JP4285272B2 - Morphological analysis method, program using the method, and information processing apparatus - Google Patents

Description

  The present invention relates to a technique for inputting a predetermined amount of text data to a computer and performing morphological analysis on the text data.

  In recent years, with the spread of the Internet, it has been proposed to perform morphological analysis on text data included in web pages, e-mails, etc., and use extracted words for the purpose of creating other documents. In addition, software that converts text data into speech data and outputs it based on the result of the morphological analysis has been developed.

In a general morpheme analysis process, a character string included in text data to be processed is searched using a dictionary in which various morphemes are registered, and a plurality of candidates are extracted. Further, an optimal candidate combination is determined based on the rules relating to the combination of morphemes, and the combination of morphemes based on the determination is output.
In such morphological analysis, in order to improve the accuracy of analysis, it is necessary to be able to extract unknown words that are not registered in the dictionary. With respect to this point, there is a prior art such as Patent Document 1 below.

JP-A-6-12453

  In the above-mentioned patent document 1, an unknown word is extracted by comparing an unregistered word and the connection relationship before and after it with various rules, and then the extracted unknown word is displayed on a monitor. Then, after the user selects an unknown word to be registered, input such as reading and part-of-speech information is accepted, and new dictionary data corresponding to these is created and registered in the memory.

  In morphological analysis, it is necessary to be able to respond quickly to new names and phrases that have been born due to the situation and fashion of the world, and words that users use among friends (hereinafter collectively referred to as “new words”). is there. In recent years, in particular, the names of talents, brand names, abbreviations, etc. are often expressed in katakana and alphabet strings, so a simple method can be used to quickly respond to these new words even in morphological analysis. It is desirable to be able to register new words at.

In the invention described in Patent Document 1 described above, when text data including a new word is subjected to morphological analysis, the user must perform a registration operation for the new word extracted as an unknown word. For this reason, there exists a problem that a user's burden becomes large. Also, if the user overlooks a new word to be registered or if the processing is canceled due to a burden on the registration work, the new word will remain unregistered, so text data containing the same new word will be processed in the next morphological analysis. In this case, the new word is treated as an unknown word again.
As described above, the invention of Patent Document 1 requires a special processing time for registering an unknown word, and presupposes that the user performs an unknown word registration operation, and performs the registration process easily. It is difficult.

On the other hand, if the system provider periodically adds new words to the dictionary for morphological analysis and provides them to each user by a method such as data distribution, the burden on the user can be reduced and the function of morphological analysis can be improved. It can also be increased. However, while there are many new words to register, the words that are actually used are likely to be part of them, and the new words that are used may vary from user to user .
As described above, the method of periodically updating the morphological analysis dictionary is inefficient and causes a problem that it does not meet the profit of the system provider.

  By the way, when creating a document such as an e-mail, a general user is likely to input a new word that he / she is interested in. In addition, when accessing an e-mail or a web page from another person, it is considered that there is a high possibility of browsing data including a new word of interest.

  A software for kana-kanji conversion processing is generally provided with a learning function for automatically registering a character string after conversion when the character string is not registered in the internal dictionary. Also, with this kind of software, a function that instructs conversion to a specific character type other than Kana-Kanji (alphabet, katakana, etc.) is assigned to a predetermined key, and by operating that key, the input character string is assigned to a specific character type. It is converted to a character string. This operation for instructing conversion to a character string of a specific character type is called “post-conversion operation”.

In this specification, the term “post-conversion operation” includes an operation from an instruction to instruct conversion to a specific character type to an operation to determine a character string converted according to the operation.
The new words represented by the above-described katakana and alphabet can be input by a conversion operation thereafter. In addition, since the character string determined by the conversion operation can be registered by the learning function described above, the next time the same new word is input, the kana character string before conversion is input and the normal conversion is performed. By performing the operation, it is possible to call the previously registered character string. As described above, in the character input process, it is possible to add a new word to the dictionary data in the course of a normal character input operation without the user performing a special registration operation.

This invention pays attention to the above points and enables the dictionary data accumulated in the character conversion process to be used in the morphological analysis process, so that the user can perform a special registration work or periodically set up a dictionary for morphological analysis. It is an object of the present invention to be able to naturally improve the function for morphological analysis without updating it automatically.
Another object of the present invention is to improve convenience in information processing by enabling extraction of new words according to the interests of individual users in morphological analysis.

