JP5849675B2 - Character input program and information processing apparatus - Google Patents

Description

  The present invention relates to a program for causing a computer to function as a character input device for inputting characters to a running application, and an information processing device having a character input function.

  In a general character input system (IME), a plurality of post-conversion character string candidates are extracted by searching a dictionary in the system based on the input pre-conversion character string, a list of candidates is displayed, and a selection operation is performed. Accept. In addition, there is a system provided with a learning dictionary that stores a converted character string determined by selection in association with an input order and part-of-speech type in order to easily call a word that is frequently used. Furthermore, in order to improve the accuracy of conversion, character strings based on combinations of syntactic and ancillary words are stored by separating the initial part (syntactic) and the ending part (adjunct), and character strings that represent verbs. Has been proposed to be stored in combination with its utilization form (see Patent Document 1).

  Furthermore, in recent years, using a communication system such as the Internet, a function for displaying a candidate based on a post-conversion character string transmitted from the conversion system by transmitting the pre-conversion character string to an external conversion system to perform conversion processing. (For example, refer to paragraphs 0075 to 0084 of Patent Document 2 and paragraphs 0047 to 0052 of Patent Document 3.) Hereinafter, this function is referred to as an “external conversion function”, and a candidate acquired by the function is referred to as an “external conversion candidate”.

JP-A-8-69459 JP 2008-250606 A JP 2009-20865 A

According to the external conversion function, even a device that does not have a large-capacity dictionary or a device that does not have any dictionary can acquire a considerable number of conversion candidates for the pre-conversion character string. Moreover, it is possible to easily cope with the case of inputting a new word.
Moreover, if the learning dictionary mechanism is used, even if the external conversion candidate is acquired by the external conversion function, if it is selected by the user and registered in the learning dictionary, the mobile terminal device It can be called by the conversion function.

  However, if the part-of-speech classification adopted in the external conversion system is different from the classification adopted in the apparatus, it is difficult to accurately specify the part-of-speech type of the external conversion candidate. For this reason, if the selected external conversion candidate is learned in association with an incorrect part-of-speech type, there is a possibility that an incorrect conversion is performed in the conversion process using the learning result.

  For example, if a verb is recognized as a noun, it is recognized that the word has an attached word, so that the ending of the word itself cannot be changed, and misrecognition occurs in the relationship with the subsequent character string. There is a case. For example, if the word “sleep” acquired from the outside is learned as a noun for the pre-conversion character string “nemuru”, the pre-conversion character string “nemu ...” It becomes impossible to convert to "". In addition, in the conversion process for the reading character string “Nemurumori”, it is determined that “mo” is a related particle for the noun “Nemuru”, and the pre-conversion character string is separated from “Nemuru-mo / ri”. It may be converted to “Sleep Motori”.

  The present invention pays attention to the above problem, and learns so that the part of speech classification can be estimated with high accuracy and correct conversion can be performed even for words such as words acquired from an external system that cannot uniquely identify the part of speech classification. This is the issue.

  The program according to the present invention includes an input unit that accepts an input operation for a pre-conversion character string, a search unit that searches for a post-conversion character string candidate corresponding to the pre-conversion character string, and one of the candidates extracted by the search unit. A confirmation processing means for confirming and outputting a candidate converted character string selected in accordance with an operation for selecting one, and a learning dictionary in which the confirmed converted character string is associated with a part of speech type that matches the character string The computer is caused to function as a character input device provided with a learning processing means for storing the data.

  This character input device is provided with a ending change pattern table in which a plurality of part of speech types set by classifying part of speech whose ending changes based on the ending change pattern are registered together with the ending change pattern.

The learning processing means collates the converted character string with the ending change pattern table in response to confirmation of the converted character string in which the part of speech type is not specified, and finds a ending that matches the rear part of the converted character string. A part of speech type having a pattern of ending change that includes the part of speech type of the converted character string, an estimation means for estimating that the matching ending is the ending of the converted character string, and the estimated part of speech type and ending Storage means for saving. The estimating means combines the converted character string in which the part of speech type and the ending are estimated, with a converted character string that is confirmed after the converted character string and is forward-matched with a portion excluding the estimated ending (the beginning portion). , together with the stores to create and store means a set of endings for each combination, the estimated out of parts of speech classification, the converted character a word class type having a pattern conforming inflection to the set of endings corresponding to the set The part-of-speech type corresponding to the column is stored in the storage means.

