JP2007171275A - Language processor and language processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a language processing method with high portability, capable of correctly providing reading. <P>SOLUTION: A language processor comprises: a detecting means for detecting a character string which is not registered in a word dictionary, from objects to be processed including a plurality of character strings; an acquiring means for acquiring a candidate of reading of each character in the character strings detected by the detecting means by using a Chinese character dictionary; a creating means for creating the candidate of the reading for the whole character strings detected by the detecting means from the candidates of the reading; and a selection means for selecting the reading of the character strings from the candidates of the reading of the character strings by using a pronunciation dictionary. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a language processing method for reading a character string.

  It is necessary to read the text correctly in the text-to-speech synthesis technology for converting the text into speech or the recognition grammar in the speech recognition technology or the automatic creation of the pronunciation dictionary. Conventionally, as a method of performing this reading, as shown in FIG. 3, a method of performing morphological analysis using a word dictionary and a single kanji dictionary has been widely used. For example, when the text “Today is good weather” is input, the morphological analysis unit 301 refers to the word dictionary 302 to read “Today (Kyo) / Ha (wa) / Good (Yoi) / Weather (Tenki) / The result of “(de) / su (su)” is obtained. In parentheses, reading is used, and / indicates a word break. Since the technique for obtaining this result is known, the details are omitted. In this example, since all the words in the input text are present in the word dictionary, a correct reading result can be obtained. However, in the example of “My name is Yamamoto Yuka” (assuming that Yuka is correct to read “Sayaka”), the name part “Yuka” is not in the word dictionary, that is, an unregistered word Is determined. In this case, “Rika” is read using 303 single kanji dictionary. Now, suppose that the single kanji dictionary has the readings of “Sha”, “Ya”, and “Kou”. In this case, “I (I) / No (No) / Name (Namae) / Ha (Wa) / Yamamoto (Yamamoto) / Rin (Sha) / Rin (Ya) / Incense (Kou) / de (de) / su The result “(su)” is obtained, and the part of “Mika” is misread as “Shayako”.

  For such a problem, in Patent Document 1, after detecting an unregistered word, a combination candidate for reading is obtained using a single kanji dictionary, and one of the combination candidates is selected by referring to the rule. If the selected one result reading is in the word dictionary, this reading is used. Here, the rule is, for example, “a certain character is often read aloud rather than read aloud, and when it is read aloud, the character behind it is often read aloud”.

Further, Patent Document 2 discloses an analysis method that uses word reading tendency information. The word reading tendency information is information (included in the word dictionary) of the influence of the number of characters, the number of mora, the grammar, the accent type, the preceding and following characters, the sound lesson, the clarity, and the like. For example, if “Kuromori” is detected as an unregistered word, the reading tendency of the word “mori” (the two kanji nouns with “mori” at the end are accent type 2 and the first character is kunomi To determine the reading of “black” and the overall accent type.
JP-A-8-185197 Japanese Patent Laid-Open No. 11-249866

  In Patent Document 1, reading of an unregistered word is determined by referring to a word dictionary. The word dictionary usually stores words of various parts of speech and fields necessary for analysis of tens of thousands to hundreds of thousands of words. On the other hand, many of the words that are judged as unregistered words are related to proper nouns, especially people's names, so if you use a word dictionary that contains a lot of words other than proper nouns, you will mistakenly read general nouns and attached words. There are things to do. For example, “Yui” matches “Decision” and becomes “Ketsui” (correctly “Yui”), “Nanami” matches “Nanami” and becomes “Sitami” (correctly “Nanami”), etc. There is a possibility that reading appropriate for a person's name will not be made. In other words, in order to accurately identify readings from reading candidates obtained from a single kanji dictionary, rather than using a word dictionary containing various words, a pronunciation dictionary that considers the category of words that are unregistered words is used. It is necessary to use it.

  Further, in Patent Documents 1 and 2, reading is determined by a rule-based heuristic method. For this reason, there is a problem that it takes a lot of time to create and update rules when there is a trend or abolishment such as a change in the field of text to be analyzed or a name name notation. The present invention has been made in view of the above-described problems, and an object of the present invention is to provide a language processing method that has high portability and can be read accurately.