The morpheme analysis method according to the present invention includes a conversion standard dictionary in which a predetermined number of converted character strings are registered, a learning dictionary for registering a new converted character string, and a morpheme in which a predetermined number of morphemes are registered. It has a memory that stores an analysis dictionary, and after receiving an input of a kana character string from the operation unit, it converts with an input kana character string in response to an operation instructing conversion processing based on dictionary search. A standard dictionary for learning and a dictionary for learning, and a candidate for a character string after conversion, and after receiving input of a kana character string from the operation unit, one of katakana, hiragana, alphabet, number A function to convert the input kana character string into a character string of the selected character type in response to accepting a post-conversion operation that selects a character type and instructs conversion; Is converted in accordance with the work, which is executed in a computer that operates the converted character string not registered in the conversion standard dictionary and learning dictionary as a character conversion means for and a function of registering in the dictionary for the learning Step A for extracting a character string of a predetermined length from the text data when the input of the text data to be processed (text data including a character string mixed with kanji and kana. The same applies hereinafter) is received . Step B for searching the morpheme analysis dictionary using the character string extracted in Step A to extract morpheme candidates, and passing the character string extracted in Step A to the character conversion means A step of performing a search of the learning dictionary and accepting the character string extracted by the search from the character conversion means as a morpheme candidate. And C after multiple cycles executed, executes Step D for selecting and outputting a predetermined number of the candidates obtained in step B and C of each cycle. In step C, if the character string extracted in step A is a hiragana character string, a processing result indicating that a morpheme candidate could not be obtained without passing this character string to the character conversion means is set. . In step D, the katakana character string, alphabetic character string, and numeric character string obtained in step C are selected in preference to the candidate obtained in step B.

In the above, in the conversion process executed by the character conversion means , the input kana character string (in many cases it is a hiragana character string) is converted into a specific character such as alphabet, number, katakana, hiragana in addition to the kanji character string. It can be converted to a character string by character type. Note that the kanji character string referred to in this specification includes not only a character string consisting of only kanji characters but also a character string combining kanji and kana (mainly hiragana).

  In the standard dictionary for conversion, a post-conversion character string associated with reading / part of speech / frequency of use can be stored as dictionary data. Although the dictionary data having the same configuration can be stored in the learning dictionary, the learning dictionary may be initially empty. In addition to these two types of dictionaries, a dictionary (user dictionary) for storing dictionary data created by user setting operations may be provided.

  The morpheme analysis dictionary can store dictionary data in which morphemes are associated with reading / part of speech / dependency relationships with other morphemes. In the morpheme analysis process, the input text data is divided at a predetermined position to extract a character string as a search keyword, and the search keyword is used to search the morpheme analysis dictionary and the learning dictionary to extract candidates. it can.

According to the above morphological analysis method, the post-conversion character string stored in the learning dictionary according to the post-conversion operation performed when the normal kana character string conversion processing is executed is also used in the morphological analysis processing. Therefore, the morpheme analysis function can be improved without performing a special registration process or updating the dictionary for morpheme analysis. In addition, when a kana character string is converted to a new word such as a name or abbreviation by a post-conversion operation, and the new word is registered in the learning dictionary, text data including the registered new word is input in the morphological analysis. The new word can be extracted by a search using the learning dictionary.
Also, in this method, katakana character strings, alphabetic character strings, and numeric character strings extracted from the learning dictionary are selected preferentially over the candidates extracted from the morphological analysis dictionary, so they are registered in the morphological analysis dictionary. Uncharacterized katakana character strings, alphabetic character strings, and numerical character strings can be easily extracted by morphological analysis processing.

Next, if the character string extracted from the text data received for the morpheme analysis process is a hiragana character string, a morpheme candidate can be obtained in step C without passing this character string to the character conversion means. The reason why there is no processing result is that the hiragana character string is often input for the purpose of specifying “sentence” and the like, and is unlikely to represent an important meaningful word . In this way, only the katakana character strings, alphabetic character strings, and numerical character strings that are likely to represent names and abbreviations are extracted by the morphological analysis process from the character strings registered in the learning dictionary by the post-conversion operation. The time required for the search can be shortened.

Next, a program for morphological analysis according to the present invention includes a memory in which a standard dictionary for conversion in which a predetermined number of converted character strings are registered and a learning dictionary for registering new converted character strings are stored. The standard dictionary for conversion and the learning dictionary are searched with the input kana character string in response to receiving an operation for instructing conversion processing based on dictionary search after receiving the input of the kana character string from the operation unit. Then, after accepting the input of the kana character string from the operation unit and the function of extracting the converted character string candidates, select one of the katakana, hiragana, alphabet, and number character types to instruct conversion In response to accepting a post-conversion operation, the input kana character string is converted into a character string of the selected character type, and converted according to the post-conversion operation, and the conversion standard It is introduced dictionary and the converted character string that is not registered in the learning dictionary to the computer to operate as a character conversion means for and a function of registering in the dictionary for the learning.

The above program causes the computer to execute each step of a first step of receiving input of text data to be processed; a second step of decomposing the input text data into morphemes and outputting a combination of the morphemes. And electronic data constituting the morphological analysis dictionary described above. In the second step , a step A for extracting a character string of a predetermined length from the text data received in the first step, and a morpheme analysis dictionary is searched by using the character string extracted in the step A. The character string extracted in step A and the character string extracted in step A are passed to the character conversion means to search the learning dictionary using the character string, and the character string extracted by the search is used as a morpheme candidate. After executing step C received from the character conversion means for a plurality of cycles, step D for selecting and outputting a predetermined number of candidates obtained in step B and step C of each cycle is executed.

