In addition, (3) using the likelihood information for each kanji character string in the dictionary that matches the leading part of the reading sentence string using (4) or frequency information for at least a part of the partial character string of the reading character string. It is also characterized by prioritizing the retrieved Kanji Kana notation and inputting Japanese prediction.

According to the present invention, there is provided a Japanese predictive input system comprising: a dictionary storing Kanji Kana notation constituting a part of a Japanese sentence and a reading character string thereof; a means for inputting a reading character string; Means for extracting part or all of the partial character string group from the read character string, means for searching the dictionary for a kanji kana notation whose reading matches the extracted partial character string in front of the character string, and (1) which is included in the dictionary. Using the frequency information for the Kanji Kana written character string and the frequency information for the part or all of the partial character string of the reading character string, (2) or the part or all of the partial character string of the reading character string in the dictionary Means for prioritizing kanji kana notations searched using the frequency information for. Also, (3) using likelihood information for each of the read character strings that match forward with respect to the kanji kana written character string in the dictionary, (4) likelihood information for each of the read character strings that match forward with respect to the read character string Means for prioritizing the kanji kana notation searched by using.

The recording medium of the present invention is used in a Japanese predictive input system for predicting and inputting a character string to be input based on already input characters, and a step of capturing an input reading character string. Extracting at least a partial character string group from the read character string or the predicted and complemented read character string, and using a dictionary to search for a kanji kana notation in which the cut-out partial character string and the reading match in front. (1) frequency information for the kanji kana written character string in the dictionary and frequency information for at least a part of the partial character string of the read character string, or frequency information for at least a part of the read character string in the dictionary. Prioritizing the kanji kana notation searched by using, or (2) the kanji kana notation character string in the dictionary Prioritizing the kanji kana notation searched using the likelihood information for each reading character string that matches each other, or the likelihood information for each reading character string that matches forward with respect to the reading character string in the dictionary; Is programmed and recorded.

[0011] As a result, the result of automatically predicting the input from an arbitrary position in the text input can be given to the user with high accuracy prioritized and presented to the user. Can input Japanese.

[0012]

FIG. 1 is a block diagram showing an embodiment of a Japanese predictive input system according to the present invention.

In the figure, 101 is an input unit, 102 is a control unit, 103 is a prediction control unit, 104 is a kana-kanji conversion unit,
105 is a kana-kanji conversion dictionary search unit, 106 is a kana-kanji conversion dictionary, 107 is a reading character string cutout unit, 108 is a prediction dictionary search unit, 109 is a prediction dictionary, 110 is a prediction candidate evaluation unit, 111 is a prediction candidate narrowing unit, and 112 is a prediction candidate narrowing unit. Output unit.

From the input unit 101, various operation information from the user, including reading information, is taken in. The acquired information is sent to the control unit 102 and analyzed, and then passed as necessary information to each processing unit. The processing result of each processing unit is also returned to the control unit 102, and necessary information such as a character string mixed with kanji and kana is output through the output unit 112. The reading information of the input information is first passed from the control unit 102 to the prediction control unit 103. The fetched information is analyzed by the prediction control unit 103 and then passed as necessary information to each processing unit. The processing result of each processing unit is also returned to the prediction control unit 103, and the prediction control unit 10
After the processing in step 3, the required information, such as the predicted kanji kana notation and the read character string, is returned to the control unit 102. The processes after the prediction control unit 103 will be described later in detail.

Next, the control unit 102 outputs the input reading information and the predicted kanji kana notation as information to be presented to the user, for example, by displaying it on a display screen.
Here, operation information input by the user according to the displayed information, such as adoption / non-adoption of the displayed prediction candidate, selection of a desired candidate, etc., is taken in from the input unit 101 and sent to the control unit 102. The processing of this part of the control unit 102 will be described later in detail. Also, the control unit 102, when instructed to perform the kana-kanji conversion process, when a character that is determined to enter the conversion process in advance is input, or when the result of the input prediction is adopted by the user, The input reading information and, if there is a predicted and complemented reading character string that follows it, are sent to the kana-kanji conversion unit 104. The processing of this part of the control unit 102 will be described later in detail.

The kana-kanji conversion unit 104 converts the read character string sent from the control unit 102 into a kana-kana-kana mixed character string while referring to the kana-kanji conversion dictionary 106 by the kana-kanji conversion dictionary search unit 105. Since the kana-kanji conversion method is widely known, it will not be described here. The conversion result is output from the output unit 112 through the control unit 102,
For example, it is displayed on the display. In accordance with the displayed information, the user performs operations such as inputting a reading character string or determining a desired Kanji Kana notation.

Here, the control unit 102 sends the continuously input character string to the prediction control unit 103. At this time, in addition to the input read character string, if there is a predictive and complemented read character string continuous with the input read character string, the control unit 102 combines the read character strings and sends the read character string to the prediction control unit 103.
The processing of this part of the control unit 102 will be described later in detail.

As described above, the input of the Japanese sentence is completed by a series of user operations and the processing of the system. The above is the outline of the processing flow by the Japanese predictive input system of the present invention.

Next, an outline of the flow of processing after the prediction control unit 103 will be described with reference to FIG.

The prediction control unit 103 converts the read character string received from the control unit 102 into a read character string cutout unit 107.
Send to The read character string cutout unit 107 cuts out some partial character strings necessary for predictive dictionary search from the received read character string, and returns the partial character strings to the prediction control unit 103. Next, the prediction control unit 103 sends each read character string to the prediction dictionary search unit 108. The prediction dictionary search unit 108 outputs the prediction dictionary 10
A head-match search is performed on the reading information corresponding to the headword registered in No. 9, and the kanji kana notation of the headword matching the head word and its reading information and various attached information are obtained.
For example, the information of the part of speech and the frequency is taken out and the prediction control unit 103
To return. Next, the prediction control unit 103 sets each prediction candidate
(Headword) to the prediction candidate evaluation unit 110.

