JP5853595B2 - Morphological analyzer, method, program, speech synthesizer, method, program - Google Patents

Description

  The disclosed technology relates to a morpheme analyzer, a morpheme analysis method, a morpheme analysis program, a speech synthesizer, a speech synthesis method, and a speech synthesis program.

  Morphological analysis is one of the basic technologies of natural language processing, and it divides text described in natural language into morphemes (the smallest language forms that have meaning) and analyzes the reading of texts. . The following techniques are known as techniques using morphological analysis (Patent Documents 1 to 4).

  For example, in text mining that extracts useful information from a text as a keyword, a technique for extracting a compound noun as a keyword using morphological analysis is known. This technology first analyzes morpheme of document data, then refers to the part-of-speech combination rule with the part-of-speech information of the morpheme. Get candidate data. Next, for each of the first morpheme and the last morpheme constituting the compound noun candidate, the front score of the first morpheme and the rear score of the last morpheme are obtained by referring to the character string frequency data. When both scores are larger than the score setting value, the character string from the first morpheme to the last morpheme is extracted as a compound noun.

  In addition, for example, a technique for performing extraction and narrowing of keywords used for text search using morphological analysis is also known. This technology performs simple morphological analysis on text to generate a simple noun set, and when there is a compound noun element that is sandwiched between compound noun sets, katakana noun sets, and noun elements, A combined noun set connected through the integrated particle “no” is generated.

  In addition, for example, other techniques for extracting keywords used for text search using morphological analysis are also known. In order to avoid omission of keyword extraction, this technology calculates the morpheme appearance probability by analyzing the input text, and calculates the frequency information of each word based on the importance of the morpheme corresponding to the morpheme appearance probability. Then, a word set with frequency information is generated.

  Further, for example, another technique is known in which a keyword used for text search is extracted using morphological analysis. This technique decomposes a character string into a plurality of morphemes, determines the part of speech of each morpheme, extracts each morpheme group composed of morphemes of a first number and a predetermined part of speech from each morpheme, A combination of each morpheme constituting the morpheme group and each part of speech of each morpheme is calculated. Then, among the calculated combinations, when the appearance frequency of a combination common among a plurality of morpheme groups is equal to or higher than a predetermined threshold, the common combination is combined with a plurality of morphemes included in an arbitrary character string. Extract as a compound rule.

JP 2011-059754 A JP 2006-004283 A JP 2006-243976 A JP 2010-009355 A

  Morphological analysis uses a word dictionary in which words are registered in association with reading, etc., compares the text to be analyzed with the word dictionary, decomposes the word registered in the word dictionary as a unit, and then analyzes the text to be analyzed. The above includes processing for evaluating the connectivity of adjacent word pairs. For this reason, the analysis accuracy of morphological analysis depends on what words are registered in the word dictionary.

  Here, when a compound noun combining words of a plurality of nouns is present on the text to be analyzed, the compound noun is analyzed as a combination of a plurality of words registered in the word dictionary in the morphological analysis. The On the other hand, in natural languages such as Japanese, there are words that have the same notation in the text but different readings (and meanings), and such words are associated with different readings and duplicated in the word dictionary. be registered. For this reason, if a compound noun that exists in the text to be analyzed contains words that have the same notation in the text but different readings, the reading of the compound noun may not be correctly analyzed by morphological analysis There is.

  On the other hand, the above-mentioned technology using morphological analysis is a technology that extracts keywords including compound nouns from text based on the part of speech and the appearance frequency, etc., and contributes to improving the analysis accuracy of compound noun reading in morphological analysis. Not what you want.

  An object of the disclosed technique is to improve the analysis accuracy of reading a set of words of nouns that are adjacent to each other in a text to be analyzed and registered in a word dictionary.

The disclosed technology stores a word dictionary in which a plurality of words are registered in association with at least the part of speech of each word in the first storage unit. In addition, the second storage unit sets a group of noun words that are registered in the word dictionary and are adjacent to the specific word across the specific word on the learning text as a group of connectable words. To remember. In addition, a connection evaluation value representing connection possibility is obtained for a pair of words that are registered in the word dictionary and are adjacent to each other in the text to be analyzed. Moreover, it connects about the word group of the noun class which is not memorize | stored in the 2nd memory | storage part as a word group which can be connected among the word groups which are each registered in the word dictionary and are adjacent on the analysis object text. The connection evaluation value is decreased so that the evaluation value is not included in the word combination having the maximum evaluation value. Then, the morphological analysis of the text to be analyzed is performed based on the set of words having the maximum connection evaluation value.

  The disclosed technique has an effect that the analysis accuracy of reading a set of words of nouns that are adjacent to each other on the text to be analyzed and registered in the word dictionary is improved.

It is a functional block diagram of the speech synthesizer described in the embodiment. It is a schematic block diagram of a computer that functions as a speech synthesizer. It is the schematic which shows an example of a 1st word dictionary. It is the schematic which shows an example of a 2nd word dictionary. It is the schematic which shows an example of the connection evaluation value table between word parts of speech. It is a flowchart of a connection possibility table production | generation process. It is the schematic for demonstrating a connection possibility table production | generation process. It is the schematic for demonstrating determination of the group of the word which can be connected. It is the schematic for demonstrating the calculation of the deduction point value with respect to the group of the word which cannot be connected. It is the schematic for demonstrating determination of the group of the word which can be connected. It is the schematic for demonstrating the calculation of the deduction point value with respect to the group of the word which cannot be connected. It is a flowchart of a morphological analysis process. It is the schematic for demonstrating the morphological analysis using a connection possibility table. It is the schematic for demonstrating the morphological analysis using a connection possibility table. It is the schematic for demonstrating the morphological analysis using a connection possibility table. It is the schematic for demonstrating the morphological analysis using a connection possibility table. It is the schematic for demonstrating the problem of the morphological analysis which does not use a connection possibility table. It is the schematic for demonstrating the problem of the morphological analysis which does not use a connection possibility table.

  Hereinafter, an example of an embodiment of the disclosed technology will be described in detail with reference to the drawings. FIG. 1 shows a speech synthesizer 10 according to the present embodiment. The speech synthesizer 10 performs a process of synthesizing speech that reads out the text from the text. The speech synthesizer 10 includes a morphological analyzer 12, a dependency analyzer 38, a phonetic character string generator 40, and a speech synthesizer 42. The morpheme analyzer 12 includes a morpheme analyzer 14, a word connection possibility determination unit 16, a first storage unit 22 that stores a first word dictionary 18 and a second word dictionary 20, and a second storage that stores a word connection possibility table 24. A storage unit 26 is provided. The morpheme analyzer 12 includes a third storage unit 30 that stores the inter-word-part-of-speech connection evaluation value table 28, a fourth storage unit 36 that stores the analysis target text 32 and the learning text corpus 34.

  The speech synthesizer 10 can be realized by a computer 70 shown in FIG. 2, for example. The computer 70 includes a CPU 72, a memory 44, a nonvolatile storage unit 46, a keyboard 48, a mouse 50, a display 52, and a speaker 54, which are connected to each other via a bus 56. The storage unit 46 can be realized by an HDD (Hard Disk Drive), a flash memory, or the like. In the storage unit 46 as a recording medium, a speech synthesis program 58 for causing the computer 70 to function as the speech synthesizer 10, the first word dictionary 18, the second word dictionary 20, the word connection possibility table 24, and connection between word parts of speech. Each evaluation value table 28 is stored. The storage unit 46 also stores an analysis target text 32 and a learning text corpus 34. The CPU 72 reads the speech synthesis program 58 from the storage unit 46 and expands it in the memory 44, and sequentially executes the processes of the speech synthesis program 58.