  According to the above configuration, if a word such as a verb or adjective that causes a change in the ending is determined in a state where the part of speech type cannot be uniquely specified, the part of speech type or ending is checked by checking the word against the ending change pattern table. Can be estimated. Even if the part of speech type estimated here is not specified, if the same word is confirmed with different endings and plural endings are stored, the ending change pattern is adapted to the set of these endings. Part-of-speech types can be narrowed down as correspondence targets. Therefore, based on the narrowed part-of-speech type ending pattern, it becomes possible to accurately perform the processing of segmenting the reading character string and transforming the ending.

  In the above-mentioned ending change pattern table, the part-of-speech type set by classifying the part-of-speech whose ending does not change based on the accessory-word pattern that can be attached to the word-of-speech word, and the accessory-word pattern as the ending-change pattern At least one can be registered. According to this configuration, it becomes easy to distinguish a part of speech such as a noun whose ending does not change and a part of speech whose ending changes. In addition, a word belonging to a part of speech whose ending does not change can be converted into a form with an attached word that matches the part of speech.

  In one embodiment of the character input device, the search means searches the learning dictionary with the pre-conversion character string, and the post-conversion character string associated with the part-of-speech type registered in the ending change pattern table by this search. Is extracted, a plurality of types of candidates are created by combining the ending change pattern corresponding to the part-of-speech type and the beginning part of the extracted converted character string. In this way, it is possible to create a candidate based on a character string in which the end of the stored phrase is deformed as well as a phrase that completely matches the phrase stored in the learning dictionary.

  In another embodiment of the character input device, the search means requests an external conversion system to perform conversion processing using a pre-conversion character string, and acquires post-conversion character string candidates by a response from the conversion system in response to the request. It has a function. In addition, when the candidate for the converted character string acquired from the external conversion system is confirmed, the estimating means collates the converted character string with the ending change pattern table to estimate the part of speech type and the ending, and after the conversion The character string is divided into an estimated ending and an initial portion excluding the ending, and stored in the learning dictionary in association with the estimated part of speech type.

  According to the above embodiment, even if the part-of-speech information of the converted character string obtained and confirmed from the external system is unknown, the part-of-speech type and ending of the character string are estimated, and the conversion is performed based on the estimation result. It can be divided into the beginning part and the ending part of the subsequent character string and stored together with the estimation result of the part of speech type. Therefore, when a character string that has the same initial part as the saved character string and a different ending is confirmed, it is easy to find a character string to be combined with the confirmed character string from the stored character strings, and combine the two. Can do. In addition, it is possible to efficiently narrow down the part of speech classification for the part of speech classification associated with the combined character string.

  According to the present invention, even if a converted character string that does not conform to the part of speech type employed in the character input device is determined as an input character string to the application, the part of speech that is highly likely to conform to the converted character string. Since learning can be performed in association with the type, it is possible to prevent conversion into a character string having an incorrect form by a conversion process using the learned converted character string. Therefore, the accuracy and convenience of the character input process can be improved.

It is a functional block diagram of a character input system to which the present invention is applied. It is a figure which shows the data structural example of an end change pattern table. It is a flowchart which shows the schematic procedure of the process in a character input system. It is a figure explaining the structure of the data preserve | saved at each table in a learning dictionary, and the relationship between tables. It is a figure explaining the structure of the data preserve | saved at each table in a learning dictionary, and the relationship between tables. It is a flowchart which shows the detailed procedure of a prediction candidate list creation process. It is a flowchart which shows the detailed procedure of a candidate confirmation process.

FIG. 1 shows a configuration example of a character input system to which the present invention is applied.
A character input system 1 according to this embodiment is for inputting characters into an application 101 (such as a memo pad or a mailer) running on the apparatus in accordance with a user operation within an information processing apparatus 100 such as a mobile phone. . Specifically, the system 1 includes a kana-kanji conversion processing unit 10, a user interface 11, a part-of-speech estimation processing unit 12, a conversion dictionary 13, a learning dictionary 14, a ending change pattern table 15, and the like. All of these are set by incorporating a dedicated program into the information processing apparatus 100.

  In the conversion dictionary 13, dictionary data of a plurality of words selected by the developer is registered. The dictionary data includes information such as word part (Kana character string before conversion) and notation (character string after conversion), part of speech type, priority, and the like. The conversion dictionary 13 of this embodiment is composed of a plurality of data files 131. However, the present invention is not limited to this, and can be combined into one data file.

