JP5906869B2 - Accent imparting apparatus, accent imparting method, and program - Google Patents

Description

  The embodiments discussed herein relate to speech synthesis techniques that convert text to speech.

In speech synthesis in which text is converted into speech, giving a correct accent to read out greatly affects the level of understanding of the user who hears the speech.
In a so-called compound word in which a plurality of words are combined to form a single word, there is a case where an accent coupling phenomenon occurs in which accents of individual words change and accents different from the accents of the original words. There are a plurality of accent change patterns in this accent coupling phenomenon, and a uniform rule does not apply to all cases.

  In order to correctly analyze such an accent coupling phenomenon, several techniques for analysis using word attributes and accent coupling rules set for each attribute are known. Specifically, rules for numbers and classifiers, rules for genus, rules for each part of speech, rules for isolated words, rules for conjunctions and katakana or kanji nouns, and rules for each part of speech of the word to which the morpheme connects The technique of using such as is known. In addition to this, there is also known a technique of using rules when adjunct words are combined with independent words, rules based on semantic information and part-of-speech information of adjacent words, and the like.

  In addition, as a technique for acquiring the accent of other compound words, a technique of registering the compound word itself in the dictionary, or searching for a compound word similar to a word not registered in the dictionary, and combining the accents of the similar words A technique of using rules is known.

  As another background art, a technique is known in which an accent can be given to an appropriate position for an exceptional case that cannot be processed by an integrated rule when generating an output sound accent.

  In addition, a technique is known in which accent information for a compound word from a synthesis target text is determined based on learning data created based on frequency information of a compound word of a morpheme sequence included in a text in a text corpus prepared in advance. It has been.

  Furthermore, at the time of dividing / combining accent phrases, accent phrase information associated with a combination of a plurality of accent phrases composed of morphemes is generated based on the accent combining rules of the phrases, and an arbitrary character string is generated from the information. A technique for searching for information to be selected is known.

Japanese Patent Laid-Open No. 11-231898 Japanese Patent Application Laid-Open No. 08-185195 Japanese Patent Laid-Open No. 02-186397 Japanese Patent Laid-Open No. 02-005097 JP 63-094298 A JP-A 61-006693 JP 58-181098 A Japanese Patent Laid-Open No. 04-036799 JP 2008-164785 A JP 2008-116650 A Japanese Patent Laid-Open No. 08-190397 JP 2002-258484 A JP 2005-242029 A

  The various techniques described above for obtaining the accent of a compound word are highly effective for compound words composed of two words, but have a problem that they cannot sufficiently cope with compound words composed of three or more words. Semantic connections are important for compound words of three or more words, and the semantic relationship between each word must be taken into account. Therefore, simple word parts and attributes can be used to combine accents of three or more words. This is because it is difficult to judge correctly. In addition, such semantic relationships often cannot be analyzed without background knowledge. Furthermore, since there are an enormous number of combinations of three or more compound words, it is difficult to prepare a compound word dictionary in advance.

  In view of the above-described problems, the accent applying apparatus described later in this specification enables an accent combining phenomenon to be appropriately analyzed even with Japanese text consisting of three or more words.

  One of the accent assignment devices described later in this specification includes a text input unit, a morpheme analysis unit, a verification phrase generation unit, an example search unit, an accent combination determination unit, and a result output unit. . Here, the text input unit accepts input of Japanese text. The morpheme analysis unit performs morpheme analysis on the accepted Japanese text and outputs a morpheme string divided into a plurality of morphemes. The verification phrase generation unit generates a plurality of types of verification phrases for the morpheme string by inserting the character “NO” into any one of the adjacent morpheme boundaries in the morpheme string. The example search unit counts the number of times the generated verification phrase appears in the sentence corpus. The accent combination determination unit determines whether or not an accent combination phenomenon occurs in a composite word obtained by combining two morphemes adjacent to the insertion position of the character “NO” by the verification phrase generation unit in the morpheme string. The determination is made based on the number of appearances described above for each type of verification phrase. Note that the accent combining phenomenon is a phenomenon in which the accent of the composite word changes from that for each of the two morphemes. The result output unit changes the accent information for each of the plurality of morphemes constituting the morpheme string based on the determination result in the accent combination determination unit, and creates an accent for the accepted Japanese text. Output.

  In addition, one of the accenting methods described later in this specification first generates a morpheme string divided into a plurality of morphemes by performing morphological analysis on the input Japanese text. Next, a plurality of types of verification phrases are generated for the generated morpheme sequence by inserting the character “NO” into any one of the adjacent morpheme boundaries in the generated morpheme sequence. Next, the number of appearances of the generated plural kinds of verification phrases in the sentence corpus is counted. Next, there is an accent combining phenomenon in the composite word obtained by combining two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when generating multiple types of verification phrases. Determine if it occurs. The accent combining phenomenon is a phenomenon in which the accent of the composite word changes from that of each of the two morphemes, and this determination is based on the number of appearances described above for each of a plurality of types of verification phrases. Done. Then, the accent information for each of the plurality of morphemes constituting the generated morpheme sequence is changed based on the determination result of whether or not the accent coupling phenomenon occurs, and the accent for the input Japanese text is changed. Create and output.

  One of the programs described later in this specification causes a computer to perform the following processing. In this process, first, morpheme analysis is performed on the input Japanese text to generate a morpheme string divided into a plurality of morphemes. Next, a plurality of types of verification phrases are generated for the generated morpheme sequence by inserting the character “NO” into any one of the adjacent morpheme boundaries in the generated morpheme sequence. Next, the number of appearances of the generated plural kinds of verification phrases in the sentence corpus is counted. Next, there is an accent combining phenomenon in the composite word obtained by combining two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when generating multiple types of verification phrases. Determine if it occurs. The accent combining phenomenon is a phenomenon in which the accent of the composite word changes from that of each of the two morphemes, and this determination is based on the number of appearances described above for each of a plurality of types of verification phrases. Done. Then, the accent information for each of the plurality of morphemes constituting the generated morpheme sequence is changed based on the determination result of whether or not the accent coupling phenomenon occurs, and the accent for the input Japanese text is changed. Create and output.

  The accent assignment device described later in this specification has an effect that an accent combining phenomenon can be appropriately analyzed even in Japanese text composed of three or more words.

It is a functional block diagram of one Example of an accent provision apparatus. It is a figure showing the example of the morpheme table. It is explanatory drawing of an example of the method of a morphological analysis. It is a figure showing the example of the accent joint rule. It is a figure showing the hardware structural example of one Example of the accent provision apparatus. It is the flowchart which illustrated the processing content of the accent provision process.

First, FIG. 1 will be described. FIG. 1 is a functional configuration diagram of an embodiment of an accent applying apparatus.
1 includes a text input unit 11, a morpheme analysis unit 12, a morpheme modification unit 13, a verification phrase generation unit 14, an example search unit 15, an accent combination determination unit 16, and a result output unit 17. .

The text input unit 11 receives input of Japanese text (text of hiragana, katakana, kanji, etc.) and outputs it to the morpheme analysis unit 12.
The morpheme analysis unit 12 performs morpheme analysis on the Japanese text received by the text input unit 11 and outputs a morpheme string divided into a plurality of morphemes.

  Any method may be used as a method for morphological analysis of the Japanese text. An example of a morphological analysis technique that can be used in the morphological analysis unit 12 of FIG. 1 will be briefly described.

  In this morphological analysis method, a word dictionary 21 prepared in advance in the accent assignment device 10 is used. FIG. 2 shows an example of a morpheme table that the word dictionary 21 has.

  In the morpheme table of FIG. 2, the “notation” of a character string that is a morpheme includes “reading” of the morpheme, “accent” when the morpheme is pronounced alone, and the morpheme represented by the character string. “Part of speech” is associated and registered. In addition, in the morphemes registered in the morpheme table, not only nouns but also verb stems, inflection endings, particles and auxiliary verbs are registered.

  In the following description, the position of the accent nucleus for a certain character string is indicated by adding the apostrophe “'” to the reading of the character string written in katakana, thereby obtaining the accent information for this character string. Try to express. Also in the morpheme table of FIG. 2, the “accent” data uses this expression method.