The prediction candidate evaluation unit 110 calculates and assigns various evaluation points to the sent headwords based on various kinds of attached information, and further calculates and calculates representative evaluation points representative of them.
And returns it to the prediction control unit 103. Next, the prediction control unit 1
03 sends the various evaluation points of each prediction candidate to the prediction candidate narrowing unit 111. The prediction candidate narrowing unit 111 refers to thresholds that can be set for various evaluation points, selects only candidates that are equal to or larger than the threshold for which various evaluation points are set, and returns a determination result to the prediction control unit 103. The prediction control unit 103 sorts the candidates satisfying the determination conditions of the prediction candidate narrowing unit 111 in the order of the representative evaluation points, and sends the upper candidate to the control unit 102.

Next, the details of each of the above-described processes will be described with reference to flowcharts and tables.

First, the flow of processing of the read character string cutout unit 107 will be described with reference to the flowchart of FIG. First, in step 201, a reading character string list is initialized. Next, in step 202, the reading character string buffer is initialized. In step 203, the read character string received from the prediction control unit 103 is stored in the read character string buffer. In step 204, the length of the character string stored in the read character string buffer is set in a variable len. Step 205
Then, the variable i is set to 0 and the variable start is set to 1. In step 206, the value of the variable start is compared with the value of the variable len. If the value of the variable start is equal to or less than the value of the variable len, the process proceeds to step 207. In step 207, the characters from the first character to the lenth character of the character string in the reading character string buffer are copied to the ith element of the reading character string list, and the process proceeds to step 208. In step 208, the variable i and the variable start are each incremented by 1, and the process returns to step 206.

On the other hand, in step 206, the variable start
Is larger than the value of the variable len, the process ends. From the reading character string received by the above processing, a partial reading character string including the last character of the reading character string is stored in the reading character string list (the reading character string is stored in the reading character string list. Are partial reading strings that include the last character of the received reading string, respectively.)

FIG. 3 shows an example of the format of the read character string list and the processing result. This is the prediction control unit 10
This is the result of processing when a reading character string “Kanakanji” is received from No.3. The column of numbers is the number corresponding to the variable i in the flowchart shown in FIG. 2, and the column of partial reading character strings is the partial reading character string stored in step 207. For example, the second reading character string list stores a character string "kanjihen" which is a partial reading character string of the reading character string "kanakanhen".

FIG. 4 shows an embodiment of the information registered in the prediction dictionary 109. The prediction dictionary 109 stores reading information, kanji kana notation information, part of speech information, notation frequency information, and reading frequency information. For example, for the notation “kana-kanji conversion”, the reading “kanakanjihenkan” and the part of speech “sa-hen” are given, and the numerical value representing the notation frequency is “60”, indicating the reading frequency. "12980, 890, 230,
90, 80, 80, 80, 80 ". Here, the notation frequency information" 60 "is a numerical value indicating the frequency at which the notation" kana-kanji conversion "is used. It can be represented by the number of times the notation “kana-kanji conversion” appears in the corpus. In addition, the reading frequency information is delimited by “,”, respectively.
In this example, nine readings of “Kanakanjihenkan” are assigned with eight numerical values for eight partial character strings, and the first “12980” is a partial reading character string of “Kana”. Frequency, the second “890” is the reading frequency of “Kanaka”, and thereafter “230” is “Kanakan”, “90” is “Kanakan”, “80” is “Kanakanji”, and the next “80” is a numerical value indicating the frequency of reading “Kanakanjikan”, the next “80” is “Kanakanjikan”, and the next “80” is a numerical value indicating the frequency of reading “Kanakanjikan”. Specifically, for example, it can be represented by the number of appearances of a character string such as “Kana”, “Kanaka”, “Kanakan”, etc. in a corpus to which a certain corpus is given a reading.

Similarly, for the notation "Kanagawa prefecture", the pronunciation of "Kanakawa prefecture", the part of speech of "place name", the frequency of the notation "1050" of "Kanagawa prefecture", and the reading of "Kanagawa prefecture" The respective frequency information “1298” for the partial reading character strings “kana”, “kanaga”, “kanagawa”, “kanakaze”, and “kanagahan”
0 "," 1720 "," 1590 "," 1050 ",
“1050” is registered as “stock company”, and the respective information is registered.

FIG. 5 shows another embodiment of the information registered in the prediction dictionary 109. In this example, the information of “reading”, “notation”, “part of speech”, and “notation frequency” are the same as the prediction dictionary shown in FIG. 4, but different from the prediction dictionary shown in FIG. Instead, “probability information” is added. In this method, instead of the frequency of each partial reading character string, a probability (conditional probability) that a corresponding notation appears when the partial reading character string appears according to Bayes' formula or the like. In this example, the probability information shown in FIG. 5 is obtained by dividing the notation frequency information by the numerical value of each partial reading character string frequency information and displaying the result as a percentage in the example of FIG. 4 described above. For example, "kana-kanji conversion: kanakanjihenkan"
To "12980" and "89" respectively.
0 "," 230 "," 90 "," 80 "," 80 ",
Values obtained by multiplying the value obtained by dividing by “80” and “80” by 100 and rounding the decimal point are “0”, “7”, “26”,
"67", "75", "75", "75", "75". These are `` Kana-Kanji conversion '' when the readings `` Kana '', `` Kanakan '', `` Kanakan '', `` Kanakanji '', `` Kanakanji '', `` Kanakanjihen '', `` Kanakanjihenka '' and `` Kanakanjihenkan '' appear, respectively. Can be used in subsequent processing as an approximate value of the probability of appearance.

FIG. 6 shows an example of a dictionary search result list in which the results of forward matching search of the prediction dictionary 109 shown in FIG. 5 by the prediction dictionary search unit 108 are stored. FIG. 6 (a)
Is a storage example of the result of performing a dictionary search based on the read character string information “Kanakanji”. In the predictive dictionary shown in FIG. 5, the heading that starts with the reading character string “Kanakanjihen” (= corresponding to the condition of the prefix search) is only one heading “Kana-Kanji conversion”, so one item is included in the list. Only is stored. Reading, notation, part of speech,
The notation frequency information and the probability information are copies of the information in the dictionary, and the match length information is the length of the read character string found by the prefix search, that is, in this case, the reading of "kanakanjihen" Stores the length. FIG. 6B is an example of storing a result of a dictionary search using the read character string information “Hen”. Similarly, since the corresponding heading is only “conversion device”, one item is stored in the list. The matching length information is the length of the reading “unsatisfactory” that matches the beginning. FIG. 6C is an example of storing a result of a dictionary search using the read character string information “kana”. In this case, since there are two corresponding headings, “Kana-Kanji conversion” and “Kanagawa prefecture”, two items are stored in the list.

