JP5229448B2 - Reading imparting device and program - Google Patents

Description

  The present invention relates to a reading imparting device that outputs information about how to read a keyword.

  Conventionally, there has been a reading device as shown below. The reading device is provided with a language model in which pronunciation of words (part of speech information is unnecessary) and a language model that distinguishes and gives connection probabilities between words having the same notation and different pronunciations. Then, the hypothesis generation / deletion unit of the reading device adds a hypothesis of a reading candidate for the Japanese text by the dictionary, and the hypothesis calculation unit connects the words in the hypothesis based on the language model. Computes the product of probabilities, merges so that only the hypothesis with the highest connection probability product between words remains, selects the hypothesis with the highest product of the connection probabilities between words at the end of the sentence, and its pronunciation sequence Is output (for example, see Patent Document 1).

In general, word reading information is indispensable for systems such as speech recognition, speech synthesis, and machine translation. The word reading information is usually incorporated into the system in the form of a reading dictionary.
Japanese Unexamined Patent Publication No. 2003-132052 (first page, FIG. 1 etc.)

  However, in conventional reading imparting devices and the like, there are many words that are not described in the dictionary, such as newly created keywords and proper nouns, which cause processing failures such as speech recognition, speech synthesis, and machine translation. It was.

  The reading assigning device according to the first aspect of the present invention relates to a receiving unit that receives a keyword, and a reading dictionary that stores two or more term reading information having a term and reading information indicating reading of the term. A search dictionary search unit that searches using a target character string that is a keyword received by the user or two or more divided character strings that are character strings obtained by dividing the keyword, and acquires two or more reading information, and the target character string And a document group using each acquisition target character string reading information, which is a set of each reading information among the two or more reading information acquired by the reading dictionary search unit, and each of the acquisition target character string reading information A co-occurrence frequency acquisition unit that acquires, for each reading information, a co-occurrence frequency that is the number of documents in which the target character string and the reading information co-occur, and a co-occurrence for each reading information acquired by the co-occurrence frequency acquisition unit Using frequency, at least A reading application device having a highest occurrence frequency with many reading information or the most co-occurrence frequency with many read output unit for outputting the combination information.

  With this configuration, keyword reading information is acquired using a group of documents that are currently described, such as a homepage on the Web, or a group of documents that reflect the current activities of human beings, so that the keyword reading information can be acquired with high accuracy. .

  Further, in the reading imparting device of the second invention, in contrast to the first invention, the reading dictionary search unit stores two or more term reading information having a term and reading information indicating the reading of the term. A reading information acquisition unit that searches the reading dictionary using the target character string and acquires two or more reading information, and if the keyword includes a katakana character string, the katakana character string is hiragana It is a reading provision apparatus provided with the conversion means to convert.

  With such a configuration, a document group such as a homepage on the Web often uses “Hiragana” rather than “katakana” for reading, and keyword reading information is acquired using the characteristics of the document group. The reading information of keywords can be acquired with high accuracy.

  Further, in the reading imparting device according to the third aspect of the present invention, in contrast to the second aspect, the conversion means determines whether or not the keyword includes a predetermined katakana and includes the katakana. Is a reading imparting device that does not convert the katakana character string into hiragana.

  With such a configuration, when “V” that does not exist as a hiragana character, or in particular, a character that describes how to read in katakana, etc., the keyword reading information is acquired using the characteristics of the character, so the keyword Reading information can be acquired.

  Further, in the reading imparting device according to the fourth aspect of the present invention, with respect to any one of the first to third aspects, the reading dictionary search unit divides the keyword received by the receiving unit into units of one or more characters, Keyword dividing means for obtaining two or more divided character strings that are divided character strings, and one or more reading information paired with a term that matches each of the two or more divided character strings, from the reading dictionary, the divided characters This is a reading imparting device including a reading information acquisition unit that searches for each column and acquires two or more acquisition target character string reading information having divided character strings and reading information.

  With such a configuration, it is possible to acquire reading information of a keyword with high accuracy even for a long keyword, a keyword composed of compound words, and the like.

  Further, in the reading imparting device according to the fifth invention, in contrast to the fourth invention, the keyword dividing means acquires the character string length of the keyword received by the receiving unit, and the character string length is determined in advance. If the character string length is longer than a predetermined character string length or longer than a predetermined character string length, the keyword is not divided, and the reading information acquisition unit accepts the keyword when the keyword is not divided. This is a reading imparting device that acquires one or more reading information paired with a term that matches the keyword from the reading dictionary.

  With this configuration, it is possible to prevent an enormous amount of processing, and to acquire keyword reading information in an appropriate time.

  Further, in the reading imparting device according to the sixth aspect of the present invention, in contrast to the fourth aspect, the keyword dividing means is configured such that the keyword received by the receiving unit in order from the case where the number of cuts N (N is 0 or more) is small. Is divided into units of one character or more, one or more divided character strings that are the divided character strings are obtained, and division of the keyword is repeated until an end instruction is given from the reading information obtaining means, thereby obtaining the reading information The means acquires, for each divided character string, one or more reading information that is paired with a term that matches the one or more divided character strings acquired by the keyword dividing means, and acquires one or more divided characters. When reading information corresponding to all of the columns cannot be acquired, the keyword dividing unit is instructed to acquire two or more divided character strings that are division patterns, and the one or more Split character If the reading information corresponding to all of the one or more divided character strings can be acquired, the reading information corresponding to all of the one or more divided character strings is at least temporarily stored in the storage medium, and the keyword division is performed. An end instruction is passed to the means, or two or more divided character strings that are other division patterns having the same number of cuts as when reading information corresponding to all of the one or more divided character strings can be obtained The reading giving device that instructs the dividing unit to acquire, and after completing the reading information acquisition process for the other division patterns of the same number of cuts, passes the end instruction to the keyword dividing unit.

  With such a configuration, it is possible to acquire reading information of a keyword with high accuracy even for a long keyword, a keyword composed of compound words, and the like.

  Further, in the reading imparting device according to the seventh aspect of the present invention, with respect to any one of the first to sixth aspects, the reading dictionary search unit determines whether or not the target character string includes Arabic numerals. In the case of including a number, the reading imparting device further includes an Arabic numeral processing unit that generates digit number reading information indicating a reading with the Arabic numeral as a digit, and adds the digit reading information as the reading information. is there.

  With this configuration, reading information can be obtained with high accuracy even for keywords including Arabic numerals.

  Further, in the reading imparting device according to the eighth aspect of the present invention, with respect to any one of the first to seventh aspects, the reading dictionary search unit determines whether or not the target character string includes an alphabet, If included, the reading imparting apparatus further includes alphabet processing means for generating Roman character reading information indicating the Roman character reading corresponding to the alphabet and adding the Roman character reading information as the reading information.

  With this configuration, reading information can be acquired with high accuracy even for keywords including alphabets.

  In addition, in the ninth aspect of the present invention, in the first to eighth aspects of the invention, the co-occurrence frequency acquisition unit includes two or more acquisition target character strings acquired by the reading dictionary search unit. The document group is searched using the reading information, and the keyword or the divided character string included in each of the acquisition target character string reading information and the co-occurrence frequency that is the number of documents in which the reading information co-occurs are obtained. Co-occurrence frequency acquisition means for acquiring each information, and the acquisition target character string reading by parameterizing the co-occurrence frequency corresponding to the acquisition target character string reading information having the keyword or two or more divided character strings received by the reception unit For each combination of information or character string reading information to be acquired, score calculating means for calculating a score is provided, and the output unit uses the score acquired by the co-occurrence frequency acquisition unit and at least has the highest co-occurrence frequency. Reading A reading application device for outputting a combination of information or reading information.

  With this configuration, it is possible to output information on appropriate reading candidates.

  According to the reading imparting apparatus of the present invention, it is possible to acquire keyword reading information with high accuracy.

Hereinafter, embodiments of a reading imparting device and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.
(Embodiment 1)

  In this embodiment, in response to a keyword input, a reading dictionary is searched to obtain information on reading candidates (referred to as reading information), and a group of documents (for example, a set of web pages on the Web) using the keyword and reading information as keys. ) Is acquired, the co-occurrence number of the keyword and the reading candidate is acquired, the reading information to be output is determined using the co-occurrence number, and the reading imparting apparatus 1 that outputs the reading information will be described.