  For example, it is assumed that a Japanese text “Today is fine” is input to the morphological analysis unit 12. The morpheme analyzer 12 refers to the word dictionary 21 and expands this Japanese text into possible morpheme combinations. FIG. 3 shows a state where the morpheme analysis unit 12 develops “Today is fine” into various morpheme combinations using the word dictionary 21. Next, the morpheme analysis unit 12 makes a rule for connection between morphemes for each set of combinations of the listed morphemes (for example, a verb inflection ending is easy to connect after the verb stem, and after the noun. Various rules are applied, such as rules that are difficult to connect) and rules that prioritize morphemes with a large number of characters. Then, by quantifying the conformity to these rules, a combination of morphemes that seems to be optimal is selected, and the selected combination of morphemes is set as a result of morpheme analysis. In FIG. 3, the morpheme combinations “Today”, “Ha”, and “Sunny” selected in this way are represented by bold arrows as a result of the morphological analysis of “Today is sunny”.

  The morpheme modification unit 13 includes a plurality of morphemes constituting the morpheme sequence output by the morpheme analysis unit 12 and does not have a lexical meaning by itself and is used with other morphemes. It works when In this case, the morpheme correcting unit 13 corrects the morpheme sequence by treating a composite word obtained by combining such a morpheme and the other morpheme as one morpheme. For example, when affixes are included in a plurality of morphemes constituting a morpheme string, the morpheme modification unit 13 uses the morpheme that is the affix and the morpheme that is the stem of the affix among the plurality of morphemes. A combined word obtained by combining and is treated as one morpheme. The affix is a prefix or suffix, for example, and is combined with the morpheme that is the original word (stem) to supplement or change the meaning of the original word, or the part of speech from the original word. The morpheme to change. For example, a classifier is a suffix that is combined with a word representing a number to form a numeral, and is an example of a suffix. In this embodiment, information indicating whether or not a morpheme is an affix is also registered in the morpheme table of the word dictionary 21, and the morpheme correction unit 13 uses this information to correct the morpheme string. Shall be performed.

  The verification phrase generator 14 receives the morpheme string output from the morpheme analyzer 12. However, when the morpheme correction unit 13 corrects the morpheme sequence output from the morpheme analysis unit 12, the morpheme sequence corrected by the morpheme correction unit 13 is input to the verification phrase generation unit 14. The verification phrase generation unit 14 inserts the character “NO” as a particle into any one of the adjacent morpheme boundaries in the input morpheme sequence, thereby generating a plurality of types of verification phrases for the morpheme sequence. Generate.

  The example search unit 15 counts the number of appearances that appear in the sentence corpus 22 prepared in advance in the accent assignment device 10 for each of a plurality of types of verification phrases generated by the verification phrase generation unit 14. In this embodiment, a large-scale sentence corpus 22 having several hundred to several billion general Japanese sentences is used. Note that the sentence corpus 22 is not necessarily prepared in advance and is not necessarily held in the accent assigning apparatus 10, but may be configured to refer to a sentence corpus existing outside via a network, for example.

  The accent combination determination unit 16 determines whether or not an accent combination phenomenon occurs when two morphemes adjacent to the insertion position of the character “NO” in the morpheme string input to the verification phrase generation unit 14 are combined. . The accent combining phenomenon is a phenomenon in which the accent of a composite word obtained by combining these two morphemes changes from that for each of the two morphemes. In addition to the determination based on the accent combination rule 23 between the two morphemes respectively associated with the two morphemes, the accent combination determination unit 16 further performs a plurality of types of verification by the example search unit 15. This is also based on the result of counting the number of appearances for each phrase.

The accent combining rule 23 is prepared in advance in the accent assigning device 10. FIG. 4 shows an example of the accent combination rule 23.
The table of the accent combination rule 23 illustrated in the figure includes fields of “entry word”, “position”, “influence”, “part of speech to be combined”, “combination presence / absence”, “deformation type”, and “exception”. Is provided.

“Headword” stores a morpheme that constitutes a composite word that is a target for determining whether or not an accent coupling phenomenon has occurred.
“Position” stores data indicating whether the position of the morpheme that is the “entry word” is “front” or “rear” in the composite word that is the determination target.

  In the “influence”, when the morpheme that is the “entry word” and the other morpheme that is combined with the morpheme and constitutes a composite word conflict with each other in the “combination presence / absence” information described below, Information for determining which “combination presence / absence” information is to be used is stored. More specifically, data representing “large” or “small” is stored in this “influence”. Here, when the information on “combination presence / absence” conflicts between the two morphemes that compose the compound word, the information on “combination presence / absence” of the morpheme whose “influence” data is “large” The determination result is whether or not an accent coupling phenomenon occurs in the synthesized word. In addition, when the “influence” data is the same between two morphemes, the “combined presence / absence” information of the morpheme whose position in the compound word composed of the two morphemes is backward is: It is determined whether or not an accent coupling phenomenon occurs in the synthesized word.

“Combined part-of-speech” stores information on the part-of-speech of the other party's morpheme that is combined with the morpheme that is the “entry word” to form a composite word.
In “Combination presence / absence”, data indicating “Yes” or “No” is stored as data indicating whether or not an accent combining phenomenon occurs when a composite word is configured using a morpheme that is “headword”. Is done.

  The “deformation type” stores information indicating the change in the accent of the morpheme that is the “entry word” when the accent coupling phenomenon occurs. For example, a morpheme whose “deformation type” is “1 type” indicates that an accent is given to the first (that is, the top) in the mora (mora: beat in Japanese) of the morpheme. In addition, a morpheme whose “deformation type” is “N-type” represents that an accent is given to the Nth (ie, the end) mora in the morpheme having N mora. Furthermore, a morpheme whose “deformation type” is “N-1 type” is the N-1th mora in the morpheme having N mora (that is, the end is also a position returned from the mora by 1 mora). Indicates that an accent is given. On the other hand, a morpheme whose “deformation type” is “flattened (zero-type)” represents that the accent attached to the morpheme alone is eliminated (flattened). Note that a morpheme whose “deformation type” is “non-deformation” indicates that even if an accent combining phenomenon occurs in a compound word composed using the morpheme, the morpheme itself does not change the accent. Yes.

  An “exception” is a list of counterpart morphemes that are combined with a morpheme that is a “headword” to form a compound word, and from the above rules for determining whether or not an accent combining phenomenon occurs for the compound word This is a list of things that come off. The determination of the presence or absence of an accent combining phenomenon for a compound word composed of these listed morphemes and the morpheme of “headword” is “exception” when the “participant part of speech” in the morpheme of the “headword” is “exception”. This is performed based on the information of “combination presence / absence” in the recorded record.

  Details of a method for determining whether or not an accent coupling phenomenon has occurred based on the accent coupling rule 23 and the number of appearances in the sentence corpus 22 for each verification phrase by the accent coupling determination unit 16 will be described later.

  The result output unit 17 creates and outputs accent information about the Japanese text received by the text input unit 11. The creation of the accent information is performed by changing the accent information for each of the plurality of morphemes constituting the morpheme string input to the verification phrase generation unit 14 based on the determination result in the accent combination determination unit 16. Done.

  The accent assigning apparatus 10 configured as described above estimates Japanese language text composed of a plurality of morphemes by estimating the meaning breaks between the morphemes constituting the Japanese text using the example of the sentence corpus 22. Into a compound word consisting of two morphemes. By doing in this way, the accent imparting apparatus 10 can appropriately analyze the accent coupling phenomenon even in Japanese text composed of three or more words.

  In addition, since the accent assigning device 10 includes the morpheme correction unit 13, it is possible to search for an example based on an error in a meaning break that may occur by handling a morpheme that does not have a lexical meaning by itself. As a result, the accuracy of analysis of the accent coupling phenomenon is improved.

  Further, when the accent combination determination unit 16 in the accent assigning apparatus 10 combines two morphemes adjacent to the insertion position of the character “NO” in the morpheme string input to the verification phrase generation unit 14, the accent combination phenomenon occurs. The determination of whether or not it occurs is performed as follows.