Next, the flow of processing of the prediction control unit 103 will be described with reference to the flowchart shown in FIG. First, in step 701, the read character string received from the control unit 102 is sent to the read character string cutout unit 107, and step 7 is executed.
02, the resulting read character string list and the number of partial read character strings, for example, 7 (= that is, FIG.
In the flowchart of the above, the value of i) is received and the variable max
To the number of partial reading character strings, that is, the number of items in the partial reading character string list. Next, the routine proceeds to step 703, where 0 is substituted for the variable i and 0 for the variable cnt. Step 704
Then, the value of the variable i is compared with the value of the variable max. If the value of the variable i is smaller than the value of the variable max, the process proceeds to step 705. Step 7 when the value of the variable i is equal to or greater than the variable max
Proceed to 08. In step 705, the i-th read character string in the partial read character string list is sent to the prediction dictionary search unit 108. As described above, the prediction dictionary search unit 108 performs a forward match search of the dictionary based on the sent partial reading character string, stores the search result in the dictionary search result list, and returns it to the prediction control unit 103. For example, in the example of the partial reading character string list shown in FIG. 3, the 0th partial reading character string “Kanakanjihen” is sent to the prediction dictionary search unit 108, and as shown in FIG. A search result list is created. The prediction control unit 103 proceeds to the process of step 706, copies n search results stored in the search result list in order from the cnt-th to the cnt + n-1-th in the prediction candidate list, and then adds the value of cnt by n. I do. An example of the format of the prediction candidate list will be described later. Next, step 707
, The variable i is incremented by 1, and the process proceeds to step 704. As described above, i and m in step 704 are used.
As long as the condition is satisfied by the comparison of ax, the processing up to step 707 is repeated. For example, when i = 5, the reading “Hen” is sent to the prediction dictionary search unit 108, and FIG. The search result as described above is created.

FIG. 8 shows an example of the prediction candidate list.
As shown in the figure, the format of the prediction candidate list is, for each prediction candidate, the number, reading, notation, part of speech, notation frequency information, probability information, match length information, likelihood, usefulness, and po.
It is composed of information of int and judgment. Of these, reading, notation, part of speech, notation frequency information, probability information, and match length information are copied from the dictionary search result list,
Information on likelihood, usefulness, point, and determination is given in step 708 and subsequent steps. In the example of FIG. 8, first, when i = 0 in FIG. 7, the information of FIG. 6A is copied first in FIG. 8 at step 706, and when i = 5, the information of FIG. The information of b) is copied second in FIG. When the value of i is 1, 2, 4, or 6, the corresponding headline is not registered in the prediction dictionary 109, so that the search result is n = 0, and nothing is stored in the prediction candidate list. By the above processing, step 708 in FIG.
By now, the prediction candidate list as shown in FIG. 8 has been created.

Returning to FIG. 7, the processing after step 708 will be described. If the condition is not satisfied in step 704, the process proceeds to step 708, where 0 is substituted for the variable j. Next, in step 709, the values of the variable j and the variable cnt are compared. If j is smaller, the process proceeds to step 710. In step 710, the j-th candidate in the prediction candidate list is sent to the prediction candidate evaluation unit 110, and the result is written in the prediction candidate list. Write. For example, FIG. 9A shows an example in which the first information of the prediction candidate list in FIG. 8 is transmitted to the prediction candidate evaluation unit 110, and the transmitted result is written in the list. (B) is an example in which information is sent to the prediction candidate narrowing unit 111 and the result is written.

Here, the flow of processing of the prediction candidate evaluation unit 110 will be described with reference to the flowchart shown in FIG. First, in step 1101, the length of a candidate read character string in the prediction candidate information sent from the prediction control unit 107 is checked and assigned to the variable len. For example, when the first prediction candidate in FIG. 8 has been sent, len = 9 because len is the length of “Kanakan Henkan”.
Next, in step 1102, the matching length information is extracted from the prediction candidate similarly sent to the variable base and substituted. For example, in the case of the first candidate in FIG.
Becomes In step 1103, the (base-1) -th value of the probability information is assigned to the corresponding "likelihood" column of the prediction candidate list. For example, in the case of the first prediction candidate in FIG. 8, “the likelihood is“ 0 ”“ 7 ”“ 26 ”“ 67 ”“ 7
"5", "75", "75"
At this time, if a prediction dictionary in the format of FIG. 4 is used, the value of “likelihood” is calculated here. That is, the value of the notation frequency is calculated as (base− 1) A value obtained by multiplying the value obtained by dividing the value by the 100th value by 100 is assigned to “likelihood”.

Next, at step 1104, "usefulness" = len-base is calculated by the given formula, and is substituted in the corresponding "usefulness" column of the prediction candidate list. For example,
In the case of the first prediction candidate in FIG. 8, 9−7 = 2. Next, in step 1105, a representative evaluation point representing these evaluation points (“likelihood” and “usefulness”) is calculated based on the given equation. Figure 1
Although it is calculated by the formula as shown in FIG. 1, various calculation methods are possible, such as changing the value of a constant multiple for each item as needed. Here, since this equation is adopted, for example, for the first prediction candidate in FIG.
* 10 + 60 + 2 = 752. Similar processing is performed when the second candidate in FIG. 8 is sent, and as a result, the second result in FIG. 10A is obtained.