In the above morpheme analysis program, if the character string extracted in step A is a hiragana character string in step C, the character string cannot be obtained without passing this character string to the character conversion means. Set the processing result. In step D, the katakana character string, alphabetic character string, and numeric character string obtained in step C are selected in preference to the candidate obtained in step B.

According to said program, the morphological analysis process which can extract the word input by post-conversion operation can be performed. Once this program is installed, the converted character string stored in the learning dictionary according to the post-conversion operation can be used in the morphological analysis without performing the morphological analysis dictionary update process. In other words, the user does not need to perform special registration work for morphological analysis or install updated data for the morphological analysis dictionary, and improves the ability of morphological analysis by simply performing normal character conversion processing. Can be made. Also, the system developer can respond to each user's request without regularly updating the morphological analysis dictionary.

This morphological analysis program can be activated in response to a user operation or the like when an application such as a web browser or a mail reader (hereinafter referred to as “upper application”) is in operation. In step D in this case, the determined combination of morphemes can be output to the upper application. In addition, the determined combination of morphemes can be passed to an application (for example, an audio output system) other than the upper application.

In addition to a personal computer, the above program can be incorporated into a computer that constitutes a control unit of a mobile terminal such as a mobile phone or a PDA. The program, a method of storing in a storage medium such as a CD-ROM, by a method of transmitting a telecommunication line, can be provided to the user. A combination of the morphological analysis program and the character conversion processing program is composed of one package software (consisting of one or a plurality of storage media storing the above two types of programs. Can be provided at the same time or selectively, or can be provided to a predetermined computer via a telecommunication line.

Next, the information processing apparatus according to the present invention, an operation unit, a character converting accepts input and conversion instruction operation of kana string from the operation unit, the kana character string input into a string of other forms of It comprises conversion means and morpheme analysis means for accepting input of text data to be processed and decomposing the text data into morphemes. The character conversion means receives a conversion standard dictionary in which a predetermined number of converted character strings are registered, a learning dictionary for registering a new converted character string, and an input of a kana character string from the operation unit. Candidates for searching the conversion standard dictionary and learning dictionary with the input kana character string and extracting the converted character string candidates in response to receiving an operation to instruct conversion processing based on dictionary search In response to accepting a post-conversion operation for selecting a character type from katakana, hiragana, alphabet, and numbers after receiving input of a kana character string from the search means and the operation unit Te, a conversion operation at processing unit after converting the character string by character type instructed to Do string or entered, is converted by the following conversion operations during processing unit, a standard dictionary and learning the conversion The converted character string is not registered in the dictionary and registers the learning dictionary, when receiving the character string to be searched from the morphological analysis unit searches the learning dictionary by the character string extraction by the search Learning dictionary processing means for passing the converted character string to the morphological analysis means . Meanwhile, the morphological analysis means, a morphological analysis dictionary a predetermined number of morphemes is registered, when receiving an input of text data to be processed, the steps A to extract a character string having a predetermined length from the text data , Searching the morpheme analysis dictionary using the character string extracted in step A to extract morpheme candidates , passing the character string extracted in step A to the learning dictionary processing means of the character conversion means, A search unit that executes a plurality of cycles of receiving a character string passed from a learning dictionary processing unit that has performed a search of a learning dictionary by a character string as a morpheme candidate; and step B of each cycle by the search unit; and selecting a predetermined number of the candidates obtained in step C and an output means for outputting.

Further, when the character string extracted in step A is a hiragana character string, the search means can obtain a morpheme candidate in step C without passing the extracted character string to the learning dictionary processing means. Set the processing result that did not exist. Further, the output means selects the katakana character string, alphabetic character string, and numeric character string obtained in step C in preference to the candidate obtained in step B.

  In the above information processing apparatus, both the character conversion means and the morphological analysis means are set in the control computer of the information processing apparatus by a program. The character conversion means can operate in conjunction with an application such as a word processor or an e-mail editor. Further, the morphological analysis means can be configured to be activated in response to a command from the user or a user's call operation when the web browser or the mail reader is operating.

According to the present invention, the post-conversion character string automatically learned according to the post-conversion operation of the user can be used as dictionary data for morphological analysis processing. The function of morphological analysis can be naturally improved without performing registration work or fetching update data of the morphological analysis dictionary.

Further, according to the present invention, hiragana character strings that are highly likely not to have an important meaning among post-conversion character strings registered in the learning dictionary in accordance with the post-conversion operation are prevented from being extracted as morpheme candidates. Thus, the morpheme candidates extracted from the learning dictionary can be limited to katakana character strings, alphabetic character strings, and numerical character strings . In addition, the candidate extracted from the learning dictionary is selected in preference to the candidate extracted from the morphological analysis dictionary. Therefore , the character string based on the specific character type (katakana, alphabet, number) input by the post-conversion operation is given priority. Can be included in the morpheme analysis results. Therefore, for example, in a mobile phone, when the received mail is morphologically analyzed and a reply mail is created using the words extracted by the processing, the user's interest is extracted. It is possible to extract new words according to the accuracy. In addition, the convenience of information processing can be greatly improved, such as each user can extract new words according to his / her interest without providing a service for updating the morphological analysis dictionary by the system developer. .