  Further, in the reading assigning apparatus 1 of the present embodiment, usually, keywords are divided, divided character strings are acquired, reading dictionaries are searched using the divided character strings, and reading candidate information (referred to as reading information) is obtained. get. Then, the co-occurrence number of the divided character string and the reading candidate is acquired, the reading information to be output is determined using the co-occurrence number, and the reading information is output. In addition, the process which does not perform a division | segmentation when a keyword is more than predetermined length is also demonstrated. In the reading imparting apparatus 1 according to the present embodiment, a description will be given of processing that normally performs processing in order from the smallest number of keyword cuts, and stops processing at the read number of cuts.

  Moreover, in the reading provision apparatus 1 of this Embodiment, hiragana is mainly used for reading information. Moreover, in the reading provision apparatus 1 of this Embodiment, when the predetermined character (for example, "V") is included, reading information is set to katakana.

  In addition, in the reading assigning apparatus 1 according to the present embodiment, a process of generating a reading with a digit and adding it to a candidate in the case of Arabic numerals will be described. In addition, in the case of alphabets, a process for adding a romaji reading will also be described.

  Further, description will be given of ranking each divided keyword using the number of co-occurrence and obtaining a predetermined number of reading candidates using the ranking.

  FIG. 1 is a conceptual diagram of a system including a reading imparting apparatus 1 according to the present embodiment. This system includes a reading dictionary device 2 and a server device 3. The reading dictionary device 2 receives a term and outputs reading information corresponding to the term. Two or more reading information may be output. In addition, there may be cases where there is no output reading information. The reading dictionary device 2 stores a reading dictionary. The reading dictionary is a dictionary that stores two or more term reading information having a term and reading information indicating the reading of the term. The server device 3 stores a document group that is a set of one or more documents. The document group is, for example, a homepage on the Web or a set of newspaper articles. Further, the documents constituting the document group may be described in a description language such as HTML, XML, or SGML, may be text data, or may be a table or a record constituting a database. That is, the data structure, format, etc. of the documents constituting the document group do not matter. The reading assigning device 1, the reading dictionary device 2, and the server device 3 are connected by a network 4 such as the Internet.

  In FIG. 1, the reading imparting device 1 does not include the reading dictionary device 2 or the server device 3, but may include it.

  FIG. 2 is a block diagram of the reading imparting apparatus 1 in the present embodiment. The reading assigning apparatus 1 includes a receiving unit 11, a reading dictionary search unit 12, a co-occurrence frequency acquisition unit 13, and an output unit 14.

  The reading dictionary search unit 12 includes a keyword dividing unit 121, a reading information acquisition unit 122, a conversion unit 123, an Arabic numeral processing unit 124, and an alphabet processing unit 125.

  The co-occurrence frequency acquisition unit 13 includes co-occurrence frequency acquisition means 131 and score calculation means 132.

  The reception unit 11 receives a keyword. A keyword means a character string. The keyword may be a word, a compound word composed of a plurality of words, or a sentence including the word or compound word. In addition, when the reception part 11 receives a sentence, it is suitable to divide into a word (a compound word is included), and to perform the process which acquires the reading information mentioned later. The keyword is usually a noun, but may be a character string of another part of speech such as a verb or an adjective. Here, accepting includes accepting input from a user, accepting from another processing unit (module), receiving from another device, reading from a storage medium (such as a hard disk, DVD, or FD). The keyword input means may be anything such as a keyboard, mouse or menu screen. The accepting unit 11 can be realized by a device driver for input means such as a keyboard, control software for a menu screen, or the like.

  The reading dictionary search unit 12 searches the reading dictionary using the target character string, and acquires two or more reading information. The reading dictionary stores two or more terms reading information. The term reading information includes a term and reading information. Reading information is information indicating the reading of the corresponding term. The term reading information includes, for example, the term “Narita” and reading information “Narita”. The reading dictionary may be held by the reading assigning device 1 or may be held by an external device (reading dictionary device 2). Further, the reading dictionary may be one type or two or more types. The reading dictionary is, for example, one or more dictionaries among a general dictionary, a linkage dictionary, and a personal name dictionary. The general dictionary includes, for example, a place name dictionary, a national language dictionary, a Hanwa dictionary, a katakana English word dictionary, and the like. The target character string is a keyword received by the receiving unit 11 or two or more divided character strings. The divided character string is a character string obtained by dividing the keyword received by the receiving unit 11. Acquisition (division) of the divided character string is performed by the keyword division unit 121 described later. The search performed by the reading dictionary search unit 12 is a process of obtaining the reading information from the reading dictionary device 2 by passing the target character string to a search engine held by the reading dictionary device 2, for example. The search engine may be included in the reading dictionary search unit 12, may be included in another device, or may be another processing unit (not shown).

  The reading dictionary search unit 12 can usually be realized by an MPU, a memory, or the like. The processing procedure of the reading dictionary search unit 12 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The keyword dividing unit 121 divides the keyword received by the receiving unit 11 in units of one or more characters, and acquires two or more divided character strings that are divided character strings. The keyword dividing unit 121 acquires the character string length of the keyword accepted by the accepting unit 11, and when the character string length is greater than the predetermined character string length or greater than or equal to the predetermined character string length, It is preferable not to divide the keyword.

  Further, the keyword dividing unit 121 determines that the keyword received by the receiving unit 11 is a unit of one or more characters from the case where the number of cuts N (N is 0 or more) is small (that is, “0”, “1”, “2” in order). It is preferable to repeat the keyword division until one or more divided character strings, which are divided character strings, are acquired, and an end instruction is received from the reading information acquisition unit 122. Then, the divided character string obtained by dividing the keyword is passed to the reading information acquisition means 122. When the number of cuts N = 0, the keyword dividing unit 121 does not perform keyword division.

  The keyword dividing unit 121 can usually be realized by an MPU, a memory, or the like. The processing procedure of the keyword dividing unit 121 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The reading information acquisition unit 122 searches the reading dictionary using the target character string, and acquires two or more reading information. Note that the reading dictionary stores two or more term reading information having a term and reading information indicating the reading of the term. The reading information acquisition unit 122 searches the reading dictionary for one or more reading information paired with a term that matches two or more divided character strings for each divided character string, and obtains the divided character string and the reading information. Two or more acquisition target character string reading information is acquired. Also, the reading information acquisition unit 122 acquires, from the reading dictionary, one or more reading information that is paired with a term that matches the keyword received by the receiving unit 11 when the keyword is not divided.

  The reading information acquisition unit 122 attempts to acquire, from the reading dictionary, one or more pieces of reading information paired with a term that matches one or more divided character strings acquired by the keyword dividing unit 121 for each divided character string. When the reading information acquisition unit 122 cannot acquire reading information corresponding to all of the one or more divided character strings, the reading information acquisition unit 122 instructs the keyword dividing unit 121 to perform two or more divisions which are the next division patterns. Tells to get a string. When the reading information acquisition unit 122 acquires reading information corresponding to all of the one or more divided character strings, the reading information acquisition unit 122 stores at least the reading information corresponding to all of the one or more divided character strings in the storage medium. Acquires reading information to be stored. Then, the reading information acquisition unit 122 may immediately give an end instruction to the keyword dividing unit 121 when the acquisition process is performed, or reading information corresponding to all of one or more divided character strings can be acquired. Instructs the keyword dividing unit 121 to acquire two or more divided character strings that are other division patterns having the same number of cuts as the case, and obtains reading information for other division patterns having the same number of cuts. After the processing is completed, an end instruction is given to the keyword dividing unit 121. That is, there may be at least two pieces of end information of the above process. Further, as many division patterns as possible may be configured, and reading information acquisition processing may be performed.

  The reading information acquisition unit 122 can be usually realized by an MPU, a memory, or the like. The processing procedure of the reading information acquisition unit 122 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  When the target character string includes a katakana character string, the conversion unit 123 converts the katakana character string into hiragana. In addition, the conversion unit 123 determines whether or not the katakana character string includes a predetermined katakana (for example, “V”). If the katakana character string includes the katakana, the conversion means 123 does not convert the katakana into hiragana. Is preferred.