Here, the prediction dictionary shown in FIG. 4 holds the notation frequency information in the dictionary. Alternatively, the notation frequency information may not be held in the dictionary. In this case, the prediction candidate evaluation unit 110 replaces the last value of the reading frequency information with (ba
A value obtained by multiplying the value obtained by dividing the value by the (se-1) -th value by 100 is substituted for the “likelihood”. The prediction dictionary shown in FIG.
As the probability information in the dictionary, the information of the value obtained by dividing the notation frequency information by the frequency information of the partial reading character string is held, but separately from this, the reading frequency information of the heading is divided into the frequency information of each partial reading character string. It is also possible to hold the information of the value divided by. In this case, evaluation is performed using the prediction candidate evaluation unit 110 and the likelihood information.

In the method that does not use the notation frequency information, the likelihood information is not the likelihood of the notation character string for the partial reading character string but the likelihood of the headline reading character string for the partial reading character string. If there is a word, or if there is another segmentation break that results in the same reading character string as the heading reading character string, the likelihood of the likelihood information decreases,
If the heading of the prediction dictionary is a phrase-level expression such as an idiomatic expression and is a sufficiently long reading character string, only the frequency information of the heading reading character string and the partial reading character string can be used as the likelihood information. Can be used.

The flow of processing for narrowing down prediction candidates will be described with reference to the flowchart shown in FIG. First, in step 1201, the value of the likelihood information transmitted from the prediction control unit 103 is compared with a predetermined value. Here, if the likelihood value is 10 or more, the process proceeds to step 1202, and if it is smaller than 10, the process proceeds to step 1204. In step 1202, the value of the sent usefulness information is compared with a predetermined value. If the value of the usefulness is 2 or more, the process proceeds to step 1203. If the value is less than 2, the process proceeds to step 1204. In step 1203, the judgment result is returned to the prediction control unit 103 as “○”. In step 1204, the determination result is returned to the prediction control unit 103 as “x”. As a result, the likelihood is 10 or more and the usefulness is 2 or more, so that a judgment “” ”is given to the first candidate in the prediction candidate list in FIG. Therefore, the judgment “x” is given because the likelihood is smaller than 10. That is, FIG.
Is set as the lower limit value for the likelihood information calculated in step 1101, and only when the value of the likelihood is equal to or more than this lower limit is a candidate for prediction. Step 110 of FIG.
Only when the usefulness information calculated in step 2, ie, the length of the remaining reading character string that matches the partial reading character string of the predicted candidate reading character string, is equal to or greater than the set lower limit, the prediction is performed. Candidate.

Returning to FIG. 7, steps 712 and thereafter will be described. That is, after various evaluation points (likelihood and usefulness) and a representative evaluation point are assigned to each prediction candidate by the processing up to step 711, the processing proceeds to step 712 by the determination in step 709. Here, in step 712, from the candidate list, only those having a judgment result of ○ are selected, sorted in ascending order of the value of point, and the top x pieces of information are assigned to the output buffer in the sorting order. Since the format of the output buffer is the same as the format of the prediction candidate list, the description is omitted here. Here, x may be any positive integer such as 1, 3, 10, etc.
Here, it is assumed to be 3. In the prediction candidate list shown in FIG. 10B, since only one first prediction candidate has a judgment of “○”, this prediction candidate, that is, information such as “Kana-Kanji conversion” Is set in the output buffer. In step 713, the information of the output buffer is sent from the prediction control unit 103 to the control unit 102.

Next, another embodiment of the information registered in the prediction dictionary will be described with reference to FIG.

In this format, the parts of reading, notation, part of speech, and notation frequency information are the same as the prediction dictionaries shown in FIGS. The following prediction start character string length information is a feature of this mode. When the predictive dictionary search unit performs a dictionary search, the lower limit value of the number of characters to be read ahead of the received partial read character string is set. This is set for each registered headline. Specifically, the ordinal number of the column in which the value of the probability information column in FIG. 5 is 10 or more, that is, the number of characters to be read in which the probability information is 10 or more is given. That is, in this case, the reading frequency information and the probability information are not added to the prediction dictionary, but the prediction start character string length information is added instead.

Next, referring to FIG. 14, the prediction dictionary search unit 1 in the case of using the prediction dictionary according to the embodiment shown in FIG.
The flow of the process 08 will be described. First, in step 1401, a normal prefix search is performed in the prediction dictionary using the read character string received from the prediction control unit 103 as a key, and all candidates are copied to a buffer. The format of the buffer is the same as the search result list shown in FIG.
A column for prediction start character string length information is provided instead of the column for probability information. The number of candidates is substituted for the variable kouho, and the process proceeds to step 1402. In step 1402, the variable i
And 1 to a variable j. Next, step 1403
Then, the value of the variable i is compared with the value of the variable kouho. If the variable i is equal to or smaller than the variable kouho, the process proceeds to step 1404. If it is larger than the variable kouho, the process ends.
In step 1404, the length of the read-out character string of the i-th candidate in the buffer, that is, the matching length information, is substituted into the variable len1, and the predicted start character string length information of the i-th candidate in the buffer is substituted into the variable len2. I do. Proceeding to step 1405, the values of the variables len1 and len2 are compared, and if the variable len1 is greater than or equal to the variable len2, step 1406
Otherwise, if it is smaller, the process proceeds to step 1408. In step 1406, the i-th prediction candidate in the buffer is copied to the j-th search result list. The format of the search result list is the same as the format of the buffer described above. Next, after adding 1 to j in step 1407, the process proceeds to step 1408. In step 1408, i is added by 1, and the process returns to step 1403. As a result of the above processing, only a heading set for each registered heading and whose front part matches at least the lower limit of the number of characters to be read that should start with the received partial reading character string is left as a prediction candidate.

Next, referring to FIG. 15, the prediction control unit 103 in the case of using the prediction dictionary according to the embodiment shown in FIG.
The flow of the process will be described. First, in step 1501, the received read character string is sent to the read character string cutout unit 107. In step 1502, the result is received.
In step 05, information is sent to the prediction dictionary search unit 108, and
At 506, the result is set in the candidate list. Step 15
The processing from 02 to 1506 is the same as the processing from step 701 to step 706 in the flowchart shown in FIG. It should be noted that the candidate list does not need to include columns for information on “likelihood”, “priority”, “point”, and “judgment”. When i becomes equal to or greater than the value of max in step 1504, the process proceeds to step 1508.
In step 08, the candidate list is sorted in the order of the notation frequency, the top candidate is set in the output buffer, and the information of the output buffer is sent to the control unit 102 in step 1509. That is, in this case, the prediction candidate evaluation unit 110 that reads and calculates “likelihood”, “priority”, and “point”, and the prediction candidate narrowing unit 111 that narrows down the prediction candidates are not required.