  The part-of-speech type in this embodiment is classified into smaller units than the grammatical part-of-speech. For example, parts of speech such as verbs whose endings change are classified based on patterns of change (utilization forms), such as “ka-line five-stage inflection verbs”, “la-line five-stage inflection verbs”, “one-step verbs”, etc. A plurality of part-of-speech types are set. As for nouns, part-of-speech types such as “common nouns” and “sa-changing nouns” are set according to the classification based on the pattern of attached words that can be attached. However, there are cases where a part of speech type is set for each part of speech without being classified, such as a single adjunct or an impression verb.

  In the learning dictionary 14, words / phrases (hereinafter referred to as “determined words / phrases”) that are determined by the user's selection operation and input to the application 101 are accumulated up to a certain number in the input order. Specifically, in the learning dictionary 14, in addition to the used word history table 141 for accumulating definite phrases, the part-of-speech information table 142 that stores part-of-speech types that match each definitive word-phrase, and the coincidence used to specify the part-of-speech types A matching pattern table 143 in which patterns are stored is included. The configuration of these tables 141, 142, and 143 and stored data will be described later in detail.

  The user interface 11 starts up an operation unit including a character key on a display unit (not shown), and communicates the contents of the key operation to the kana-kanji conversion processing unit 10 while processing by the kana-kanji conversion processing unit 10 (reading character string assembly process, The result of candidate list creation processing, candidate confirmation display, etc.) is displayed. The operation unit is not limited to an image, and an operation unit using real keys may be provided.

  The kana-kanji conversion processing unit 10 cooperates with the user interface 11 to assemble and convert a read character string. Further, the kana-kanji conversion processing unit 10 of this embodiment passes a reading character string to the external kana-kanji conversion service system 2 via the network 3 such as the Internet and requests conversion processing, and the kana-kanji conversion service system 2 responds to this request. The external conversion candidate can also be acquired by the response.

  Candidates extracted by various searches are displayed as a list by the user interface 11. When the user interface 11 accepts an operation for selecting one of the candidates, the kana-kanji conversion processing unit 10 finalizes the selected candidate character string and outputs it to the application 101, and saves the confirmed phrase in the learning dictionary 14. . Further, the learning dictionary 14 is searched for the fixed phrase, and a character string that may be input next is connected and extracted as a prediction candidate.

  The part-of-speech estimation processing unit 12 executes a process of estimating the part-of-speech type for a character string that cannot specify the part-of-speech type, such as an external conversion candidate provided from the kana-kanji conversion service 2. In the ending change pattern table 15, the ending change pattern of the part of speech type that is frequently used is registered for each type for this estimation process. In this embodiment, the kana-kanji conversion processing unit 10 creates another character string having a different ending from the candidate character string read from the learning dictionary 14 as a candidate (prediction candidate list creation process described later). A change pattern table 15 is used.

  FIG. 2 shows an example of registration information in the ending change pattern table 15. This table 15 stores a set of ending patterns of phrases including words belonging to the type as ending change patterns for each part of speech type. Each ending pattern is represented by a combination of a “˜” symbol representing the beginning of the word and a character string at the ending. For part-of-speech types in which the end of the word itself changes, such as a verb, a ending pattern representing the ending ending is set. On the other hand, for part-of-speech types in which the end of a word does not change, such as a noun, a ending pattern representing an appendix that can be attached to the word is set.

FIG. 3 shows a schematic procedure of processing executed by the character input system 1 described above.
In this process, while accepting an hourly key operation (step S1), a process corresponding to the operation content is executed.

  Processing associated with various operations will be described. First, when a kana indicating reading is input (“YES” in step S2), a reading character string before conversion is assembled (step S6), the learning dictionary 14 is searched using the reading character string, and a prediction candidate list is obtained. Is created (step S7), and each candidate in the list is displayed (step S11).

  When the conversion operation is performed in a state where the reading character strings are assembled (“YES” in step S3), the conversion dictionary 13 and the learning dictionary 14 are searched using the reading character strings to create a conversion candidate list. Then, each candidate in the list is displayed (step S11). In step S8, a word string that matches the reading character string is extracted, and the converted character string candidate is classified by dividing the reading character string into a head part and a tail part based on the part of speech type of the extracted word. Create Furthermore, if necessary, it communicates with the Kana-Kanji conversion service system 2 to acquire external conversion candidates. The acquisition of external conversion candidates may be performed by a second conversion operation.