  First, the accent combination determination unit 16 pays attention to the verification phrase having the maximum number of appearances counted by the example search unit 15 among the plurality of types of verification phrases generated by the verification phrase generation unit 14. The verification phrase attracting attention at this time is referred to as a “determination target verification phrase”. Next, the accent combination determination unit 16 determines the number of appearances for the verification phrase other than the determination target verification phrase among the plurality of types of verification phrases generated by the verification phrase generation unit 14 for the determination target verification phrase. It is determined whether or not the ratio is larger than a predetermined determination threshold in any verification phrase other than the determination target verification phrase. In this determination, when the number of appearances for a verification phrase other than the determination target verification phrase is zero, the ratio of the number of appearances described above with respect to the verification phrase other than the determination target verification phrase for the determination target verification phrase. Assume that it is determined that is greater than a predetermined determination threshold.

  If the accent combination determination unit 16 determines that the above-described ratio is larger than the predetermined determination threshold in any of the verification phrases other than the determination target verification phrase, Note the insertion position of the character “NO” inserted by the verification phrase generator 14. Here, the accent combination determination unit 16 determines that the insertion position of the character “NO” in the determination target verification phrase is the rearmost boundary among the adjacent morpheme boundaries in the morpheme string input to the verification phrase generation unit 14. It is determined whether or not. In the following description, this rearmost boundary is referred to as “the last boundary”.

  Here, if the insertion position of the character “NO” is not the last boundary, the accent connection determination unit 16 determines that the accent connection phenomenon does not occur at this insertion position. On the other hand, if the insertion position of the character “NO” is the last boundary, the accent combination determination unit 16 determines whether or not the accent combination phenomenon occurs at the last boundary. Is determined according to the accent combining rule 23 for two morphemes adjacent to.

  By the way, even when the accent combination determination unit 16 determines that the above-described ratio is smaller than the predetermined determination threshold in any of the verification phrases other than the determination target verification phrase, Whether or not the coupling phenomenon occurs is determined in the same manner as described above. That is, even in this case, the accent joint determination unit 16 determines whether or not an accent joint phenomenon occurs at the tail boundary according to the accent joint rule 23 for two morphemes adjacent to the tail boundary. To do.

  Note that the number of appearances counted by the example search unit 15 for each of a plurality of types of verification phrases generated by the verification phrase generation unit 14 may be all zero. In this case, the accent combination determination unit 16 determines that an accent combination phenomenon occurs in all of the boundary positions in the morpheme string input to the verification phrase generation unit 14.

  In addition, it is determined that there are only two adjacent morpheme boundaries in the morpheme string input to the verification phrase generation unit 14 and that no accent combining phenomenon occurs in one of them. There may be a case where the accent combination determination unit 16 makes a decision. In this case, the accent coupling determination unit 16 determines that an accent coupling phenomenon occurs at the other position.

  As described above, the accent combination determination unit 16 determines whether an accent combination phenomenon occurs at the insertion position of the character “NO” in the morpheme string input to the verification phrase generation unit 14. By performing this determination by the above-described method, it is possible to estimate a break in meaning from the number of appearances in the example sentence corpus 22. In addition, in this way, when a break in meaning is estimated at the tail boundary, an accent joint phenomenon that emphasizes the accent joint rule 23 associated with the morpheme adjacent to the tail boundary is performed. Analysis can be performed. Therefore, the accent coupling determination unit 16 determines whether or not the accent coupling phenomenon occurs as described above, thereby improving the accuracy of the accent coupling phenomenon analysis.

  Even after the accent combination determination unit 16 determines whether or not the accent combination phenomenon occurs at the insertion position of the character “NO” as described above, the determination is performed at the position of another boundary in the morpheme string. There may still be things that have not been left. In this case, the verification phrase generation unit 14 may newly generate a plurality of types of verification phrases for the morpheme string by inserting the character “NO” into any one of the remaining boundaries. Good. In this case, the example search unit 15 counts the number of appearances for a newly generated verification phrase. The accent combination determination unit 16 determines whether or not an accent combination phenomenon occurs at the remaining boundary position based on the number of appearances of the newly generated verification phrase. Here, when the verification phrase generation unit 14 newly generates a plurality of types of verification phrases, the morpheme sequence is determined at the boundary position where it is already determined that the accent combination phenomenon does not occur in the morpheme sequence. Is divided into two. Then, the verification phrase is newly generated for each of the two divided morpheme strings.

  Since each component of the accent assigning apparatus 10 operates as described above in the above-described case, the presence / absence of the accent coupling phenomenon is sequentially determined even in Japanese text including four words or more. The coupling phenomenon can be analyzed appropriately.

Next, FIG. 5 will be described. FIG. 5 shows an example of a hardware configuration of the accent applying apparatus 10 of FIG.
In this configuration example, the accent imparting apparatus 10 is configured by a computer 30. The computer 30 includes an MPU 31, a ROM 32, a RAM 33, a hard disk device 34, an input device 35, an output device 36, an interface device 37, and a recording medium drive device 38. Note that these components are connected via a bus line 39, and various data can be exchanged under the management of the MPU 31.

An MPU (Micro Processing Unit) 31 is an arithmetic processing unit that controls the operation of the entire computer 30.
A ROM (Read Only Memory) 32 is a read-only semiconductor memory in which a predetermined basic control program is recorded in advance. The MPU 31 can control the operation of each component of the computer 30 by reading out and executing this basic control program when the computer 30 is activated. As the ROM 32, a memory such as a flash memory whose storage data is nonvolatile may be used.

  A RAM (Random Access Memory) 33 is a semiconductor memory that can be written and read at any time and used as a working storage area as necessary when the MPU 31 executes various control programs.

  The hard disk device 34 is a storage device that stores various control programs executed by the MPU 31 and various data. The MPU 31 can perform various control processes by reading and executing a predetermined control program stored in the hard disk device 34. When the computer 30 is used to configure the accent assigning apparatus 10, the word dictionary 21, the sentence corpus 22, and the accent combining rule 23 shown in FIG. 1 are stored in advance in the hard disk device 34.

  The input device 35 is, for example, a keyboard device or a mouse device. For example, when operated by the user of the accent imparting device 10, the input device 35 acquires and acquires various information input from the user associated with the operation content. The input information is sent to the MPU 31. Japanese text that is input to the accenting device 10 is received by the input device 35, for example.

  The output device 36 is, for example, a speaker or a liquid crystal display, and generates synthesized speech and displays various texts and images according to output data sent from the MPU 31. The accent of the input Japanese text, which is the output of the accent assigning device 10, is given to the synthesized speech of the Japanese text to be pronounced and output from the output device 36, or, for example, as described above It is displayed on the output device 36 by the notation.

  The interface device 37 manages the exchange of various information with various devices connected to the computer 30. Japanese text that is input to the accent assignment device 10 may be output from another device and received by the interface device 37, for example. Further, for the accent of the input Japanese text, which is the output of the accent assigning device 10, for example, the accent information may be output from the interface device 37 and sent to another device.

  The recording medium driving device 38 is a device that reads various control programs and data recorded on the portable recording medium 40. The MPU 31 can read out and execute a predetermined control program recorded on the portable recording medium 40 via the recording medium driving device 38, thereby performing an accent applying process described later. As the portable recording medium 40, for example, a flash memory equipped with a CD-ROM (Compact Disc Read Only Memory), a DVD-ROM (Digital Versatile Disc Read Only Memory), or a USB (Universal Serial Bus) standard connector. and so on.

  In order to configure the accent applying apparatus 10 using such a computer 30, for example, a control program for causing the MPU 31 to perform an accent applying process described later is created. The created control program is stored in advance in the hard disk device 34 or the portable recording medium 40. Further, the word dictionary 21, the sentence corpus 22, and the accent combination rule 23 are stored in the hard disk device 34 in advance. Then, a predetermined instruction is given to the MPU 31 to read and execute the control program. In this way, the computer 30 is caused to function as the text input unit 11, the morpheme analysis unit 12, the morpheme correction unit 13, the verification phrase generation unit 14, the example search unit 15, the accent combination determination unit 16, and the result output unit 17. Is possible.