FIG. 1 shows a configuration example of an information processing apparatus to which the present invention is applied.
This information processing apparatus is incorporated in a mobile phone (not shown), and includes a kana-kanji conversion processing unit 1 and a morpheme analysis processing unit 2. These processing units 1 and 2 are both set in the control unit (CPU) of the mobile phone by a program. The kana-kanji conversion processing unit 1 operates together with an application for document creation such as electronic mail. On the other hand, the morphological analysis processing unit 2 is activated in response to an operation for designating a morphological analysis process when a higher-level application such as an application for browsing an electronic mail or a web page is running.

  The kana-kanji conversion processing unit 1 is provided with three types of dictionaries, a standard dictionary 16, a user dictionary 17, and an automatic learning dictionary 18, an input / output unit 11, and a conversion control unit 12, and a search for the dictionary is performed for each dictionary. And processing units 13, 14, and 15 for registration are provided. In any of the dictionaries 13, 14, and 15, dictionary data in which a word notation (character string after conversion) is associated with reading, use frequency, or the like is stored.

  The input / output unit 11 receives input of operation keys while exchanging information with a human interface (not shown) developed on the display unit of the mobile phone. For example, while a kana character string (pre-conversion character string) is being input for Kana-Kanji conversion processing, the input character is recognized based on the type of key operated and the number of operations, and the recognized character is a human interface. Output to. When a conversion operation is performed after the input of the kana character string, an unconfirmed converted character string at that time is output to the human interface. Further, when an unconfirmed converted character string is confirmed, the input / output unit 11 outputs the confirmed character string to an application (such as a mail editor) that is a character input target.

In the operation unit of the mobile phone in which the information processing apparatus is introduced, specific keys are set for the conversion operation and the post-conversion operation described later, in addition to the conversion operation and the operation for determining the post-conversion character string. When the post-conversion key is operated, the input / output unit 11 displays a character type selection screen on the human interface, and selects any one of katakana, hiragana, alphabet, and numbers on this screen. I am doing so. For katakana, alphabets, and numbers, full-width or half-width characters can be selected.
In the following description, “post-conversion operation” refers to the operation from the post-conversion key operation to the operation of selecting the final character type.

  The conversion control unit 12 receives an input of a kana character string via the input / output unit 11 and passes it to the processing units 13, 14, 15 of each dictionary to execute search processing. Then, based on the search results returned from the processing units 13, 14, 15, the kana character string is converted into a predetermined converted character string, and the converted character string is output to the input / output unit 11.

  Of the three types of dictionaries, the standard dictionary 16 stores dictionary data of words selected in advance by the system designer. Although new dictionary data cannot be written in the standard dictionary 16, the frequency of use can be updated each time a character string of dictionary data is adopted.

  The user dictionary 17 is for registering a word designated by the user. When the conversion control unit 12 receives input of a converted character string, reading, part-of-speech information, and the like for a word to be registered, the conversion control unit 12 passes the input to the user dictionary processing unit 14 and registers it in the user dictionary 17.

The automatic learning dictionary 15 is for registering a character string after conversion by the operation when a post-conversion operation instructing conversion to a character type other than the kanji character string is performed. The registration process in the automatic learning dictionary 18 is also controlled by the conversion control unit 12, but the direct registration process is performed by the automatic learning dictionary processing unit 15.
Note that the usage frequency of the user dictionary 17 and the automatic learning dictionary 18 can be updated each time registered dictionary data is used.

  Next, the morpheme analysis processing unit 2 includes an input / output unit 21, an analysis control unit 22, a dictionary search unit 23, a morpheme analysis dictionary 24, a candidate control unit 25, and the like. The morpheme analysis dictionary 24 stores dictionary data in which attribute data such as part-of-speech information and a dependency relationship with another morpheme is associated with the morpheme notation. This morphological analysis dictionary 24 is also created by the system designer like the standard dictionary 16 of the kana-kanji conversion processing unit 1, and is not set to register new dictionary data.

  The input / output unit 21 communicates with the above-described upper application, inputs text data to be processed, and outputs a result of morpheme analysis on the text data (a sequence of extracted morphemes together with attribute data). To be set.

  The analysis control unit 22 receives the text data from the input / output unit 21 and cuts out a search keyword from the text data. The search keyword is output to the dictionary search unit 23, the candidate control unit 25, and the user dictionary processing unit 14 of the kana-kanji conversion processing unit 1 in order to search for a morpheme whose notation matches the character string.

  When the search keyword is passed from the analysis control unit 22, the dictionary search unit 23 searches the morpheme analysis dictionary 24 and extracts dictionary data whose morpheme notation matches the search keyword. Similarly, the user dictionary processing unit 14 searches the user dictionary 17 using the search keyword passed from the analysis control unit 22, and extracts dictionary data whose notation of the converted character string matches the search keyword. Any dictionary data extracted by these searches is passed to the analysis control unit 22.