Next, referring to FIG. 16, a display example of a prediction candidate and the movement of the control unit 102 related thereto will be described. FIG.
(C) is an example in which the control unit 102 receives the prediction result from the prediction control unit 103, displays the input reading information and the predicted kanji notation on a display screen, and presents it to the user. Here, the reading character string is input in the order of “KA”, “NA”, and “KA”, and when the input proceeds up to “KANA”, the input is performed by the processing of the prediction control unit 103 as described above. An example in which a prediction candidate is output is shown. According to the processing of the prediction dictionary 109 and the prediction control unit 103 described above, when the “?” Is input, the prediction control unit 1
No prediction candidates are sent from 03 to the control unit 102, and no prediction candidates are displayed as shown in FIG. Next, when "kana" is input, a reading character string "kana" is sent to the prediction control unit 103, and a prediction candidate "kanagawa prefecture" is returned from the prediction control unit 103 to the control unit 102. A candidate is displayed as shown in FIG. Here, when the next “KA” is input, the previous candidate display is cancelled, and the next read character string “KANAKA” is displayed in the prediction control unit 10.
3 and the prediction candidate “kana-kanji conversion” is returned from the prediction control unit 103 to the control unit 102, and FIG.
Is displayed as shown in FIG.

Here, when the user presses a prediction candidate adoption key, for example, the up arrow key ↑, the inverted prediction candidate among the displayed prediction candidates (if there is only one prediction candidate, the prediction candidate is used).
Is adopted, and the display of the display is changed as shown in FIG. Since the number of prediction candidates is not limited to one, a plurality of displays may be performed. At that time, the prediction buffer is displayed in the prediction candidate window in order from the top of the output buffer of the prediction control unit 103, and the first candidate is initially highlighted. Therefore, for example, when the user presses a prediction candidate moving key, for example, a right arrow key →, the prediction candidates to be highlighted are moved one by one, and when the desired key is pressed when the desired candidate is highlighted, the candidate is adopted. Is done.

Next, the processing of the control unit 102 relating to the transfer of the read character string to the kana-kanji conversion unit 104 and the prediction control unit 103 will be described with reference to FIG. First step 1
In step 701, the contents of the conversion target read character string buffer and the prediction target read character string buffer are initialized. Step 1702
Then, the process waits for a user's keyboard input. It is checked whether or not the input from the keyboard is a reading character input. If the input is not a reading character string, the flow proceeds to step 1704, and if the input is a reading character string, the flow proceeds to step 1707. Whether or not the input is a reading character can be performed by processing such as referring to a table given in advance. Here, it is assumed that the punctuation is not a reading character string. If the input is a so-called Roman character input, buffering is performed here until an input corresponding to one character is read.
In step 1707, the input one-character reading is added to the conversion-target reading character string buffer.
Proceed to 08. In step 1708, one character input to the predicted target character string buffer is added, and the process proceeds to the next step.
In step 1709, the read character string information in the predicted read character string buffer is sent to the prediction control unit 103, and the prediction process starts. The prediction control unit 103 performs the processing described above, and returns a result to the control unit 102. Prediction control unit 1
The control unit 102 having received the result from step 03 executes step 17
The process proceeds to step S10, where the result display process as described above is performed, and the process returns to step S1702.

In step 1704, it is checked whether or not the keyboard input is a predictive selection key, for example, the above-mentioned arrow key 、. If so, the process proceeds to step 1711; otherwise, the process proceeds to step 1705. . In step 1711, a part of the prediction candidate selected by the selection key that is not the input reading character string from the keyboard, that is, the prediction-completed reading character string is added to the conversion target reading character string buffer. For example, in the examples of FIGS. 16C to 16D, the steps up to the reading character string of “Kanaka”
The reading character string stored in the conversion target reading character string buffer 707 is added to the reading character string “Kanjihenkan” in step 1711. As a result, the reading character string “Kanakanjikan” becomes the prediction target reading character. Stored in the column buffer. Similarly, in step 1712, the predicted and complemented reading character string is added to the prediction target reading character string buffer. In the examples of FIGS. 16C to 16D, “Kanaka” has already been set.
The reading character string up to the reading character string is stored in the prediction target reading character string buffer in step 1707, and as a result of adding the reading character string “kanjihankan” in step 1711, the reading character “kankanjihenkan” is obtained. The character string is stored in the predicted reading character string buffer.

Next, in step 1713, the control unit 102 sends the information on the character string buffer to be converted to the kana-kanji conversion unit 104, and the kana-kanji conversion process starts. The kana-kanji conversion unit 104 returns the result to the control unit 102. Control unit 1
02 proceeds to step 1714 to perform display processing. For the Kana-Kanji conversion processing unit and the related display processing,
Since it is widely known, it will not be described here. Next, the process proceeds to step 1718, and after the content of the conversion target read character string buffer is cleared, the process returns to step 1702. Next, in step 1705, if the input from the keyboard is a conversion instruction key, for example, a conversion key, the process proceeds to step 1713, and so on. Otherwise, the process proceeds to step 1706. Next, in step 1706,
When the input from the keyboard is the punctuation key, the process proceeds to step 1715. If the input is any other key, the process proceeds to another process defined separately. In step 1715, the information of the conversion target read character string buffer is sent to the kana-kanji conversion unit 104, and in step 1716, the result is received and a display process is performed. Subsequently, after clearing the prediction target read character string buffer in step 1717, the process proceeds to step 1717.
The program proceeds to step S18, where the information in the conversion target read character string buffer is cleared, and the procedure returns to step S1702.