Next, an accent imparting process performed by the MPU 31 of the computer 30 will be described. FIG. 6 is a flowchart illustrating the processing contents of this accent assignment processing.
When the processing of FIG. 6 is started, first, in S101, the MPU 31 performs text input processing. This process is a process of acquiring input of Japanese text (text of hiragana, katakana, kanji, etc.) to the computer 30 (for example, the input device 35 or the interface device 37). The MPU 31 that performs this process functions as the text input unit 11.

  Next, in S102, the MPU 31 performs morphological analysis processing. This process is a process for obtaining a morpheme string divided into a plurality of morphemes by performing morphological analysis on the Japanese text acquired by the process of S101. In this embodiment, the word dictionary 21 stored in advance in the hard disk device 34 is used, and the MPU 31 performs this morpheme analysis process using the morpheme analysis method already described with reference to FIGS. The MPU 31 that performs this process functions as the morphological analyzer 12.

  Next, in S103, the MPU 31 performs a morpheme correction process. This process extracts morphemes that do not have a lexical meaning by themselves and are used with other morphemes, and the other morphemes, from a plurality of morphemes constituting the morpheme string obtained by the processing of S102 Thus, both are combined and handled as one morpheme. For example, when affixes are included in a plurality of morphemes constituting a morpheme string, this process results in the morpheme that is the affix and the morpheme that is the stem of the affix among the plurality of morphemes. The combined word obtained by combining is handled as one morpheme. The MPU 31 that performs this process functions as the morpheme correction unit 13.

  Next, in S104, the MPU 31 performs verification phrase generation processing. This process is a process for generating a plurality of types of verification phrases from the morpheme string obtained by the process of S102 or, when the morpheme is corrected by the process of S103, from the corrected morpheme string. Generation of a plurality of types of verification phrases is performed by inserting the letter “NO” as a particle at any one of the boundaries of adjacent morphemes in the morpheme string to be processed. The MPU 31 that performs this process functions as the verification phrase generator 14.

  Next, in S105, the MPU 31 performs an example search process. This process is a process of counting the number of appearances appearing in the sentence corpus 22 stored in advance in the hard disk device 34 for each of a plurality of types of verification phrases generated by the process of S104. The MPU 31 that performs this process functions as the example search unit 15.

  Next, in S106, the MPU 31 performs an accent joint determination process. This process is a process for determining whether or not an accent combining phenomenon occurs when two morphemes adjacent to the insertion position of the character “NO” in the process of S104 are combined. This determination is performed based on the accent combination rule 23 between the two morphemes respectively associated with the two morphemes and the count result of the number of appearances for each of the plurality of types of verification phrases by the process of S105. . The accent combination rule 23 is stored in the hard disk device 34 in advance. The MPU 31 that performs this process functions as the accent coupling determination unit 16, and the above-described determination method by the accent coupling determination unit 16 is used to determine whether or not an accent coupling phenomenon occurs due to this processing.

  Next, in S107, it is determined whether or not there is still a boundary that has not yet been determined whether or not the accent coupling phenomenon occurs at the morpheme boundary in the morpheme sequence that is the target of the determination process in S106. Processing is performed by the MPU 31.

  When the MPU 31 determines in the determination process of S107 that such an undetermined boundary remains (when the determination result is Yes), the MPU 31 returns the process to S104 and repeats the above-described process. At this time, in the verification phrase generation process of S104, a process of newly generating a plurality of types of verification phrases for the morpheme string is performed by inserting the character “NO” into any one of the undetermined boundaries. Then, in the subsequent example search process of S105, a process of counting the number of appearances described above is performed for a newly generated verification phrase. Then, in the subsequent accent joint determination process of S106, the character “NO” is inserted at the position of the undetermined boundary where the character “NO” is inserted based on the number of appearances of the newly generated verification phrase counted in the process of S105. A determination process is performed to determine whether or not an accent coupling phenomenon occurs.

On the other hand, when the MPU 31 determines in the determination process of S107 that there is no undetermined boundary (when the determination result is No), the process proceeds to S108.
In S108, the MPU 31 performs a result output process. This process is a process of creating accent information for the Japanese text acquired by the process of S101 and outputting the accent information from the output device 36 or the interface device 37 of the computer 30, for example. The creation of the accent information is performed by changing the accent information for each of the plurality of morphemes constituting the morpheme sequence to be processed in S104 based on the determination result by the accent combination determination process in S106.

When the process of S108 described above is completed, the accent imparting process in FIG. 6 ends.
When the MPU 31 performs the above accent assignment processing, the computer 30 operates as the accent assignment device 10 and can create and output accent information for the input Japanese text.

Hereinafter, the above-described accent assignment process of FIG. 6 will be described in detail using examples of some Japanese texts input to the accent assignment apparatus 10.
[1] Case of “Senior House Reconstruction Special Committee” First, the case where the Japanese text “Senshuin Reconstruction Special Committee” is input to the accent assignment device 10 will be described.

  When the Japanese text “Special Committee for Reconstruction of the House of Representatives” is acquired by the text input process of S101, next, a morphological analysis process is performed in S102. It is assumed that an analysis result of the morpheme string “Shrine | Reconstruction | Special | Committee | Meeting” is obtained by this processing. The symbol “|” represents the boundary position of adjacent morphemes in the morpheme string.

  Next, the morpheme correction process of S103 is performed on the morpheme string “Shu-Shin | Reconstruction | Special | Committee | Meeting”. Here, it is assumed that the morpheme table of the word dictionary 21 includes information indicating that the morpheme “kai” is a suffix. In this case, the morpheme “committee” and the morpheme “meeting” are combined by the morpheme correction processing, and the obtained composite word “committee” is treated as one morpheme. As a result, the morpheme sequence “Senshui | Reconstruction | Special | Committee | Meeting” is revised to the morpheme sequence “Senshui | Reconstruction | Special | Committee”.

  Next, the verification phrase generation process of S104 is performed on the morpheme string “Shu-Shin | Reconstruction | Special | Committee” obtained as described above. Through this process, the character “no” is inserted between “Representative House” and “Reconstruction”, between “Reconstruction” and “Special”, and between “Special” and “Committee”. The verification phrases “Special Committee for Reconstruction of the House of Representatives”, “Special Committee for Reconstruction of the House of Representatives”, and “Special Committee for Reconstruction of the House of Representatives” are generated.

Next, the example search process of S105 is performed for the three verification phrases obtained as described above. Here, it is assumed that the count results of the example search process for the number of appearances of the three verification phrases in the sentence corpus 22 are as follows.
・ “Special Committee for Reconstruction of the House of Representatives” ... 104 times ・ “Special Committee for the Reconstruction of the House of Representatives” ... …… 0 times ・ “Special Committee for Reconstruction of the House of Representatives” ……… 0 times

Next, the accent combination determination process of S106 is performed for the morpheme string “Senshui | Reconstruction | Special | Committee”.
In this process, first, among the above three verification phrases, the verification phrase “Special Committee for Reconstruction of the House of Representatives” whose count value by the example search process in S105 is 104 times is the verification target verification phrase described above. Attention.

  Then, the ratio of the number of occurrences of the verification phrase “Senior Reconstruction Special Committee” and each of the other verification phrases “Special Committee for Reconstruction of the Lower House” and “Special Committee for Reconstruction of the Lower House” It is determined whether or not it is larger than the threshold value. However, here, the count values obtained by the example search processing of the other verification phrases “Special Committee for Reconstruction of the Lower House” and “Special Committee for Reconstruction of the Lower House” are both zero. Therefore, it is determined that the ratio of the number of appearances described above with respect to the other verification phrases for the determination target verification phrase is larger than a predetermined determination threshold value.

  Then, the insertion position of the character “NO” in the verification target verification phrase “Senshuin Reconstruction Special Committee” is noticed, and this insertion position is the last of the morpheme sequence “Senshui | Reconstruction | Special | Committee”. It is determined whether it is a boundary. In this example, the insertion position of the character “no” is not the last boundary in the morpheme sequence “Shrine House | Reconstruction | Special | Committee”, so the morphemes “Shrine House” and “ It is determined that the accent coupling phenomenon does not occur between “Reconstruction” and “Reconstruction”.