On the other hand, the candidate control unit 25 passes the search keyword to the automatic learning dictionary processing unit 15 of the kana-kanji conversion processing unit 1 to execute a search using the automatic learning dictionary 18. However, if the search keyword is a hiragana character string, the candidate control unit 25 returns the search result “no candidate” to the analysis control unit 22 without passing it to the automatic learning dictionary processing unit 15. Since the search keyword other than the hiragana character string is given to the automatic learning dictionary processing unit 15, the automatic learning dictionary processing unit 15 searches the automatic learning dictionary 18 using the given search keyword, and the notation of the converted character string is described above. Dictionary data that matches the search keyword is extracted. The extracted dictionary data is passed to the analysis control unit 22 via the candidate control unit 25.
Note that the use of the automatic learning dictionary 18 and the user dictionary 17 in the morphological analysis processing is preferably only for reading data and not for writing data.

  Although the details will be described later, the analysis control unit 22 sets a plurality of search keywords, and obtains a morpheme candidate for each of these search keywords by the above three searches. Further, the analysis control unit 22 determines an optimum combination of morphemes from candidates obtained for each search keyword, and data representing the structure of the text data based on the determination (hereinafter referred to as “analysis data”). Is output. Note that the analysis data includes morpheme separation positions and part of speech information of each morpheme.

  Hereinafter, among the processes executed by the kana-kanji conversion processing unit 1 and the morpheme analysis processing unit 2, processes related to the present invention will be described in detail. In the flowchart of FIG. 2 and subsequent figures, each step is abbreviated as “ST”, and each step is indicated by a three-digit number.

  FIG. 2 shows a procedure of character input processing by the kana-kanji conversion processing unit 1. In the procedure of this embodiment, a necessary phrase can be corrected until a final confirmation operation is performed after a character string for a plurality of phrases is input and batch conversion is performed.

  This procedure starts when a kana character input operation is started. First, in the first ST101, the content of the operation by the user is determined. A user operates a key to which a predetermined kana character is assigned for character input. Each time the user operates, the process proceeds from ST101 to ST115, and processing corresponding to the operation (for example, input characters are temporarily stored in a memory). Execute processing for storing in the storage area, processing for displaying the stored character string, etc.).

  When a conversion operation is performed when a predetermined number of kana characters are input, the process proceeds from ST101 to ST102, and an accumulated input character string (kana character string) is acquired. In the next ST103, a process for converting the input character string (kana-kanji conversion process) is executed. Since this kana-kanji conversion process is a known technique, the details are omitted, but the input character string is divided into a predetermined number of phrases and searched by the standard dictionary processing unit 16, the user dictionary processing unit 17, and the automatic learning dictionary processing unit 18. To extract the best candidate for each phrase. In this embodiment, it is assumed that the converted character string obtained by combining the extracted candidates for each phrase is displayed on the display unit as an undetermined character string. At this time, the first phrase is initialized as a noticeable phrase, and identification display with a cursor is performed.

  In this embodiment, a post-conversion flag is set for each clause included in the unconfirmed post-conversion character string. This post-conversion flag is for storing the fact that the conversion process by the post-conversion operation has been performed. In ST104 immediately after the kana-kanji conversion process, a process for resetting the post-conversion process for all the phrases is executed.

  Here, if the converted character string matches the character string intended by the user, a confirmation operation is performed. By this operation, ST105 and 106 become “YES”, the process proceeds to ST113, and an automatic learning process to be described later is executed. In the next ST114, the unconfirmed converted character string is confirmed and output. Thereby, the process for the input character string is completed.

  On the other hand, if there is a clause including a character string that is not intended by the user in the unconfirmed converted character string, the user moves the cursor to the clause and executes a correction operation. Each time these operations are performed, “YES” in ST105 and “NO” in ST106, the process proceeds to ST107 and subsequent steps.

  First, when a cursor movement operation is performed, ST107 becomes “YES”, the process proceeds to ST108, and the phrase of interest is changed according to the operation. At this time, the cursor also moves to the attention phrase according to the movement operation.

  Next, when a post-conversion operation is performed on the target phrase, ST109 becomes “YES” and the process proceeds to ST110, where the target phrase is changed to a character type corresponding to the operation. Further, in the next ST111, the post-conversion flag of the target phrase is turned on.

  If an operation other than the post-conversion operation (an operation for selecting another conversion candidate, an operation for canceling conversion, or the like) is performed on the target phrase, the process proceeds to ST112 and processing corresponding to the operation is executed. To do. When the correction process is completed and a confirmation operation is performed, the above-described ST113 and 114 are executed, and the process is terminated.

