JPH09274613A - Character string predicting method and document preparing device using the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a character string intended by a user as a high-priority predictive candidate by storing the receding candidate selected by the user, inspecting connectivity with the preceding selected candidate in the case of prediction and narrowing down the predictive candidates. SOLUTION: When a character string to be a predictive object is inputted through a character string input part 101, a dictionary retrieval part 105 retrieves character strings having that input character string at their leading parts out of a word dictionary 106 as the candidates of a predictive character string. A predictive candidate control part 107 refers to a connection rule table 112 and a preceding candidate table 113 when any additional word is included in this retrieved candidate of the predictive character string, and the additional word connectable to this candidate selected the last time is narrowed down as the candidate of the predictive character string.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character string prediction method for predicting a character string subsequent to an input character string based on a character string input by an input device, and a method of using the character string prediction method. Document creation device.

[0002]

2. Description of the Related Art Conventionally, in a document creating apparatus such as a Japanese word processor, when inputting a character string using a handwritten character recognition apparatus or the like, one character is input unlike a keyboard input. Since it takes time, the burden on the input person was large.

Therefore, as an attempt to save the trouble of inputting characters, there has been a method of registering a long character string with a short character string in kana-kanji conversion. However, in this method, the character string had to be registered in advance.

A method called a predictive input was conceived for this. For example, as reported in the literature (48th IPSJ National Convention Proceedings 4J-11 et al.), The first few characters There is a method of searching a word dictionary at the time when is input and extracting a character string (word) that starts with a few characters.

This is because, for example, in inputting a character string "Kanji dictionary", when the input person inputs "Kanji", "Dictionary" is predicted as a character string following the "Kanji", and finally "Kanji". It outputs "Kanji dictionary".

[0006]

By the way, in the predictive input method as described above, the greater the number of input characters at the time of starting the prediction, the higher the prediction rate becomes, and therefore the character string desired by the user is output. be able to. However, the user is forced to make a large number of inputs, which impairs the original purpose of reducing the user's burden of input.

On the other hand, the smaller the number of input characters at the time of starting the prediction, the less the burden on the user. However, as the number of input characters is small, it becomes difficult to narrow down the candidates for the predicted character string, and many candidate character strings will appear. In this case, if the display screen is small, it is not possible to display all prediction candidates at once. Therefore, there is a problem that the character string intended by the user is not displayed preferentially.

The present invention has been made in view of the above points, in which the immediately preceding candidate selected by the user is stored, and when predicting, the connectivity with the immediately selected candidate is checked,
An object of the present invention is to provide a character string prediction method capable of obtaining a character string intended by a user as a prediction candidate having a high priority by narrowing down the prediction candidates, and a document creation apparatus using this character string prediction method. And

[0009]

[Means for Solving the Problems]

(1) The present invention has a word dictionary that stores words, a connection rule table that stores connection rules for independent words and attached words, and a pre-candidate table that stores information about previously selected candidates. By inputting the target character string, a character string that has the input character string at the beginning is searched from the above word dictionary as a candidate for the predicted character string, and an auxiliary word is included in the searched candidate for the predicted character string. In this case, referring to the connection rule table and the previous candidate table, the auxiliary word connectable to the previously selected candidate is used as the predicted character string candidate.

According to such a configuration, when predicting a character string, only the auxiliary words connectable to the word (previous candidate) input immediately before are selected as the prediction candidates, thereby narrowing down the prediction candidates. be able to.

(2) In the present invention, when a candidate for a predictive character string includes an adjunct word, the candidate for the predictive character string is rearranged so that the adjunct word has priority over the independent word. It was done.

According to such a configuration, when an adjunct word exists in the prediction candidates, the adjunct word can be obtained with priority over the independent word. (3) The present invention has a word dictionary storing words and a pre-candidate table that stores information about previously selected candidates, and when a character string to be predicted is input, the input character string is placed at the head part. Is searched as a candidate for a predicted character string from the word dictionary, and if the candidate word for the predicted character string thus searched contains an adjunct word, whether or not the preceding candidate table is empty is determined. If it is checked and the previous candidate table is empty, only independent words excluding the adjuncts are used as candidates for the predicted character string.