  In order to achieve the above object, the language processing apparatus of the present invention uses the detection means for detecting a character string not registered in the word dictionary from the processing target including a plurality of character strings, and the detection using the single kanji dictionary. Obtaining means for obtaining reading candidates for each character in the character string detected by the means; generating means for generating reading candidates for the entire character string detected by the detecting means from the reading candidates; and pronunciation Selecting means for selecting reading of the character string from the character string reading candidates using a dictionary;

  In order to achieve the above object, the language processing apparatus of the present invention detects a character string not registered in the word dictionary from a processing target including a plurality of character strings, and the detection unit detects the character string. Candidates for reading each character in the character string detected by the detecting means using an acquisition means for acquiring attribute information indicating the attribute of the character string and at least one single kanji dictionary corresponding to the attribute information A character string reading candidate generating means for generating reading candidates for the entire character string from the reading candidates, and at least two pronunciation dictionaries corresponding to the attribute information. Selecting means for selecting reading of the character string from reading candidates.

  In order to achieve the above object, the language processing method of the present invention uses a detection step of detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings, and the single kanji dictionary. An acquisition step of acquiring reading candidates for each character in the character string detected in the detection step; a generation step of generating reading candidates for the entire character string detected in the detection step from the reading candidates; And a selection step of selecting reading of the character string from the character string reading candidates using a pronunciation dictionary.

  In order to achieve the above object, the language processing method of the present invention includes a detection step of detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings, and the detection of the character string detected in the detection step. A candidate for reading each character in the character string detected in the detection step using an acquisition step for acquiring attribute information indicating the attribute of the character string, and at least one single kanji dictionary corresponding to the attribute information A character string reading candidate generating step for generating reading candidates for the entire character string from the reading candidates, and at least two pronunciation dictionaries corresponding to the attribute information, and the character string Selecting a reading of the character string from reading candidates.

  According to the present invention, it is possible to read a character string to be processed more accurately.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

  FIG. 2 is a block diagram illustrating the configuration of the language processing apparatus according to the first embodiment of the present invention. A CPU 201 performs various controls in the language processing apparatus according to the present embodiment in accordance with a control program stored in the ROM 202 or a control program loaded from the external storage device 204 to the RAM 203. The ROM 202 stores various parameters, a control program executed by the CPU 201, and the like. The RAM 203 provides a work area when various controls are executed by the CPU 201 and stores a control program executed by the CPU 201. Reference numeral 204 denotes an external storage device such as a hard disk, a floppy (registered trademark) disk, a CD-ROM, a DVD-ROM, or a memory card. When the external storage device is a hard disk, various programs installed from a CD-ROM, a floppy (registered trademark) disk or the like are stored. Reference numeral 205 denotes an input device for inputting and selecting text information by an external operation such as a numeric keypad, a button, a touch panel, a keyboard, a mouse, a microphone, and a pen. The input device may be directly attached to the language processing device, or may be operated from the outside of the language processing device using a remote controller, computer, mobile phone, etc. via communication such as infrared, wireless LAN, Internet, telephone line, etc. A form may be sufficient (the apparatus regarding a communication part is abbreviate | omitted). Moreover, the form which combined these may be sufficient. An output device 206 includes a CRT, a liquid crystal display, a speaker, and the like. Reference numeral 207 denotes a bus for connecting the above-described units. The text to be subjected to language processing may be stored in 202, 203, 204 in addition to being input or selected by 205, or communication such as infrared, wireless LAN, Internet, telephone line, etc. It may be obtained via.

  FIG. 1 is a block diagram showing the module configuration of the language processing method. A character string detection unit 101 analyzes the text to be analyzed using the word dictionary 105 and detects a character string not included in 105. A character reading candidate generation unit 102 generates a reading candidate for each character by using the single kanji dictionary 106 for the character string detected in 101. A character string reading candidate generation unit 103 generates reading candidates for the entire character string from the character candidates generated in 102. A character string reading selection unit 104 selects a character string reading using the pronunciation dictionary 107 for the character string reading candidates generated in 103, and obtains an analysis result. For the character string that can be analyzed in 101, the analysis result is obtained from 101 without performing the processing from 102 to 104.