FIG. 3 shows a detailed procedure for the automatic learning process of ST113.
In this process, after the first phrase is set as the target phrase in ST201, the loops of ST202 to 208 are repeated to process each phrase in turn. However, when the post-conversion flag is off, the focused phrase is simply moved to the next phrase, and substantial processing (ST 203 to 206) is executed only for the phrase for which the post-conversion flag is set. .

  When the post-conversion flag is set for the noticeable phrase, in ST203, dictionary data that associates the reading of the noticeable phrase (kana character string before conversion) with the converted character string is set. In the next ST204, it is checked whether or not there is a free area in the automatic learning dictionary 18. If there is a free area, ST205 is skipped and the process proceeds to ST206 to register new dictionary data in the free area. At the time of registration, a usage frequency and a registration order are added to the dictionary data.

  If there is no free space in the automatic learning dictionary 18, the process proceeds from ST204 to ST205, where the oldest dictionary data in the dictionary is extracted based on the registration order described above and deleted. However, when the use frequency of the dictionary data is equal to or higher than a predetermined value, old dictionary data is extracted next. As a result, it is possible to delete the least frequently used dictionary data from the relatively long elapsed time after registration. Thereafter, the process proceeds to ST206, and new dictionary data is registered in an area freed by the deletion.

The dictionary data in the automatic learning dictionary 18 may be appropriately sorted based on the registration order and usage frequency.
The procedure of the character input process is not limited to FIGS. 2 and 3 described above. For example, the input character string may be determined for each phrase after batch conversion. In this case, each time a confirmation operation is performed, the post-conversion flag of the clause related to the operation is checked. If the conversion flag is on after that, processing for registering the confirmed character string in the automatic learning dictionary 18 Will do.

  FIG. 4 shows a data configuration example of the automatic learning dictionary 18. "Reading" in the figure corresponds to the input character string described above, and "notation" corresponds to the converted character string. Reading includes not only hiragana characters but also single-byte numeric characters. The notation stores a character string according to any of the character types of katakana, hiragana, alphabet, and numbers according to the selection in the post-conversion operation.

Next, the outline of the morphological analysis process will be described with reference to FIG.
In this embodiment, the text data to be processed is divided into sentence units, and the longest match method, which is one morphological analysis method, is executed for each sentence. That is, each character (in the case of the example in FIG. 5, “morning”, “day”, “ga”, “wealth”, etc. in a character string (hereinafter referred to as “processing target character string”) constituting the sentence to be processed. (Each character) is searched in order, and the search is executed. In this search, a character string that matches a character string from the first character to a character at a predetermined position is extracted by using a character string from the attention character to the end of the sentence as a search target.

  For example, the fourth character “wealth” in the processing target character string in FIG. 5 will be described as an example. The character string from this “wealth” to the last character (not shown) is used as the search target character string, and the “wet” A search for extracting a character string that matches the character string from the character to the character at the predetermined position is executed. As a result, in the illustrated example, “Mt. Fuji” is extracted as the longest character string, and hereinafter, two types of character strings “Fuji” and “Wealth” are extracted.

  Each character string extracted for each character in the processing target character string is stored in the memory as a morpheme candidate starting from that character. Hereinafter, a memory area in which candidates extracted for each character are stored is referred to as a “search result list”, and a memory area including a search result list for all characters is referred to as a “candidate list”.

  When the candidate list is created in this way, each search result list is checked along the character sequence of the processing target character string, and a predetermined candidate is selected as a morpheme. In this case, in the longest match method, candidates are selected in preference to those with a long character string, and candidate selection is skipped for characters included in the previously selected candidates. However, the selection of candidates is not limited to one set, and a plurality of combinations of candidates are set, and the one having the best morpheme separation is selected from the combinations.

  FIG. 6 shows a detailed processing procedure in the morpheme analysis processing unit 2. This procedure is for one sentence in the text data. When the text data includes a plurality of sentences, the procedure of FIG. 6 is executed for each sentence.

  First, in the first ST301, the first character of the processing target character string is initially set as a target character. In the next ST302, the memory area for setting the candidate list is cleared, and then the process proceeds to ST303.

  In ST303, a character string from the attention character to the last character of the processing target character string is cut out and set as a search target. In subsequent ST304, a search process using the character string to be searched is executed. Details of this search processing will be described later.

  When the search ends, in ST305, it is determined whether or not a candidate has been extracted. If this determination is “NO”, the flow proceeds to ST306, and a character string consisting only of the attention character is extracted as a candidate. If candidates are extracted in the search process, the process skips ST306 and proceeds to ST307. In ST307, the extracted candidates are stored in the search result list of the target character. Even when ST306 is executed, in ST307, the character string composed of the target character is stored in the search result list.

  Thereafter, the processing of ST303 to 307 is repeated while moving the target character to the last character of the processing target character string in ST308 and 309. Thereafter, the process proceeds from ST308 to ST310, a candidate combination is set from the above candidate list, and a process of selecting an optimum one from the combination is executed. In ST311, analysis data is created based on the selected combination of candidates, and this is output to a higher-level application.