According to such a configuration, when predicting a character string, the adjunct word is not made a prediction candidate immediately after the punctuation mark, and only the independent word is made a prediction candidate, so that the prediction candidates can be narrowed down. it can.

(4) The present invention has a word dictionary in which words are stored, and a utilization table in which inflection endings according to inflection types and inflection forms of each word are listed, and a character string to be predicted is input. Thus, a character string having the input character string at the beginning is searched as a predicted character string candidate from the word dictionary, and when a word is included in the searched predicted character string candidates,
The adjective presents only the stem, and when the adjective is selected from the candidates for the predicted character string, it refers to the inflection table and presents the inflection corresponding to the adjective.

With such a configuration, when predicting a character string, if the prediction candidate includes a noun, only the word stem is presented and the user performs the candidate selection process to predict the noun. It is possible to narrow down the prediction candidates by presenting the corresponding inflection ending at the selected time.

(5) The present invention has a word dictionary that stores words and a symbol correspondence table that enumerates pairs of symbols, and by inputting a character string to be predicted, the input character A character string having a column at the beginning is searched from the word dictionary as a candidate for a predictive character string, and it is checked whether or not the symbol is a symbol registered in the symbol pair table according to the input of the symbol. If the symbol is a symbol registered in the symbol pair table, a symbol paired with the symbol is detected, and the detected symbol is added to the predicted character string candidates.

According to this structure, when predicting a character string, when one of a pair of symbols such as a start angle bracket """and an end angle bracket""" is input. Can obtain the other symbol paired with the symbol as a prediction candidate in the next prediction process.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a document creation device according to an embodiment of the present invention. This device is a Japanese word processor with a character string prediction function. In FIG. 1, a character string input unit 101 inputs a character string to be predicted. The input method in this case includes a key input method or a handwriting input method.

FIG. 2A shows the configuration of the character string input unit 101 assuming a key input method. The character string input unit 101 is
The keyboard 11 includes a kana-kanji conversion unit 12 and a conversion dictionary 13.

The keyboard 11 is an input device for keying in a character string. The keyboard 11 has a prediction key 11a for instructing the start of a prediction operation, in addition to various input keys such as a "character" key and a "numeric" key. The kana-kanji conversion unit 12 converts kana-kanji into a character string input through the keyboard 11. The character string (conversion result) obtained by this kana-kanji conversion is temporarily stored in the character string buffer 104 described later, and then cleared when the input person confirms the character string. However, when the start of the prediction operation is instructed by pressing the prediction key 11a, the kana-kanji conversion unit 12 causes the character string buffer 104
The character string stored in is extracted and passed as a prediction target character string to the dictionary conversion unit 105 described later. Conversion dictionary 13
Is a dictionary used in the kana-kanji conversion unit 12 and stores various information necessary for kana-kanji conversion processing, such as reading and headwords corresponding to the reading.

FIG. 2B shows the structure of the character string input unit 101 assuming a handwriting input method. Character string input unit 101
Is composed of a tablet 21, a recognition unit 22, and a recognition dictionary 23.

The tablet 21 is an input device for handwriting input of a character string. The tablet 21 uses a pen for handwriting input, and performs handwriting input of a character string by a writing operation of the pen. Further, here, a prediction button 21a for instructing the start of the prediction operation is provided.
The recognition unit 22 character-recognizes the character string handwritten and input through the tablet 21. A character string (recognition result) obtained by this character recognition is a character string buffer 104 described later.
After being temporarily stored in, it is cleared when the user confirms the character string. However, when the start of the prediction operation is instructed by pressing the prediction button 21a, the recognition unit 22 extracts the character string stored in the character string buffer 104 and uses this as the prediction target character string, which will be described later in the dictionary search unit 105. Pass to. The recognition dictionary 23 is a dictionary used in the recognition unit 22, and stores various kinds of information necessary for character recognition processing, such as information on strokes forming a character.

Further, in FIG. 1, the symbol pair detecting unit 102
Output buffer 116 when a symbol is input
When the symbol is registered in the symbol pair table 103, the symbol pair table 103 stores the symbol in the symbol pair table 103. It is detected and stored in the symbol pair buffer 115.