  The conversion unit 123 can be usually realized by an MPU, a memory, or the like. The processing procedure of the conversion means 123 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The Arabic numeral processing means 124 determines whether or not the target character string includes Arabic numerals. If the target character string includes Arabic numerals (for example, “123”), the Arabic numeral processing means 124 includes digits indicating the reading with the Arabic numerals as digits. Number reading information (for example, “Hyaku Niju-san”) is generated, and the number reading information with digits is added as reading information. Note that whether or not the target character string includes Arabic numerals can be determined based on whether or not the target character string includes characters “0” to “9”. Moreover, the process which produces | generates the number reading information with a digit from an Arabic number is a well-known technique. In other words, reading rules corresponding to the number of digits (for example, “Hyaku” for the third digit, “10” for the second digit, etc.) and information on how to read the numerical values (“1” is “1”, “2” is Digit reading information with digits is generated from Arabic numerals using a specific rule (for example, “1” after the second digit is not added to the reading). The Arabic numeral processing means 124 can be usually realized by an MPU, a memory or the like. The processing procedure of the Arabic numeral processing means 124 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The alphabet processing unit 125 determines whether or not the target character string includes an alphabet. If the target character string includes the alphabet, the alphabet processing unit 125 generates Roman character reading information indicating the Roman character reading corresponding to the alphabet, and uses the Roman character reading information as the reading information. Add. A process for determining whether or not the target character string includes an alphabet and a process for generating information on reading Roman characters from the alphabet are also known techniques. Further, the alphabet processing means 125 determines whether or not the target character string includes an alphabet. If the target character string includes the alphabet, the alphabet processing unit 125 generates character unit reading information when the alphabet is read out character by character. It is preferable to add reading information as reading information. That is, when the target character string is “ABC”, the alphabet processing unit 125 preferably generates character unit reading information “Ebisu-shi” and adds it as reading information. The alphabet processing means 125 can be usually realized by an MPU, a memory, or the like. The processing procedure of the alphabet processing means 125 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The co-occurrence frequency acquisition unit 13 searches the document group using the target character string and the reading information acquired by the reading dictionary search unit 12, and the co-occurrence is the number of documents in which the target character string and the reading information co-occur. The frequency is acquired for each reading information. Information on a set of the target character string and reading information is also referred to as acquisition target character string reading information. In other words, the co-occurrence frequency acquisition unit 13 uses each acquisition target character string reading information that is a set of the target character string and each of the reading information of the two or more reading information acquired by the reading dictionary search unit 12. Then, the document group is searched, and the co-occurrence frequency that is the number of documents in which the reading information co-occurs with the target character string included in each acquisition target character string reading information is acquired for each reading information. Here, the document group may exist locally or on a network such as the Web (external). A document group is a set of documents. The document may be anything such as a homepage on the web, a record in a database, or an e-mail document. A document that is a target for determining whether or not to co-occur is usually one home page (one file) on the Web, but may be a plurality of files such as one site. The acquisition target character string reading information is a keyword and its reading information, or a divided character string and its reading information.

  The co-occurrence frequency acquisition unit 13 can usually be realized by an MPU, a memory, or the like. The processing procedure of the co-occurrence frequency acquisition unit 13 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The co-occurrence frequency acquisition means 131 searches the document group using each of the two or more acquisition target character string reading information acquired by the reading dictionary search unit 12, and includes keywords or divided character strings included in the acquisition target character string reading information. And the co-occurrence frequency, which is the number of documents in which reading information co-occurs, is acquired for each acquisition target character string reading information (for each reading information).

  The co-occurrence frequency acquisition unit 131 can be usually realized by an MPU, a memory, or the like. The processing procedure of the co-occurrence frequency acquisition unit 131 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The score calculation means 132 parameters the co-occurrence frequency corresponding to the acquisition target character string reading information having the keyword or two or more divided character strings received by the receiving unit 11, and acquires the acquisition target character string reading information or the acquisition target character string. A score is calculated for each combination of reading information.

  The calculation of the score is usually a calculation by an increasing function using the co-occurrence frequency as a parameter. For example, the formula for calculating the score is an expression for calculating the sum of co-occurrence frequencies corresponding to each divided character string of two or more divided character strings, an expression for calculating an average of co-occurrence frequencies corresponding to each divided character string, An expression for obtaining the median value of the co-occurrence frequencies corresponding to the divided character strings. It can be said that the score is a numerical value corresponding to the certainty and likelihood that would be read as such.

  The score calculation unit 132 can be usually realized by an MPU, a memory, or the like. The processing procedure of the score calculation means 132 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  Using the co-occurrence frequency for each piece of reading information acquired by the co-occurrence frequency acquisition unit 13, the output unit 14 outputs at least reading information with the highest co-occurrence frequency or a combination of reading information with the highest co-occurrence frequency. Using the score acquired by the co-occurrence frequency acquisition unit 13, the output unit 14 outputs at least reading information with the highest co-occurrence frequency or a combination of reading information with the highest co-occurrence frequency. For example, it is preferable that the output unit 14 outputs a predetermined number (for example, “3”) of reading information from the side with the higher co-occurrence frequency. Even in such a case, it is more preferable that the output unit 14 outputs the reading information of the fourth place or lower of the same score as the third place. Here, output refers to display on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage in a recording medium, other processing device or other program, etc. It is a concept including delivery of processing results. The output unit 14 may be considered as including or not including an output device such as a display or a speaker. The output unit 14 can be realized by output device driver software, or output device driver software and an output device.

  Next, the operation of the reading imparting apparatus will be described using the flowchart of FIG.

  (Step S301) The reception unit 11 determines whether a keyword has been received. If a keyword is accepted, the process goes to step S302, and if no keyword is accepted, the process returns to step S301.

  (Step S302) The reading dictionary search unit 12 performs reading candidate generation processing using the keyword received in step S301. The reading candidate generation process will be described with reference to the flowchart of FIG.

  (Step S303) The co-occurrence frequency acquisition unit 131 of the co-occurrence frequency acquisition unit 13 determines whether or not the reading information (reading candidate information) generated in step S302 is empty. If it is empty, go to step S301, and if it is not empty, go to step S304. In addition, when going to step S301, it is preferable to output an error message indicating that reading information cannot be output.

  (Step S304) The co-occurrence frequency acquisition means 131 searches the document group using the keyword received in step S301 and the reading information acquired in step S302, and acquires the co-occurrence frequency of the keyword and the reading information. When two or more pieces of reading information are acquired in step S302, the co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency of the keyword and the reading information corresponding to each reading information. Further, when the keyword reading information (reading information when the number of cuts is 0) cannot be acquired in step S302, this step and step S305 may be skipped and the process may be performed in step S306.

  (Step S305) The co-occurrence frequency acquisition unit 13 determines whether or not all the co-occurrence frequencies among the one or more co-occurrence frequencies acquired in step S304 are “0”. If all the co-occurrence frequencies are “0”, the process goes to step S306, and if any of the co-occurrence frequencies is not “0”, the process goes to step S307.

  (Step S306) The co-occurrence frequency acquisition unit 13 searches the document group using the reading information for the divided character string acquired in step S302, and acquires the co-occurrence frequency corresponding to the divided character string. Such a process is called a divided character string search process. The divided character string search process will be described with reference to the flowchart of FIG.

  (Step S307) The score calculation means 132 calculates a score for each reading information of the divided character string, using the co-occurrence frequency acquired in step S306. Such processing is called score calculation processing. The score calculation process will be described with reference to the flowchart of FIG.

  (Step S308) The output unit 14 constitutes information to be output. The information to be output has reading information or a combination of reading information having the highest co-occurrence frequency. Also, the information to be output is, for example, reading information consisting of N reading information with the highest score (N is determined in advance) or a combination of reading information with the highest score N.

  (Step S309) The output unit 14 outputs the information configured in step S308. The process returns to step S301.

  In the flowchart of FIG. 3, the process ends when the power is turned off or the process is terminated.

  Next, the reading candidate generation processing in step S302 will be described using the flowchart in FIG.

  (Step S401) The keyword dividing unit 121 acquires the number of characters of the keyword.

  (Step S402) The keyword dividing unit 121 determines whether or not the number of characters acquired in step S401 satisfies a predetermined relationship by comparing with a predetermined threshold (for example, “15”). If the predetermined relationship is satisfied, the process proceeds to step S403. If the predetermined relationship is not satisfied, the process proceeds to step S404. The case where the predetermined relationship is satisfied is, for example, a case where the number of characters is greater than the threshold or the number of characters is equal to or greater than the threshold.

  (Step S403) The reading information acquisition means 122 searches the reading dictionary using a keyword. That is, for example, the reading information acquisition unit 122 passes a keyword to a search processing unit (search engine) (not shown), and obtains reading information corresponding to a term that matches the keyword. The number of reading information obtained is an integer of 0 or more. Then, the reading information acquisition unit 122 temporarily stores the obtained reading information in a storage medium. Return to upper process.

  (Step S404) The keyword dividing unit 121 performs an initialization process. The initialization process is a process of substituting 0 for the counter i, 1 for the counter j, and substituting OFF (for example, “0”) for the end flag.

  (Step S405) The keyword dividing unit 121 determines whether or not there is a keyword dividing pattern having the number of cuts i. This determination is, for example, a determination as to whether “the number of characters in the keyword <= i”. If there is a division pattern with the number of cuts i, the process goes to step S406, and if there is no division pattern with the number of cuts i, the process goes to step S417.

