JP2000099635A - Character string prediction device and character string prediction method - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To appropriately limit the number of candidates for a predicted character string. A pattern recognition unit (103) that recognizes an input character pattern by referring to a pattern recognition dictionary (104) and outputs recognition candidate characters, a word dictionary (108) storing at least readings and corresponding headings, Dictionary 10
8, a base character string is generated by combining a dictionary search unit 107 that searches for a predicted character string having a base character string at the beginning with a recognition candidate character output by the pattern recognition unit 103, and generates the generated base character string. Of the recognition candidate characters constituting the base character string, and instructs the dictionary search unit 107 to search for the predicted character string by instructing the dictionary search unit 107 on a base character string that satisfies a preset prediction start condition regarding the number of characters for each character type. And a prediction control unit 106.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character string predicting apparatus and a character string predicting method for predicting a character string to be input by a user following the character string based on the input character string.

[0002]

2. Description of the Related Art Input prediction refers to searching a word dictionary or the like when the first few characters are input, as reported in the literature (the 48th Information Processing Society of Japan National Convention Proceedings 4J-11 and others). Is a character input method of extracting and presenting a character string (word) starting with the several characters. When an arbitrary character string is selected from the presented (displayed) character string, the character string is determined and displayed as an input character string.

[0003] The purpose of input prediction is to input only the first few characters, and the subsequent character strings are complemented, so that it is not necessary to input all the character strings intended by the user, and input operations are reduced. It is in.

However, in the case of character input by pattern recognition such as handwritten character recognition or voice recognition, the input character cannot be obtained because of ambiguity at the time of recognition (recognition characters are not uniquely determined and a plurality of recognition candidate characters are obtained). Since the column is not determined, there remains a problem that a character correction operation for correcting an erroneously recognized character to a correct character and determining an input character string must be performed.

To solve such a problem, there is a technique (Japanese Patent Application No. 8-243292) for predicting a character string following all combinations of recognition candidate characters as a result of recognition of an input pattern.

[0006]

However, if the character string following each of the combinations of the candidate character as a result of recognition of the input pattern is predicted using the conventional technique, the input character string is erroneously recognized. Even if is not determined, it is easier to obtain the character string intended by the user,
There is a problem in that the number of predicted character string candidates becomes enormous.

[0007] For example, if there are n recognition candidate characters in the recognition result for each input pattern, the combinations are the number of input characters x n. Also, how to set the timing of starting the input prediction is one of the factors that causes the number of predicted character strings to be searched to be enormous.

For example, if prediction is performed at the time of input of one character, in the case of kanji, there are not many problems because there are many types of kanji, so there is not much problem. In the case of, the character string searched from the dictionary may be hundreds or thousands.

The present invention has been made in view of the above circumstances, and provides a character string prediction apparatus and a character string prediction method capable of appropriately limiting the number of candidates for a predicted character string. Aim.

[0010]

According to the present invention, a character is input by recognizing an input character pattern with reference to a pattern recognition dictionary, and the input character string is used as a base character string, and a subsequent character string is input. The prediction is performed by referring to the word dictionary, and among the base character strings formed by combining the recognition candidate characters of the recognition result, the character types of the recognition candidate characters constituting the base character string are checked, and the number of characters is preset for each character type. It is characterized in that a predicted character string is searched from a word dictionary for a base character string that satisfies a prediction start condition.

[0011] Thus, in the base character string formed by combining the recognition candidate characters as a result of recognition of the input pattern, a condition relating to the number of characters for starting prediction for each character type is set, and the base character string satisfying the condition is set. By predicting the character string for, the number of predicted character strings to be searched is limited, and the user can easily input the intended character string.

Further, according to the present invention, a character is input by recognizing an input character pattern with reference to a pattern recognition dictionary, and the input character string is used as a base character string, and subsequent character strings are referred to a word dictionary. The evaluation value is calculated for each base character string formed by combining the recognition candidate characters of the recognition result, and the base character string having a high evaluation so that the number of predicted character strings searched is within a preset range. In which a predicted character string is searched from the word dictionary.