By repeating the above processing, the control unit 102 converts the information on the input read character string and the predicted and complemented read character string into the kana-kanji conversion unit 104 and the prediction control unit 10.
3 and receive the result to proceed to the next process.
For example, when the reading character string “so” is input according to the state of FIG. 16D, the reading character string “so” is added to the prediction target reading character string buffer in step 1708, and in step 1709 The read character string “Kanakanjikankanso” is sent to the prediction control unit 103.

With the processing described above, the prediction control unit 103 creates a prediction candidate list as shown in FIG. 18 for the partial character string “ Returned to control unit 102 via FIG. 1 shows the result of the display processing performed in step 1710.
9 (a), the display after this is selected by the の key is shown in FIG.

Here, for example, as shown in FIG. 10 (b), for the input of "to Kanagawa", the "converter"
The candidate "" is not displayed because the determination is "x", but when the input is determined up to "Kanakan Henkanso", the candidate "conversion device" is determined to be ○ and the candidate is displayed. Also, for example, "sa""i""ki""n""no"
When the input of “KA”, “NA” and “KA” is advanced, step 1
In step 709, the reading character string "Saikin no Kanaka" is sent to the prediction control unit 103, and the result of the processing on the partial reading character string "Kanaka" is performed in the same manner as in the previous input example of "Kanaka". The prediction result returns from the prediction control unit 103 and is displayed as shown in FIG. 20A. If a candidate is selected by using a prediction adoption selection key, for example, the ↑ key, the process proceeds to step 1713 and proceeds to step 1713. The reading character string “Nonkanakanjihenkan” is sent to the kana-kanji conversion unit 104, and as a result, a conversion result “recent kana-kanji conversion” is displayed as shown in FIG.

Here, the prediction control unit 103 shown in FIG.
In the process of step 1712, after the process of step 1712, step 1
In 713, the information of the conversion target reading character string buffer is sent to the Kana-Kanji conversion unit. As another form, as shown in the flowchart of FIG. 24, Kana-Kanji conversion is not performed immediately after Step 1712, and the process proceeds to Step 1719. , On the display, the part of the reading character string in the conversion target reading character string buffer that is not included in the prediction candidate is read,
It is also possible to perform a process of displaying the prediction candidates in kanji kana notation. For example, "sa""i""ki""n"
In the example of “no”, “ka”, “na”, and “ka”, when the prediction adoption selection key is pressed after the display as shown in FIG.
As shown in FIG. 25, it is displayed as “Saiki no kana kanji conversion”. Thereafter, the process returns to step 1702, and the rest of the readings for which the input from the user was a conversion instruction key or a punctuation mark for which the remaining Kanji Kana notation was undetermined are also converted to Kanji Kana notation.

Next, another form of the prediction dictionary in the case where a read character string including information on a read character string that has been predicted and complemented is transferred to the prediction control unit 103 will be described with reference to FIG. In this figure, the format from the reading information to the probability information is the same as that of the prediction dictionary shown in FIG. 5, but in addition, each heading has information on a dividable position.

FIG. 22 shows the format of the read target character string buffer described in the flowchart of FIG. 17 when a prediction dictionary having dividable position information is used. here,
The prediction target read character string buffer has dividable position information in addition to the information of the read character string buffer. Step 171
In step 2, when adding the predicted and completed reading character string to the prediction target reading character string buffer, reference is made to the dividable position information of the corresponding prediction candidate, and the corresponding position of the reading character string in the prediction target reading character string buffer is read. The information is calculated, and the dividable position information is written as shown in FIG. For example, when a prediction candidate of “kana-kanji conversion” is selected in response to the input of “kana”, in step 1712, a reading character string of “njihenkan” is added to the reading character string buffer, and dictionary information “ It is checked whether "Kanakanjikan" starts from the m-th character of "Kanakanjikan" in the reading character string buffer (in this case, m = 0), and the value of m + 1, that is,
In this case, “1” and the dividable position “6” which is the dictionary information
Is added to m, in this case, "6", and the value obtained by adding the length of the read character string "Kanakanjikan" to m + 1, in this case, "10" is written in the dividable position information.

FIG. 23 is a flowchart showing the flow of processing of the read character string cutout unit 107 when a prediction target read character string buffer having dividable position information is used. The processing from step 2301 to step 2306 in the figure is the same as that from step 201 to step 206 in FIG. Step 2307
Then, it is checked whether or not the value of the variable “start” matches the dividable position information of the prediction target read string buffer. If the value of the variable “start” matches, the process proceeds to step 2308 to start from the start character of the read string buffer. Up to the len-th character is copied to the i-th character string in the read character string buffer. If step 230
If the value does not match in step 7, the process directly proceeds to step 2309. In step 2309, the value of start is set to 1
Then, the process returns to step 2306.

By the above processing, the character string up to the character string "Kanakanjikankan" has been determined.
According to the reading character string cutout processing shown in FIG. 2, when the reading of “Yomi” is input, “Kanakan Henkanso”, “Nakanji Henkanso”, “Kanji Henkanso”, “Kanji Henkanso”, Nine partial reading character strings of "Henkanso", "Henkanso", "Kanso", "Nenso" and "So" are cut out,
Since a predictive dictionary search is performed for each of these, unnecessary dictionary search processing time is required. In addition, for example, there is a candidate such as "dryer" in the predictive dictionary, which satisfies the conditions and is displayed as a predictive candidate. In this case, as shown in FIG. 16D, a candidate “dryer” is displayed where a candidate “kana-kanji conversion” is already displayed, and unnecessary candidates are displayed. That means
According to the processing shown in FIG. 23, “Kanakan Henkanso”
Since only three partial reading character strings “Henkanso” and “So” are cut out and a prediction dictionary search is performed, unnecessary processing time is omitted and unnecessary prediction candidate display is suppressed. And the quality of processing is improved.

Although the present invention has been described by exemplifying only Japanese input, the present invention is not limited to this and can be applied to, for example, machine translation.