  The speech synthesis program 58 includes a morphological analysis process 60, a word connection possibility determination process 62, a dependency analysis process 64, a phonetic character string generation process 66, and a speech synthesis process 68. The CPU 72 operates as the morphological analysis unit 14 illustrated in FIG. 1 by executing the morphological analysis process 60. Further, the CPU 72 operates as the word connection possibility determination unit 16 illustrated in FIG. 1 by executing the word connection possibility determination process 62. The CPU 72 operates as the dependency analysis unit 38 shown in FIG. 1 by executing the dependency analysis process 64. The CPU 72 operates as the phonetic character string generation unit 40 shown in FIG. 1 by executing the phonetic character string generation process 66. The CPU 72 operates as the speech synthesizer 42 shown in FIG. 1 by executing the speech synthesis process 68.

  The speech synthesis program 58 is an example of a speech synthesis program in the disclosed technology, and the morpheme analysis process 60 included in the speech synthesis program 58 is an example of a morpheme analysis program in the disclosed technology.

  When the speech synthesizer 10 is realized by the computer 70, the storage unit 46 that stores the first word dictionary 18 and the second word dictionary 20 is used as the first storage unit 22, and a partial area of the memory 44 is a first word. The dictionary 18 and the second word dictionary 20 are used. The storage unit 46 that stores the word connection possibility table 24 and the word-part-of-speech connection evaluation value table 28 is used as the second storage unit 26 and the third storage unit 30, and a partial region of the memory 44 is a word connection possibility. The table 24 and the word part-of-speech connection evaluation value table 28 are used. The storage unit 46 that stores the analysis target text 32 and the learning text corpus 34 is used as the fourth storage unit 36, and a partial area of the memory 44 is used as a storage area for the analysis target text 32 and the learning text corpus 34. It is done. As a result, the computer 70 that has executed the speech synthesis program 58 functions as the speech synthesizer 10.

  As shown in FIG. 3 as an example, in the first word dictionary 18, “notation”, “part of speech”, and “reading” information are registered in association with each other for a plurality of words. FIG. 3 shows only some of the words registered in the first word dictionary 18, and the first word dictionary 18 is selected in consideration of the analysis accuracy of morphological analysis described later. Information on a large number of words is registered respectively. In the second word dictionary 20, for example, as shown in FIG. 4, “notation”, “part of speech”, and “reading” information are registered in association with each other for a plurality of words. The second word dictionary 20 is a word dictionary for additionally registering information on unregistered words in the first word dictionary 18. For example, the first word among the words in the text included in the learning text corpus 34 or the like. Information on unregistered words is additionally registered in the dictionary 18.

  In the word part-of-speech connection evaluation value table 28, as shown in FIG. 5 as an example, connection evaluation values representing the possibility of connection on the text are registered for combinations of words of various parts of speech. For example, in FIG. 5, “-100” is used as the connection evaluation value between the common noun word and the suffix word (the possibility that the suffix word is present after the common noun word in the text). However, "-100" is set as the connection evaluation value between the prefix word and the common noun word. Further, in the combination indicated by “*” in FIG. 5, a connection evaluation value (“-1000000” as an example in FIG. 5) indicating that there is no connection possibility is set.

  The morpheme analysis unit 14 of the morpheme analyzer 12 performs the following morpheme analysis on the morpheme analysis target text (the analysis target text 32 or the text of the learning text corpus 34). That is, the morpheme analysis unit 14 extracts, from the first word dictionary 18 or the second word dictionary 20, the words included in the morpheme analysis target text and registered in the first word dictionary 18 or the second word dictionary 20. Identify the part of speech of each individual word. The morpheme analysis unit 14 also evaluates the connection possibility of each pair of adjacent words on the morphological analysis target text based on the connection evaluation value registered in the word-part-of-speech connection evaluation value table 28. Then, the morpheme analysis unit 14 determines reading of the text to be analyzed for morpheme based on the connection possibility evaluation result. The morpheme analyzer 14 functions as an example of a morpheme analyzer in the disclosed technology.

  The word concatenation possibility determination unit 16 identifies a noun class word pair (a pair of words) that can be concatenated on the text from the text in the learning text corpus 34, and is capable of word concatenation in which the identified word pair is registered. A sex table 24 is generated. The word connection possibility determination unit 16 functions as an example of a connection possibility determination unit in the disclosed technology. Details of the processing by the word connection possibility determination unit 16 and the contents of the word connection possibility table 24 will be described later.

  The analysis target text 32 is a text that is subjected to morpheme analysis by the morpheme analysis unit 14 of the morpheme analyzer 12 and is then synthesized with speech that reads out the text through dependency analysis and generation of a phonetic character string to be described later. is there. In this embodiment, an aspect in which the analysis target text 32 is stored in advance in the fourth storage unit 36 will be described. However, the analysis target text 32 is input by the user via the keyboard 48 and stored in the memory 44, for example. It may be configured. The analysis target text 32 is input through an input device provided in an external information processing apparatus, for example, and the text is received from the external information processing apparatus by communication and stored in the memory 44. May be.

  The learning text corpus 34 is a text group for acquiring unregistered words in the first word dictionary 18, and the text corresponding to the purpose of registering words in the second word dictionary 20 and the use of the second word dictionary 20. Is selected. For example, when it is desired to register a new word or current vocabulary that is not registered in the first word dictionary 18 in the second word dictionary 20, a text such as a newspaper article or a magazine article is selected as the learning text corpus 34. Further, for example, when a term in a specific field that is not registered in the first word dictionary 18 is to be registered in the second word dictionary 20, a text such as a paper or manual in the specific field is selected as the learning text corpus 34. . In the present embodiment, a mode in which the learning text corpus 34 is stored in advance in the fourth storage unit 36 will be described. However, the learning text corpus 34 is received by communication from an external information processing apparatus, for example, and is stored in the memory 44. The structure memorize | stored in may be sufficient. Further, the text group as the learning text corpus 34 is not limited to being collectively stored in the fourth storage unit 36 or the memory 44. For example, individual texts are distributed on a computer network such as the Internet. It may be stored.

  The dependency analysis unit 38 performs dependency analysis on the analysis target text 32 based on the result of the morphological analysis of the analysis target text 32 performed by the morphological analysis device 12. The dependency analysis unit 38 functions as an example of a dependency analysis unit in the disclosed technology. The phonetic character string generation unit 40 also adds at least an accent or a phrase to the character string representing the reading of the analysis target text 32 based on the result of the dependency analysis of the analysis target text 32 by the dependency analysis unit 38. A phonetic character string to which information representing a delimiter is added is generated. The phonetic character string generation unit 40 functions as an example of the phonetic character string generation unit in the disclosed technology. The speech synthesizer 42 performs speech synthesis processing for synthesizing speech that reads the generated phonogram string based on the phonogram string generated by the phonogram string generator 40. The speech synthesizer 42 functions as an example of a speech synthesizer in the disclosed technology.