Thus, among the base character strings formed by combining the recognition candidate characters obtained as a result of recognizing the input pattern, the base character strings having the highest evaluation within the preset number of predicted character strings are successively arranged. Is predicted, the number of predicted character strings to be searched is limited, and the user can easily input the intended character string.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a character string prediction device according to the present embodiment. The character string prediction device in the present embodiment is, for example, a CD-ROM,
It is realized by a computer which reads a program recorded on a recording medium such as a DVD or a magnetic disk and whose operation is controlled by the program. In the present embodiment, handwritten character recognition will be described as an example of pattern recognition.

As shown in FIG. 1, a character string predicting apparatus according to this embodiment includes an input device 101, a character input unit 102,
Pattern recognition unit 103, pattern recognition dictionary 104, input buffer 105, prediction control unit 106, dictionary search unit 10
7, a word dictionary 108, a candidate buffer 109, an output control unit 110, an output buffer 111, and an output device 112.

The input device 101 is a handwriting tablet for inputting a character pattern as coordinate data. For example,
Character data is input by writing on the tablet surface with a pen or the like, and coordinate data representing a character pattern is input.

The character input unit 102 cuts out a character pattern input by the input device 101, causes the pattern recognition unit 103 to execute character recognition for the character pattern, and stores a recognition candidate character obtained as a recognition result in the input buffer 105. I do. In the present embodiment, it is assumed that ten recognition candidate characters are obtained by character recognition by the pattern recognition unit 103.

The pattern recognition unit 103 includes the character input unit 10
Under the control of 2, the input character pattern is recognized with reference to the pattern recognition dictionary 104, and a recognition candidate character is output as a recognition result.

The pattern recognition dictionary 104 stores characters and their corresponding character patterns. The prediction control unit 106 has a control function for appropriately restricting the number of candidates for a predicted character string, and generates a base character string by combining recognition candidate characters stored in the input buffer 105.

Subsequently, in the case of the first embodiment described later, the prediction control unit 106 checks the character type of the recognition candidate character constituting the base character string, and determines the number of characters for each character type in the generated base character string. The dictionary search unit 107 instructs the dictionary search unit 107 to search for a predicted character string that satisfies the prediction start condition set in advance. For example, in the case of the first embodiment, prediction is started with one character when the recognition candidate character is a kanji character, and with two characters when the recognition candidate character is any other character type.

Further, the prediction control unit 106 controls a second
In the embodiment, an evaluation value is set for each base character string, and a predicted character string is searched from a base character string having a high evaluation value such that the number of predicted character strings to be searched is within a preset range. For example, in the case of the second embodiment, the evaluation value of the base character string is an average value of the evaluation values for each character obtained at the time of character recognition of the recognition candidate character forming the base character string.

The prediction control unit 106 stores a predicted character string obtained by searching the word dictionary 108 by the dictionary search unit 107 in the candidate buffer 109. In this embodiment, the maximum number of predicted character strings to be searched is set to 50.

The dictionary search unit 107 searches the word dictionary 108 for a character string having the base character string at the beginning, and stores this character string in the candidate buffer 109 as a predicted character string.
The word dictionary 108 stores information such as headings and readings for each word (character string).

The output control unit 110 controls the candidate buffer 109
Presents (displays) the predicted character string stored in
A candidate selection process for determining a character string to be input is performed, and the predicted character string selected by the user is stored in the output buffer 111. After that, the output control unit 110 outputs the predicted character string stored in the output buffer 111 to the output device 1.
12 to be displayed on the screen.

The output device 112 is a screen display device including a CRT, an LCD display, and the like, and allows the user to interactively proceed with character input processing.

Next, the operation of the first embodiment will be described. FIG. 2 is a flowchart illustrating a flow of processing from character input to output of a predicted character string of the character string prediction device according to the first embodiment.

First, the character pattern input from the input device 101 is cut out by the character input unit 102 for each character. The cut-out character pattern is recognized by the pattern recognition unit 103 with reference to the pattern recognition dictionary 104 (steps A1 and A2), and recognition candidate characters of the recognition result are stored in the input buffer 105 (step A3). In the present embodiment, the recognition candidate character of the recognition result is 10
Characters.