[0057]

As described above, according to the present invention, the prediction control unit 10
3, a list of partial reading character strings corresponding to the reading character string sent to the reading character string cutout unit 107 is created, and the prediction dictionary search unit 108 predicts a prediction candidate having a forward match for each of them. , A prediction candidate list is created, and further, the frequency information of the partial character string of the reading character string, the frequency information of the heading kanji kana notation character string, or the heading for the partial reading character string stored in the prediction dictionary 109 Based on the probability information of the Kanji Kana notation, the likelihood information of the heading Kanji Kana notation with respect to the read character string that matches forward is calculated by the prediction candidate evaluation unit 110, and the evaluation points based thereon are calculated, and the prediction control unit 103 predicts the prediction candidates in the order of the evaluation points. Are prioritized, and the top n places are presented to the user, thereby automatically performing input prediction from an arbitrary position during text input. By being able to present the results to the user prioritizes high precision, increases the efficiency of document input, it is possible to perform Japanese input easily.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a Japanese predictive input system according to the present invention;

FIG. 2 is a flowchart showing a processing flow of a reading character string cutout unit according to the Japanese predictive input method of the present invention;

FIG. 3 is a diagram showing a format of a reading character string list and a storage example of a processing result;

FIG. 4 is a diagram showing an embodiment of information registered in a prediction dictionary,

FIG. 5 is a diagram showing another embodiment of information registered in a prediction dictionary,

FIG. 6 is a diagram showing a format of a dictionary search result list and a storage example of a processing result;

FIG. 7 is a flowchart showing a processing flow of a prediction control unit for realizing the Japanese prediction input method of the present invention;

FIG. 8 is a diagram showing a format and a storage example of a prediction candidate list;

FIG. 9 is a diagram showing a storage example of a prediction candidate list;

FIG. 10 is a diagram showing a storage example of a prediction candidate list;

FIG. 11 is a flowchart showing a processing flow of a prediction candidate evaluation unit for realizing the Japanese prediction input method of the present invention;

FIG. 12 is a flowchart showing a processing flow of a prediction candidate narrowing-down unit for realizing the Japanese prediction input method of the present invention;

FIG. 13 is a diagram showing another embodiment of information registered in the prediction dictionary;

FIG. 14 is a flowchart showing a processing flow of a predictive dictionary search unit for realizing the Japanese predictive input method of the present invention;

FIG. 15 is a flowchart showing a processing flow of a prediction control unit for realizing the Japanese prediction input method of the present invention;

FIG. 16 is a diagram showing a display example of prediction candidates;

FIG. 17 is a flowchart showing a processing flow of a control unit for realizing the Japanese predictive input method of the present invention;

FIG. 18 is a diagram showing a storage example of a prediction candidate list,

FIG. 19 is a diagram showing a display example of prediction candidates.

FIG. 20 is a diagram showing a display example of prediction candidates;

FIG. 21 is a diagram showing another embodiment of information registered in the prediction dictionary,

FIG. 22 is a diagram showing a format and a storage example of a prediction target read character string buffer;

FIG. 23 is a flowchart showing a processing flow of a reading character string cutout unit for realizing the Japanese predictive input method of the present invention;

FIG. 24 is a flowchart showing a processing flow of a control unit for realizing the Japanese predictive input method of the present invention;

FIG. 25 is a diagram showing a display example of prediction candidates.

[Explanation of symbols]

101 input unit, 102 control unit, 103 prediction control unit, 104 kana-kanji conversion unit, 105 kana-kanji conversion dictionary search unit, 106 kana-kanji conversion dictionary, 107 reading character string cutout unit, 108 prediction dictionary search unit , 109: prediction dictionary, 110: prediction candidate evaluation unit, 111: prediction candidate narrowing unit

Claims (22)