  Next, the operation of this embodiment will be described. In order to ensure the reading accuracy when synthesizing the speech that reads the analysis target text 32 from the analysis target text 32, it is important to improve the accuracy of the morphological analysis that identifies the words included in the analysis target text 32. It is essential to improve the word dictionary used for analysis. However, adding words to the word dictionary in the dark cloud may adversely affect the analysis accuracy of morphological analysis, so it is necessary to select registered words so that appropriate results can be obtained in morphological analysis, which is efficient. It is difficult to improve the word dictionary. On the other hand, there are no limit to words such as proper nouns and technical terms that appear in text written in natural language, and it is also required to efficiently register a large number of words in a word dictionary.

  Based on the above, in the present embodiment, the first word dictionary 18 and the second word dictionary 20 are provided. In the first word dictionary 18, information on the word selected in consideration of the analysis accuracy of the morphological analysis is stored in advance. Each word is registered, and a word not registered in the first word dictionary 18 is additionally registered in the second word dictionary 20. For the second word dictionary 20, for example, a process of automatically collecting “notation”, “part of speech”, and “reading” of words from the learning text corpus 34 and automatically registering the information can be applied. For example, if the text included in the learning text corpus 34 includes the character string “bone organ”, it is likely to indicate the notation and reading of the word. And can be registered in the second word dictionary 20.

  However, as described above, a word additionally registered in the second word dictionary 20 may adversely affect the morphological analysis. Hereinafter, specific examples will be described with reference to FIGS. In the following specific example, the information shown in FIG. 3 is registered in the first word dictionary 18, the information shown in FIG. 4 is registered in the second word dictionary, and the word-part-of-speech connection evaluation value table 28 has the information shown in FIG. It is assumed that the information shown in is registered.

  FIG. 17 shows a case where a morphological analysis is performed on the text “striking bone device”. In the morphological analysis, first, all the words registered in the first word dictionary 18 or the second word dictionary 20 and included in the text “striking boner” are extracted from the first word dictionary 18 and the second word dictionary 20. The In the example of FIG. 17, only the common noun word “bone” is extracted from the second word dictionary 20, and other words are extracted from the first word dictionary 18. In the morphological analysis, next, connection evaluation values between parts of speech are obtained by referring to the word part-of-speech connection evaluation value table 28 for all word pairs that may be connected. For example, the common noun word “Dasei” and the suffix word “Bone” have a connection evaluation value = −100, which indicates that the connection possibility is relatively high. On the other hand, the word “hit (da)” at the beginning and the suffix is unlikely to have the suffix at the beginning of the sentence, so the connection evaluation value = −1000000, indicating that the possibility of connection is very low.

When connection evaluation values are obtained for all word pairs, a word combination (path) having the maximum connection evaluation value is obtained, and a reading corresponding to the obtained path is output as an analysis result of morphological analysis. In the example of FIG.
`` First-common noun `` made '' (dase) ''-common noun `` bone '' (last)
Since the connection evaluation value of the path is the maximum (= −2150), the reading “Dassy Kokki” corresponding to this path is output as the analysis result of the morphological analysis.

  FIG. 18 shows a case where morphological analysis is performed on the “encoding method”. First, all the words included in the text “encoding method” are extracted from the first word dictionary 18 and the second word dictionary 20. In the example of FIG. 18, only the common noun word “formation (keho)” is extracted from the second word dictionary 20, and the other words are extracted from the first word dictionary 18. Next, by referring to the word part-of-speech connection evaluation value table 28 for all possible word pairs, the connection evaluation value between parts of speech is obtained, and the word having the maximum connection evaluation value is obtained. Find a combination (path). The results are as follows.

・ Pass 1
"First-common noun" sign (Fugo) "-common noun" Kaho "-last"
Total connection evaluation value = -2150
・ Pass 2
`` First-common noun `` sign '' (fugo)-suffix `` ka ''-suffix `` ho ''-end
Total connection evaluation value = -2200
In this case, even though the reading “Fuga Kaho” corresponding to the path 2 is correct, the reading “Fuga Keho” corresponding to the path 1 is output as an analysis result of the morphological analysis.

  As described above, when the second word dictionary 20 is used, the accuracy of morphological analysis can be improved in a scene such as the example shown in FIG. 17, but morphological analysis is performed in another scene such as the example shown in FIG. Adversely affects the accuracy of In order to solve this, it is conceivable to use the second word dictionary 20 with a lower priority than the first word dictionary 18, but if the priority of the second word dictionary 20 is lowered, the second word dictionary 20 is used. The effect of using is also reduced.

  On the other hand, nouns (common nouns, proper nouns, prefixes, etc.) before and after specific words such as “no” “ga” “ni” “ni” “na” “do” “do” on a certain text. If a word such as a suffix is present, the noun class word set may be connected on another text. On the other hand, if there is no text in which a pair of words of a certain noun class is located before and after a specific word, the pair of words is unlikely to be connected on another text. It can be judged. Considering the above, in the present embodiment, the word storage possibility table 24 is stored in the second storage unit 26 and the word connection possibility determination unit 16 for registering information in the word connection possibility table 24 is provided. Hereinafter, first, the connection possibility table generation processing realized by the morphological analysis unit 14 and the word connection possibility determination unit 16 will be described with reference to FIG. This connection possibility table generation process is performed prior to the morphological analysis unit 14 performing morphological analysis on the analysis target text 32.

  In step 100 of the concatenation possibility table generation process, the morphological analysis unit 14 tries to acquire one unit of text from the learning text corpus 34 stored in the first storage unit 22. In the next step 102, the morpheme analyzer 14 determines whether or not one unit of text has been acquired from the learning text corpus 34 in step 100. If one unit of text can be acquired from the learning text corpus 34 at step 100, the determination at step 102 is affirmed and the routine proceeds to step 104. If the determination in step 102 is negative, the morphological analysis unit 14 ends the connection possibility table generation process.

  In step 104, the morpheme analysis unit 14 registers the first word dictionary 18 and the second word as the morpheme analysis in the first word dictionary 18 or the second word dictionary 20 and the words included in the text acquired in step 100. A process of extracting all from the word dictionary 20 is performed. In this embodiment, the total number of words extracted by the morphological analysis in step 104 is “N”.

  In the next step 106, the word connection possibility determination unit 16 obtains the analysis result of the morpheme analysis by the morpheme analysis unit 14 (information on the total number of N words), and then sets the variable i for identifying individual words. Substitute 0. In step 108, the word connection possibility determination unit 16 determines whether or not the variable i is smaller than a value obtained by subtracting 2 from the total number N of words. If the determination in step 108 is affirmative, the process proceeds to step 110, and the word concatenation possibility determination unit 16 determines that the part of speech of the i-th word among the N words is a noun, and the i + 1-th word is a specific word. Then, it is determined whether or not the part of speech of the i + 2nd word is a noun class. Note that the specific word in the determination in step 110 includes any of the above-mentioned “NO”, “GA”, “NI”, “ON”, “NA”, “YES”, “YES”, and the like.

  If the determination in step 110 is negative, the i-th word and the i + 2th word are different because the part of speech of at least one of the i-th and i + 2 words is not a noun or the i + 1-th word is not a specific word. It is not possible to judge that there is a high possibility of being connected on the text. Therefore, if the determination in step 110 is negative, the word connection possibility determination unit 16 proceeds to step 118, increments the variable i by 1, and returns to step 108.

If the determination in step 110 is affirmative , the i-th word and the i + 2nd word are both nouns and the i + 1th word is a specific word. Is likely to be connected on another text. Therefore, the word connection possibility determination unit 16 proceeds to step 112 when the determination in step 110 is affirmed, and determines whether or not the i-th word or the i + 2nd word is registered in the second word dictionary 20. judge.