  In the following description, a symbol “/” is attached to the boundary of a morpheme that is determined not to cause an accent coupling phenomenon in a morpheme string, and two morphemes adjacent to the symbol are non-coupled. Try to express that. Therefore, the morpheme sequence “Senshui | Reconstruction | Special | Commission”, in which it was determined that the accent coupling phenomenon does not occur between “Senshuin” and “Reconstruction” by the above processing, | Special | Committee].

  Corresponding to the expression of the morpheme sequence with the symbol “/” described above, the boundaries of the morphemes where it is determined that the accent coupling phenomenon occurs in the morpheme sequence are the symbols “/” and “|”. A morpheme string is written without both.

  Next, if the morpheme string “Senior / Reconstruction | Special | Committee” is subjected to the determination process of S107, there is still a boundary that has not yet been determined whether or not the accent coupling phenomenon occurs. Is made. Therefore, in this case, the morpheme sequence “Senior / Reconstruction | Special | Committee” is divided at the boundary of non-bonding to obtain the morpheme “Senior House” and the morpheme sequence “Reconstruction | Special | Committee”. Of these, the morpheme “Shionin” has no morpheme boundary that is subject to the determination of whether or not an accent coupling phenomenon has occurred. Therefore, the verification phrase generation process of S104 is performed next for the other morpheme string “reconstruction | special | committee”. Through this process, the characters “no” are inserted between “reconstruction” and “special” and between “special” and “committee”, respectively, and the verification phrases “special committee for reconstruction” and “reconstruction” A special committee '' is generated.

Next, the example search process of S105 is performed for the two verification phrases obtained as described above. Here, it is assumed that the count results of the example search process for the number of appearances of the two verification phrases in the sentence corpus 22 are as follows.
・ "Reconstruction Special Committee" ......... 110 times ・ "Reconstruction Special Committee" ......... 0 times

Next, the accent combination determination process of S106 is performed for the morpheme string “reconstruction | special | committee”.
In this process, first, of the above two verification phrases, the verification phrase “Special Committee for Reconstruction” whose count value by the example search process in S105 is 110 times is the attention as the above-described determination target verification phrase. Is done.

  Then, it is determined whether the ratio of the number of appearances of the determination target verification phrase “reconstruction special committee” and the other verification phrase “reconstruction special committee” is greater than a predetermined determination threshold. However, here, the count value by the example search process of the other verification phrase “reconstruction special committee” is zero. Therefore, it is determined that the ratio of the number of appearances described above with respect to the other verification phrases for the determination target verification phrase is larger than a predetermined determination threshold.

  Then, the insertion position of the character “no” in the verification phrase “reconstruction special committee” is noticed, and this insertion position is the last boundary in the morpheme sequence “reconstruction | special | committee” It is determined whether or not. In this example, the insertion position of the character “no” is not the last boundary in the morpheme sequence “Reconstruction | Special | Committee”, so the morphemes “Reconstruction” and “Special” in the Japanese text “Reunion Special Committee” Is determined that no accent coupling phenomenon occurs. Therefore, the morphological sequence “Reconstruction | Special | Committee” becomes “Reconstruction / Special | Committee”.

  By the way, the morpheme sequence “Reconstruction | Special | Committee” has only two borders between adjacent morphemes, and one of them is the border between the morpheme “Reconstruction” and “Special”. Has been determined that the accent combining phenomenon does not occur by the above-described processing. Therefore, in this case, it is judged that the accent coupling phenomenon occurs at the position of the boundary between the morpheme “special” and “committee”, which is the other boundary in the morpheme sequence “reconstruction | special | committee”. Is done.

  As a result of the processing described so far, the final result of the accent combination determination process for the morpheme string “Senior House | Reconstruction | Special | Committee” is obtained as “Senior House / Reconstruction / Special Committee”. The determination result of the determination process of S107 performed after this is No, and the result output process of S108 is performed.

In the result output process, first, the process of acquiring the accents for each of the morphemes “Shrine”, “Reconstruction”, “Special”, and “Committee” from the morpheme table of the word dictionary 21 is performed. Here, it is assumed that the accent acquisition results for each of these morphemes are as follows.
・ "Shrine" ......... "Shoe-in"
・ "Reconstruction" ......... "Fucco"
・ "Special" ......... "Tokubetsu"
・ "Committee" …… "Lee Inkai"

  Note that the “committee” is a compound word that is treated as one morpheme by the morpheme correction processing in S103, but the morpheme table also includes information on such a compound word treated as one morpheme. It is assumed that it is prepared in advance.

  Here, as a final result of the above-described accent combination determination process, it is determined that an accent combination phenomenon occurs at the boundary between the morphemes “special” and “committee”. Therefore, processing for changing the individual accents of the morphemes “special” and “committee” according to the accent combining rule 23 is performed as follows.

  First, the morpheme “special” is noted. This morpheme “special” is positioned “forward” in the compound word “special committee” with the morpheme “committee”. Further, it is found from the information of the morpheme table of the word dictionary 21 that the part of speech of the morpheme “committee” to which the morpheme “special” is combined is not at least “suffix”. Therefore, a record in which “headword” in the table of the accent combination rule 23 in FIG. 4 is “special”, “position” is “forward”, and “part of speech to be combined” is “other” is referred to. . Then, it can be seen that “Yes” is stored in this record, “No change” is stored in “Deformation type”, and “No” is stored in “Exception”. Therefore, the morpheme “special” is left unchanged from the accent “Tokubetsu” registered in the morpheme table.

  Next, the morpheme “Committee” attracts attention. This morpheme “committee” is located “backward” in the compound word “special committee” with the morpheme “special”. Therefore, a record in which “headword” in the table of the accent combination rule 23 in FIG. 4 is “special” and “position” is “backward” is referred to. Since the “joint predecessor” of this record is “arbitrary”, the process of checking the part of speech of the morpheme “committee” is not necessary. It can be seen that this record stores “Yes”, “No change” for “Deformation type”, and “No” for “Exception”. Therefore, the morpheme “committee” is also left undeformed from the accent “I'inkai” registered in the morpheme table.

  It should be noted that the “combination presence / absence” in the reference record for the morpheme “special” and the reference record for the morpheme “committee” are both “Yes” and are not in conflict. Therefore, as the accent information of the compound word “Special Committee”, the unmodified accent “Tokubetsu” for the morpheme “Special” and the unmodified accent “I'Inkai” for the morpheme “Committee” were combined. “Tokubetsui Inkai” is created.

  As a result of the above processing, the accent information “Shoein Fukuko Tokubei 'Inkai” of the Japanese text “Special Committee for Reconstruction of the House of Representatives” is created and output.

[2] Case of “Recruitment Information Posting” Next, a case where the Japanese text “Recruitment Information Posting” is input to the accent assignment device 10 will be described.

  When the Japanese text “recruitment information posting” is acquired by the text input processing of S101, next, morphological analysis processing is performed in S102. It is assumed that an analysis result of morpheme string “adoption | information | publishing” is obtained by this processing.

Next, although the morpheme correction process of S103 is performed, the morpheme string is not corrected for the morpheme string “adopted | information | published”.
Next, the verification phrase generation process of S104 is performed for the morpheme string “adopt | information | post”. Through this process, the characters “no” are inserted between “recruitment” and “information” and between “information” and “publish”, respectively, and the verification phrases “recruitment information publication” and “recruitment information” are inserted. ”Is generated.