When the character recognition is completed, the prediction control unit 106
In step A4, a base character string combining the recognition candidate characters is generated (step A4). Here, it is checked whether or not any of the base character strings satisfies the prediction start condition (step A5). In the present embodiment, the prediction start condition is set such that the prediction starts when one character is input for a kanji character, and starts when two characters are input for other character types. Shall be. As a result of checking the base character string, if there is no base character string that satisfies the prediction start condition, the process returns to the character input processing.

On the other hand, if there is a base character string that satisfies the prediction start condition, the prediction start dictionary search unit 107 searches the predicted character string with reference to the word dictionary 108 (step A6). Is stored in the candidate buffer 109 (step A7).

The search processing of these predicted character strings is performed for all base character strings that satisfy the prediction start condition (steps A6 to A8). When the search for the predicted character string is completed, the prediction control unit 106 causes the output device 112 to display the predicted character string via the output control unit 110, and presents the predicted character string corresponding to the input character to the user. A candidate selection process is performed (step A9).

As a result, if the predicted character string has not been selected, the process returns to the character input processing to input the next character. On the other hand, if any of the displayed predicted character strings is selected, the output buffer 1
11 stores the selected predicted character string (step A1).
0, A11), the predicted character string selected by the output control unit 110 is displayed on the screen (step A12).

Finally, each buffer is initialized in preparation for the next input (step A13). FIG. 3 shows a state from the character input to the character string prediction of the character string prediction device in the first embodiment.

In the first embodiment, the number of recognition candidate characters in the recognition result is ten. The prediction start condition regarding the number of characters for each character type is one for kanji characters, and two for other character types.

In the example of the display screen shown in FIG. 3, an area where a character pattern is input by handwriting and an area where a character string whose input is confirmed are displayed are provided. Further, an area for displaying a character string obtained by the input prediction processing (predicted character string display) is provided as necessary.

As shown in FIG. 3A, the katakana "ko"
It is assumed that, when a character pattern intended to be input is input, the recognition result is “ko, a, yu, u, 2], 5, yo, yo, sword”. When one character is input, only the kanji is the base character string that satisfies the prediction start condition. Therefore, among the recognition candidate characters, “5” and “sword” are the base character strings that satisfy the prediction start condition.

Therefore, a predicted character string having "5" and "sword" as base character strings is searched. That is, "five" and "sword"
Character strings starting with the characters
08 by retrieving it.

The obtained predicted character string is shown in FIG.
Is displayed as a list of predicted character strings.
In the example shown in FIG. 3B, eight predicted character strings are displayed. However, when more than eight predicted character strings are searched, another predicted character string is changed by an instruction to the scroll bar. Can be displayed.

Subsequently, when a character pattern intended for katakana "n" is input, the recognition result is shown in FIG.
As shown in ", d, n, si,;, i, j, et al.,>,
" When two characters are input, the two kana characters also satisfy the prediction start condition, and are generated by combining a kana character in the first candidate character and a kana character in the second candidate character. A predicted character string is searched based on the base character string to be obtained.

Here, it is assumed that a predicted character string has been obtained by a search based on the base character strings of “koni”, “kon”, and “koshi” as shown in FIG. The acquired predicted character strings are displayed in a list as a predicted character string display as shown in FIG.

When the two characters shown in FIG. 3C are input, a predicted character string having "5" or "sword" as a base character string is out of the candidates. This is because, among the recognition candidates for, there was no one that succeeded in the dictionary search among the base character strings formed in combination with “5” or “sword”.

As described above, in the base character string formed by combining the recognition candidate characters obtained as a result of recognition of the input pattern, a prediction start condition relating to the number of characters for starting prediction for each character type, for example, 1 for kanji A prediction start condition is set such that prediction starts when a character is input, and in the case of other character types, prediction starts when two characters are input, and a character string is set for a base character string that satisfies the condition. By performing the prediction, the number of predicted character strings to be searched is limited, so that the user can more easily input an intended character string.

In the above description of the first embodiment,
The description has been made assuming that the prediction start condition starts when one character is input for a kanji character and two characters are input for other character types, but other prediction start conditions may of course be used.

For example, the length of a word (character string) stored in the word dictionary 108 is 2 to 2 as shown in FIG.
In the case of about four characters, if the prediction start condition is set to two or three characters, the effect of input prediction cannot be obtained, but if the length of the character string stored in the word dictionary 108 is longer, It is assumed that an appropriate number of characters can be set as a prediction start condition according to how many input characters are searched and the number of predicted character strings is narrowed down.