  The word connection possibility determination unit 16 proceeds to step 114 when the determination in step 112 is affirmed. In step 114, a deduction value of the connection evaluation value is calculated for a set of words that cannot be connected and one of the words registered in the first word dictionary 18 is different from the set of the i-th word and the i + 2th word. . Here, the deduction value of the connection evaluation value with respect to a group of words that cannot be connected is the connection evaluation value when the group of words that cannot be connected appears in the morpheme analysis by the morphological analysis unit 14. It is calculated so as not to be included in the word combination having the maximum value.

  Specifically, from the words extracted by the morphological analysis in step 104, the word (second word dictionary registered word) registered in the second word dictionary 20 among the i-th word and the i + 2th word, Extract all words that correspond to the same part of the text. Next, as a common noun word, if the second word dictionary registered word is the i + 2nd word, the immediately preceding word is included, and if the second word dictionary registered word is the i th word, the immediately following word is included. Connection evaluation values are obtained for all word paths including the word dictionary registered word or extracted words. Then, the connection evaluation values obtained for each path are compared, and when the connection evaluation value of the path including the immediately preceding word or the immediately following word and the second word dictionary registered word is the maximum, the connection evaluation value of the path Finds the deduction value for decreasing to the second and subsequent values in descending order.

  In step 116, the word connection possibility determination unit 16 registers the combination of the i-th word and the i + 2 word in the word connection possibility table 24 as a combination of words that can be connected, and calculates in step 114. The deducted points are also registered in association with each other. Then, the word connection possibility determination unit 16 increments the variable i by 1 in the next step 118 and returns to step 108. If the determination in step 112 is negative, the word connection possibility determination unit 16 skips steps 114 and 116, proceeds to step 118, increments the variable i by 1, and returns to step 108.

  The connection possibility table generation process described above will be further described with an example. FIG. 7 shows an example of text included in the learning text corpus 34 and an example of words registered in the first word dictionary 18 and the second word dictionary 20. Further, FIG. 8 shows a case where the connection possibility table generation processing is performed on the first text “from a large number of punched bone organs ...” among the texts shown in FIG. 7. FIG. 9 shows an example of calculating a deduction value in this case.

When the morphological analysis unit 14 performs the morphological analysis of Step 104 on the first text, only the common noun word “bone” from the second word dictionary 20 among the words shown in FIG. The other words are extracted from the first word dictionary 18. In addition, the following word strings are extracted as word strings that satisfy the condition that “the part of speech of the i-th and i + 2 words is a noun and one is registered in the second word dictionary 20 and the i + 1-th word is a specific word”. The
"Common noun" Dasei "-particle" no "-common noun" bone "

  It can be determined that the combination of the i-th word and the i + 2nd word in the word string satisfying the above condition is highly likely to be connected on another text, such as “stamping bone device (Dusse Kokki)”. For this reason, as shown in FIG. 7, a set of the i-th word “done” and i + 2 word “bone” in the word string satisfying the above condition is stored in the word connection possibility table 24. be registered. As shown in FIG. 7, the word concatenation possibility table 24 is configured to register the second word dictionary registered word as a main word in the group of connectable words, and the first word dictionary registered word is the i-th word. If so, the first word dictionary registered word is registered as a preceding possible word for the main word. If the first word dictionary registered word is the i + 2nd word, the first word dictionary registered word is registered as a succeedable word. In this example, since the i + 2 word “bone” is registered in the second word dictionary 20, the i + 2 word “bone” is the main word, and the i-th word “cook”. “Made” is registered in the word concatenation possibility table 24 as a preceding possible word.

  Next, with reference to FIG. 9, the calculation of the deduction point value in this example will be described. As shown in FIG. 9A, the words corresponding to the same part of the text as the second word dictionary registered word (main word) in this example include the suffix “bone” and the suffix “instrument (key ) ”, Common noun“ bone ”, and common noun“ Utsuwa ”are extracted from the first word dictionary 18. In this example, the first word dictionary registered word is a word that can be preceded. Next, all word strings (word paths) including the word immediately before the common noun and including the main word or the extracted word are connected. Each evaluation value is calculated. The results are as follows (see also FIG. 9 (B)).

・ Pass 1
"Common nouns-suffix" bone "-suffix" vessel "": total connection evaluation value = -200
・ Pass 2
"Common nouns-suffix" bone "-suffix" vessel "": total connection evaluation value = -250
・ Pass 3
"Common nouns-common nouns" bone organs "": total connection evaluation value = -150
In the above, the common noun `` Utsuwa '' has a lower connection evaluation value with the previous word than the suffix `` Uki '', so the path through the common noun `` Utsuwa '' is clearly indicated. Omitted.

  From the above result, when a noun word is present immediately before the word “bone” on the text to be analyzed 32, even if the word is a word other than the predecessor word “Dasei”, the second The connection evaluation value of the path (path 3) including the word “bone (cock)” registered in the word dictionary 20 is maximized. Therefore, when the word immediately before the word “bone” is other than the word “striking”, a deduction value for reducing the connection evaluation value of pass 3 to the second and subsequent values in descending order is obtained. . Specifically, as shown in FIG. 9C, from the connection evaluation value of path 3 (= −150), the connection evaluation value of the path having the largest connection evaluation value other than path 3 (connection evaluation value of path 1 = -200) is subtracted, and a value (= 51) obtained by adding a predetermined value (“1” in the example of FIG. 9C) is used as the deduction point value. In this example, since a word that can be connected to the main word is a preceding possible word, the calculated deduction value is registered in the word connecting possibility table 24 in association with the preceding possible word.

FIG. 10 shows a case where the connectability table generation processing is performed on the second text “... is a multivalent function” among the texts shown in FIG. 7. 11 shows an example of calculating a deduction value in this case. When the morphological analysis unit 14 performs the morphological analysis of Step 104 on the second text, only the common noun word “multivalent (hawk)” from the second word dictionary 20 among the words shown in FIG. The other words are extracted from the first word dictionary 18. In addition, the following word strings are extracted as word strings that satisfy the condition that “the part of speech of the i-th and i + 2 words is a noun and one is registered in the second word dictionary 20 and the i + 1-th word is a specific word”. The
"Common noun" multivalent (taka) "-particle" no "-common noun" function (kansu) ""

  It can be determined that the combination of the i-th word and the i + 2nd word in the word string satisfying the above condition is highly likely to be connected on another text as in the “multivalent function (Taka Kansu)”. For this reason, as shown in FIG. 7, a set of the i-th word “multivalent (hawk)” and the i + 2th word “function (kansu)” in the word string satisfying the above condition is registered in the word connection possibility table 24. Is done. In this example, since the i-th word “multivalent (hawk)” is registered in the second word dictionary 20, the i-th word “multivalent (hawk)” is the main word, and the i + 2th word. “Function (Kang)” is registered in the word connection possibility table 24 as a subsequent possible word.

  Next, with reference to FIG. 11, the calculation of the deduction point value in this example will be described. As shown in FIG. 11A, the word corresponding to the same part of the text as the second word dictionary registered word (main word) in this example includes the prefix “ta” and the adjective stem “ ) ”And the“ Sai ”variable noun“ value ”are extracted from the first word dictionary 18 respectively. In this example, since the first word dictionary registered word is a followable word, next, all word strings (word paths) including the word immediately after the common noun and including the main word or the extracted word are connected. Each evaluation value is calculated. The results are as follows (see also FIG. 11 (B)).