Next, the example search process of S105 is performed for the two verification phrases obtained as described above. Here, it is assumed that the count results of the example search process for the number of appearances of the two verification phrases in the sentence corpus 22 are as follows.
・ "Recruitment information posting" ......... 26 times ・ "Recruitment information posting" ......... 283 times

Next, the accent combination determination process of S106 is performed for the morpheme string “adopt | information | post”.
In this process, first, of the above two verification phrases, the verification phrase “post employment information” having a maximum count value of 283 times in the example search process of S105 is noted as the above-described determination target verification phrase. The

  Then, it is determined whether or not the ratio of the number of appearances of the determination target verification phrase “recruitment information posting” and the other verification phrase “recruitment information posting” is greater than a predetermined determination threshold. Here, if the determination threshold is set to 3, for example, the ratio of the number of appearances of the determination target verification phrase and the other verification phrases is 283/26, that is, about 11, and thus is larger than the determination threshold. A decision is made.

  Then, in this case, the insertion position of the character “no” in the verification target verification phrase “posting employment information” is noticed, and this insertion position is the last in the morpheme string “recruitment | information | posting”. It is determined whether it is a boundary. In this example, the insertion position of the character “NO” is the last boundary in the morpheme string “adopt | information | published”. Accordingly, whether or not an accent coupling phenomenon occurs at the last boundary is determined according to the accent coupling rule 23 for two morphemes adjacent to the last boundary.

  First, the morpheme “information” is noted. This morpheme “information” is positioned “forward” in the compound word “information publication” with the morpheme “publication”. Further, it is found from the information in the morpheme table of the word dictionary 21 that the part of speech of the morpheme “published” that is the combination destination of the morpheme “information” is not “suffix”. Therefore, in the table of the accent combination rule 23 in FIG. 4, the “influence” of the record whose “entry word” is “information”, “position” is “forward”, and “participant part-of-speech” is “other”. And “bonded”. Then, it can be seen that “small” is stored in the “influence”, and “present” is stored in the “combination presence / absence” in this record.

  Next, the morpheme “publishing” attracts attention. This morpheme “published” is located “backward” in the compound word “information published” with the morpheme “information”. Therefore, a record in which “headword” in the table of the accent combination rule 23 in FIG. 4 is “published” and “position” is “backward” is referred to. Note that since the “combined antecedent” of this record is “arbitrary”, the process of examining the part of speech of the morpheme “information” is unnecessary. Therefore, it can be seen that “Influence” is “Large”, and “Combination presence / absence” is “None”. In other words, in this case, the information on “combination presence / absence” is contradictory between the two morphemes constituting the composite word “information posting”. Therefore, in this case, based on the data of “combination presence / absence” for the morpheme “publishing” whose “influence” is “large”, it is judged that the accent combining phenomenon does not occur in the composite word “information posting”. Is done. Therefore, the morpheme string “adoption | information | publication” becomes “adoption | information / publication”.

  By the way, the morpheme string “adopt | information | publishing” has only two borders between adjacent morphemes, and at the position of the border between the morpheme “information” and “publishing” which is one of them. Therefore, it is determined that the accent coupling phenomenon does not occur by the above-described processing. Therefore, in this case, it is determined that an accent coupling phenomenon occurs at the position of the boundary between the morpheme “adopt” and “information”, which is the other boundary in the morpheme string “adopt | information | publication”. .

  As a result of the above processing, “adopting information / publishing” is obtained as the final result of the accent combination determination processing for the morpheme string “adopting | information | posting”. The determination result of the determination process of S107 performed after this is No, and the result output process of S108 is performed.

In the result output process, first, a process of acquiring accents for each of the morphemes “adopted”, “information”, and “published” from the morpheme table of the word dictionary 21 is performed. Here, it is assumed that the accent acquisition results for each of these morphemes are as follows.
・ “Recruitment” ……… “Saiyo”
・ "Information" ......... "Joe Ho"
・ "Publishing" ......... "Keisai"

  Here, as a final result of the above-described accent combination determination process, it is determined that an accent combination phenomenon occurs at the boundary between the morphemes “adopted” and “information”. Therefore, processing for changing the individual accents of the morphemes “adopt” and “information” in accordance with the accent combining rule 23 is performed as follows.

  First, the morpheme “adoption” is noted. This morpheme “recruitment” is positioned “forward” in the compound word “recruitment information” with the morpheme “information”. Further, it is found from the information in the morpheme table of the word dictionary 21 that the part of speech of the morpheme “information” that is the combination destination of the morpheme “adopt” is not “suffix”. Therefore, a record in which “headword” is “adopted”, “position” is “front”, and “participant part-of-speech” is “other” in the table of the accent combining rule 23 in FIG. . Then, it can be seen that “Yes” is stored in this record, “No change” is stored in “Deformation type”, and “No” is stored in “Exception”. Therefore, the morpheme “adopted” is left unchanged from the accent “saiyo” registered in the morpheme table.

  Next, the morpheme “information” is noted. This morpheme “information” is positioned “backward” in the compound word “recruitment information” with the morpheme “recruitment”. Therefore, a record in which “headword” in the table of the accent combination rule 23 in FIG. 4 is “information” and “position” is “backward” is referred to. Since the “joint predecessor” in this record is “arbitrary”, it is not necessary to check the part of speech of the morpheme “adopted”. In this record, “Yes / No” is stored for “Availability”, “Type 1” is stored for “Deformation type”, and “None” is stored for “Exception”. Therefore, in the morpheme “committee”, an accent is given to the first mora for the accent “Joe Ho” registered in the morpheme table, and the accent information is changed to “Joho Ho”.

  Note that the “combination presence / absence” in the reference record for the morpheme “adopted” and the reference record for the morpheme “information” are both “Yes” and are not in conflict. Therefore, as the accent information of the composite word “recruitment information”, the unmodified accent “Saiyo” for the morpheme “recruitment” and the accent “Jojoho” changed to the type 1 for the morpheme “information” are combined. "Saiyo Jo Ho" is created.

  As a result of the above processing, the accent information “Saiyo Jo'-Hockey” of the Japanese text “Recruitment Information” is created and output.

[3] Case of “Reduction of Parliamentary member constant” Next, a case where the Japanese text “Reduced member of the congressional member” is input to the accent assignment apparatus 10 will be described.

  When the Japanese text “representative constant reduction” is acquired by the text input process of S101, next, a morphological analysis process is performed in S102. It is assumed that an analysis result of the morpheme string “Members | constant | reduction” is obtained by this processing.

Next, although the morpheme correction process of S103 is performed, the morpheme string is not corrected for the morpheme string “Members | constant | reduction”.
Next, the verification phrase generation process of S104 is performed for the morpheme string “Meeting member | constant | reduction”. By this process, the characters “no” are inserted between “Members” and “Constant”, and between “Constant” and “Reduction”, respectively, and the verification phrases “Members constant reduction” and “Members constant” are inserted. Reduction ”is generated.

Next, the example search process of S105 is performed for the two verification phrases obtained as described above. Here, it is assumed that the count results of the example search process for the number of appearances of the two verification phrases in the sentence corpus 22 are as follows.
・ “Reduction of the number of members of the Diet” ……… 497 times ・ “Reduction of the number of members of the Diet” ……… 398 times

Next, the accent combination determination process of S106 is performed on the morpheme string “Meeting member | constant | reduction”.
In this process, first, among the above two verification phrases, the verification phrase “representative constant reduction” having a maximum value of 497 times in the example search process of S105 is noted as the above-described determination target verification phrase. The

  Then, it is determined whether or not the ratio of the number of appearances of the determination target verification phrase “representative constant reduction” and the other verification phrase “representative constant reduction” is greater than a predetermined determination threshold. Here, if the determination threshold value is set to 3, for example, the ratio of the number of appearances of the determination target verification phrase and other verification phrases is 497/398, that is, about 1.2, and therefore is smaller than the determination threshold value. Is determined.

  Then, in this case, whether or not an accent coupling phenomenon occurs at the boundary between the morpheme “constant” and “reduction”, which is the last boundary in the morpheme sequence “legal member | constant | reduction”, is the morpheme “constant”. ] And “reduction” are determined according to the accent combination rule 23. This determination is performed in the same manner as in the example described above.

  The accent combining rule 23 for the morpheme “constant” and “reduction” is not specified in the table of FIG. Therefore, here, it is assumed that the determination that the accent combining phenomenon does not occur in the composite word “constant reduction” is made based on the accent combining rule 23.