  (Step S406) The keyword dividing unit 121 determines whether there is a dividing pattern of the j-th keyword with the number of cuts i. For example, when the number of characters is 5 and the number of cuts is 1, there are four division patterns. For example, in the case of a keyword with 5 characters and a cut number of 2, there are 6 division patterns.

  (Step S407) The keyword dividing unit 121 acquires all the divided character strings in the j-th division pattern with the number of cuts i, and arranges them on the memory.

  (Step S408) The keyword dividing unit 121 assigns 1 to the counter k.

  (Step S409) The keyword dividing unit 121 determines whether or not the k-th divided character string exists among the two or more divided character strings acquired in step S407. If the kth divided character string exists, the process goes to step S410, and if the kth divided character string does not exist, the process goes to step S415.

  (Step S410) The reading information acquisition unit 122 searches the reading dictionary using the k-th divided character string to obtain zero or more reading information. Here, for example, when there is only one dictionary constituting the reading dictionary, the keyword dividing unit 121 simply uses the divided character string as a key, and zero or more reading information paired with a term matching the divided character string. Get. Further, for example, when the dictionary constituting the reading dictionary is composed of “general dictionary”, “rendition dictionary”, “person name dictionary”, the reading information acquisition means 122 searches, for example, “general dictionary” and reads the reading information. If “No” is not obtained, a “relationship dictionary” may be searched, and if no reading information is obtained, a “person name dictionary” may be searched. Further, for example, when the dictionaries constituting the reading dictionary are composed of a “general dictionary”, “rendition dictionary”, and “person name dictionary”, the reading information acquisition unit 122 searches all three dictionaries, for example, Corresponding reading information may be acquired and placed on the memory. Note that the “Random Dictionary” is used when it is not the first divided character string.

  (Step S411) The reading information acquisition unit 122 determines whether the reading information acquired in step S410 is empty (“0”). If it is empty, go to step S412, and if it is not empty, go to step S413.

  (Step S412) The keyword dividing unit 121 increments the counter j by 1, and returns to Step S406.

  (Step S413) The reading dictionary search unit 12 performs an exception process. The exception process will be described with reference to the flowchart of FIG.

  (Step S414) The reading information acquisition unit 122 increments the counter k by 1, and returns to Step S409.

  (Step S415) The keyword dividing unit 121 temporarily stores at least the j-th division pattern with the number of cuts i and one or more reading information in each divided character string.

  (Step S416) The keyword dividing unit 121 sets the end flag to ON (for example, “1”). Go to step S412.

  (Step S417) The keyword dividing unit 121 determines whether or not the end flag is ON. If the end flag is ON, the process returns to the host process, and if the end flag is not ON, the process goes to step S418.

  (Step S418) The keyword dividing unit 121 increments the counter i by 1, and returns to step S405.

  In the flowchart of FIG. 4, processes such as step S401 and step S413 are not essential.

  Next, the divided character string search process in step S306 will be described with reference to the flowchart of FIG.

  (Step S501) The co-occurrence frequency acquisition unit 131 assigns 1 to the counter i.

  (Step S502) The co-occurrence frequency acquisition unit 131 determines whether or not the i-th division pattern exists. If the i-th division pattern exists, the process goes to step S503, and if the i-th division pattern does not exist, the process returns to the upper level process. The divided pattern is a divided pattern corresponding to two or more divided character strings in which reading information of each divided character string is temporarily stored in step S415 of the flowchart of FIG. The division pattern has, for example, two or more divided character strings.

  (Step S503) The co-occurrence frequency acquisition unit 131 assigns 1 to the counter j.

  (Step S504) The co-occurrence frequency acquisition unit 131 determines whether or not the j-th acquisition target character string reading information exists in the i-th division pattern. Here, the acquisition target character string reading information includes a target character string and reading information. If the jth acquisition target character string reading information exists, the process proceeds to step S505. If the jth acquisition target character string reading information does not exist, the process proceeds to step S509.

  (Step S505) The co-occurrence frequency acquisition unit 131 searches the document group using the target character string and the reading information included in the j-th acquisition target character string reading information as keys. The search for a document group is to give a target character string and reading information to a so-called Web search engine and start the search engine (or issue a start instruction).

  (Step S506) The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency corresponding to the j-th acquisition target character string reading information as a result of the search in step S505.

  (Step S507) The co-occurrence frequency acquisition unit 131 stores the co-occurrence frequency acquired in step S506 at least temporarily in a memory or the like in association with the j-th acquisition target character string reading information.

  (Step S508) The co-occurrence frequency acquisition unit 131 increments the counter j by 1, and returns to step S504.

  (Step S509) The co-occurrence frequency acquisition unit 131 increments the counter i by 1, and returns to step S502.

  Next, the score calculation process of step S307 will be described using the flowchart of FIG.

  (Step S601) The score calculation means 132 substitutes 1 for the counter i.

  (Step S602) The score calculation unit 132 determines whether or not the i-th division pattern exists. If the i-th division pattern exists, the process goes to step S603, and if the i-th division pattern does not exist, the process returns to the upper level process.

  (Step S603) The score calculation means 132 substitutes 1 to the counter j.

  (Step S604) The score calculation unit 132 determines whether there is a combination of j-th reading information in the i-th divided pattern. If the j-th reading information combination exists, the process goes to step S605, and if not, the process goes to step S609.

  (Step S605) The score calculation means 132 acquires all the co-occurrence frequencies corresponding to each reading information included in the j-th reading information combination.

  (Step S606) The score calculation means 132 calculates a score using the two or more co-occurrence frequencies acquired in step S605. The score calculation method is, for example, the sum of the ranks (numerical values) of the co-occurrence frequencies in the same divided character string. In this case, priority is given to the smaller score.

  (Step S607) The score calculation unit 132 temporarily stores at least the score calculated in Step S606 in association with the combination of the j-th reading information of the i-th divided pattern.

  (Step S608) The score calculation means 132 increments the counter j by 1, and returns to step S604.

  (Step S609) The score calculation means 132 increments the counter i by 1, and returns to step S602.

  Next, the exception process in step S413 (FIG. 4) will be described with reference to the flowchart of FIG.

  (Step S701) The conversion unit 123 determines whether or not the divided character string includes katakana. If katakana is included, the process proceeds to step S702, and if katakana is not included, the process proceeds to step S704.

  (Step S702) The conversion unit 123 determines whether or not the divided character string includes a predetermined katakana character. If a predetermined katakana character is included, the process returns to the upper process, and if it does not include a predetermined katakana character, the process goes to step S703.

  (Step S703) The conversion unit 123 converts the divided character string into hiragana, and uses the converted hiragana character string as reading information of the divided character string. Return to upper process.

  (Step S704) The alphabet processing means 125 determines whether or not the divided character string includes alphabets. If the alphabet is included, the process proceeds to step S705, and if the alphabet is not included, the process proceeds to step S707.

  (Step S705) The alphabet processing means 125 extracts an alphabetic character string of the divided character string, generates Roman character reading information indicating Roman reading from the character string, and adds the Roman character reading information as reading information.

  (Step S706) If an alphabet is included, character unit reading information when the alphabet is read out character by character is generated, and the character unit reading information is added as reading information. Return to upper process.

  (Step S707) The Arabic numeral processing means 124 determines whether or not the divided character string includes Arabic numerals. If Arabic numerals are included, the process proceeds to step S708, and if Arabic numerals are not included, the process returns to the higher-level processing.

  (Step S708) The Arabic numeral processing means 124 generates digit reading information with digits indicating reading with Arabic numerals as digits, and adds the digit reading information with digits as reading information. Return to upper process.

  Hereinafter, a specific operation of the reading imparting apparatus according to the present embodiment will be described. FIG. 1 is a conceptual diagram of a system including the reading imparting device 1.

Assume that the dictionary shown in FIGS. 8 to 10 is given as a reading dictionary. The dictionary in FIG. 8 is a general dictionary, for example, a dictionary created from a place name dictionary, a national language dictionary, a Hanwa dictionary, a Katakana English word dictionary, and the like. Further, the dictionary of FIG. 9 is a rendaku dictionary, and is created for words that can be reconstituted in the vocabulary of the general dictionary of FIG. Furthermore, FIG. 10 is a personal name dictionary. Here, the reading dictionary search unit 12 first checks a “general dictionary” for a given character string, and acquires reading information paired with a term that matches the character string. Then, if the character string is other than the beginning of the keyword, the “relationship dictionary” is examined to obtain reading information paired with a term matching the character string, and the acquired reading information is added. If none exists, the “person name dictionary” is checked to obtain reading information paired with a term matching the character string. The reading information corresponding to the character string is obtained by the algorithm as described above. Hereinafter, the operation of the reading imparting apparatus 1 will be described using two specific examples.
(Specific example 1)

  First, it is assumed that the keyword “Narita” is given to the reading assigning apparatus 1. Then, the reception unit 11 receives the keyword “Narita”.