・ Pass 1
“Prefix“ Ta (Ta) ”-Sa-Variable Noun“ Value (Tai) ”-Common Noun”: Sum of Connection Evaluation Values = -250
・ Pass 2
"Common noun" multivalent (hawk) "-common noun": total connection evaluation value = -150
In the above, the adjective stem `` man '' is lower than the prefix `` man '' because the adjective stem `` man '' is lower than the prefix `` man ''. The path passing through is omitted.

  From the above result, if a noun word is present immediately after the word “multivalent” on the text to be analyzed 32, the second word even if the word is not a succeedable word “function (kansu)”. The connection evaluation value of the path (path 2) including the word “multivalent (hawk)” registered in the dictionary 20 is maximized. For this reason, when the word immediately after the word “multivalent” is other than the word “function (kansu)”, a deduction value for reducing the connection evaluation value of pass 2 to the second and subsequent values in descending order is obtained. Specifically, as shown in FIG. 11C, the connection evaluation value of the path having the largest connection evaluation value other than the path 2 (connection evaluation value of the path 1 = pass 150) from the connection evaluation value of the path 2 (= −150). -250) is subtracted, and a value (= 101) obtained by adding a predetermined value ("1" in the example of FIG. 11C) is set as the deduction point value. In this example, since the word that can be connected to the main word is a subsequent word, the calculated deduction value is registered in the word connection possibility table 24 in association with the subsequent word.

  Similarly, for example, by performing the connectability table generation process for the third text “... the oldest bone found ...” shown in FIG. 7, “oldest (psycho)” becomes “bone” It is registered in the word concatenation possibility table 24 as a preceding possible word corresponding to “container (cock)”. Further, for example, by performing a connectability table generation process for the fourth text “… the skeleton has been manufactured ...” shown in FIG. 7, the “production (seisaku)” is changed to “the skeleton”. ) "Is registered in the word concatenation possibility table 24 as a subsequent possible word. Further, a corresponding deduction value (= 51) is calculated and registered in the word connection possibility table 24 in association with the succeedable word of the main word “bone organ”. Further, for example, by performing a connectability table generation process for the fifth text “... , “Jinshin” is registered in the word connection possibility table 24 as a preceding possible word. In addition, a corresponding deduction value (= 51) is calculated and registered in the word concatenation possibility table 24 in association with the preceding possible word of the main word “formation method (Kehoe)”.

  Subsequently, the morpheme analysis process performed by the morpheme analysis unit 14 after the above-described connection possibility table generation process is performed and the word connection possibility table 24 is generated will be described with reference to FIG.

  In step 130 of the morpheme analysis process, the morpheme analysis unit 14 tries to acquire one unit of text from the analysis target text 32 stored in the first storage unit 22. In the next step 132, the morpheme analyzer 14 determines whether or not one unit of text has been acquired from the analysis target text 32 in step 130. If one unit of text can be acquired from the analysis target text 32 at step 130, the determination at step 132 is affirmed and the routine proceeds to step 134. If the determination in step 132 is negative, it can be determined that the morphological analysis has been completed up to the end of the analysis target text 32, and the morphological analysis unit 14 ends the morphological analysis processing.

  In step 134, the morphological analyzer 14 is registered in the first word dictionary 18 or the second word dictionary 20, and the words included in the text acquired in step 130 are converted from the first word dictionary 18 and the second word dictionary 20. Extract all. In the present embodiment, the total number of words extracted in step 134 is “N”. In the next step 136, the morphological analysis unit 14 substitutes 0 for a variable i for identifying each word. In the next step 138, the word connection possibility determination unit 16 determines that the variable i is the total number N of words N. Or less.

  If the determination in step 138 is affirmed, the morpheme analysis unit 14 proceeds to step 140, and the i-th word and the words before and after the N words (i−1th and i + 1th words). Are obtained from the connection evaluation value table 28 between word parts-of-speech. In the next step 142, the morphological analysis unit 14 determines whether or not the i-th word is a noun class and is registered in the second word dictionary 20. If the determination in step 142 is negative, the morpheme analyzer 14 proceeds to step 152, increments the variable i by 1, and returns to step 138.

  If the determination in step 142 is affirmed, the morpheme analysis unit 14 proceeds to step 144, where the word immediately before the i-th word is a noun class and the preceding possible word corresponding to the i-th word It is determined whether or not the word is not registered in the word connection possibility table 24. If the determination in step 144 is affirmative, it can be determined that the i-th word and the immediately preceding word (i-1th word) are unlikely to be connected (connected) on the text. For this reason, the morphological analysis unit 14 proceeds to step 146 when the determination in step 144 is affirmed, and the deduction value associated with the preceding possible word of the i-th word as the main word is stored in the word concatenation possibility table. Read from 24. Then, the previously read deduction value is subtracted from the connection evaluation value acquired in the previous step 140 for the set of the i-th word and the immediately preceding word (i-1th word). Then, the process proceeds from step 146 to step 148.

  On the other hand, when the part of speech of the word immediately before the i-th word is not a noun, or when the immediately preceding word is registered in the word concatenation possibility table 24 as a predecessor word of the i-th word, It can be determined that there is a possibility of connecting (connecting) with the word immediately before in the text. Therefore, if the determination in step 144 is negative, the morpheme analyzer 14 proceeds to step 148 without performing the process in step 146. In this case, the connection evaluation value of the set of the i-th word and the immediately preceding word (i-1th word) is maintained as the connection evaluation value acquired in the previous step 140.

  In the next step 148, the morphological analysis unit 14 does not register the word immediately after the i-th word in the word concatenation possibility table 24 as a subsequent possible word corresponding to the i-th word whose part of speech is a noun class. Determine if it is a word. If the determination in step 148 is affirmative, it can be determined that the i-th word and the immediately following word (i + 1-th word) are unlikely to be connected (connected) on the text. For this reason, the morphological analysis unit 14 proceeds to step 150 when the determination in step 148 is affirmed, and the deduction value associated with the succeedable word of the i-th word as the main word is used as the word connection possibility table. Read from 24. Then, the previously read deduction value is subtracted from the connection evaluation value acquired in the previous step 140 for the set of the i-th word and the immediately following word (i + 1-th word). Then, the process proceeds from step 150 to step 152.

  On the other hand, when the part of speech of the word immediately after the i-th word is not a noun, or when the word immediately after is registered in the word concatenation possibility table 24 as a succeedable word of the i-th word, It can be determined that the word immediately after that has the possibility of being connected (connected) on the text. For this reason, if the determination in step 148 is negative, the morphological analysis unit 14 proceeds to step 152 without performing the process in step 150. In this case, the connection evaluation value of the set of the i-th word and the immediately following word (i + 1-th word) is maintained as the connection evaluation value acquired in the previous step 140.

  The morphological analysis unit 14 repeats the above-described steps 138 to 152 until the value of the variable i becomes equal to or greater than the total number N of words and the determination in step 138 is negative. If the determination in step 138 is negative, the process proceeds to step 154, and the morphological analysis unit 14 extracts a word combination (path) that maximizes the total connection evaluation value. In step 156, the morpheme analysis unit 14 reads the reading corresponding to the word combination (pass) extracted in step 154 in units of words from the first word dictionary 18 or the second word dictionary 20, and outputs the result as an analysis result. Return to step 130.