  By the way, the morpheme string “Members | constant | reduction” has only two borders between adjacent morphemes, and at the boundary between the morpheme “constant” and “reduction”. Therefore, it is determined that the accent coupling phenomenon does not occur by the above-described processing. Therefore, in this case, it is determined that an accent coupling phenomenon occurs at the position of the boundary between the morpheme “Meetors” and “Constant”, which is the other boundary in the morpheme string “Members | constant | reduction”. .

  As a result of the above processing, the “decision member constant / reduction” is obtained as the final result of the accent combination determination process for the morpheme string “representative | constant | reduction”. The determination result of the determination process of S107 performed after this is No, and the result output process of S108 is performed.

In the result output process, first, the process of acquiring the accents for each of the morphemes “Meetors”, “Constant”, and “Reduction” from the morpheme table of the word dictionary 21 is performed. Here, it is assumed that the accent acquisition results for each of these morphemes are as follows.
・ "Representative" ......... "Gui-in"
・ "Constant" ......... "Teth"
・ "Reduction" ... "Sakugen"

  Here, as a final result of the above-described accent combination determination process, it is determined that an accent combination phenomenon occurs at the boundary between the morpheme “Meetors” and “Constants”. Therefore, the process of changing the individual accents of the morpheme “Meetors” and “Constants” according to the accent combining rule 23 is performed in the same manner as in the above-described example.

  The accent combining rule 23 for the morphemes “Meetors” and “Constants” is not specified in the table of FIG. Therefore, here, the accent of the morpheme “constant” is changed to “Guin” and the accent of the morpheme “constant” is changed to “tein” based on the “unknown word” rule representing the general rule of the accent combination rule 23. It is assumed that it has been changed to 'Susu'. Therefore, as the accent information for the compound word “Meeting member constant”, “Guinte” is a combination of the accented “Guin” after the change of the morpheme “Meeting member” and the accented “Te'su” after the change of the morpheme “Constant”. Is created.

  As a result of the above processing, the accent information “Guinte 'Sue Sakugen” of the Japanese text “Reduction of the Diet member constant” is created and output.

[4] Case of “International Football Federation” Next, the case where the Japanese text “International Football Federation” is input to the accenting device 10 will be described.

  When the Japanese text “International Soccer Federation” is acquired by the text input process of S101, next, a morphological analysis process is performed in S102. It is assumed that the analysis result of the morpheme string “international | soccer | federation” is obtained by this processing.

Next, although the morpheme correction process of S103 is performed, the morpheme string is not corrected for the morpheme string “international | soccer | federation”.
Next, the verification phrase generation process of S104 is performed on the morpheme string “international | soccer | federation”. By this process, the characters “no” are inserted between “international” and “soccer” and between “soccer” and “federation”, respectively, and the verification phrases “international soccer federation” and “international soccer” The Federation of ”is generated.

Next, the example search process of S105 is performed for the two verification phrases obtained as described above. Here, it is assumed that the count results of the example search process for the number of appearances of the two verification phrases in the sentence corpus 22 are as follows.
・ “International Football Federation” ……… 0 times ・ “International Football Federation” ……… 0 times

  Next, the accent combination determination process of S106 is performed on the morpheme string “international | soccer | federation”. Here, the number of appearances in the sentence corpus 22 for each of the verification phrases generated by the verification phrase generation process is all zero. Therefore, in this case, all boundaries of morphemes in “international | soccer | federation” in the morpheme sequence, ie, between “international” and “soccer”, and between “soccer” and “federation” In both cases, it is determined that an accent coupling phenomenon occurs.

  As a result, the final result of the accent combination determination process for the morpheme string “international | soccer | federation” is the “international soccer federation” in which there is no non-bond between the morphemes. The determination result of the determination process of S107 performed after this is No, and the result output process of S108 is performed.

In the result output process, first, a process of acquiring accents for each of the morphemes “international”, “soccer”, and “federation” from the morpheme table of the word dictionary 21 is performed. Here, it is assumed that the accent acquisition results for each of these morphemes are as follows.
・ "International" ......... "Kokusai"
・ "Soccer" ......... "Sucker"
・ "Alliance" ......... "Renme"

  Here, as a final result of the above-described accent combination determination process, it is determined that an accent combination phenomenon occurs at both the boundary between the morpheme “international” and “soccer” and the boundary between the morpheme “soccer” and “federation”. ing. Therefore, the process of changing the individual accents of the morphemes “international”, “soccer”, and “federation” in order from the previous morpheme to the subsequent morpheme according to the accent combination rule 23 is performed in the same manner as in the above-described example.

  The accent combining rules 23 for the morphemes “international”, “soccer”, and “federation” are not specified in the table of FIG. Therefore, here, based on the accent combining rule 23, the accent of the morpheme “international” is not changed to “Kokusai”, the morpheme “soccer” is changed to “sucker”, and the combined accent is “ It becomes "Kokusaisa 'Ker". Furthermore, in the connection with the “Federation”, the “Kokusaisa-Ker” is flattened to become “Kokusai Soccer”, and the accent of “Renme” is changed to “Lemme”. Therefore, “Kokusai soccer label” is created as the accent information of the composite word “International Soccer Federation”.

  As a result of the above processing, the accent information “Kokusai soccer label” of the Japanese text “International Soccer Federation” is created and output.

  As described above, the accent assigning apparatus 10 in FIG. 1 estimates Japanese language texts composed of a plurality of morphemes by estimating the meaning breaks between the morphemes constituting the Japanese text using the example of the sentence corpus 22. Into a compound word consisting of two morphemes. In this way, the accent assigning apparatus 10 enables an appropriate analysis of the accent combining phenomenon even in Japanese text consisting of three or more words.