  Next, a processing flow in the present embodiment will be described. Here, a case where the text “My name is Yuka Yamamoto” is analyzed and read is described as an example.

  FIG. 4 is a flowchart for analyzing this text. First, in step S401, a text to be analyzed is acquired. Next, in step S402, pre-processing prior to text analysis, such as sentence deletion and deletion of unnecessary characters such as quotation marks, is performed. In step S403, the pre-processed text is divided into word morphemes (morpheme analysis) using the word dictionary 105.

  An example of the word dictionary is shown in FIG. The first to fifth columns are the word ID, notation, reading, part of speech, and score, respectively. The score is a logarithm of the probability of occurrence of each word, and is obtained by a word unigram using some kind of language corpus. Any method of morphological analysis may be used, but a word lattice is generated from what is searched in a word dictionary, and this is used as a longest match method or a minimum cost method (in the case of the score in FIG. 6, the maximum score method). An analysis result can be obtained using the above criteria.

  FIG. 7 shows an example of a word lattice when a morphological analysis is performed on the text “My name is Yuka Yamamoto” using the word dictionary of FIG. Here, “name” is searched as “name” (word ID = 3), or “name” (word ID = 4) and “previous” (word ID = 5), and “yamamoto” is searched for “yamamoto” (word ID = 7) or “mountain” (word ID = 8) and “book” (word ID = 9). In addition, since “Sayaka” is not in the word dictionary, it is an unregistered word. If the longest match method or the minimum cost method is applied to this word lattice, the path indicated by the solid line in FIG. 7, that is, the word division of “I / no / name / hae / Yamamoto / Sayaka / de / su”. Is made. Here, since the word dictionary of FIG. 6 has reading and part-of-speech information in addition to the notation, in addition to word division, reading and part-of-speech information can also be obtained as an analysis result for words other than “Sayaka”. it can.

  Next, in step S404, the counter i is set to 1, and since the number of words is 8 in I, including “Sayaka” in this example, I = 8. Next, in step S405, it is determined whether or not the analysis result output in step S410 has been made for all words. If i> I, the process ends. Otherwise, the process proceeds to step S406. In step S406, it is determined whether or not the i-th word w (i) is an unregistered word. If it is not an unregistered word, the process proceeds to step S410 and an analysis result (notation, reading, part of speech, etc.) is output. If it is an unregistered word (in the above example, “Sayaka” (w (6))), step The process proceeds to S407. In step S407, the single-kanji dictionary 106 is searched to generate reading candidates for each character of the unregistered word. Next, in step S408, using the character reading candidates generated in S407, reading candidates for the character strings of the entire unregistered word are generated. Next, the pronunciation dictionary 107 is searched in step S409, and the character string reading is identified and selected from the character string reading candidates generated in S408. Next, the analysis result is output in S410, and in the case of “Sayaka”, “Sayaka / Sayaka / unregistered word” or the like is output. Next, in step S411, the counter i is incremented by 1, and the process returns to S405.