[Claims] 1. A reading character string is input, at least a part of a partial character string group is extracted from the input or predicted and completed reading character string, and a kanji kana notation in which the extracted partial character string and the reading match in front is obtained. A search is performed using a dictionary, and priority is given to the kanji kana notation searched using the frequency information for the kanji kana notation character string in the dictionary and the frequency information for at least a part of the partial character string of the reading character string. A Japanese predictive input method characterized by Japanese predictive input. 2. A reading character string is input, at least a partial character string group is extracted from the input or prediction-completed reading character string, and a kanji kana notation in which the extracted partial character string and the reading are in front of each other is obtained. Searching using a dictionary, prioritizing the kanji Kana notation searched using the likelihood information for each of the kanji kana notation character strings in the dictionary that matches the leading character string, and inputting Japanese prediction A Japanese predictive input method characterized in that: 3. The likelihood information for each reading character string that matches forward with respect to a kanji kana written character string in the dictionary is expressed by frequency information for the kanji kana written character string and frequency for a partial character string of the read character string. 3. The method according to claim 2, further comprising: calculating from the information, and prioritizing the kanji kana notation retrieved using the calculated likelihood information. 4. A reading character string is input, and at least a partial character string group is extracted from the input or prediction-completed reading character string. Searching using a dictionary, prioritizing the searched kanji kana notation using frequency information for at least a part of the partial character string of the reading character string in the dictionary, and performing Japanese prediction input. Japanese prediction input method. 5. A reading character string is input, and at least a partial character string group is extracted from the input or prediction-completed reading character string, and a kanji kana notation in which the extracted partial character string and the reading match in front is obtained. Searching using a dictionary, prioritizing the kanji kana notation searched using the likelihood information for each of the reading character strings that match the head of the reading character string in the dictionary, and performing Japanese prediction input Japanese input prediction method characterized by the following. 6. The likelihood information for each of the reading character strings that match the head of the kanji kana notation character string in the dictionary is calculated from the frequency information for the partial character string of the reading character string, and this likelihood information is used. 6. The method according to claim 5, wherein priorities of the kanji kana notations searched are determined. 7. The Japanese predictive input method according to claim 1, further comprising the step of outputting the top n places (n is arbitrary) in the prioritized Kanji Kana notation. 8. A dictionary in which a kanji kana notation constituting a part of a Japanese sentence and its reading character string are stored, means for inputting a reading character string, and a partial character string group based on the input or prediction-completed reading character string. Means for retrieving a part or the whole, means for retrieving from the dictionary a vague notation whose reading is in agreement with the retrieved partial character string, frequency information for a kanji kana notation character string in the dictionary, Means for prioritizing kanji kana notations retrieved using frequency information for all or part of a partial character string of a reading character string. 9. A dictionary in which a kanji kana notation constituting a part of a Japanese sentence and its reading character string are stored, means for inputting the reading character string, and a partial character string group based on the input or prediction-completed reading character string Means for retrieving a part or all of the characters, means for retrieving from the dictionary a kanji kana notation whose reading matches the retrieved partial character string, and a reading which matches the kanji kana notation character string in the dictionary. Means for prioritizing kanji kana notations retrieved using likelihood information for each character string. 10. The likelihood information for each reading character string whose head matches with the Kanji Kana notation character string in the dictionary,
Means for calculating from the frequency information for the kanji kana notation character string and frequency information for the partial character string of the reading character string, and means for prioritizing the kanji kana notation searched using the likelihood information The Japanese predictive input system according to claim 9, comprising: 11. A dictionary that stores Kanji Kana notation and its reading character string that constitutes a part of a Japanese sentence, a means for inputting a reading character string, and a partial character string group based on the input or prediction-completed reading character string. Means for retrieving a part or all of the part, means for retrieving from the dictionary a kanji kana notation whose reading matches the retrieved partial character string in front, and a part of the partial character string of the read character string in the dictionary or Means for prioritizing kanji kana notations retrieved using frequency information for all of them. 12. A dictionary storing kanji kana notation and its reading character string constituting a part of a Japanese sentence, means for inputting a reading character string, and a partial character string group based on the input or prediction-completed reading character string. Means for retrieving a kanji kana notation whose pronunciation matches with the retrieved partial character string from the dictionary, and likelihood information for each of the reading character strings whose prefix matches the reading character string in the dictionary. Means for prioritizing kanji kana notations retrieved by using a Japanese language predictive input system. 13. A means for calculating likelihood information for each reading character string whose head matches with a kanji kana notation character string in the dictionary from frequency information for a partial character string of the reading character string, 13. The Japanese predictive input system according to claim 12, further comprising means for prioritizing kanji kana notations searched using the information. 14. The Japanese predictive input system according to claim 8, further comprising: means for outputting the n most significant places (n is arbitrary) in the prioritized Kanji Kana notation. 15. A method of predicting a character string to be inputted by using a character already inputted as a clue, a step of capturing a reading character string to be used, which is used in a Japanese predictive input system for inputting; Extracting at least a partial character string group from the predicted and completed reading character string, searching for a kanji kana notation in which the cut-out partial character string and the reading match in front using a dictionary, and Kanji Kana notation searched using frequency information for a certain Kanji notation character string and frequency information for at least a part of a partial character string of the reading character string, or frequency information for at least a part of the reading character string in the dictionary And computer-readable recording medium in which the step of prioritizing is programmed and recorded. 16. A step of predicting a character string to be inputted by using a character already inputted as a clue, used in a Japanese predictive input system for inputting, and capturing a reading character string to be inputted; Extracting at least a partial character string group from the predicted and completed reading character string, searching for a kanji kana notation in which the cut-out partial character string and the reading match in front using a dictionary, and The likelihood information of each kanji kana spelling character string that matches forward with respect to a certain kanji character string, or the kanji kana notation searched using the likelihood information of each kanji kana notation character string that matches the leading characters of a reading character string in the dictionary. A computer-readable recording medium in which the step of prioritizing is programmed and recorded. 17. A step of initializing a read character string list and a read character string buffer, a step of setting an input read character string in the read character string buffer, and a step of setting a character string length set in the read character string buffer. Storing a partial reading character string including the last character of the reading character string one by one from the beginning of the input reading character string in the above reading character string list. 17. The recording medium according to claim 15, wherein: 18. A step of receiving the contents of a reading character string list and the number of partial reading character strings thereof, using the number of partial reading character strings as a variable, performing a forward match search of a dictionary with the partial character strings sent one by one, Storing the search result in a dictionary search result list; copying the n search results stored in the search result list into a prediction candidate list; Performing a prediction evaluation based on the reading frequency information or by calculating at least the likelihood in addition to the information to be copied, and writing the predicted evaluation in a corresponding portion of the prediction candidate list. Item 17. The recording medium according to Item 15 or 16. 19. A step of examining the length (len) of a candidate read character string and extracting matching length information (base) from the prediction candidate information sent via the prediction candidate list, -1) or the notation frequency included in the dictionary is read as (base-
(1) calculating a value as a likelihood;
17. The computer-readable program according to claim 16, wherein a step of calculating usefulness by calculating (n-base) and a step of performing arbitrary point calculation based on likelihood and usefulness and performing prediction evaluation are programmed and recorded. recoding media. 20. A step of comparing the transmitted likelihood information with a predefined first value, and comparing the transmitted usefulness information with a predefined second value. 17. The computer-readable recording medium according to claim 16, wherein a step and a step of narrowing down prediction candidates according to a result of the comparison are programmed and recorded. 21. Performing a forward match search using a dictionary using the received read character string as a key and copying candidates to a buffer; and using the forward match candidate as a variable, an i-th candidate forward match read character in the buffer. Column length and buffer i
A step of comparing the i-th candidate with the predicted start character string length information stored in the search result list, and, depending on the result of the comparison, reflecting the i-th content of the buffer as a predicted candidate in a corresponding portion of the search result list The computer-readable recording medium according to claim 15, wherein the steps of programming and recording are programmed and recorded. 22. A Japanese predictive input system capable of predicting and inputting a character string to be input based on a character already input as a clue, wherein a conversion target reading character string buffer and a prediction target reading character string buffer are used. Initializing, checking the input of the reading character, adding the input reading character to the conversion target reading character string buffer, and when instructing prediction other than the reading character input, the conversion target reading character string buffer The predicted reading is
A step of adding a predicted and completed reading character to a prediction target reading character string buffer; a prediction instruction other than a reading character input;
17. A computer-readable program according to claim 15, wherein when a conversion instruction is issued, a step of performing a kana-kanji conversion process of a reading character string in a conversion target reading character string buffer and displaying a result on a screen is programmed and recorded. recoding media.