  Next, the reading dictionary search unit 12 performs reading candidate generation processing as follows. That is, first, the keyword dividing unit 121 acquires the number of characters “2” of the keyword “Narita”. Next, the keyword dividing unit 121 determines that the acquired number of characters “2” does not satisfy a predetermined relationship (for example, “threshold <number of characters”) as compared with a predetermined threshold (15). .

  Next, the keyword dividing unit 121 obtains a keyword dividing pattern “Narita” with 0 cuts. Then, the reading information acquisition unit 122 searches the reading dictionaries of FIGS. 8 to 10 using the target character string “Narita” as a key. Here, the reading information acquisition unit 122 searches the reading dictionary of FIG. 8 using the target character string “Narita” as a key. Then, the reading information acquisition unit 122 obtains “Naru”, “Naruda”, and “Naruta”.

  Next, an exception process (described in FIG. 7) is to be performed, but it is determined that the exception process is not targeted.

  Then, the reading information acquisition unit 122 obtains the reading candidate management table shown in FIG. The reading candidate management table has “divided pattern”, “divided character string”, and “reading information”. “Division pattern” is information indicating a division pattern, and here, a keyword in which “|” is entered at a cutting point (division point) such as “Narita” (no division) or “Narita | Limousine Bus | Platform”. Information. Further, the “divided character string” has the same meaning as the target character string here. Furthermore, “reading information” is information indicating the reading of the corresponding divided character string.

  Next, the co-occurrence frequency acquisition unit 131 of the co-occurrence frequency acquisition unit 13 determines that the generated reading information (reading candidate information) is not empty (see FIG. 11).

  Then, the co-occurrence frequency acquisition unit 131 searches the document group using the received keyword “Narita” and the acquired first reading information “Narita”, and acquires the co-occurrence frequency of the keyword and the reading information. . That is, here, the co-occurrence frequency acquisition unit 131 passes “Narita, Narita” to a so-called Web search engine to execute a search. The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency (“116000” in this case) of the character string “Narita” and the character string “Narita” from the search engine. Then, the co-occurrence frequency acquisition unit 131 temporarily stores the character string “Narita”, the character string “Naruto”, and the co-occurrence frequency “116000” in the memory.

  Next, the co-occurrence frequency acquisition unit 131 searches the document group using the received keyword “Narita” and the acquired second reading information “Naruda”, and acquires the co-occurrence frequency of the keyword and the reading information. To do. The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency (“38400” in this case) of the character string “Narita” and the character string “Naruda” from the search engine. Then, the co-occurrence frequency acquisition unit 131 temporarily stores the character string “Narita” and the character string “Naruda” in the memory with the co-occurrence frequency “38400” as a pair.

  Further, the co-occurrence frequency acquisition unit 131 searches the document group using the received keyword “Narita” and the acquired second reading information “Naruta”, and acquires the co-occurrence frequency of the keyword and the reading information. . The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency (“1010” in this case) of the character string “Narita” and the character string “Naruta” from the search engine. The co-occurrence frequency acquisition unit 131 temporarily stores the character string “Narita”, the character string “Naruta”, and the co-occurrence frequency “1010” in the memory.

  Next, the score calculation means 132 calculates a score for each reading information of the divided character string using the acquired co-occurrence frequency. Here, it is assumed that the score is a rank. In other words, the score calculation means 132 has a score “1” for the character string “Narita” and the character string “Narita”, a score “2” for the character string “Narita” and the character string “Naruda”, a character string Score “3” is obtained for “Narita” and the character string “Naruta”. Here, the lower the score, the higher the priority as a reading candidate.

Next, the output unit 14 configures the information to be output as “1. is, 2. is, 3. is”. Then, the output unit 14 outputs two reading candidates for “Narita” in descending order of priority. That is, the output unit 14 outputs “1. Here, it is assumed that the output unit 14 stores, for example, the number “2” of reading information to be output.
(Specific example 2)

  Next, it is assumed that the keyword “Narita Limousine Bus Stop” is given to the reading imparting device 1. And the reception part 11 receives the keyword "Narita limousine bus platform".

  Next, the reading dictionary search unit 12 performs reading candidate generation processing as follows. That is, first, the keyword dividing unit 121 acquires the number of characters “13” of the keyword “Narita Limousine Bus Station”. Next, the keyword dividing unit 121 determines that the acquired number of characters “13” does not satisfy a predetermined relationship (for example, “threshold <number of characters”) as compared with a predetermined threshold (15). .

  Next, the keyword dividing unit 121 obtains a keyword division pattern “Narita Limousine Bus Platform” with a cut number of 0 (see division pattern ID = 1 in FIG. 12). Then, the reading information acquisition unit 122 searches the reading dictionaries of FIGS. 8 to 10 using the target character string “Narita Limousine Bus Station” as a key. Here, the reading information acquisition unit 122 searches the reading dictionary of FIG. 8 using the target character string “Narita Limousine Bus Stop” as a key. And the reading information acquisition means 122 cannot acquire the reading information corresponding to the object character string “Narita Limousine Bus Station”.

  Next, the keyword dividing unit 121 acquires a keyword dividing pattern with the number of cuts of one. That is, first, the keyword dividing unit 121 obtains the first division pattern “Sei | Ta Limousine Bus Station” with the number of cuts 1 (see division pattern ID = 2 in FIG. 12). “Sei | Ta Limousine Bus Station” means a divided character string “Nari” and a divided character string “Ta Limousine Bus Station”.

  And the reading information acquisition means 122 searches the reading dictionary of FIGS. 8-10 using the division | segmentation character string "sei", but reading information cannot be obtained.

  Next, the keyword dividing unit 121 obtains information that the divided character string “sei” cannot obtain reading information from the processing result of the reading information acquisition unit 122, and obtains the divided character string “Narita” and the divided character string “limousine”. "Bus stop" is acquired (see division pattern ID = 3 in FIG. 12).

  Then, the reading information acquisition unit 122 searches the reading dictionary of FIGS. 8 to 10 using the divided character string “Narita”, and reads the reading information “Naru”, “Naruda” and “Naruda” from the reading dictionary of FIG. Get ".

  Next, the reading information acquisition unit 122 searches the reading dictionaries of FIGS. 8 to 10 using the divided character string “Limousine bus stop”, but cannot read the reading information.

  Similarly, the keyword dividing unit 121 sequentially obtains divided character strings from the division pattern IDs “4” to “11” in FIG. Then, in order, the reading information acquisition unit 122 searches the reading dictionary using the divided character string acquired by the keyword dividing unit 121, but reading information cannot be obtained.

  Next, the keyword dividing unit 121 acquires the dividing pattern of the keyword with 2 cuts. That is, first, the keyword dividing unit 121 obtains the first division pattern “Narita | Li | Mujin Bus stop” with the number of cuts 1 (see division pattern ID = 12 in FIG. 12).

  Next, for the divided character string “Narita”, the reading information acquisition unit 122 obtains reading information “Narita”, “Naruda”, and “Naruta”. However, the reading information acquisition unit 122 searches the reading dictionary of FIGS. 8 to 10 for the divided character string “Li”, but cannot read the reading information.

  The above processing is repeated from the division pattern IDs “13” to “16” in FIG. 12, but reading information of all divided character strings cannot be obtained for each division pattern.

  Then, the keyword dividing unit 121 acquires the division pattern “Narita | Limousine bus | Place” (see division pattern ID = 17 in FIG. 12).

  Next, for the divided character string “Narita”, the reading information acquisition unit 122 obtains reading information “Narita”, “Naruda”, and “Naruta”. Then, the reading information acquisition unit 122 obtains the reading information “Rimujinbas” for the divided character string “limousine bus”. Further, for the divided character string “Platform”, “Park” and “Norijo” are obtained. That is, the reading information acquisition unit 122 acquires the reading information management table shown in FIG. 13 and arranges it on the memory. The reading information management table is a table that manages reading candidates, and includes “divided character strings” and “reading information”. Note that the split character strings “Narita” and “Platform” are not subject to exception processing. Further, the conversion unit 123 determines that the divided character string “limousine bus” does not include the predetermined character string “V”, and converts the “limousine bus” into the hiragana “Riminbus”. This conversion process may not be performed. This is because the reading information “Rimujinbas” has already been acquired. In addition, when “Limousine Bus, Rimu Jinba” does not exist in the reading dictionary of FIG. 8, exception processing for converting katakana into hiragana functions.