If any candidate is selected in a state where the candidate is displayed (step S4 is “YES”), the character string of the selected candidate is confirmed (step S9). Further, a word candidate list (connection prediction list) following the confirmed character string (determined word) is created (step S10), and each candidate in the list is displayed (step S11).
When a termination operation is performed at a predetermined time (step S5), the character input process is terminated by erasing the display on the operation unit.

  The process flow itself shown in FIG. 3 is the same as that of the conventional system. However, the prediction candidate list creation process (step S7) and candidate determination process (step S9) indicated by thick frames have unique characteristics. In the following, the configuration and processing procedures related to these processes will be described.

FIG. 4 specifically shows the configuration of each table 141, 142, 143 of the learning dictionary 14 and the contents of data stored in these tables.
The used word history table 141 is stored in accordance with the order in which the reading and notation of the fixed word each time are fixed. Both readings and notations are divided into a head part and a tail part (a hyphen symbol (-) is provided between them), but depending on the content of the fixed phrase, it may be stored without this division. In that case, the entire fixed phrase is the beginning part and the ending is “none”.

  Each fixed character string is further associated with a use frequency, a use order, and two types of codes. The use frequency indicates the frequency with which the fixed word has been selected so far, and the use order indicates what number the fixed word has been input in a series of character inputs.

  One of the two types of code indicates corresponding information in the part-of-speech information table 142, and the other code indicates corresponding information in the matching pattern table 143. In the part-of-speech information table 142, the part-of-speech type that matches the fixed phrase is combined with the code information and stored. In the coincidence pattern table 143, a pattern (hereinafter referred to as “match pattern”) that matches the ending of the fixed word phrase among the ending patterns registered in the ending change pattern table 141 is stored in combination with the code information. .

  The information in the part-of-speech information table 142 and the matching pattern table 143 is not associated one-to-one with the confirmed word / phrase, but common information is stored for each confirmed word / phrase having the same head part. The part-of-speech information table 142 stores information for all confirmed words / phrases, but the information in the matching pattern table 143 is a confirmed word / phrase for which the part-of-speech information cannot be uniquely specified at the time of the first learning, specifically, external A matching pattern for a phrase that is captured and confirmed as a conversion candidate is stored.

The configuration of each table in FIG. 4 and the relationship between the tables will be described below with reference to FIG.
A and B in FIG. 4 show sentence data input to the application 101 in a format in which each fixed phrase is separated by a slash mark “/”. In this embodiment, the first “I am” is extracted from the conversion dictionary 13, the second “Run” and the third “You” are also provided from the Kana-Kanji conversion service system 2, and the last “ “Run” is assumed to have been created from “run” data stored in the learning dictionary 14.

Each table at the time when the character input process is started and the sentence “I am running” shown in A in the figure is input is as shown in (A1), (A2), and (A3) in the figure.
The first definite phrase “I am” is derived by matching with “I” registered as a common noun in the conversion dictionary 13, so “I” and “Ha” are based on the search results. It is divided and stored in the used word history table 141. In addition, “common noun” of the part of speech type associated with “I” is written in the part of speech information table 142, and “001” of the writing destination is stored in the used word history table 141. Since the part-of-speech information is uniquely specified, the matching pattern of the first definite word / phrase is not stored in the matching pattern table 143.

  The second “run” was an external conversion candidate, so the part of speech type cannot be specified. In this embodiment, with respect to such a character string, the part-of-speech type including the ending pattern that matches “run” is extracted with reference to the ending change pattern table 15. Specifically, a backward matching search is performed using each ending pattern. This search is performed by the part of speech estimation processing unit 12 of FIG.

  Referring now to FIG. 2, the ending change pattern table 15 stores two part-of-speech types (La-line five-step verb and one-step verb) in which the ending pattern “˜RU” is set. Therefore, by the above search, these two part-of-speech types are extracted and stored in the part-of-speech information table 142 in association with the common code “002”. Also, “˜” of the matched ending pattern is stored in the matched pattern table 143 in association with the code “102”.

  Further, based on the above processing result, the definite word “run” is divided into “run” and “ru” and stored in the used word history table 141. Furthermore, the codes “002” and “102” of the correspondence information in the part-of-speech information table 142 and the matching pattern table 143 are stored in association with the fixed word / phrase.