The symbol pair table 103 is a table listing a set of symbols such as a start bracket symbol and an end bracket symbol. Examples of this symbol include the following.

"" And """""and""""(" and ")""[" and "]""{" and "}" The character string buffer 104 stores the character string before being determined. Used for temporary storage. The dictionary search unit 105 searches the word dictionary 10 for a character string having the input character string at the beginning.
6 is searched as a candidate for the predicted character string. Word dictionary 10
In 6 are stored various words.

The prediction candidate control unit 107 instructs the prediction candidate determination unit 108 to narrow down the prediction candidates. In addition, the prediction candidate control unit 107 presents the prediction candidate to the user and performs a candidate selection process, stores the prediction candidate selected by the user in the output buffer 116, and stores the prediction candidate in the previous candidate table 113. Particularly, when the punctuation mark is input, the previous candidate table 113 is cleared.

The prediction candidate determination unit 108 is a dictionary search unit 10
When the search result obtained in step 5 includes an adjunct word, the connection rule table 112 is referred to, connection determination with the immediately preceding candidate stored in the previous candidate table 113 is performed,
Only connectable attached words are stored in the candidate buffer 111. In particular, when the previous candidate table 113 is empty, it is determined that the input character string is the beginning of the sentence, and all the adjunct words cannot be connected.

The output order adjusting unit 109 is provided in the candidate buffer 11
When the attached word is stored in 1, the prediction candidates are rearranged so that the attached word has priority over the independent word. The prediction candidate addition unit 110 immediately uses the inflection ending table 11 when the candidate selected by the user is the stem of a noun.
4 is searched, and the inflection ending corresponding to the stem of the said idiom is presented. Alternatively, if there is a symbol stored in the symbol pair buffer 115, the symbol is added to the candidate buffer 111 in which the search result of the dictionary search unit 105 is stored. The candidate buffer 111 stores predicted character string candidates.

The connection rule table 112 is a table listing connection rules for independent words and adjunct words. The previous candidate table 113 stores the candidate selected last time by the user, and is cleared when a punctuation mark is input. The inflection ending table 114 is a table that enumerates inflection endings according to the inflection type for each inflection type.

The symbol pair buffer 115 includes the symbol pair detector 1.
Store the symbol detected by 02. The output buffer 116 stores the candidates selected by the user.
The character string output unit 117 displays the character string stored in the output buffer 116 on the screen.

Next, the operation of the embodiment will be described. FIG.
And FIG. 4 is a flowchart showing the flow of the character string prediction process in the same embodiment. First, the character string input section 1
01, input buffer 104, candidate buffer 11
1. The output buffer 116 is initialized (step 20)
1). After that, the character strings sequentially input through the character string input unit 101 are buffered in the input buffer 104 until they become prediction targets (steps 202 to 205). However, if the symbol is entered (N in step 203)
o), the processing after step 226 is performed without performing the prediction processing.

The character string input through the character string input unit 101 is a character string that has undergone kana-kanji conversion by key input or a character string that has been recognized by handwriting.

The timing for determining the input character string as a prediction target is, for example, when the user gives a prediction instruction by pressing the prediction button, or when a predetermined time has elapsed. The timing is not limited to these.

When the character string stored in the input buffer 104 becomes a prediction target due to a prediction instruction or the like, the prediction target character string stored in the input buffer 104 is sent to the dictionary retrieval unit 105. The dictionary search unit 105 searches the word dictionary 106 for a character string having the character string at the beginning, and sends the search result to the prediction candidate control unit 107 (step 206).

The prediction candidate control unit 107 checks whether the prediction candidate selected immediately before by the user is stored in the previous candidate table 113 (step 20).
7). When the previous candidate is stored in the previous candidate table 113 (Yes in step 207), the prediction candidate control unit 107 refers to the information stored in the previous candidate table 113 and the connection rule table 112 through the prediction candidate determination unit 108. Then, the connection with the previous candidate is determined (step 208).
As a result, candidates that can be connected to the previous candidate are selected in the candidate buffer 11
1 (step 209). The same processing as above is repeated for all dictionary search results (step 208-).
210).