  The morphological analysis process described above will be further described with an example. FIG. 13 shows an example in which the above-described morphological analysis process is performed on one unit of the text “striking boner” extracted from the analysis target text 32. In the following, the words shown in FIG. 7 are registered in the first word dictionary 18 and the second word dictionary 20, and the word pairs shown in FIG. Description will be made assuming that the connection evaluation values shown in FIG. 5 are registered in the evaluation value table 28.

  In the example of FIG. 13, as words included in the text “striking boner”, from the first word dictionary 18, the common noun “striking (dasei)”, the suffix “striking (da)”, and the suffix “manufacturing (sasei) ) ”, The suffix“ bone ”, the suffix“ vessel ”, the common noun“ bone ”, and the common noun“ vessel ”. In the example of FIG. 13, the common noun “bone organ” is extracted from the second word dictionary 20 as a word included in the text “striking bone device”.

  Next, for all word pairs adjacent on the text “striking boner”, the corresponding connection evaluation value is acquired from the word-part-of-speech connection evaluation value table 28 based on the part of speech of each word. , Connectability is evaluated. For example, the connection evaluation value between the head and the common noun `` Daisei '' is -1000, the connection evaluation value between the head and the suffix `` Da '' is -1000000, and the suffixes (`` Da '' ”And“ Made ”,“ Bone ”and“ Vessel ”,“ Made ”and“ Bone ”) have a connection evaluation value of -100. Also, the connection evaluation value of common nouns and suffixes (`` Dasei '' and `` Bone '', `` Bone '' and `` Ki '')) is -100, and common nouns ( The connection evaluation value of “done” (dose) and “bone organ (cock)”, “done (dase)” and “bone”) is −150.

  Here, referring to the word connection possibility table 24 for the common noun “bone” extracted from the second word dictionary 20, the common noun “Dasei” is registered as the preceding possible word. The connection evaluation value between “made (made)” and “bone device (cock)” remains -150. In addition, regarding the possibility of connection with the suffix `` made '' as the preceding word of the common noun `` bone organ '', the part of speech of the suffix `` made '' is a noun class, but the word The connection evaluation value is reduced because it is not registered as a preceding possible word in the connection possibility table 24. That is, since the deduction value associated with the preceding possible word of the main word “bone organ” is “51”, the deduction value = 51 is subtracted from the initial connection evaluation value = −150, and the suffix The connection evaluation value between “made” and the common noun “bone” is changed to “−201”. The combination (path) of the word that gives the maximum total connection evaluation value is a combination of “common noun“ Dasei ”-common noun“ bone ”), and the corresponding correct reading“ "Dassy Kokki" is output as the result of morphological analysis.

  FIG. 14 shows an example in which the morpheme analysis process described above is performed on one unit of text “encoding method” extracted from the analysis target text 32. In the example of FIG. 14, as words included in the text “encoding method”, from the first word dictionary 18, the common noun “code (fugo)”, the suffix “mark (fu)”, the suffix “go (go)” , The suffix “ka”, the suffix “ho”, the verb stem “ka”, and the common noun “ho” are extracted. In the example of FIG. 14, the common noun “formation method (keho)” is extracted from the second word dictionary 20 as a word included in the text “encoding method”.

  Next, for all word pairs adjacent on the text “encoding method”, the corresponding connection evaluation value is acquired from the word-part-of-speech connection evaluation value table 28 based on the part of speech of each word, Connectability is evaluated. For example, the connection evaluation value between the head and the common noun `` Code (Fugo) '' is -1000, the connection evaluation value between the head and the suffix `` F (fu) '' is -1000000, and the suffixes (`` mark (F) '' And “No. (go)”, “No. (go)” and “Chemical (F)”, “Chemical (F)” and “Law (Ho)”) have a connection evaluation value of -100. In addition, for example, the connection evaluation value of the common noun `` Code (Fugo) '' and the suffix `` Ka '' is -100, the connection evaluation of the common noun `` Code (Fugo) '' and the common noun `` Kaho '' The value is -150.

  Here, as the preceding words of the common noun “Keho” extracted from the second word dictionary 20, the common noun “Code (Fugo)” and the suffix “No. (Go)” can be considered. Although the common noun “code” and the suffix “go” are both nouns, they are registered in the word concatenation possibility table 24 as the preceding possible words of the common noun “Keho”. Not. Accordingly, the connection evaluation value between the common noun “sign (Fugo)” and the suffix “No. (Go)” and the common noun “Kakeho (Keho)” is reduced. That is, the deduction value associated with the preceding possible word of the main word “Keho” is “51”, and the deduction value = 51 is subtracted from the initial connection evaluation value = −150. The connection evaluation value of “Fugo” and the suffix “No. Go” and the common noun “Kaho” is changed to “−201”. And the combination (path) of the word that gives the maximum total connection evaluation value is a combination of “common noun“ sign (Fugo) ”-suffix“ ka (ka) ”-suffix“ ho (ho) ”. Therefore, the corresponding correct reading “Fugo Ka Ho” is output as the analysis result of the morphological analysis.

  FIG. 15 shows an example in which a morphological analysis process is performed on one unit of the text “deepening method” extracted from the analysis target text 32. In the example of FIG. 15, the common noun “Jin Shin”, the single Chinese character “Jin”, the adjective stem “Fuka”, and the suffix “ ”), The suffix“ ho (ho) ”, the verb stem“ bake (ba) ”, and the common noun“ ho (ho) ”are extracted from the first word dictionary 18. In the example of FIG. 15, the common noun “formation (keho)” is extracted from the second word dictionary 20 as a word included in the text “deepening method”.

  Next, for all word pairs adjacent on the text “deepening method”, the corresponding connection evaluation value is acquired from the word-part-of-speech connection evaluation value table 28 based on the part of speech of each word, Connectability is evaluated. For example, the connection evaluation value between the head and the common noun `` Jin Shin '' is -1000, the connection evaluation value between the head and the single kanji `` Jin '' is -1000, and the suffixes (`` ”And“ go (go) ”) have a connection evaluation value of −100. Also, for example, the connection evaluation value of the common noun `` Jinshin '' and the suffix `` Ka '' is -100, the common noun `` Jin Shin '' and the common noun `` Kaho '' The connection evaluation value is -150.

  Here, as the preceding words of the common noun “Keho” extracted from the second word dictionary 20, the common noun “Jinshin” and the adjective stem “Fuka” can be considered. Among these, the adjective stem “Fuka” is excluded from the determination by the word connection possibility table 24 because the part of speech does not belong to the noun class. Further, the common noun “Jinshin” is registered in the word concatenation possibility table 24 as a preceding possible word for the common noun “Keho”. Therefore, the connection evaluation value between the common noun “Jinshin” and the common noun “Keho” remains -150. Since the combination (path) of the word with the maximum total connection evaluation value is a combination of “common noun“ Jinshin ”-common noun“ Keho ””, the corresponding correct reading “Jinshin Keho” is output as the result of morphological analysis.