  FIG. 5 is a flowchart of processing for reading unregistered words in S407 to S409. Processing for obtaining a reading result “Sayaka” from “Sayaka” of the unregistered words will be described in detail. In step S501, the counter j is set to 1, and if J is the number of characters of w (i), that is, “Sayaka” of w (6), there are three characters, so J = 3. Next, in step S502, it is determined whether reading candidates have been generated for all characters. If j> J, the process proceeds to step S505; otherwise, the process proceeds to step S503. In S503, a reading r (j) for the j-th character c (j) is generated by searching the single kanji dictionary 106. An example of a single kanji dictionary is shown in FIG. The first column to the third column are a single Chinese character ID, notation, and reading, respectively, and when there are a plurality of readings, a plurality of “/” are described as delimiters. The reading candidate r (1) for c (1), that is, “sha” is generated as “sha” or “sa” from FIG. In step S504, the counter j is incremented by 1, and the process returns to S502. By repeating the same processing, the reading candidates for “耶” become “ya”, “ja”, and “sha”, and the reading candidates for “scent” are “Kou”, “Kyo”, “K”, “Kao”. In step S505, w (i) reading candidates, that is, character string reading candidates t (k) are generated from these character reading candidates. For example, in the case of the above example, the character reading candidates are expressed as pronunciation lattices as shown in FIG. 10A, and the 24 types of character string reading candidates shown in FIG. 10B in which all paths of the lattice are expanded are shown. Is t (k). In step S506, the counter k is set to 1, K is set to the number of reading candidates for w (6), that is, K = 24, and the maximum pronunciation score maxScore and its index maxID are set to maxScore = −1000 and maxID = 1, respectively. And so on. In step S507, it is determined whether or not all character string reading candidates have been searched. If k> K, the process proceeds to step S512. Otherwise, the process proceeds to step S508. In step S508, the pronunciation dictionary 107 is searched for the score of t (k).

  An example of the pronunciation dictionary is shown in FIG. The first column to the third column are a pronunciation ID, a reading, and a score, respectively. The larger the value of the score, the easier the reading occurs. When searching for t (k) in the order of candidate IDs in FIG. 10B, the pronunciation of t (1) is not included in the pronunciation dictionary. Assuming that the score when t (k) reading does not exist in the pronunciation dictionary is −1000, the score curScore = −1000 is obtained. As a result of the determination in step S509, the process proceeds to step S511, and the counter k is set to 1. Is incremented by one and the process returns to S507. Since the pronunciation from t (2) to t (4) is not included in the pronunciation dictionary, the same process is repeated. Next, since “Sayako” at t (5) exists in the pronunciation dictionary, the score is set as cureScore = −30. In this case, since the condition of S509 is satisfied, maxScore = −30 and maxID = 5 are set. The pronunciation of the next t (6) is not included in the pronunciation dictionary, and “Sayaka” of t (7) exists in the pronunciation dictionary and curScore = −7. Therefore, in S510, maxScore = −7 and maxID = 7 are set. Is done. Hereinafter, since t (8) to t (24) do not exist in the pronunciation dictionary, in S512, the reading of w (6) = “Sayaka” is identified as t (7) = “Sayaka”, and the process ends.

  In the present embodiment, the highest score is output in S512. However, the present invention is not limited to this, and a plurality of candidates can be output from the highest score. In this embodiment, an example in which a single kanji dictionary and a pronunciation dictionary are used is described. However, the present invention is not limited to this, and a plurality of single kanji dictionary and pronunciation dictionary can be used. is there. For example, in reading human names, if you prepare two types, a single kanji dictionary and pronunciation dictionary set for male names, and a single kanji dictionary and pronunciation dictionary set for female names, each reading candidate can be selected. Can be used to identify readings. Similarly, in addition to gender, it is also possible to prepare a plurality of single kanji and pronunciation dictionaries according to differences in generations, regions, etc., and perform processing using these. In this embodiment, a sentence “My name is Yuka Yamamoto” was the text to be analyzed. However, the present invention is not limited to this, and only one word such as “Mika” or one phrase is used. Similar processing can be applied.

  The reading results obtained in this embodiment can be used for reading unregistered words in text-to-speech synthesis. Furthermore, the reading result obtained in this embodiment can be used to create a pronunciation dictionary or a speech recognition grammar for an unregistered word in speech recognition.

  As is clear from the above description, according to the present embodiment, candidate readings for unregistered words are generated using a single kanji dictionary, and the candidate is identified using a pronunciation dictionary different from the word dictionary. Therefore, portability is high and reading can be performed accurately.

  In addition, the quality of text-to-speech synthesis and the accuracy in automatic creation of a pronunciation dictionary for speech recognition are improved. Furthermore, it is possible to easily cope with changes in the field of analysis target text and changes in the notation method.