  Next, the keyword dividing unit 121 sequentially acquires other division patterns having a cut number of 2. The keyword dividing means 121 is, for example, “Narita | Limousine bus boarding | ground”, “Narita | Limousine bus boarding | field”, “Narita Ri | Mu | Jinbus boarding”, “Narita rim | Jimbus boarding”, “ Get Narita Limousine | Non Bus Stop, Narita Limousine | Bus Bus Stop, etc. in order. Then, the reading information acquisition unit 122 determines whether reading information of all divided character strings included in each divided pattern can be acquired, and the divided pattern capable of acquiring reading information of all divided character strings is “Narita | Limousine bus | It is determined that there is no place other than "Platform".

  Next, the co-occurrence frequency acquisition unit 131 acquires the reading information “Narita” of the first target character string “Narita” in the first division pattern “Narita | Limousine Bus | Place”.

  Next, the co-occurrence frequency acquisition unit 131 passes the target character string “Narita” and the reading information “Narita” included in the first acquisition target character string reading information to a search engine (regardless of the search engine). By operating a search engine, a document group (here, a home page) on the Web is searched. The co-occurrence frequency acquisition means 131 gives “Narita” and “Narita” as arguments to an API (for example, search ()) of the search engine and executes it (for example, search (Narita, Narita)). Run). The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “116000” in the document group on the Web for “Narita” and “Narita”. Then, the co-occurrence frequency acquisition unit 131 arranges “Narita, Narita, 116000” on the memory.

  Next, the co-occurrence frequency acquisition unit 131 acquires the second reading information “Naruda” of the first target character string “Narita” in the first division pattern “Narita | Limousine bus | Place”. Then, the co-occurrence frequency acquisition means 131 passes the terms “Narita” and “Naruda” to the search engine, and operates the search engine to search for a document group (here, a home page) on the Web. Then, the co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “38400” in the document group on the Web for “Narita” and “Naruda”. Then, the co-occurrence frequency acquisition unit 131 arranges “Narita, Naruda, 38400” on the memory.

  Next, the co-occurrence frequency acquisition unit 131 acquires the third reading information “Naruta” of the first target character string “Narita” in the first division pattern “Narita | Limousine Bus | Place”. Then, the co-occurrence frequency acquisition unit 131 passes the terms “Narita” and “Naruta” to the search engine, and operates the search engine to search a document group (here, a home page) on the Web. The co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “1010” in the document group on the Web of “Narita” and “(regardless of the search engine)”. Then, the co-occurrence frequency acquisition unit 131 arranges “Narita, Naruta, 1010” on the memory.

  Next, the co-occurrence frequency acquisition means 131 acquires the reading information “Rimujinbas” of the second target character string “limousine bus” in the first division pattern “Narita | Limousine bus | Place”. The co-occurrence frequency acquisition means 131 passes the terms “limousine bus” and “Rim Jinba” to the search engine, and operates the search engine to search a document group (here, a home page) on the Web. Then, the co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “61900” in the document group on the Web for “Limousine bus” and “Rim Jinbus”. Then, the co-occurrence frequency acquisition unit 131 arranges “Limousine bus, Rimujinsuda, 61900” on the memory.

  Next, the co-occurrence frequency acquisition unit 131 acquires the first reading information “Platform” of the third target character string “Platform” in the first division pattern “Narita | Limousine Bus | Platform”. Then, the co-occurrence frequency acquisition means 131 searches the document group on the Web (here, a home page) by passing the terms “depot” and “pick” to the search engine and operating the search engine. Then, the co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “62500” in the document group on the Web for “Platform” and “Platform”. Then, the co-occurrence frequency acquisition unit 131 arranges “Platform, platform, 62500” on the memory.

  Next, the co-occurrence frequency acquisition means 131 acquires the second reading information “Norijo” of the third target character string “Platform” in the first division pattern “Narita | Limousine Bus | Platform”. Then, the co-occurrence frequency acquisition means 131 searches the document group on the Web (here, the homepage) by passing the terms “Place” and “Norijo” to the search engine and operating the search engine. Then, the co-occurrence frequency acquisition unit 131 acquires the co-occurrence frequency “1” in the document group on the Web for “Platform” and “Norijo”. Then, the co-occurrence frequency acquisition unit 131 arranges “Platform, Norijo, 1” on the memory.

  Next, the score calculation means 132 calculates a score as follows, for example. That is, the score calculation unit 132 obtains the ranking of the reading information for the term “Narita” from the co-occurrence frequency “116000, 38400, 1010” of each reading information. That is, the score calculation unit 132 obtains “1: become”, “2: become”, and “3: become”.

  Next, similarly, the score calculation means 132 obtains the information “1: Rimujinba” of the ranking of the reading information for the term “rimujinba”.

  Furthermore, the score calculation means 132 obtains information “1: platform” and “2: route” of the ranking of the reading information for the term “depot”.

  Next, the score calculation means 132 calculates scores for all combinations of reading information. Here, it is assumed that the score is the sum of the ranks of the reading information. That is, the score calculation unit 132 calculates scores for all combinations of reading information, and obtains a score management table shown in FIG. The score of the score management table is obtained by adding the scores of the three reading information constituting the reading information of “Narita Limousine Bus Station”. In FIG. 14, the smaller the score, the higher the priority as a reading candidate.

Next, the output unit 14 configures information to be output. Here, the output unit 14 obtains one reading information with the highest priority, “Narimu Mujinba Stop”, and configures the data structure to be output. Next, the output unit 14 outputs the configured information “Naritar Mujinba Bus Stop”.
(Experimental result)

  Hereinafter, experimental results of the reading imparting apparatus 1 will be described. In this experiment, many landmarks listed in the guidebook were selected as keywords. And the experimental result using the method of the reading provision apparatus 1 is shown in FIG. In FIG. 15, “985.53%”, which is the sum of the “ratio” of the portions surrounded by the ellipse, is the correct answer rate. That is, the reading imparting device 1 can acquire reading information with a very high accuracy rate. In FIG. 15, the first value “23.98%” surrounded by an ellipse is estimated to be an effect of the above division processing. The second value “69.17%” surrounded by an ellipse is estimated to be obtained as correct reading information when two or more reading information candidates exist. Further, the third value “5.38%” surrounded by an ellipse is estimated to be obtained as correct reading information due to the effect of the large vocabulary dictionary.

  As described above, according to the present embodiment, it is possible to acquire keyword reading information with very high accuracy.

  According to the present embodiment, an algorithm for dividing a keyword and obtaining two or more divided character strings is not limited to the algorithm described above. A divided character string may be obtained by a procedure such as

  Furthermore, the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded and distributed on a recording medium such as a CD-ROM. This also applies to other embodiments in this specification. Note that the software that implements the information processing apparatus according to the present embodiment is the following program. That is, the program accepts the computer with respect to a reading dictionary storing a keyword and a reading dictionary storing two or more term reading information including a term and reading information indicating reading of the term. A keyword dictionary or a character string obtained by dividing the keyword and a target character string that is two or more divided character strings, and obtains two or more reading information; the target character string; A document group is searched using each acquisition target character string reading information that is a set of each reading information among two or more reading information acquired by the reading dictionary search unit, and each acquisition target character string reading information has. The co-occurrence frequency acquisition unit that acquires the co-occurrence frequency, which is the number of documents in which the target character string and the reading information co-occur, for each reading information, and the co-occurrence frequency for each reading information acquired by the co-occurrence frequency acquisition unit. Use A program, to function as an output unit for outputting a combination of more reading information or reading information of at least the most co-occurrence frequency.

  In the above program, the reading dictionary search unit searches the reading dictionary storing two or more term reading information having a term and reading information indicating the reading of the term using the target character string. A computer having a reading information acquisition means for acquiring two or more reading information and a conversion means for converting the katakana character string into hiragana when the target character string includes a katakana character string; It is preferable that the program is for causing the program to occur.

  Further, in the above program, the conversion means determines whether or not the target character string includes a predetermined katakana, and if it includes the katakana, does not convert the katakana character string into hiragana It is preferable that the program is for causing a computer to function.

  In the above program, the reading dictionary search unit divides the keyword received by the receiving unit in units of one character or more, and obtains two or more divided character strings that are divided character strings; One or more reading information paired with a term matching each of the two or more divided character strings is retrieved from the reading dictionary for each divided character string, and the acquisition target character string reading information having the divided character string and the reading information Is preferably a program for causing a computer to function as a reading information acquisition means for acquiring two or more.