Further, as shown in FIG. 4B, when a sentence “You also ran” is input after “I ran”, each table 141, 142, 143 is displayed as (B1) in the figure. (B2) (B3).
Since “you too” is an external conversion candidate like the previous “run”, the part-of-speech type including the ending pattern matching “you too” is extracted from the ending change pattern table 15, and this is stored in the part-of-speech information table 142. Saved. Referring to FIG. 2 again, since the ending change pattern table 15 stores two part-of-speech types (common nouns and sa-changing nouns) that include the ending pattern “~ mo” that matches “you too”. Two part-of-speech types are extracted and stored in the part-of-speech information table 142 in association with a common code “003”. In addition, the matched ending pattern “˜mo” is stored in the matched pattern table 143 in association with the code “103”.

  Based on the result of the above processing, the fixed phrase “you” is stored in the used word history table 141 after being divided into “you” and “also”. Furthermore, the codes “003” and “103” of the correspondence information in the part-of-speech information table 142 and the matching pattern table 143 are stored in association with the fixed phrase.

  The fourth fixed character string “run” is extracted as a variation of “run” by searching the learning dictionary 14, so that the beginning portion “run” and “ Are stored in the used word history table 141.

  For “run”, no new code is created in the part-of-speech information table 142 or the match pattern table 143, and the same code as “run-run” (the match pattern matches the code “002” in the part-of-speech information table) The table code “102”) is stored. However, since the ending is different from that of “Run”, this ending pattern “~” is added as a matching pattern.

  Also, according to FIG. 2, the part of speech type in which “to” and “to” associated with the code “102” are included in the ending change pattern is only “La line five-step verb”. Therefore, the “single verb” in the information of the code 002 in the part of speech information table 142 is deleted.

  In the prediction candidate list creation process (step S7) shown in FIG. 3, a plurality of candidates are created by changing the endings of the words extracted by the search of the learning dictionary 14 using the read character strings based on the respective part of speech types. To do. Further, in the conversion candidate list creation process (step S8), the reading character string to be converted is divided into a head part and a tail part by collation with the conversion dictionary 13 and the learning dictionary 14, and the notation (kana-kanji) is applied. Therefore, when the part of speech type of each word in the dictionary is not specified, erroneous conversion occurs.

  However, in this embodiment, when a character string whose part-of-speech type cannot be specified is stored in the learning dictionary 14, the beginning part and the ending part are identified by collation with the ending change pattern table 15 and can correspond to the character string. The type of part of speech can be estimated. Therefore, it is possible to extract correctly converted candidates based on the estimated part of speech type.

  For example, in the example of FIG. 4, when the reading character string “Hashi” is input for the fourth “Run” input, the part of speech stored for the previously determined “Run” is entered. Each ending pattern of the type “La line five-step verb” and “one-step verb” is applied to “Run”. Therefore, at this stage, incorrect conversion candidates based on the ending change pattern of the first verb such as “do not run” or “run” are created, but the correct conversion candidate is also created based on the ending change pattern of the la-row five-step verb. . In addition, when “Run” is selected from the candidates created, the part-of-speech type corresponding to “Run” will be narrowed down to only the five-stage verb of the line. No erroneous conversion occurs when the word is entered.

  In this way, it is possible to create conversion candidates in which the endings of the external conversion candidate character strings provided from the Kana-Kanji conversion service system 2 are correctly changed in the same manner as the words registered in the conversion dictionary 13. become. In addition, part-of-speech types can be identified with a broad concept, such as identifying verbs and nouns.

  Note that the definitive words / phrases for which the part-of-speech type is to be estimated are not limited to external conversion candidates, but can be re-converted by changing the hiragana character string determined from the read character string or the beginning / end delimiter set by the conversion process. It is possible to target all words that are not registered in the conversion dictionary 13 such as a kana-kanji character string, a katakana character string converted according to a post-conversion operation, and an alphabetic character string.

  In the example of FIG. 4, one ending pattern that matches the confirmed word / phrase is specified by collation with the ending change pattern table 15, but this is not always the case, and a plurality of ending patterns may be extracted. In that case, the part-of-speech information table 142 and the matching pattern table 143 store information with different codes for each extracted ending pattern.

FIG. 5 shows information stored in the tables 141, 142, and 143 when the word “hahaha” is confirmed.
The confirmed word / phrase “hahaha” in this example is the one in which the reading character string is confirmed as it is and the part of speech type is unknown, and therefore the ending pattern matching the fixed word / phrase is searched using the ending change pattern table 15. Processing is performed. According to FIG. 2, there are two change patterns that match the definite word “hahaha”: “~ ha” and “˜ (no attached word)” of common nouns and sa variable nouns.