The preceding candidate and the connectable candidate are attached words such as "de", "de", and "de" as a character string starting with "de". in this way,
When the previous candidate exists, an adjunct word connectable to the previous candidate is included as one of the prediction candidates of the input character string (claim 1 or 6, FIG. 5B). Note that the adjunct word is a word such as a particle or auxiliary verb that cannot be a phrase by itself and is always attached to an independent word.

If there is an adjunct word in the prediction candidates stored in the candidate buffer 111, the output order adjusting unit 109 rearranges the prediction candidates so that the adjunct word has priority (step 211). As a result, the adjunct word is preferentially output among the candidates, and the user can obtain the adjunct word as the first candidate (see claim 2 or 7, FIG. 5 (c)). .

On the other hand, in step 207, when the previous candidate table 113 is empty, the prediction candidate control unit 10
7 stores only the independent word in the dictionary search result in the candidate buffer 111 through the prediction candidate determination unit 108 (steps 212 to 214).

As described above, when there is no previous candidate, the input character string is the beginning of the sentence, and there is no attached word as a prediction candidate, so only the independent word is the candidate buffer 1.
11 (see claim 3 or 8, FIG. 6B). An independent word is a word that can form one phrase with one word, such as a noun, a verb, an adjective, an adjective, an adverb, an adnominal, a conjunction, and a verb.

When the process of storing the prediction candidates in the candidate buffer 111 is completed, the prediction candidate adding unit 110 causes the symbol pair buffer 1
It is checked whether a symbol is stored in 15 (step 215). When the symbol is stored in the symbol pair buffer 115 (Yes in step 215), the prediction candidate adding unit 110 adds the symbol to the candidate buffer 111 as one of the prediction candidates of the input character string (step 2).
16).

Subsequently, the prediction candidate control unit 107 presents the prediction candidates stored in the candidate buffer 111 to the user and performs a candidate selection process (step 217). At that time, when the prediction candidate includes a noun, only the stem of the word is presented to the user, and when the user selects the noun (Yes in step 218), the prediction candidate control is performed. The unit 107 searches the inflection ending table 114 for an inflection ending corresponding to the relevant noun through the prediction candidate adding unit 110 (step 2
18, 219), the inflection ending is presented to the user, and the candidate selection process is performed again (step 220). Note that a verb is a word having an inflected form such as a verb.

As described above, when the prediction candidate is a noun, first, only its stem is presented, and then the corresponding inflection ending is presented by the user's selection (claim 4). Or 9, see FIG. 7 (b).

If the prediction candidate selected by the user is a symbol (Yes in step 221),
The prediction candidate control unit 107 clears the symbol pair buffer through the prediction candidate adding unit 110 (steps 221 and 22).
2). After that, the prediction candidate control unit 107 stores the part-of-speech information of the selected candidate in the previous candidate table 113 (step 223) and outputs the selected candidate to the output buffer 1.
16 (step 224). Character string output unit 11
7 displays the selection candidates stored in the output buffer 116 on the screen (step 230).

If a symbol is input in the character string input process (No in step 203), the prediction candidate control unit 107 stores the input symbol in the output buffer 116 (step 225). , Symbol pair detector 10
The symbol pair table 103 is searched through 2 (step 226). As a result of the search, the relevant symbol is the symbol pair table 1
If the symbol is registered in 03 (step 227)
Yes), the prediction candidate control unit 107 stores the symbol paired with the symbol in the symbol pair buffer 115 (step 22).
7, 228) and the previous candidate table 113 is cleared (step 229). The character string output unit 117 displays the symbols stored in the output buffer 116 on the screen (step 230).

As described above, when one of the pair of symbols is input, the other symbol forming a pair with the symbol is temporarily stored in the symbol pair buffer 115, and then, at the time of the next prediction process. , Will be stored in the previous candidate table 113 as one of the prediction candidates of the input character string (step 21).
5, 216). Thereby, for example, when there is a symbol "", the symbol "" is included as one of the prediction candidates of the input character string (claim 5).
Or 10, see FIG. 8).