  In the above program, the keyword dividing unit obtains the character string length of the keyword received by the receiving unit, and the character string length is larger than a predetermined character string length or a predetermined character string length. In the above case, the keyword is not divided, and the reading information acquisition means is one or more reading information paired with a term that matches the keyword received by the receiving unit when the keyword is not divided. Is obtained from the reading dictionary, and is preferably a program for causing a computer to function.

  In the above program, the keyword dividing means divides the keyword received by the receiving unit in order of one character or more in order from the case where the number of cuts N (N is 0 or more) is small, and the divided character string One or more divided character strings are acquired, and the keyword division is repeated until an end instruction is received from the reading information acquisition unit, and the reading information acquisition unit includes one or more divisions acquired by the keyword dividing unit. When one or more reading information paired with a term matching the character string is obtained from the reading dictionary for each divided character string, and reading information corresponding to all of the one or more divided character strings cannot be obtained. Instructs the keyword dividing means to acquire two or more divided character strings which are the next division patterns, and reads information corresponding to all of the one or more divided character strings. Is acquired, the reading information corresponding to all of the one or more divided character strings is read at least temporarily stored in a storage medium, and an end instruction is passed to the keyword dividing means, Alternatively, two or more divided character strings, which are other division patterns having the same number of cuts as when reading information corresponding to all of the one or more divided character strings has been acquired, are acquired from the keyword dividing unit. A program for causing a computer to function as an instruction to pass an end instruction to the keyword dividing means after completion of reading information acquisition processing for other division patterns of the same number of cuts. Is preferred.

  In the above program, the reading dictionary search unit determines whether or not the target character string includes Arabic numerals. If the target character string includes Arabic numerals, the reading dictionary search section includes digits indicating the reading with the Arabic numerals as digits. It is preferable that the program is for causing a computer to function as further comprising Arabic numeral processing means for generating numeral reading information and adding the digit reading information with digits as reading information.

  Further, in the above program, the reading dictionary search unit determines whether or not the target character string includes an alphabet, and if it includes the alphabet, generates romaji reading information indicating a romaji reading corresponding to the alphabet, A program for causing a computer to function is preferable as further comprising alphabet processing means for adding the Roman character reading information as reading information.

In the above program, the co-occurrence frequency acquisition unit searches for a document group using each of the two or more acquisition target character string reading information acquired by the reading dictionary search unit, and each of the acquisition target character string reading information A co-occurrence frequency acquisition unit that acquires, for each acquisition target character string reading information, a co-occurrence frequency that is the number of documents or reading character co-occurrence, and a keyword received by the reception unit, A score calculation means for calculating a score for each combination of the acquisition target character string reading information or the acquisition target character string reading information by setting the co-occurrence frequency corresponding to the acquisition target character string reading information having two or more divided character strings. And the output unit outputs at least reading information or a combination of reading information having the highest co-occurrence frequency using the score acquired by the co-occurrence frequency acquiring unit. As the, it is preferable that a program for causing a computer to function.
(Embodiment 2)

  In the present embodiment, an application example of the reading imparting apparatus 1 described in the first embodiment will be described.

  First, the reading imparting device 1 can be used for constructing a dictionary of a speech synthesizer, a speech recognition device, or a machine translation device. That is, the reading imparting device 1 functions when creating a speech synthesis dictionary having a plurality of read speech information having a pair of “reading information” and “speech” necessary for the speech synthesizer. In addition, the reading imparting device 1 functions when creating a speech recognition dictionary having a plurality of speech recognition information having a pair of “reading information”, “term”, and “speech” necessary for the speech recognition device. Furthermore, the reading imparting apparatus 1 functions when the word information of a pair of “English words” made up of proper nouns such as “person names” and “place names” and Roman letters is configured, which is necessary for the machine translation device. That is, the reading information becomes parallel translation information, and the reading assigning apparatus 1 can be used to construct a translation dictionary. For example, the reading imparting apparatus 1 acquires “Sensoji” from “Asakusa Temple”, obtains the Roman character “Sensoji” from the “Sensoji” hiragana by a known means (not shown), and “(Japanese) Sensoji Temple (English ) Sensoji "translation information.

The speech synthesizer can use the reading imparting device 1 as follows. That is, the speech synthesizer holds a speech database. This speech database stores speech corresponding to hiragana notation for each hiragana notation. Preferably, the speech database stores speech corresponding to katakana notation for each katakana notation. Then, the speech synthesizer of the speech synthesizer reads speech corresponding to the reading information (kana notation) output by the reading imparting device 1 from the speech database, synthesizes speech, and acquires an analog speech signal. The speech output unit of the speech synthesizer then synthesizes speech by the speech synthesizer and outputs the acquired analog speech signal.
In the case of a speech translation apparatus, the reading imparting apparatus 1 can be used as follows. A speech translation apparatus usually includes a speech recognition unit and a translation unit. Then, reading information is obtained from a speech recognition unit included in the speech translation apparatus. Next, the translation unit determines the meaning of the word using the reading information. For example, “Koufu” is translated as “worker”, and “Kufuu” is translated as “device”. Also, “Sawara” is translated as “sawara”, and “Sahara” is translated as “sahara”. Furthermore, “how” is translated as “person”, and “how” is translated as “direction”.

  Furthermore, the machine translation apparatus can use the reading provision apparatus 1 as follows. In other words, the machine translation device holds a translation dictionary. The translation dictionary includes a pair of reading information of a Japanese word and a word of a target language (for example, English or Chinese). The translation dictionary preferably includes a pair of a Japanese word, its reading information, and a target language word. The input reception unit receives a Japanese sentence to be translated. Next, the morphological analysis unit of the machine translation device performs morphological analysis on the sentence received by the input reception unit of the machine translation device, and breaks it down into morphemes. And the reading provision apparatus 1 acquires the reading information of the said keyword by using the word of the output of a morphological analysis part as an input keyword. Next, the word search unit of the machine translation device uses the keyword reading information acquired by the reading assigning device 1 to acquire a word in the target language paired with the reading information from the translation dictionary. And the translated sentence structure part of a machine translation apparatus obtains the translation sentence of the sentence which the input reception part received using the word of the target language which the result of the morphological analysis part acquired by the word search part. The translated sentence output unit of the machine translation device outputs the translated sentence acquired by the translated sentence constituent unit.

  As described above, the reading imparting device 1 can be used in various devices.

  In each of the above embodiments, each process (each function) may be realized by centralized processing by a single device (system), or by distributed processing by a plurality of devices. May be.

  FIG. 16 shows the external appearance of a computer that executes the program described in this specification to realize the reading imparting apparatus and the like of the various embodiments described above. The above-described embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 16 is an overview diagram of the computer system 340, and FIG. 17 is a block diagram of the computer system 340.

  In FIG. 16, a computer system 340 includes a computer 341 including an FD (Flexible Disk) drive and a CD-ROM (Compact Disk Read Only Memory) drive, a keyboard 342, a mouse 343, a monitor 344, and a microphone 345. .

  In FIG. 17, in addition to the FD drive 3411 and the CD-ROM drive 3412, a computer 341 includes a CPU (Central Processing Unit) 3413, a bus 3414 connected to the CD-ROM drive 3412 and the FD drive 3411, and a boot-up program. ROM (Read-Only Memory) 3415 for storing programs such as a RAM, and a RAM (Random Access Memory) 3416 connected to the CPU 3413 for temporarily storing application program instructions and providing a temporary storage space , An application program, a system program, and a hard disk 3417 for storing data. Although not shown here, the computer 341 may further include a network card that provides connection to the LAN.

  A program that causes the computer system 340 to execute the function of the reading imparting device according to the above-described embodiment is stored in the CD-ROM 3501 or the FD 3502, inserted into the CD-ROM drive 3412 or the FD drive 3411, and further stored in the hard disk 3417. May be forwarded. Alternatively, the program may be transmitted to the computer 341 via a network (not shown) and stored in the hard disk 3417. The program is loaded into the RAM 3416 at the time of execution. The program may be loaded directly from the CD-ROM 3501, the FD 3502, or the network.

  The program does not necessarily include an operating system (OS), a third-party program, or the like that causes the computer 341 to execute the functions of the reading provision apparatus according to the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 340 operates is well known and will not be described in detail.

  Further, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the reading imparting apparatus according to the present invention has an effect that it can acquire keyword reading information with high accuracy, and is useful as a speech synthesizer or the like.