  As shown in FIG. 8, the single kanji dictionary used in the above embodiment includes single kanji ID, notation, and reading information. However, the present invention is not limited to this, and score information is included in the single kanji dictionary. It can be applied even in the case of including.

  FIG. 11 is an example of a single kanji dictionary with score information. The first column to the fourth column are a single Chinese character ID, a notation, a reading, and a score, respectively. When there are a plurality of readings, a plurality of “/” is described as a delimiter. The score indicates that the larger the value is, the easier the reading occurs. The reading process for an unregistered word when score information is included in the single kanji dictionary is basically the same as in the previous embodiment, and only the differences in FIG. 5 will be described.

  First, in S503, the reading candidate r (j) of c (j) is generated using the single kanji dictionary 106 with score information as shown in FIG. FIG. 12 is an example of a character string reading candidate obtained from the single kanji dictionary of FIG. 11, and is an example of generating w (i) reading candidate t (k) in S505. The score of FIG. 12 is the sum of the scores for the reading of a single kanji character in FIG. 11. For example, the score of “SHAYAKO” with candidate ID = 1 is the score of “SHA”, “YA”, “KO” in FIG. Are −4, −5, and −3, respectively, so that −4−5−3 = −12. In S508, the sum of the score in the pronunciation dictionary 107 and the score obtained from the single kanji dictionary is curScore. Here, the score of the pronunciation dictionary when the reading of t (k) does not exist in the pronunciation dictionary is -500 or the like. As a result, when not all reading candidates exist in the pronunciation dictionary, the pronunciation of w (i) is identified based on the score of the single kanji dictionary.

  In the present embodiment, the sum of the score obtained from the single kanji dictionary and the score of the pronunciation dictionary is used, but the present invention is not limited to this, and the score can be calculated by performing any operation such as weighted sum and product. Good. That is, only the score obtained from the single kanji dictionary may be used without using the score of the pronunciation dictionary at all. In this case, it is not necessary to store the score shown in FIG. 9 in the pronunciation dictionary.

  In the above embodiment, text analysis using the word dictionary 105 is performed on the text to be analyzed, and character strings determined as unregistered words are read. Here, if the character strings of unregistered words can be obtained, for example, those related to names of people, places related to places, organizations related to company names, etc., use the pronunciation dictionary corresponding to these attributes. Therefore, it is possible to perform reading with higher accuracy. FIG. 13 is a block diagram showing a module configuration of a language processing method for acquiring an attribute of a character string of an unregistered word and performing reading using the attribute.

  A character string detection unit 1301 analyzes the text to be analyzed using the word dictionary 1306 and detects a character string not included in 1306. A character string attribute acquisition unit 1302 acquires an attribute for a character string not included in 1306. A character reading candidate generation unit 1303 generates a reading candidate for each character using the single kanji dictionary 1306 for the character string detected in 1301. A character string reading candidate generation unit 1304 generates reading candidates for the entire character string from the character candidates generated in 1303. Reference numeral 1305 denotes a character string reading selection unit which selects a plurality of pronunciation dictionaries (in this example, two pronunciation dictionaries 1308 and 1309) from the character string reading candidates generated in 1304 and the character string attribute information acquired in 1302. Use to identify and select the reading of the string and obtain the analysis results. For the character string that can be analyzed in 1301, the processing from 1303 to 1305 is not performed, and the analysis result is obtained after obtaining the character string attribute of 1302.

  The processing flow in the present embodiment is almost the same as that in FIGS. 4 and 5 described in the above embodiment, and only the differences will be described. Further, as in the previous embodiment, a case where the text “My name is Yuka Yamamoto” is analyzed and read will be described as an example. Further, it is assumed that the pronunciation dictionary 1 1308 and the pronunciation dictionary 2 1309 are related to a person name and a place name, respectively. By using the same word dictionary 1306 as in FIG. 6, the text is divided into words “I / no / name / ha / yamamoto / saika / de / su” and the character string “saika” is an unregistered word. Detected. Next, in 1302, an attribute related to the character string “Sayaka” is acquired. In the case of this example, the character string “Sayaka” acquires an attribute such as a person's first name or last name. There are various ways to acquire this attribute. For example, by acquiring information that the word “Yamamoto” is a surname of a person name or a surname from 1306, “Sayaka” is related to the name of a person name or a surname. You can get it. In addition, “Sayaka” is a personal name by using knowledge such as “There is an unregistered word between the word“ name ”and the word“ is ”and the character string between them is a personal name”. Can get. Besides, it is not estimated from the analysis result of text, but it is also possible to acquire the attribute of unregistered word using the attribute of the application such as text for the field related to the person name, or specify the attribute by the user It is.