  Therefore, in the used word history table 141, the words classified into “haha-ha” based on the first ending pattern “˜ha” and the second ending pattern “˜ (no attached word)” are included. On the basis of this, “hahaha” as a whole is stored in the form of the word. As for the former form, “common noun” and “sa variable noun” including “˜ha” in the ending change pattern are stored in the part-of-speech information table 142 in association with the code 004, and the matching pattern table 143 includes “˜ha” is stored in association with the code 104. Regarding the latter form, three part-of-speech types “common noun”, “sa-variant noun”, and “adverb” including “˜” in the ending change pattern are stored in the part-of-speech information table 142 in association with the code 005. “˜” is stored in the matching pattern table in association with the code 105. In the used word history table 141, “haha-ha” and “hahaha” are stored in combination with the corresponding codes in the tables 142 and 143, respectively.

In addition to the above, the words / phrases for which the part-of-speech type is estimated are also conceivable.
For example, when a new character input system is introduced and used while leaving the old character input system, the words read from the old system dictionary are confirmed as input targets, or a conversion dictionary storing new words is added to the system. Even when a word read from the new conversion dictionary is confirmed as an input target, the part of speech type of the confirmed word or phrase can be estimated or the part of speech type of the learned word can be narrowed down by the same method as in the above embodiment. it can.

Next, a detailed processing procedure of the prediction candidate list creation process (step S7 in FIG. 3) will be described with reference to FIG.
In this process, the prediction candidate list is cleared in the first step S101. As will be described below, this prediction candidate list is divided into two, one list is created in step S102, and the other list is created in steps S103 to S107.

  In step S102, a forward match search using a reading character string is executed for words registered in the conversion dictionary 13, and a first prediction candidate list is created from the extracted candidates.

  In creating the second prediction candidate list, first, a prefix match search (extracting a word head part included in a reading character string) is performed on the word head part of a word saved in the learning dictionary 14. Hereinafter, Steps S105, S106, and S107 are executed for the extracted words in order.

In step S105, the beginning part matched in the search in step S103 is extracted. In step S106, all the part-of-speech ending patterns associated with the words to be processed are read out. Here, when a plurality of part-of-speech types are associated with each other, the ending patterns of all the part-of-speech types are read out.
In step S107, a plurality of candidate character strings are created by combining each read ending pattern with the beginning part extracted in step S105. Further, these candidates are stored in the second prediction candidate list.

  When the processing for all the words extracted in step S103 is completed (step S104 is “YES”), the process proceeds to step S108, and the first and second prediction candidate lists are integrated. For example, a process is performed in which candidates that overlap between both lists are combined into one, or a priority is set for each candidate. The integrated list is displayed by the process of step S11 in FIG.

  Next, with reference to FIG. 7, a detailed procedure of candidate determination processing (step S9 in FIG. 3) accompanying the candidate selection operation will be described.

  In this process, first, the used word history table 141 of the learning dictionary 14 is searched for the selected candidate character string (definite word / phrase), and a word whose reading and notation both match the definite word / phrase in front is extracted. When the fixed phrase is extracted by searching the conversion dictionary 13 or the learning dictionary 14, it is divided into a head part and an end part at the time of the search. In step S201, a phrase having the same head part is searched. To do.

If the above search does not extract a phrase that directly matches the confirmed phrase (step S202 is “NO”), steps S203 and S204 are executed.
In step S203, the part-of-speech type having the ending pattern that matches the fixed phrase is extracted from the ending change pattern table 15. The extracted part of speech type is stored in the part of speech information table 142. Further, in step S 204, the ending pattern matched with the fixed phrase is stored in the matching pattern table 143.

The part-of-speech type and the matching pattern are all stored in association with the code. When these storages are completed, the process proceeds to step S209, and various information (including the above-described code) related to the fixed word / phrase is stored in the use word history table 141.
Thereafter, the process of clearing the reading character string (step S210) and the process of outputting the definite word / phrase to the application (step S211) are executed, and the process is terminated.

  In the search in step S201, a phrase that directly matches the confirmed phrase is extracted, but when a plurality of part-of-speech types are associated with the phrase (step S202 is “YES”, step S205 is “NO”), The part-of-speech type and matching pattern corresponding to the extracted phrase are specified by the code associated with the extracted phrase (step S206). These are the processing targets of the next step S207 and step S208.