In addition, there are two classifications of Kanji and other character types. In addition, a dictionary of other character types such as a dictionary in which English character strings are registered is prepared, and prediction is started for each dictionary of each character type. Conditions can be set.

Next, the operation of the second embodiment will be described. FIG. 4 is a flowchart illustrating a flow of processing from character input to output of a predicted character string of the character string prediction device according to the second embodiment.

First, the character pattern input from the input device 101 is cut out by the character input unit 102 for each character. The cut-out character pattern is subjected to character recognition in the pattern recognition unit 103 with reference to the pattern recognition dictionary 104 (steps B1 and B2), and recognition candidate characters of the recognition result are stored in the input buffer 105 (step B3). In this embodiment, the recognition candidate character of the recognition result is 1
Use 0 characters. Further, an evaluation value obtained by collation with the pattern recognition dictionary 104 at the time of character recognition, that is, a similarity value (for example, indicated by 256 levels) with a standard pattern stored in the pattern recognition dictionary 104 is also stored in the input buffer 105. Shall be.

When the character recognition is completed, the prediction control unit 106
, A base character string combining the recognition candidate characters is generated (step B4). Subsequently, the evaluation value of each base character string is calculated using the evaluation value (similarity value) at the time of recognition of the recognition candidate character constituting the base character string, and the base character strings are sorted in descending order of the evaluation value. (Step B5).

Next, a predicted character string is searched by referring to the word dictionary 108 in order from the base character string having the highest evaluation value (step B6). When the dictionary search is completed, it is checked whether the number of predicted character strings searched and the number of predicted character strings stored in the candidate buffer 109 are within the range of a preset number of character strings (step B7). If it is within the range, the searched predicted character string is stored in the candidate buffer 109 (step B8). On the other hand, if it exceeds the preset range, the dictionary search process is terminated at that point.

The dictionary search processing of these predicted character strings is performed until all base character strings have been searched or the number of searched predicted character strings exceeds a predetermined range (steps B6 to B9).

When the search for the predicted character string is completed, the prediction control unit 106 presents the predicted character string obtained based on the base character string to the user and performs a candidate selection process (step B1).
0).

As a result, if the predicted character string has not been selected, the process returns to the character input processing to input the next character. On the other hand, when any of the predicted character strings is selected, the selected predicted character string is stored in the output buffer 111 (steps B11 and B12), and the selected predicted character string is displayed on the screen in the output control unit 110. (Step B13). Finally, each buffer is initialized in preparation for the next input (step B14).

FIG. 5 shows a state from the character input to the character string prediction of the character string prediction device in the second embodiment. In the second embodiment, the recognition candidate character of the recognition result is 1
It is assumed to be zero. It is assumed that the evaluation value of the base character string uses an evaluation value (similarity value) obtained at the time of recognition of the recognition candidate character constituting the base character string, and the maximum number of predicted character strings to be searched is 50.
Number. The form of the display screen is the same as that used in the first embodiment.

As shown in FIG. 5 (a), when a character pattern intended for "sha" is input, the recognition result is "thanks, shrine, welfare, forest, sou, yu, ho, pray, red" , Jiang ". Since the recognition candidate characters are all Chinese characters, they can be base character strings for searching for a predicted character string. Therefore, by searching the word dictionary 108 for each character, a predicted character string having each character at the head is searched. As a result, FIG.
As shown in, among the recognition candidate characters, "Rei, Sha, So, Yu,
The predicted character string starting with “radiation, prayer, red, and river” is the word dictionary 108.
, And the number of all prediction candidates exceeds the preset number of predicted character strings of 50.

In this case, as shown in FIG. 6A, a predicted character string is searched in order from a base character string having a high evaluation value. In the example shown in FIG. 6A, first, only the predicted character string starting with the base character string “Rei” is searched. If the number of predicted character strings searched here is less than 50, a predicted character string starting with a base character string “company” having the next highest evaluation value is searched.
Hereinafter, similarly, the predicted character strings are obtained in the order of the base character strings having the higher evaluation values until the preset number of the predicted character strings is satisfied.