  Since the processes 1303 and 1304 are the same as the processes 102 and 103, respectively, description thereof will be omitted. Next, when the character string reading is identified in 1305, the pronunciation dictionary relating to the attribute acquired in 1302 is used. In this example, since the attribute of the character string “Sayaka” is a person name, the pronunciation dictionary 1 relating to the person name 1308 is used for reading.

  In the present embodiment, the pronunciation dictionary is related to a person name and a place name. However, the present invention is not limited to this. For example, attributes that can be acquired in 1302 such as names and surnames, men and women, regions, and generations. Any one can be used. Moreover, although there are two types of pronunciation dictionaries, more pronunciation dictionaries may be provided depending on the types of attributes. Further, although there is one single kanji dictionary, a plurality of single kanji dictionaries corresponding to the type of pronunciation dictionary may be used. In 1302, the attribute is uniquely determined. However, the attribute is determined probabilistically (for example, 0.9 for a person name, 0.1 for a place name, etc.), and considering this as a score or weight 1305 It is also possible to identify the reading of the string.

  In the present invention, a program for realizing each function of the above-described embodiments is supplied directly or remotely to a system or apparatus, and a computer included in the system or apparatus reads and executes the supplied program code. Can also be achieved.

  Accordingly, since the functions and processes of the present invention are implemented by a computer, the program code itself installed in the computer also implements the present invention. That is, the computer program itself for realizing the functions and processes is also one aspect of the present invention.

  In this case, the program may be in any form as long as it has a program function, such as an object code, a program executed by an interpreter, or script data supplied to the OS.

  Examples of the recording medium for supplying the program include a flexible disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, and CD-RW. Examples of the recording medium include a magnetic tape, a non-volatile memory card, a ROM, a DVD (DVD-ROM, DVD-R), and the like.

  The program may be downloaded from a homepage on the Internet using a browser on a client computer. That is, the computer program itself of the present invention or a compressed file including an automatic installation function may be downloaded from a home page to a recording medium such as a hard disk. Further, it is also possible to divide the program code constituting the program of the present invention into a plurality of files and download each file from a different home page. That is, a WWW server that allows a plurality of users to download a program file for realizing the functions and processing of the present invention on a computer may be a constituent requirement of the present invention.

  Further, the program of the present invention may be encrypted and stored in a storage medium such as a CD-ROM and distributed to users. In this case, only the user who cleared the predetermined condition is allowed to download the key information to be decrypted from the homepage via the Internet, decrypt the program encrypted with the key information, execute it, and install the program on the computer May be.

  Further, the functions of the above-described embodiments may be realized by the computer executing the read program. Note that an OS or the like running on the computer may perform part or all of the actual processing based on the instructions of the program. Of course, also in this case, the functions of the above-described embodiments can be realized.

  Furthermore, the program read from the recording medium may be written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer. Based on the instructions of the program, a CPU or the like provided in the function expansion board or function expansion unit may perform part or all of the actual processing. In this way, the functions of the above-described embodiments may be realized.