FIG. 7 shows the detailed procedure of the candidate search process in ST304, that is, the search process using the search target character string set in ST303.
First, in ST401, the search result list corresponding to the current character of interest is cleared. Next, after the length of the character string to be searched is stored in the counter L in ST402, the character string to be searched is set as a search keyword in ST403.

  In the following, the search for each dictionary described above is executed while deleting the end of the set search keyword one character at a time. First, in ST404, the search keyword is output to the candidate control unit 25 to search the automatic learning dictionary 18, and the search result is acquired in the next ST405. If a candidate is included in the search result, in ST405, processing for storing the candidate in the search result list is executed.

  In ST406, the user dictionary processing unit 14 outputs a search keyword to search the user dictionary 17, and in ST407, the search result is acquired. Even in ST407, if the search result includes a candidate, a process of storing the candidate in the search result list is executed.

  In the next ST408, the search keyword is output to the dictionary search unit 23 to search the morphological analysis dictionary 24, and in ST409, the search result is acquired. Even in ST409, if a candidate is included in the search result, processing for storing the candidate in the search result list is executed.

  In this way, when the search for each dictionary is completed, the value of L is decreased by one. If the updated L is greater than 0, the process proceeds to ST412 and after deleting the last character of the search keyword, the process returns to ST404. When the value of L becomes 0, the process ends.

  In the above morphological analysis process, a character string (hereinafter referred to as “automatically registered character string”) that has been previously determined by a post-conversion operation and registered in the automatic learning dictionary 18 is included in the processing target character string. The automatic registration character string can be stored in the search result list by the processing of ST404 and 405 when this automatic registration character string is set as a search keyword.

  However, if the search keyword is a hiragana character string, the candidate control unit 25 cancels the search process using this character string. Therefore, even if this search keyword is an automatically registered character string, the character string is a candidate. Not extracted as. That is, what is extracted as a candidate from the automatic learning dictionary 18 is limited to character strings of three types of characters, katakana, alphabet, and numbers. Moreover, since the search by the automatic learning dictionary 18 can be skipped for a hiragana character string that has a high possibility of not having an important meaning, the time required for the search can be shortened.

  In ST310 of FIG. 6, when combining the candidates of each search result list, the candidate extracted from the automatic learning dictionary 18 is given the highest priority, and then the candidate extracted from the user dictionary 17 is selected. ing. Thus, words input by the user through post-conversion operation in the character input process and words registered in the user dictionary by the user registration process can be preferentially extracted in the morphological analysis process.

  In this embodiment, only the character string converted by the post-conversion operation is registered in the automatic learning dictionary 18, but not limited to this, it is not registered in the standard dictionary 16 or the user dictionary 17. Kanji character strings can also be registered. For example, since the character string converted as one phrase is an unintended character string, when the user inputs a desired character string by performing a conversion operation for each phrase by dividing the input character string into two, The confirmed kanji character string can be registered in the automatic learning dictionary 18. When Kanji character strings are registered in the automatic learning dictionary 18 as described above, it is preferable that these kanji character strings can be extracted even by morphological analysis processing.

  In the character conversion process, the frequency of use can be updated when a character string registered in the automatic learning dictionary 18 is called and used by a normal conversion operation. Therefore, even in the morphological analysis process, when an automatically registered character string is extracted by searching the automatic learning dictionary 18, the candidate combination process may be performed in consideration of the frequency of use of the character string.

It is a functional block diagram of the information processing apparatus to which this invention is applied. It is a flowchart which shows the procedure of a character input process. It is a flowchart which shows the procedure of an automatic learning process. It is a figure which shows the data structure of an automatic learning dictionary. It is a figure explaining the outline | summary of a morphological analysis process. It is a flowchart which shows the procedure of a morphological analysis process. It is a flowchart which shows the procedure of a candidate search process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Kana-Kanji conversion processing part 2 Morphological analysis processing part 11, 21 Input / output part 12 Conversion control part 15 Automatic learning dictionary processing part 16 Standard dictionary 18 Automatic learning dictionary 22 Analysis control part 23 Dictionary search part 24 Morphological analysis dictionary 25 Candidate control part

Claims (3)