In step S207, the final pattern of the fixed phrase is added as a matching pattern to be processed. In step S208, the added matching pattern list is compared with the ending change pattern of the part of speech type to be processed, the part of speech type having the ending change pattern that matches the matching pattern list is left, and the inconsistent part of speech type is deleted. (Step S208). After this, the process proceeds to step S209 described above, and various information related to the confirmed word / phrase is stored in the used word history table 141. This stored information includes the part-of-speech type and matching pattern code specified in step S206.
Note that if the pattern at the end of the fixed phrase is already included in the matching pattern, steps S207 and S208 are skipped.

If a word that matches the definite word / phrase in the search process in step S201 is extracted and the part of speech type of the word / phrase is fixed (“YES” in steps S202 and S205), it is necessary to estimate the part of speech type and the matching pattern. Since there is no, immediately proceed to step S209. In step S209 in this case, the reading and notation of the confirmed word are stored separately for the beginning and ending identified when the confirmed word is derived as a candidate, and the same part of speech type as the word hit by the search in step S201. Associate.
Furthermore, clearing of the reading character string (step S210) and a process of outputting the fixed phrase to the application 101 (step S211) are executed, and the process is terminated.

  Note that even a word taken as an external conversion candidate and stored in the learning dictionary 14 may be registered in the conversion dictionary 13 by uploading the conversion dictionary 13 or the like. In this case, the part-of-speech type of the word in the learning dictionary 14 is also updated to be compatible with the conversion dictionary 13.

  According to the above processing, even for a word whose part of speech type is unknown or whose part of speech type cannot be uniquely specified, the part of speech type can be narrowed down while the ending pattern is changed to various forms and input is performed a plurality of times. Therefore, like the words registered in the conversion dictionary 13, step S205 is “YES”, and the learning process is facilitated.

  However, if the part-of-speech type is narrowed down so that only the ending change pattern that matches the entire matching pattern list remains in step S208 of FIG. 7, the following trouble occurs when the user makes a mistake in selecting a candidate. There is a fear.

  For example, in the example shown in FIG. 4, after “run” is stored in association with the la-row five-step verb and the one-step verb, the character string “hashi” is assembled to input “run”. In addition to predictive conversion candidates (such as “run”) based on the ending change pattern of the five-stage verb of the line, predicted conversion candidates (such as “run”) based on the ending change pattern of the one-step verb are also displayed. If the user mistakenly selects “run” from these candidates, the part-of-speech type related to “run” is narrowed down to a one-step verb, and the correspondence with the la-row five-step verb is canceled. As a result, it is impossible to correctly derive a utilization form other than “run” for “run”.

If importance is attached to this problem, it is desirable that the part-of-speech type deletion conditions in step S208 be somewhat relaxed. For example, even if the part-of-speech type has a ending change pattern that does not match all of the matching pattern lists, it may be left without being deleted if the number of mismatches reaches a predetermined allowable value. Or, without narrowing down the part of speech type until the number of patterns included in the list of matching patterns reaches a predetermined number, for each part of speech type, count the frequency of selection of a phrase that matches the matching pattern corresponding to that type, After the number of matching patterns reaches a predetermined number, it may be determined which part-of-speech type is left based on the selection frequency of each part-of-speech type.
According to these processes, it is possible to prevent the correct part-of-speech type from being deleted even if there is a slight error in the candidate selection operation.

  In the above embodiment, the matching pattern table 143 is provided in the learning dictionary 14, and by this table 143, words and phrases whose part of speech type cannot be uniquely specified are grouped for each common word head part, and for each group. Although the matching pattern has been saved, the storage location of the ending pattern is not limited to the learning dictionary 14, but can be temporarily saved in the buffer, and even if the matching pattern is deleted for the phrase whose part of speech type is narrowed down to one Good. Similarly, when the matching pattern is stored in the learning dictionary 14, information on a word / phrase in which the part of speech type is narrowed down to one may be deleted.

  In the above embodiment, the codes are included in the storage information of the part-of-speech information table 142 and the matching pattern table 143, and the information stored in each of the tables 142 and 143 and the words / phrases stored in the used word history table 141 via this code However, the present invention is not limited to this, and the association may be performed via the part-of-speech type or the address where the matching pattern is stored. Further, the part of speech type may be included in each record of the used word history table 141 without providing the part of speech information table 142.

  Finally, the character input system 1 is not limited to a portable information processing apparatus, but can be incorporated in a stationary apparatus (such as a display apparatus or a facsimile). It can also be applied to a personal computer. Further, the input target language is not limited to Japanese, and input of other languages such as English can be handled by a system having the same configuration.