As a result, as shown in FIG. 5B, a list is displayed as a predicted character string display. In the example shown in FIG. 5B, the predicted character string searched based on the base character string “Thank you” is displayed. However, in response to an instruction to the scroll bar, for example, the base character having the next highest evaluation value is displayed. The predicted character string obtained by the column "company" can be displayed.

Next, when a character pattern intended for the second character "" is inputted, the recognition result is shown in FIG.
As shown in the table, "kai, sitting, whole, bag, evil, including, tun, public,
Just in case. In this case, when a base character string is generated by combining the first candidate character string with the recognition candidate character string, "society" and "don't care" succeed in dictionary search.

In this case, as shown in FIG. 6B, the evaluation value (the average value of the evaluation value of the first character and the evaluation value of the second character)
First, only the predicted character string starting with the base character string “social” having a high value is searched. If the number of the predicted character strings searched here is less than 50, a predicted character string starting with the base character string “don't forget” having the next highest evaluation value is searched.

As described above, among the base character strings formed by combining the recognition candidate characters obtained as a result of the recognition of the input pattern, the base character strings having the highest evaluation within the preset number of predicted character strings are successively arranged. By predicting a character string, it is possible to limit the number of predicted character strings to be searched and to more easily input a character string intended by a user.

In the above description, the evaluation of the base character string is performed based on the evaluation value (indicating the similarity with the standard pattern stored in the pattern recognition dictionary 104) at the time of character recognition processing for each recognition candidate character. Although the value is calculated, the evaluation value of the base character string may be calculated based on the candidate rank of the recognition candidate character.

For example, in the present embodiment, since ten recognition candidate characters are obtained, the score of the recognition candidate having the first candidate rank is set to 10 points, and the score of the recognition candidate character having the second or lower rank is determined. Nine points, eight points,..., One point. The prediction control unit 106 calculates an evaluation value of each base character string generated by a combination of the recognition candidate characters as an average of scores for each recognition candidate character forming the base character string.

In the above description of the second embodiment, the evaluation value of the base character string is the average value of the evaluation values for each character obtained at the time of character recognition of the recognition candidate character constituting the base character string. Alternatively, an evaluation value for a base character string may be set based on an evaluation value corresponding to a word (character string) stored in the word dictionary 108. As an evaluation value corresponding to a word in the word dictionary 108, for example, there is a value indicating the number of times determined as a recognition result (use frequency value). In this case, when the input character string is selected from the displayed predicted character strings, it is assumed that the use frequency value of the corresponding character string in the word dictionary 108 has been updated.

The prediction control unit 106 searches the word dictionary 108 for a word (character string) corresponding to the base character string generated based on the recognition candidate character by the dictionary search unit 107,
A use frequency value corresponding to the corresponding character string (a character string that completely matches) is set as an evaluation value for the base character string.

Hereinafter, in the same manner as in the description of the second embodiment described above, a predicted character string is obtained by giving priority to a base character string having a high evaluation value. By using this usage frequency value as an evaluation value, a predicted character string for a word that is frequently used in document creation is displayed with priority, so that character input using a predicted character string can be performed more efficiently. It is possible to do.

The present invention can be applied not only to the handwritten character input described in the embodiment but also to a character input device having a pattern recognition function such as a voice input. In this case, the input operation can be made easier.

The method described in each of the above-described embodiments is, for example, a magnetic disk (floppy disk, hard disk, etc.), an optical disk (CD-ROM,
DVD and the like, and can be written to a recording medium such as a semiconductor memory and provided to various devices. Further, it is also possible to transmit the data via a communication medium and provide it to various devices. A computer that implements the present apparatus reads the program recorded on a recording medium or receives the program via a communication medium, and executes the above-described processing by controlling the operation of the program.

[0066]

As described above in detail, according to the present invention, among the base character strings formed by combining the recognition candidate characters obtained as a result of recognizing the input pattern, the evaluation is performed within the preset number of predicted character strings. The search is performed by predicting the subsequent character strings in order from the base character string with the highest character, or by setting the condition regarding the number of characters to start prediction for each character type, and predicting the character string for the base character string that satisfies the condition Thus, the number of predicted character strings can be limited, and the character string intended by the user can be more easily input.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a character string prediction device according to an embodiment.