It is the block diagram which showed the module structure of the language processing method which concerns on an Example. It is the block diagram which showed the hardware constitutions of the language processing apparatus carrying the language processing method which concerns on an Example. It is the block diagram which showed the module structure of the conventional language processing method. It is a flowchart of the language processing method which concerns on an Example. It is a flowchart of the process which reads with respect to an unregistered word. It is an example of a word dictionary. It is an example which expressed the text analysis result by the word lattice. It is an example of a single kanji dictionary. It is an example of a pronunciation dictionary. It is the example which represented the reading candidate of the unregistered word as (a) pronunciation lattice and (b) character string reading candidate. It is an example of the single kanji dictionary with score information. It is an example of a character string reading candidate obtained from a single kanji dictionary with score information. It is the block diagram which showed the module structure of the language processing method holding the character string attribute acquisition part which concerns on an Example.

Claims (19)

Detecting means for detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings;
Obtaining means for obtaining candidates for reading each character in the character string detected by the detecting means using a single kanji dictionary;
Generating means for generating reading candidates for the entire character string detected by the detecting means from the reading candidates;
A language processing apparatus comprising: a selection unit that selects a reading of the character string from the character string reading candidates using a pronunciation dictionary.   The language processing apparatus according to claim 1, wherein the single kanji dictionary holds a reading score for each character.   The language processing apparatus according to claim 1, wherein the pronunciation dictionary holds a reading score for a character string.   The language processing apparatus according to claim 1, wherein the selection unit identifies a reading of the character string from the character string reading candidates and selects the identified reading.   The language processing apparatus according to claim 1, wherein the selection unit selects a character string reading using at least one of a reading score for each character and a reading score for the character string.   The language processing apparatus according to claim 5, further comprising a plurality of candidate output means for outputting the one having a high score as a plurality of candidates.   The language processing apparatus according to claim 1, wherein the reading of the character string selected by the selection unit is used for reading in text-to-speech synthesis or creation of a pronunciation dictionary or speech recognition grammar in speech recognition. Detecting means for detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings;
Obtaining means for obtaining attribute information indicating an attribute of the character string detected by the detecting means;
Generating means for generating candidates for reading each character in the character string detected by the detecting means, using at least one single kanji dictionary corresponding to the attribute information;
Character string reading candidate generation means for generating reading candidates for the entire character string from the reading candidates;
A language processing apparatus comprising: selection means for selecting reading of the character string from the character string reading candidates using at least two or more pronunciation dictionaries corresponding to the attribute information.   9. The language processing apparatus according to claim 8, wherein the attribute of the character string relates to at least one of a person name, place name, organization name, gender, generation, and region. A detection step of detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings;
An acquisition step of acquiring candidates for reading each character in the character string detected in the detection step using a single kanji dictionary;
Generating a candidate for reading for the entire character string detected in the detection step from the reading candidate;
A language processing method comprising: using a pronunciation dictionary to select a reading of the character string from the character string reading candidates.   The language processing method according to claim 10, wherein the single-kanji dictionary holds a reading score for each character.   The language processing method according to claim 10, wherein the pronunciation dictionary holds a reading score for a character string.   The language processing method according to claim 10, wherein the selecting step identifies a reading of the character string from the character string reading candidates and selects the identified reading.   13. The language processing method according to claim 10, wherein the selection step selects the reading of the character string using at least one of a reading score for each character and a reading score for the character string.   The language processing method according to claim 14, further comprising a multiple candidate output step of outputting the high score as a plurality of candidates.   The language processing method according to claim 10, wherein the reading of the character string selected in the selection step is used for reading in text-to-speech synthesis or creation of a pronunciation dictionary or speech recognition grammar in speech recognition. A detection step of detecting a character string not registered in the word dictionary from a processing target including a plurality of character strings;
An acquisition step of acquiring attribute information indicating the attribute of the character string detected in the detection step;
A generation step of generating reading candidates for each character in the character string detected in the detection step using at least one single kanji dictionary corresponding to the attribute information;
A character string reading candidate generation step of generating reading candidates for the entire character string from the reading candidates;
A language processing method comprising: a selection step of selecting reading of the character string from the character string reading candidates using at least two or more pronunciation dictionaries corresponding to the attribute information.   18. The language processing method according to claim 17, wherein the attribute of the character string relates to at least one of a person name, place name, organization name, gender, generation, and region.   A control program for causing a computer to execute the language processing method according to claim 10.

Images