  FIG. 16 shows an example in which a morpheme analysis process is performed on one unit of text extracted from the analysis target text 32 “for its conversion method”. In the example of FIG. 16, as a word included in the text “How to make it”, the adjunct “that”, the suffix “Cake”, the suffix “Hou”, the verb stem “Bake” The common noun “method (ho)”, the particle “ni”, and the particle “ha” are extracted from the first word dictionary 18. Further, in the example of FIG. 16, the common noun “formation method (keho)” is extracted from the second word dictionary 20 as a word included in the text “for its formation method”. Next, the connection evaluation value corresponding to the word “participation method” is obtained from the word-part-of-speech connection evaluation value table 28 based on the part-of-speech of each word for all adjacent word pairs The connection possibility is evaluated.

  Here, the conjunctive word “that” is considered as the preceding word of the common noun “Keho” extracted from the second word dictionary 20, and the particle “ni” is considered as the following word. However, the conjunction “no” and the particle “ni” are excluded from the determination by the word connection possibility table 24 because the part of speech does not belong to the noun class. And, the combination (path) of the word having the maximum total connection evaluation value is a combination of “adjunct“ that ”—common noun“ Kakeho ”—particle“ ni ”—particle“ ha ”. Corresponding correct reading “that ・ keho ・ ni ・ ha” is output as an analysis result of morphological analysis.

  In the above description, the mode of performing only the generation of the word connection possibility table 24 in the connection possibility table generation process (FIG. 6) has been described, but the disclosed technique is not limited to this. For example, based on the text included in the learning text corpus 34, in parallel with the generation of the word connection possibility table 24, extraction of words to be registered in the second word dictionary 20 and extracted words in the second word dictionary 20 This information may also be registered.

  In the above description, the mode in which the extraction of the word to be registered in the second word dictionary 20 and the generation of the word connection possibility table 24 are both performed based on the text included in the learning text corpus 34 has been described. The disclosed technique is not limited to this. Extraction of words to be registered in the second word dictionary 20 and generation of the word connection possibility table 24 may be performed based on texts included in different learning text corpora.

  Also, in the above, the first word dictionary 18 in which information of words selected in consideration of the analysis accuracy of morphological analysis is registered, and the second word for additionally registering unregistered words in the first word dictionary 18 Although the embodiment in which the dictionary 20 is provided has been described, the disclosed technique is not limited to this. For example, the present invention may be applied to a mode in which only one word dictionary is provided and the possibility of connection of a set of noun words registered in the word dictionary is determined.

  Moreover, although the aspect which performs the speech synthesis | combination which synthesize | combines the audio | voice which reads the analysis object text based on the result of having performed the morphological analysis with respect to the analysis object text 32 was demonstrated above, It is not limited. For example, you may apply to the aspect which performs the process which adds a kana to a kanji in a text based on the result of having performed morphological analysis. For example, when performing machine translation that translates text written in Japanese into text written in another language, morphological analysis is performed for the purpose of discriminating kanji readings that have different meanings depending on the reading. It is also possible to apply to the embodiment.

  Furthermore, in the above description, the text to be analyzed and the text written in Japanese as the learning text have been described as examples. However, the present invention is not limited to this, and a language that includes a word having multiple readings (and meanings). It is also possible to apply text described in (eg Chinese).

  In the above description, the speech synthesis program 58 is stored (installed) in the storage unit 46 in advance. However, the present invention is not limited to this. For example, the speech synthesis program (and morphological analysis program) in the disclosed technology can be provided in a form recorded on a recording medium such as a CD-ROM or a DVD-ROM.

  All documents, patent applications and technical standards mentioned in this specification are to the same extent as if each individual document, patent application and technical standard were specifically and individually stated to be incorporated by reference. Incorporated by reference in the book.

  Regarding the above embodiment, the following additional notes are disclosed.

(Appendix 1)
A set of noun-like words that are registered in the word dictionary and are adjacent to the specific word across the specific word on the learning text are extracted from the learning text, and the learning 5. A connectability determination unit that stores information on a set of noun words extracted from the text for use in the second storage unit as information representing a set of connectable words. The morphological analyzer according to any one of the above.

(Appendix 2)
The connection possibility determination unit is configured to store the information of a noun class word extracted from the learning text in the second storage unit as information indicating a connectable word set, and perform the morphological analysis. In the processing by the unit, when the group of connectable words stored in the second storage unit and a group of nouns whose words are different from each other are adjacent on the text to be analyzed, A deduction value subtracted from the connection evaluation value for a set of words of nouns with different words is included in the combination of words with a maximum connection evaluation value of the word set of nouns with different partial words The morpheme analysis device according to appendix 1, wherein the set deduction value is stored in the second storage unit in association with the connectable word set.

(Appendix 3)
The first storage unit includes, as the word dictionary, a first word dictionary in which information on a word selected in consideration of analysis accuracy of the morphological analysis is registered in advance, and words that are not registered in the first word dictionary. A second word dictionary for additionally registering information;
One of the second storage units is registered in the first word dictionary and the other is registered in the second word dictionary, and is adjacent to the specific word across the specific word on the learning text. Memorize a set of nouns as a set of connectable words,
The morpheme analyzer is registered in the first word dictionary and the other is registered in the second word dictionary, adjacent on the text to be analyzed, and as the set of connectable words. The connection evaluation value of a set of noun class words that is not stored in the second storage unit is reduced so as not to be included in a word combination that maximizes the connection evaluation value. The morpheme analyzer according to any one of 1 and Appendix 2.

(Appendix 4)
One of the words is registered in the first word dictionary and the other is registered in the second word dictionary, and a noun word adjacent to the specific word is sandwiched between the specific words on the learning text. A connection possibility that a set is extracted from the learning text, and information on a noun class word extracted from the learning text is stored in the second storage unit as information representing a connectable word set The morphological analysis device according to attachment 3, further comprising a determination unit.

(Appendix 5)
The connection possibility determination unit is configured to store the information of a noun class word extracted from the learning text in the second storage unit as information indicating a connectable word set, and perform the morphological analysis. In the processing by the unit, the set of connectable words stored in the second storage unit and the set of words of nouns different from one word registered in the first word dictionary are adjacent on the text to be analyzed A deduction value subtracted from the connection evaluation value with respect to a pair of nouns having a different one word, the word combination having a different noun with a different one word has a maximum connection evaluation value. The morpheme analyzer according to appendix 4, wherein the set deduction value is stored in the second storage unit in association with the connectable word set, so as not to be included in a combination of words.

(Appendix 6)
The connection possibility determination unit is configured such that one word registered in the first word dictionary is a set of words that can be connected from the other word registered in the second word dictionary on the learning text. If the set is a previous word set, the set deduction point value is a set of adjacent words on the text to be analyzed, the word set that can be connected, and the first set. A word that is registered in the word dictionary and stored in the second storage unit as the deduction value to be applied to a set of words of different nouns in which one word positioned before in the text to be analyzed is connected to the word Is a set of words in which one word registered in the first word dictionary is positioned after the other word registered in the second word dictionary on the learning text. Indicates the set deduction value as the text to be analyzed. A set of words adjacent to each other, wherein the set of connectable words and a set of words of a noun class that are registered in the first word dictionary and one of the words positioned later on the text to be analyzed are different The morpheme analyzer according to claim 5, which is stored in the second storage unit as the deduction point value applied to.

(Appendix 7)
The specific word is any one of “NO”, “GA”, “NI”, “ON”, “NA”, “YES”, “YES”, or any one of claims 1 to 4, and appendix 1 to appendix 6. Morphological analyzer.