Conceptual diagram of a system including a reading provision apparatus according to Embodiment 1 Block diagram of the reading device Flow chart for explaining the operation of the reading granting device A flowchart for explaining the operation of the candidate generation process A flowchart for explaining the operation of the divided character string search process A flowchart for explaining the operation of the score calculation process Flow chart explaining the operation of the exception processing Figure showing the same reading dictionary Figure showing the same reading dictionary Figure showing the same reading dictionary Figure showing the same reading candidate management table Diagram showing the same division pattern Figure showing the same reading information management table Figure showing the score management table Figure showing the results of the experiment Overview of the computer system Block diagram of the computer system

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Reading provision apparatus 2 Dictionary apparatus 3 Server apparatus 11 Reception part 12 Dictionary search part 13 Co-occurrence frequency acquisition part 14 Output part 121 Keyword division means 122 Information acquisition means 123 Conversion means 124 Arabic numeral processing means 125 Alphabetical processing means 131 Co-occurrence frequency Acquisition means 132 Score calculation means

Claims (10)

A reception unit that accepts keywords;
The keyword received by the receiving unit is divided by two or more patterns, two or more divided patterns that are sets of two or more divided character strings are acquired, and term reading information having reading information indicating a term and reading of the term A reading dictionary storing two or more, and using the two or more division patterns, a reading dictionary search unit for acquiring two or more reading information for each division pattern ;
Each of the divided character strings that constitute the divided pattern, with respect to only the divided pattern for which the reading dictionary search unit has acquired reading information corresponding to all of the two or more divided character strings that constitute the divided pattern ; A document group is searched using each acquisition target character string reading information, which is a set of each reading information among the two or more reading information acquired by the reading dictionary search unit, and each of the acquisition target character string reading information has a division A co-occurrence frequency acquisition unit that acquires the co-occurrence frequency, which is the number of documents in which character strings and reading information co-occur , for each reading information, and calculates the score for each division pattern using the co-occurrence frequency for each reading information When,
A reading imparting apparatus comprising: an output unit that outputs at least a combination of reading information having the highest co-occurrence frequency using the score acquired by the co-occurrence frequency acquiring unit. The reading dictionary search unit
A keyword dividing unit that divides the keyword received by the receiving unit in units of one or more characters and acquires two or more divided character strings that are divided character strings;
One or more reading information paired with a term matching each of the two or more divided character strings is retrieved from the reading dictionary for each divided character string, and the acquisition target character string reading information having the divided character string and the reading information The reading imparting device according to claim 1, further comprising reading information acquisition means for acquiring two or more readings,
The keyword dividing means includes:
From the case where the number of cuts N (N is 0 or more) is small, the keyword received by the receiving unit is sequentially divided into units of one or more characters, and one or more divided character strings that are divided character strings are obtained, Until the reading information acquisition means receives an end instruction, the division of the keyword is repeated,
The reading information acquisition means includes
One or more reading information paired with a term that matches one or more divided character strings acquired by the keyword dividing means is acquired for each divided character string from the reading dictionary, and all of the one or more divided character strings are obtained. If the reading information corresponding to is not acquired, the keyword dividing unit is instructed to acquire two or more divided character strings which are the next division patterns, and the one or more divided character strings are obtained. If the reading information corresponding to all of the one or more divided character strings can be acquired, the reading information corresponding to all of the one or more divided character strings is at least temporarily stored in the storage medium, and the keyword division is performed. An end instruction is passed to the means, or two or more divided character strings that are other division patterns having the same number of cuts as when reading information corresponding to all of the one or more divided character strings can be obtained 2. The reading according to claim 1, wherein the keyword dividing unit is instructed to acquire, and after completion of the reading information acquisition process for the other division patterns of the same number of cuts, an end instruction is passed to the keyword dividing unit. Granting device. The reading dictionary search unit
If the keyword includes a string of Katakana reading application device according to claim 1 or claim 2 wherein comprises a converting means for converting a character string of the Katakana to Hiragana. The converting means includes
4. The reading assigning apparatus according to claim 3 , wherein whether or not the keyword includes a predetermined katakana is determined, and if the keyword includes the katakana, the character string of the katakana is not converted into hiragana. The keyword dividing means includes:
The character string length of the keyword received by the reception unit is acquired, and the keyword is divided when the character string length is greater than a predetermined character string length or greater than a predetermined character string length. Without
The reading information acquisition means includes
If you do not split the keyword, the described terms and one or more reading information paired to match the keyword acceptance unit accepts, in claim 4 any one of claims 2 to retrieve from the read dictionary Reading imparting device. The reading dictionary search unit
It is determined whether or not the target character string includes Arabic numerals, and when the target character strings include Arabic numerals, digit number reading information indicating reading with the Arabic numerals as digits is generated, and the digit reading information with digits is generated. The reading imparting apparatus according to any one of claims 1 to 5 , further comprising an Arabic numeral processing unit that adds the information as reading information. The reading dictionary search unit
It is determined whether or not the target character string includes an alphabet. If the target character string includes an alphabet, alphabet processing means for generating Roman character reading information indicating a Roman character reading corresponding to the alphabet and adding the Roman character reading information as reading information is provided. The reading provision apparatus as described in any one of Claims 1-6 which comprises . The co-occurrence frequency acquisition unit
A document group is searched using two or more acquisition target character string reading information acquired by the reading dictionary search unit, and a divided character string included in each acquisition target character string reading information and a document in which the reading information co-occurs A co-occurrence frequency acquisition means for acquiring the co-occurrence frequency that is the number of each for each division pattern and each acquisition target character string reading information;
And parameter co-occurrence frequency corresponding to the retrieval target character string reading information having the two or more divided strings for each division pattern, comprising a score calculating means for calculating a score,
The output unit is
The reading imparting apparatus according to any one of claims 1 to 7, wherein a combination of at least reading information having the highest co-occurrence frequency is output using the score acquired by the co-occurrence frequency acquiring unit. Computer
A reception unit that accepts keywords;
The keyword received by the receiving unit is divided by two or more patterns, two or more divided patterns that are sets of two or more divided character strings are acquired, and term reading information having reading information indicating a term and reading of the term A reading dictionary storing two or more of the reading dictionary, using the two or more division patterns, to obtain two or more reading information for each division pattern;
Each of the divided character strings that constitute the divided pattern, with respect to only the divided pattern for which the reading dictionary search unit has acquired reading information corresponding to all of the two or more divided character strings that constitute the divided pattern; A document group is searched using each acquisition target character string reading information, which is a set of each reading information among the two or more reading information acquired by the reading dictionary search unit, and each of the acquisition target character string reading information has a division A co-occurrence frequency acquisition unit that acquires the co-occurrence frequency, which is the number of documents in which character strings and reading information co-occur, for each reading information, and calculates the score for each division pattern using the co-occurrence frequency for each reading information When,
A program for functioning as an output unit that outputs at least a combination of reading information with the highest co-occurrence frequency using the score acquired by the co-occurrence frequency acquisition unit. The reading dictionary search unit
A keyword dividing unit that divides the keyword received by the receiving unit in units of one or more characters and acquires two or more divided character strings that are divided character strings;
One or more reading information paired with a term matching each of the two or more divided character strings is retrieved from the reading dictionary for each divided character string, and the acquisition target character string reading information having the divided character string and the reading information The program according to claim 9 for causing a computer to function as a reading information acquisition means for acquiring two or more reading information,
The keyword dividing means includes:
From the case where the number of cuts N (N is 0 or more) is small, the keyword received by the receiving unit is sequentially divided into units of one or more characters, and one or more divided character strings that are divided character strings are obtained, Until the reading information acquisition means receives an end instruction, the division of the keyword is repeated,
The reading information acquisition means includes
One or more reading information paired with a term that matches one or more divided character strings acquired by the keyword dividing means is acquired for each divided character string from the reading dictionary, and all of the one or more divided character strings are obtained. If the reading information corresponding to is not acquired, the keyword dividing unit is instructed to acquire two or more divided character strings which are the next division patterns, and the one or more divided character strings are obtained. If the reading information corresponding to all of the one or more divided character strings can be acquired, the reading information corresponding to all of the one or more divided character strings is at least temporarily stored in the storage medium, and the keyword division is performed. An end instruction is passed to the means, or two or more divided character strings that are other division patterns having the same number of cuts as when reading information corresponding to all of the one or more divided character strings can be obtained Instruct the keyword dividing means to acquire, and after completing the reading information acquisition processing for the other division patterns of the same number of cuts, the computer functions as an instruction to pass an end instruction to the keyword dividing means The program of Claim 9 for making it do.

Images