Next, the above-described prediction process will be described with a specific example. FIG. 5 shows claims 1 or 6 and claim 2 of the present invention.
Or, it shows a prediction example corresponding to 7. FIG.
(A) shows a conventional example, (a) shows the prediction example of claim 1 or 6, and (b) shows the prediction example of claim 2 or 7.

For example, when the user inputs the character string "de" in the state where the character string "park" is already confirmed and displayed, in the conventional method, as shown in FIG. As a character string having “de” at the head, only independent words such as “Denwa (telephone)”, “Densha (train)”, “Denki (electric)” ... Are output as the prediction result.

On the other hand, according to claim 1 or 6 of the present invention, since prediction is performed in consideration of the immediately preceding word, if "de" is entered immediately after "park" as shown in FIG. The attached words such as “de”, “de”, and “de” are also output as the prediction result.

Further, in claim 2 or 7 of the present invention, considering that the number of candidates of the adjunct word is overwhelmingly smaller than that of the independent word, as shown in FIG. When an adjunct word is included, the adjunct word is output with priority over the independent word.

It should be noted that in the present invention, the attached word is also one of the prediction candidates.
Therefore, in the example of FIG. 5, the total number of candidates increases. However, in the present invention, the candidates are narrowed down only for the adjunct words that are closely related to the previous candidate, and the adjunct words having the same first character are not simply extracted. If the conventional method were to include adjuncts in the prediction candidates, all the adjuncts with the same first character would be added, and the number would be enormous.

FIG. 6 shows a prediction example corresponding to claim 3 or 8 of the present invention. FIG. 6A shows a conventional example, and FIG. 6B shows a prediction example of claim 3 or 8. For example, when the user inputs the character string “tsu” while the character string “..., so” is already confirmed and displayed, in the conventional method, as shown in FIG.
An adjunct such as "tetsu" is also predicted as a character string that starts with "tsu".

In claim 3 or 8 of the present invention, the prediction is performed in consideration of the immediately preceding punctuation mark ",". In this case, therefore, the input character string "tsu" is judged to be the beginning of the sentence, and FIG. As shown, only independent words such as "tsuyoi (strong)", "tsuki (month)", ... Are output as the prediction result.

FIG. 7 shows a prediction example corresponding to claim 4 or 9 of the present invention. FIG. 7A shows a conventional example, and FIG. 7B shows a prediction example of claim 4 or 9. For example, if a candidate is extracted by using the character string “ha” as a prediction target, and a noun (verb) such as “hasuru (runs)”, as shown in FIG. 7A, “hashira (runs)”,
Many candidates must be output, such as "Hashiri (run)", "Hashiri (run)".

Therefore, in claim 4 or 9 of the present invention,
Regarding the idiom, first, only the stem is presented, and when the idiom is selected by the user, the corresponding inflection ending is presented. That is, as shown in FIG. 7A, when "ha" is input, the verb "hashiru (run)" is also included as a prediction candidate, but at this time, only the stem "hashi (run)" is presented. . When the user selects this, the inflection ending “Ra, Ri, Ru, Re, Ro” is presented as shown in FIG.

FIG. 8 shows an example of prediction corresponding to claim 5 or 10 of the present invention. When the symbols preceding the pair of symbols, such as the opening bracket ““ ”and the closing bracket“ ””, are input, the method according to claim 5 or 1.
At 0, the backward symbol is included in the prediction candidates in the subsequent prediction processing.

For example, as shown in FIG. 8, when "to" is input after the opening bracket """isinput," Tokyo (Tokyo) "," Street "," Tomen ( For the time being, in addition to the prediction candidates such as "...", the closing bracket """is also output.

[0057]

As described above, according to claim 1 or 6 of the present invention.
According to the above, by inputting a character string to be predicted, a character string having the input character string at the beginning is searched from the word dictionary as a candidate for the predicted character string, When an adjunct word is included, the adjunct word that can be connected to the previously selected candidate is referred to as the predicted character string candidate by referring to the connection rule table and the previous candidate table. The prediction candidates can be narrowed down by using only the adjuncts that can be connected to the (previous candidate) as the prediction candidates.