A memory storing a standard dictionary for conversion in which a predetermined number of converted character strings are registered, a learning dictionary for registering new converted character strings, and a morphological analysis dictionary in which a predetermined number of morphemes are registered The standard dictionary for conversion and the learning dictionary are searched with the input kana character string in response to receiving an operation for instructing conversion processing based on dictionary search after receiving the input of the kana character string from the operation unit. Then, after accepting the input of the kana character string from the operation unit and the function of extracting the converted character string candidates, select one of the katakana, hiragana, alphabet, and number character types to instruct conversion A function of converting the input kana character string into a character string of a selected character type in response to accepting a post-conversion operation, and conversion in accordance with the post-conversion operation; In a computer operating the converted character string not registered in the standard dictionary and learning dictionary as a character conversion means for and a function of registering in the dictionary for the learning, the input text data with the text kanji and kana And executing a morpheme analysis process that decomposes the text data into morphemes ,
When input of text data to be subjected to the morphological analysis processing is received, a step A for extracting a character string of a predetermined length from the text data, and a search for the morphological analysis dictionary by the character string extracted in step A and a step B of extracting morphemes candidate, step a character string extracted by a passes to the character conversion unit to execute the search of the dictionary learning by the character string, the character string extracted by the search Step C, which is received from the character conversion means as a morpheme candidate, is executed for a plurality of cycles, and then a predetermined number of candidates obtained in Step B and Step C of each cycle are selected and output ,
In step C, if the character string extracted in step A is a hiragana character string, a processing result indicating that a morpheme candidate could not be obtained without passing the character string to the character conversion means is set. And
In the step D, the katakana character string, the alphabet character string, and the number character string obtained in the step C are selected in preference to the candidates obtained in the step B.
A morphological analysis method characterized by the above. A standard dictionary for conversion in which a predetermined number of converted character strings are registered and a memory in which a learning dictionary for registering new converted character strings is stored, and input of kana character strings is accepted from the operation unit A function for searching the standard dictionary for conversion and the learning dictionary with the input kana character string and extracting candidates for the converted character string in response to accepting an operation to instruct conversion processing based on dictionary search later; After receiving the input of the kana character string from the operation unit, the input is performed in response to receiving the conversion operation after selecting the character type of katakana, hiragana, alphabet, or number and instructing the conversion. A function for converting a kana character string into a character string of a selected character type, and a post-conversion sentence that is converted according to the post-conversion operation and is not registered in the standard dictionary for conversion and the learning dictionary On a computer running the column as a character conversion means for and a function of registering in the dictionary for the learning, a program for providing a function of the morphological analysis,
For causing the computer to execute a first step of receiving input of text data including a character string including kanji and kana; and a second step of decomposing the input text data into morphemes and outputting combinations of the morphemes And electronic data constituting a morphological analysis dictionary in which a predetermined number of morphemes are registered ,
In the second step, a step A for extracting a character string of a predetermined length from the text data received in the first step, and a search for the morpheme analysis dictionary using the character string extracted in the step A The candidate B is extracted, and the character string extracted in the step A is passed to the character conversion means to perform a search of the learning dictionary using the character string, and the character string extracted by the search is converted into a morpheme After executing a plurality of cycles of step C received from the character conversion means as candidates, a step D of selecting and outputting a predetermined number of candidates obtained in step B and step C of each cycle is executed,
In step C, if the character string extracted in step A is a hiragana character string, a processing result indicating that a morpheme candidate could not be obtained without passing the character string to the character conversion means is set. And
In the step D, the katakana character string, the alphabet character string, and the number character string obtained in the step C are selected in preference to the candidates obtained in the step B.
A program for morphological analysis characterized by this . An operation section accepts an input and conversion instruction operation of kana string from the operation unit, and a character conversion means for converting a kana character string to a character string other forms entered, a character string of kanji and kana An apparatus comprising morpheme analysis means for receiving input of text data including and decomposing the text data into morphemes,
The character conversion means includes
A standard dictionary for conversion in which a predetermined number of converted character strings are registered;
A learning dictionary for registering new converted character strings,
In response to accepting an operation to instruct conversion processing based on dictionary search after accepting an input of a kana character string from the operation unit, search the standard dictionary for conversion and the learning dictionary with the inputted kana character string. Candidate search means for extracting candidates for the converted character string;
After receiving the input of the kana character string from the operation unit, the input in response to receiving the conversion operation after selecting a character type from katakana, hiragana, alphabet, or number and instructing conversion A post-conversion operation processing means for converting the kana character string into a character string of the designated character type ,
The post-conversion operation time processing means converts the post-conversion character string that is not registered in the conversion standard dictionary and the learning dictionary into the learning dictionary, and the morpheme analysis means obtains the character string to be searched. When it receives, it comprises a learning dictionary processing means for searching the learning dictionary with this character string, and passing the converted character string extracted by the search to the morpheme analyzing means ,
The morphological analysis means includes
A dictionary for morphological analysis in which a predetermined number of morphemes are registered;
When input of text data to be processed is accepted, a step A for extracting a character string of a predetermined length from the text data, and a search for the morpheme analysis dictionary using the character string extracted in step A And the character string extracted in step A is passed to the learning dictionary processing means of the character conversion means, and is passed from the learning dictionary processing means that executes the search of the learning dictionary using the character string. Search means for executing a plurality of cycles of step C for accepting the character string as a morpheme candidate ;
Output means for selecting and outputting a predetermined number of candidates obtained in step B and step C of each cycle by the search means;
In the case where the character string extracted in step A is a hiragana character string, the search means selects a morpheme candidate in step C without passing the extracted character string to the learning dictionary processing means. Set the processing result that was not obtained,
The information processing apparatus , wherein the output unit selects the katakana character string, the alphabet character string, and the number character string obtained in step C in preference to the candidate obtained in step B.

Images