DESCRIPTION OF SYMBOLS 10 Speech synthesizer 12 Morphological analyzer 14 Morphological analyzer 16 Word connection possibility determination part 18 First word dictionary 20 Second word dictionary 22 First storage part 24 Word connection possibility table 26 Second storage part 28 Connection between word parts of speech Evaluation value table 30 Third storage unit 32 Analysis target text 34 Learning text corpus 36 Fourth storage unit 38 Dependency analysis unit 40 Phonetic character string generation unit 42 Speech synthesis unit 44 Memory 46 Storage unit 58 Speech synthesis program 60 Morphological analysis Process 70 Computer 72 CPU

Claims (9)

A first storage unit that stores a word dictionary in which a plurality of words are registered in association with at least the part of speech of each of the words;
A second set of words that are registered in the word dictionary and that are adjacent to the specific word on the learning text and that are adjacent to the specific word are stored as a set of connectable words. A storage unit;
A connection evaluation value representing a connection possibility is obtained for each set of words that are registered in the word dictionary and are adjacent to each other in the analysis target text, and is registered in the word dictionary and is stored on the analysis target text. in of the set of words that are adjacent, for a set of words of nouns such that is not stored in the second storage section as a set of possible said articulating words, the set of words that the connection evaluation value is maximized A morpheme analysis unit that reduces the connection evaluation value so as not to be included, and performs a morphological analysis of the text to be analyzed based on a set of words having the maximum connection evaluation value;
Morphological analyzer including The second storage unit is associated with the group of connectable words, and is a group of nouns whose words are partly different from the group of connectable words stored in the second storage unit. Is stored on the analysis target text, a deduction value subtracted from the connection evaluation value of a set of words of different nouns in which some of the words are different is stored,
The morphological analysis unit includes the connectable word set stored in the second storage unit as a set of noun words that are not stored in the second storage unit as the set of connectable words. From the connection evaluation value of a set of words of different nouns in which some of the words are different from each other, stored in the second storage unit in association with the set of connectable words stored in the second storage unit The morpheme analyzer according to claim 1, wherein the deduction point value is reduced. The morphological analysis unit extracts words included in the text to be analyzed from the word dictionary, and generates a set of adjacent words on the text to be analyzed from the words extracted from the word dictionary. A connection evaluation value representing the connectability of the generated word set, and a word corresponding to the text to be analyzed and having the maximum connection evaluation value based on the connection evaluation value obtained for each word set The morpheme analyzer of Claim 1 or Claim 2 which calculates | requires the group of these . In the word dictionary, a plurality of words are registered in association with the part-of-speech names of the individual words,
A third storage unit for storing a connection evaluation value representing the connectability of the word set for each combination of part of speech of the word in the word set;
The morpheme analysis unit reads the connection evaluation value corresponding to the combination of parts of speech of each word in a set of words adjacent to each other in the analysis target text from the third storage unit. The morphological analyzer according to claim 3, wherein the connection evaluation value of a set is obtained. A morphological analyzer according to any one of claims 1 to 4,
Based on the result of the morpheme analysis by the morpheme analyzer, a dependency analysis unit that performs dependency analysis on the text to be analyzed;
Based on the result of dependency analysis by the dependency analysis unit, a phonetic character string generation unit that generates a phonetic character string in which information indicating at least a paragraph break is added to a character string indicating the reading of the text to be analyzed When,
A speech synthesizer that synthesizes speech that reads out the phonogram string based on the phonogram string generated by the phonogram string generator;
A speech synthesizer. Computer
Possibility of connection with a set of words that are stored in the first storage unit and are registered in a word dictionary in which a plurality of words are registered in association with at least the part of speech of each word and are adjacent on the text to be analyzed A connection evaluation value that represents each of the word pairs registered in the word dictionary and adjacent to the analysis target text, registered in the word dictionary and specified in the learning text. A noun that is not stored as a set of connectable words in a second storage unit that stores a set of words of nouns that are adjacent to the specific word as a set of connectable words the set of word classes, a connection evaluating step of the connection evaluation value reduces the connection evaluation value so that it is not contained in the set of words having the maximum
A morpheme analysis step for performing a morpheme analysis of the text to be analyzed based on a set of words having a maximum connection evaluation value obtained in the connection possibility evaluation step;
A morphological analysis method for performing processing including: Computer
Possibility of connection with a set of words that are stored in the first storage unit and are registered in a word dictionary in which a plurality of words are registered in association with at least the part of speech of each word and are adjacent on the text to be analyzed A connection evaluation value that represents each of the word pairs registered in the word dictionary and adjacent to the analysis target text, registered in the word dictionary and specified in the learning text. A noun that is not stored as a set of connectable words in a second storage unit that stores a set of words of nouns that are adjacent to the specific word as a set of connectable words the set of word classes, a connection evaluating step of the connection evaluation value reduces the connection evaluation value so that it is not contained in the set of words having the maximum
A morpheme analysis step for performing a morpheme analysis of the text to be analyzed based on a set of words having a maximum connection evaluation value obtained in the connection possibility evaluation step;
A dependency analysis step for performing dependency analysis on the text to be analyzed based on the result of morphological analysis by the morpheme analysis step;
Based on the result of the dependency analysis in the dependency analysis step, a phonetic character string generation that generates a phonetic character string in which information representing at least a segment break is added to a character string representing the reading of the text to be analyzed Steps,
Based on the phonogram string generated in the phonogram string generation step, a voice synthesis step of synthesizing a speech that reads out the phonogram string;
A speech synthesis method for performing processing including: On the computer,
Possibility of connection with a set of words that are stored in the first storage unit and are registered in a word dictionary in which a plurality of words are registered in association with at least the part of speech of each word and are adjacent on the text to be analyzed A connection evaluation value that represents each of the word pairs registered in the word dictionary and adjacent to the analysis target text, registered in the word dictionary and specified in the learning text. A noun that is not stored as a set of connectable words in a second storage unit that stores a set of words of nouns that are adjacent to the specific word as a set of connectable words the set of word classes, a connection evaluating step of the connection evaluation value reduces the connection evaluation value so that it is not contained in the set of words having the maximum
A morpheme analysis step for performing a morpheme analysis of the text to be analyzed based on a set of words having a maximum connection evaluation value obtained in the connection possibility evaluation step;
A morpheme analysis program for executing processing including On the computer,
Possibility of connection with a set of words that are stored in the first storage unit and are registered in a word dictionary in which a plurality of words are registered in association with at least the part of speech of each word and are adjacent on the text to be analyzed A connection evaluation value that represents each of the word pairs registered in the word dictionary and adjacent to the analysis target text, registered in the word dictionary and specified in the learning text. A noun that is not stored as a set of connectable words in a second storage unit that stores a set of words of nouns that are adjacent to the specific word as a set of connectable words the set of word classes, a connection evaluating step of the connection evaluation value reduces the connection evaluation value so that it is not contained in the set of words having the maximum
A morpheme analysis step for performing a morpheme analysis of the text to be analyzed based on a set of words having a maximum connection evaluation value obtained in the connection possibility evaluation step;
A dependency analysis step for performing dependency analysis on the text to be analyzed based on the result of morphological analysis by the morpheme analysis step;
Based on the result of the dependency analysis in the dependency analysis step, a phonetic character string generation that generates a phonetic character string in which information representing at least a segment break is added to a character string representing the reading of the text to be analyzed Steps,
Based on the phonogram string generated in the phonogram string generation step, a voice synthesis step of synthesizing a speech that reads out the phonogram string;
Speech synthesis program for executing processing including

Images