FIG. 2 is a flowchart showing a flow of processing from character input to output of a predicted character string of the character string prediction device according to the first embodiment.

FIG. 3 is a diagram showing a state from character input to character string prediction of the character string prediction device according to the first embodiment.

FIG. 4 is a flowchart showing a flow of processing from character input to output of a predicted character string of a character string prediction device according to a second embodiment.

FIG. 5 is a diagram showing a state from character input to character string prediction of a character string prediction device according to a second embodiment.

FIG. 6 is an exemplary view for explaining a search for a predicted character string in the second embodiment.

[Explanation of symbols]

 101 input device 102 character input unit 103 pattern recognition unit 104 pattern recognition dictionary 105 input buffer 106 prediction control unit 107 dictionary search unit 108 word dictionary 109 candidate buffer 110 output control unit 111 output buffer 112 output device

Claims (7)

[Claims] 1. A pattern recognition unit for recognizing an input character pattern by referring to a pattern recognition dictionary and outputting recognition candidate characters, wherein the input character string is used as a base character string, A character string prediction device for predicting with reference to a dictionary, comprising: a character input unit that extracts an input character pattern, recognizes the character pattern by the pattern recognition unit, and outputs a recognition candidate character; Combining a word dictionary storing headings corresponding thereto, a dictionary search unit for searching for a predicted character string having a base character string at the beginning with reference to the word dictionary, and a recognition candidate character output by the character input unit To generate a base character string, and in the generated base character string, check the character type of the recognition candidate character constituting the base character string,
A character string prediction apparatus comprising: a prediction control unit that instructs the dictionary search unit to search for a predicted character string by instructing the dictionary search unit about a base character string that satisfies a prediction start condition preset for the number of characters for each character type. 2. A character is input by recognizing an input character pattern with reference to a pattern recognition dictionary, and the input character string is used as a base character string to predict a subsequent character string with reference to a word dictionary. A character string prediction method, wherein a character type of a recognition candidate character forming a base character string is checked among base character strings formed by combining recognition candidate characters of a recognition result, and a prediction start condition set in advance with respect to the number of characters for each character type A character string prediction method comprising: searching a word dictionary for a predicted character string for a base character string that satisfies the following. 3. A pattern recognition means for recognizing an input character pattern with reference to a pattern recognition dictionary and outputting recognition candidate characters, wherein the input character string is used as a base character string, A character string prediction device for predicting with reference to a dictionary, comprising: a character input unit that extracts an input character pattern, recognizes the character pattern by the pattern recognition unit, and outputs a recognition candidate character; Combining a word dictionary storing headings corresponding thereto, a dictionary search unit for searching for a predicted character string having a base character string at the beginning with reference to the word dictionary, and a recognition candidate character output by the character input unit To generate a base character string, set an evaluation value for each base character string, and select a base character string with a high evaluation so that the number of predicted character strings to be searched is within a preset range. Previously to string predicting apparatus and a predictive text control means for retrieving the predicted character string. 4. A character is input by recognizing an input character pattern with reference to a pattern recognition dictionary, and the input character string is used as a base character string to predict a subsequent character string with reference to a word dictionary. A character string prediction method, wherein an evaluation value is calculated for each base character string formed by combining recognition candidate characters of a recognition result, and a base having a high evaluation is set so that the number of predicted character strings to be searched is within a preset range. A character string prediction method characterized by searching for a predicted character string from a word dictionary by giving priority to a character string. 5. The character string prediction device according to claim 3, wherein the evaluation value of the base character string is calculated using an evaluation value at the time of recognition of a recognition candidate character forming the base character string. 6. The evaluation value of the base character string is referred to a dictionary in which an evaluation value is set for each character string in advance, and is set as an evaluation value corresponding to the base character string and the corresponding character string. The character string prediction device according to claim 1. 7. Each time a character is sequentially input, an evaluation value is calculated for each base character string formed by combining recognition candidate characters of a recognition result, so that the number of predicted character strings to be searched is within a set range. 4. The character string prediction device according to claim 3, wherein a process of retrieving a predicted character string from a word dictionary by giving priority to a base character string having a high evaluation is repeatedly performed.