Further, according to claim 2 or 7, when a candidate for a predictive character string includes an adjunct word, rearrangement of the candidate for the predictive character string is given priority over the independent word. Therefore, when an adjunct word exists in the prediction candidates, the adjunct word can be obtained with priority over the independent word.

Further, according to claim 3 or 8, when a character string to be predicted is input, a character string having the input character string at the beginning is searched from the word dictionary as a candidate for the predicted character string, and this search is performed. When an adjunct word is included in the candidates for the predicted character string, whether or not the preceding candidate table is empty is checked, and when the preceding candidate table is empty, the independent word excluding the above adjunct word is checked. Since only the punctuation mark is used as a candidate for the prediction character string, the adjunct word is not made a prediction candidate immediately after the punctuation mark, and only the independent word can be made a prediction candidate to narrow down the prediction candidates.

Further, according to claim 4 or 9, by inputting a character string to be predicted, a character string having the input character string at the beginning is searched from the word dictionary as a candidate for the predicted character string, and this search is performed. When a word is included in the predicted character string candidates, the word only shows the stem, and the usage table is referred to when the word is selected from the above predicted character string candidates. Since the corresponding inflection end of the relevant idiom is presented, if the idiom is included in the prediction candidates, only the stem is presented and the user performs candidate selection processing to select the idiom. By presenting the corresponding inflection ending at the time point, it is possible to narrow down the prediction candidates.

Further, according to claim 5 or 10, when a character string to be predicted is input, a character string having the input character string at the beginning is searched from the word dictionary as a candidate for the predicted character string, and the According to the input, it is checked whether the symbol is a symbol registered in the symbol pair table, and if the symbol is a symbol registered in the symbol pair table, the symbol paired with the symbol is detected, and this detection is performed. Since one of the paired symbols such as the opening bracket ““ ”and the closing bracket“ ”” is input, In the prediction process of, the other symbol paired with the symbol can be obtained as a prediction candidate.

[Brief description of drawings]

FIG. 1 is a block diagram showing a device configuration according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a specific configuration of a character string input unit in FIG.

FIG. 3 is a flowchart showing an operation of a character string prediction process in the same embodiment.

FIG. 4 is a flowchart showing an operation of a character string prediction process in the same embodiment.

FIG. 5 is a diagram showing a prediction example in the case of considering attached words.

FIG. 6 is a diagram showing a prediction example in the case of considering independent words.

FIG. 7 is a diagram showing an example of prediction in the case of taking a word into consideration.

FIG. 8 is a diagram showing a prediction example in the case of considering a pair of symbols.

[Explanation of symbols]

 101 ... Character string input unit 102 ... Symbol pair detection unit 103 ... Symbol pair table 104 ... Input buffer 105 ... Dictionary search unit 106 ... Word dictionary 107 ... Prediction candidate control unit 108 ... Prediction candidate determination unit 109 ... Output order adjustment unit 110 ... Prediction candidate addition unit 111 ... Candidate buffer 112 ... Connection rule table 113 ... Previous candidate table 114 ... Inflectional ending table 115 ... Symbol pair buffer 116 ... Output buffer 117 ... Character string output unit

Claims (10)