DESCRIPTION OF SYMBOLS 1 Character input system 2 Kana-Kanji conversion service system 10 Kana-Kanji conversion processing part 11 User interface 12 Part of speech estimation processing part 13 Conversion dictionary 14 Learning dictionary 15 End change pattern table 101 Application 141 Word usage history table 142 Part of speech information table 143 Match pattern table

Claims (5)

An input means for accepting an input operation for a pre-conversion character string, a search means for searching for a candidate for a post-conversion character string corresponding to the pre-conversion character string, and an operation for selecting one of the candidates extracted by the search means Confirmation processing means for confirming and outputting the converted character string of the candidate selected according to the above, and learning processing means for storing the confirmed converted character string in the learning dictionary in association with the part-of-speech type that matches the character string A program for causing a computer to function as a character input device comprising:
A program for incorporating in a character input device a ending change pattern table in which a plurality of part of speech types set by classifying parts of speech whose ending changes based on the ending change pattern and registered with the ending change pattern is included. And
The learning processing means collates the converted character string with the ending change pattern table in response to the confirmation of the converted character string in which the part of speech type is not specified, and the ending that matches the rear part of the converted character string. A part-of-speech type having a pattern of ending change that includes the part-of-speech type of the converted character string, an estimation means for estimating that the matching ending is the ending of the converted character string, and the estimated part-of-speech type and ending Storage means for storing
The estimation means combines the post-conversion character string whose part-of-speech type and ending are estimated with a post-conversion character string that is confirmed after the post- conversion character string and is forward-matched to a portion excluding the estimated ending part. A set of endings is created and stored in the storage means, and among the estimated part-of-speech types, part-of-speech types having a ending change pattern that matches the set of endings are converted to character strings after conversion. A program for character input, which is stored in the storage means in association with   In the ending change pattern table, the part-of-speech type set by classifying the part-of-speech whose ending does not change based on the accessory-word pattern that can be attached to the word-of-speech word, The program for character input according to claim 1, wherein at least one of them is registered.   The search means searches the learning dictionary by the pre-conversion character string, and when the post-conversion character string associated with the part-of-speech type registered in the ending change pattern table is extracted by this search, corresponds to the part-of-speech type. The program for character input according to claim 1 or 2, wherein a plurality of types of candidates are created by combining the ending change pattern to be performed and the beginning part of the extracted converted character string. The search means has a function of requesting a conversion process using a pre-conversion character string to an external conversion system and obtaining a candidate for a post-conversion character string by a response from the conversion system to the request.
When the candidate for the converted character string acquired from the external conversion system is confirmed, the estimating means compares the converted character string with the ending change pattern table to estimate the part-of-speech type and the ending, and after the conversion The character string according to any one of claims 1 to 3, wherein the character string is divided into an estimated ending part and an initial part excluding the ending part, and stored in the learning dictionary in association with the estimated part of speech type. Input program. An input means for accepting an input operation for a pre-conversion character string, a search means for searching for a candidate for a post-conversion character string corresponding to the pre-conversion character string, and an operation for selecting one of the candidates extracted by the search means A confirmation processing means for confirming the converted character string of the candidate selected according to the output and outputting it to the active application in the apparatus, and associating the confirmed converted character string with a part of speech type that matches the character string An information processing apparatus comprising learning processing means for storing in a learning dictionary,
There is further provided a ending change pattern table in which a plurality of part of speech types set by classifying parts of speech whose endings are changed based on ending change patterns are registered together with the ending change patterns,
The learning processing means collates the converted character string with the ending change pattern table in response to the confirmation of the converted character string in which the part of speech type is not specified, and the ending that matches the rear part of the converted character string. A part-of-speech type having a pattern of ending change that includes the part-of-speech type of the converted character string, an estimation means for estimating that the matching ending is the ending of the converted character string, and the estimated part-of-speech type and ending and a storage means for storing,
The estimation means combines the post-conversion character string whose part-of-speech type and ending are estimated with a post-conversion character string that is confirmed after the post- conversion character string and is forward-matched to a portion excluding the estimated ending part. a set of ending with creating and storing in said storage means, out of the estimated part of speech classification, the part of speech type having a pattern conforming inflection to the set of ending each converted character string corresponding to the set in An information processing apparatus, wherein the information is stored in the storage unit in association with each other.

Images