[Claims] 1. A prediction target having a word dictionary storing words, a connection rule table storing connection rules for independent words and adjuncts, and a previous candidate table storing information about candidates selected last time, When a character string that has the input character string at the beginning is searched as a candidate for a predicted character string from the word dictionary and an adjunct word is included in the searched candidate for the predicted character string. Further, the character string prediction method is characterized in that an adjunct word connectable to the previously selected candidate is set as a predicted character string candidate by referring to the connection rule table and the previous candidate table. 2. When a candidate for a predicted character string includes an adjunct word, the candidate for the predicted character string is rearranged so that the adjunct word has priority over an independent word. The character string prediction method according to claim 1. 3. A word dictionary storing words and a pre-candidate table storing information about previously selected candidates, and by inputting a character string to be predicted, the input character string is at the head part. When a character string is searched as a predicted character string candidate from the word dictionary, and if the searched predicted character string candidate includes an adjunct word, it is checked whether or not the previous candidate table is empty. A character string prediction method, wherein when the previous candidate table is empty, only independent words excluding the adjunct words are candidates for the predicted character string. 4. A word dictionary that stores words, and a utilization table that lists inflection endings according to inflection types and inflection forms of each adjective, and by inputting a character string to be predicted, the input characters are input. When a character string having a column at the beginning is searched as a candidate for a predicted character string from the word dictionary, and if the searched candidate for the predicted character string contains a word, that word only presents the stem. Then, when a noun is selected from the candidates for the predicted character string, the inflection corresponding to the noun is presented by referring to the inflection table. 5. A word dictionary storing words, and a symbol correspondence table enumerating pairs of symbols forming a pair, and by inputting a character string to be predicted, the input character string is set to the head part. Search for a character string that you have as a candidate for a predicted character string from the word dictionary, check whether the symbol is a symbol registered in the symbol pair table according to the input of the symbol, and register the symbol in the symbol pair table. A character string prediction method, characterized in that, if it is a recognized symbol, a symbol paired with the symbol is detected, and the detected symbol is added to the predicted character string candidates. 6. An input unit for inputting a character string to be predicted, a word dictionary storing words, and a character string having the character string input by the input unit as a head portion from the word dictionary. Dictionary search means to search as a candidate for, a connection rule table that stores the connection rules for independent words and adjunct words, a pre-candidate table that stores information about the previously selected candidate, and the predictions searched by the dictionary search means. When an adjunct word is included in a character string candidate, the adjunct word that can be connected to the previously selected candidate is referred to as the predictive character string candidate by referring to the connection rule table and the previous candidate table. A document production method comprising: a determination unit; and a prediction candidate control unit that stores a character string selected from the predicted character string candidates in the previous candidate table as a previous candidate. Equipment. 7. When the predicted character string candidates include an adjunct word, an output order adjusting means is provided for rearranging the predicted character string candidates so that the adjunct word has priority over the independent word. The document creating apparatus according to claim 1, wherein 8. A character string input means for inputting a character string, a word dictionary storing words, and a character string having the character string input by the input means at a head portion as a candidate for a predicted character string from the word dictionary. Dictionary search means to search as, a previous candidate table that stores information about the previously selected candidate, and the preceding candidate when the auxiliary word is included in the predicted character string candidates searched by the dictionary search method. It is checked whether or not the table is empty, and if the previous candidate table is empty, a prediction candidate determination means that makes only the independent word excluding the adjunct word a candidate for the prediction character string, and the prediction character string A predictive candidate control means for storing the character string selected from the candidates as a previous candidate in the previous candidate table and emptying the previous candidate table according to the input of punctuation marks. Document creating apparatus for. 9. An input unit for inputting a character string to be predicted, a word dictionary storing words, and a character string having the character string input by the input unit as a head portion from the word dictionary. Dictionary search means to be searched as a candidate, a utilization table listing the inflection endings according to the inflection type and inflection form of each adjective, and the adjective in the predicted character string candidates retrieved by the dictionary retrieval means. When included, the adjective corresponds to the adjective by referring to the prediction candidate control means that presents only the word stem and the inflection table when the adjective is selected from the candidates for the predicted character string. A document creation device, comprising: a prediction adding unit that presents an inflection ending. 10. An input unit for inputting a character string to be predicted, a word dictionary storing words, and a character string having the character string input by the input unit as a head portion from the word dictionary. Dictionary search means to be searched as a candidate of, a symbol correspondence table that lists pairs of symbols, and whether or not the symbol is a symbol registered in the symbol pair table according to the input of the symbol, If the symbol is registered in the symbol pair table, the symbol pair detecting means for detecting the symbol paired with the symbol pair, the symbol pair buffer for storing the symbol detected by the symbol pair detecting means, and the symbol pair When a symbol is stored in the buffer, a prediction adding unit that adds the symbol to the predicted character string candidate, and whether the symbol is selected from the predicted character string candidates, or , Document creation device, characterized in that when a new symbol is input to and a prediction candidate control means for clearing said pair of symbols buffer.