In addition, the following additional remarks are disclosed regarding the embodiment described above.
(Appendix 1)
A text input unit that accepts Japanese text input;
A morpheme analyzer that performs morpheme analysis on the accepted Japanese text and outputs a morpheme sequence divided into a plurality of morphemes;
A verification phrase generation unit that generates a plurality of types of verification phrases for the morpheme sequence by inserting the character “no” into any one of the adjacent morpheme boundaries in the morpheme sequence;
An example search unit that counts the number of times the generated verification phrase appears in a sentence corpus;
Accent combination in which the accent of a synthesized word obtained by combining two morphemes adjacent to the insertion position of the character “NO” by the verification phrase generation unit in the morpheme string changes from that of each of the two morphemes An accent combination determination unit that determines whether or not a phenomenon occurs based on the number of appearances for each of the plurality of types of verification phrases;
The accent information for each of the plurality of morphemes constituting the morpheme string is changed based on a determination result in the accent combination determination unit, and an accent for the accepted Japanese text is created. A result output section to output;
An accent applying apparatus comprising:
(Appendix 2)
The accent combination determination unit is configured so that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. And the verification phrase having the maximum number of appearances is generated by inserting the character “NO” at the last boundary among the boundaries in the morpheme string. If there is no accent, the accent applying apparatus according to appendix 1, wherein it is determined that an accent combining phenomenon does not occur at the insertion position of the character “NO” in the verification phrase having the maximum number of appearances.
(Appendix 3)
The accent combination determination unit is configured so that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. And the verification phrase having the maximum number of appearances is generated by inserting the character “NO” at the last boundary among the boundaries in the morpheme string. In some cases, whether or not an accent combining phenomenon occurs at the position of the rearmost boundary is determined according to a rule for combining accents for two morphemes adjacent to the rearmost boundary. 3. The accent imparting apparatus according to appendix 1 or 2, characterized in that determination is made.
(Appendix 4)
The accent combination determination unit is configured such that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. If it is smaller than the determination threshold, whether or not an accent coupling phenomenon occurs at the position of the rearmost boundary among the boundaries in the morpheme string is adjacent to the rearmost boundary prepared in advance. The accent imparting apparatus according to any one of appendices 1 to 3, wherein the determination is performed according to a rule for combining accents for two morphemes.
(Appendix 5)
The accent combination determination unit determines that an accent combination phenomenon occurs in all of the boundary positions in the morpheme sequence when the number of appearances for each of the plurality of types of verification phrases is all zero. The accent determination apparatus according to any one of supplementary notes 1 to 4, characterized in that:
(Appendix 6)
The accent combination determination unit is a case where the boundary in the morpheme sequence is only two and when it is determined that an accent combination phenomenon does not occur at one of the boundaries, 6. The accent applying apparatus according to any one of appendices 1 to 5, wherein it is determined that an accent coupling phenomenon occurs at the other position.
(Appendix 7)
If the boundary in the morpheme sequence is left undecided as to whether an accent coupling phenomenon occurs at the position of the boundary, the verification phrase generation unit A plurality of types of verification phrases are newly generated for the morpheme string by inserting the character “no” into any one of the above,
The example search unit counts the number of appearances for the newly generated verification phrase,
The accent coupling determination unit determines whether an accent coupling phenomenon occurs at the remaining boundary position based on the number of appearances for the newly generated verification phrase.
The accent imparting device according to any one of appendices 1 to 6, characterized in that:
(Appendix 8)
When the verification phrase generation unit newly generates the plurality of types of verification phrases, the verification phrase generation unit divides the morpheme string into two at a boundary position where it is already determined that the accent combination phenomenon does not occur in the morpheme string. The accent assignment apparatus according to appendix 7, wherein the verification phrase is newly generated for each of the two divided morpheme strings.
(Appendix 9)
Among the plurality of morphemes constituting the morpheme sequence output by the morpheme analysis unit, if the morpheme used in combination with other morphemes is not included in a single lexical meaning, A morpheme correction unit that corrects the morpheme string by treating a composite word obtained by combining the morpheme and the other morpheme as one morpheme,
The verification phrase generation unit generates the verification phrase for a morpheme sequence corrected by the morpheme correction unit.
The accent imparting apparatus according to any one of appendices 1 to 8, characterized in that:
(Appendix 10)
When the morpheme that is an affix is included in a plurality of morphemes that constitute the morpheme sequence output by the morpheme analysis unit, the morpheme correction unit includes the morpheme that is the affix and the morpheme The accent assignment apparatus according to appendix 9, wherein the morpheme string is corrected by treating a composite word obtained by combining a morpheme that is a stem of the affix as a single morpheme.
(Appendix 11)
Perform morpheme analysis on the input Japanese text to generate a morpheme string divided into multiple morphemes,
Generating a plurality of types of verification phrases for the generated morpheme sequence by inserting the character “no” into any one of the adjacent morpheme boundaries in the generated morpheme sequence;
Counting the number of times the generated multiple types of verification phrases appear in the sentence corpus,
An accent of a composite word obtained by combining two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when the plural types of verification phrases are generated is the two Whether or not an accent coupling phenomenon that changes from that of each morpheme is determined based on the number of appearances for each of the plurality of types of verification phrases,
The accent information for each of the plurality of morphemes constituting the generated morpheme sequence is changed based on the determination result of whether or not the accent coupling phenomenon occurs, and the input Japanese text Create and output accents for
An accenting method characterized by that.
(Appendix 12)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. In any of the phrases, the verification phrase that is greater than the predetermined determination threshold and has the maximum number of appearances is the character “NO” at the rearmost boundary among the boundaries in the generated morpheme sequence. If it is not generated by inserting the character, it is determined that the accent combining phenomenon does not occur at the insertion position of the character “NO” in the verification phrase having the maximum number of appearances. The accent imparting method described in 1.
(Appendix 13)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. In any of the phrases, the verification phrase that is greater than the predetermined determination threshold and has the maximum number of appearances is the character “NO” at the rearmost boundary among the boundaries in the generated morpheme sequence. In the case where the accent combining phenomenon occurs at the position of the rearmost boundary, whether two adjacent to the rearmost boundary are prepared. 13. The method of adding accents according to appendix 11 or 12, wherein the determination is made according to the rules for combining accents for morphemes.
(Appendix 14)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. If any of the phrases is smaller than a predetermined determination threshold, whether or not an accent coupling phenomenon occurs at the position of the rearmost boundary among the boundaries in the generated morpheme sequence is prepared in advance. 14. The accent assignment method according to any one of appendices 11 to 13, wherein the determination is performed according to a rule for combining accents for two morphemes adjacent to the rearmost boundary.
(Appendix 15)
Among the plurality of morphemes constituting the generated morpheme sequence, if the morpheme used in combination with other morphemes is not included in the lexical meaning alone, the morpheme and the other morphemes are included. The generated morpheme sequence is corrected by treating a composite word obtained by combining with a morpheme as a single morpheme,
In the generation of the plurality of types of verification phrases, the plurality of types of verification phrases are generated for the modified morpheme sequence.
The accent imparting method according to any one of appendices 11 to 14, characterized in that:
(Appendix 16)
Perform morpheme analysis on the input Japanese text to generate a morpheme string divided into multiple morphemes,
Generating a plurality of types of verification phrases for the generated morpheme sequence by inserting the character “no” into any one of the adjacent morpheme boundaries in the generated morpheme sequence;
Counting the number of times the generated multiple types of verification phrases appear in the sentence corpus,
An accent of a composite word obtained by combining two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when the plural types of verification phrases are generated is the two Whether or not an accent coupling phenomenon that changes from that of each morpheme is determined based on the number of appearances for each of the plurality of types of verification phrases,
The accent information for each of the plurality of morphemes constituting the generated morpheme sequence is changed based on the determination result of whether or not the accent coupling phenomenon occurs, and the input Japanese text Create and output accents for
A program that causes a computer to execute processing.
(Appendix 17)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. In any of the phrases, the verification phrase that is greater than the predetermined determination threshold and has the maximum number of appearances is the character “NO” at the rearmost boundary among the boundaries in the generated morpheme sequence. If it is not generated by inserting a character, it is determined that an accent combining phenomenon does not occur at the insertion position of the character “NO” in the verification phrase having the maximum number of appearances. The program described in.
(Appendix 18)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. In any of the phrases, the verification phrase that is greater than the predetermined determination threshold and has the maximum number of appearances is the character “NO” at the rearmost boundary among the boundaries in the generated morpheme sequence. In the case where the accent combining phenomenon occurs at the position of the rearmost boundary, whether two adjacent to the rearmost boundary are prepared. 18. The program according to appendix 16 or 17, characterized in that the determination is made according to an accent combining rule for morphemes.
(Appendix 19)
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. If any of the phrases is smaller than a predetermined determination threshold, whether or not an accent coupling phenomenon occurs at the position of the rearmost boundary among the boundaries in the generated morpheme sequence is prepared in advance. 19. The program according to any one of appendices 16 to 18, wherein the program is determined according to a rule of accent combination for two morphemes adjacent to the rearmost boundary.
(Appendix 20)
Among the plurality of morphemes constituting the generated morpheme sequence, if the morpheme used in combination with other morphemes is not included in the lexical meaning alone, the morpheme and the other morphemes are included. The computer further executes a process for correcting the generated morpheme sequence by treating a composite word obtained by combining the morpheme of
In the generation of the plurality of types of verification phrases, the plurality of types of verification phrases are generated for the modified morpheme sequence.
20. The program according to any one of supplementary notes 16 to 19, characterized by:

DESCRIPTION OF SYMBOLS 10 Accent assignment apparatus 11 Text input part 12 Morphological analysis part 13 Morphological correction part 14 Verification phrase generation part 15 Example search part 16 Accent combination determination part 17 Result output part 21 Word dictionary 22 Sentence corpus 23 Accent combination rule 30 Computer 31 MPU
32 ROM
33 RAM
34 hard disk device 35 input device 36 output device 37 interface device 38 recording medium drive device 39 bus line 40 portable recording medium

Claims (11)

A text input unit that accepts Japanese text input;
A morpheme analyzer that performs morpheme analysis on the accepted Japanese text and outputs a morpheme sequence divided into a plurality of morphemes;
A verification phrase generation unit that generates a plurality of types of verification phrases for the morpheme sequence by inserting the character “no” into any one of the adjacent morpheme boundaries in the morpheme sequence;
An example search unit that counts the number of times the generated verification phrase appears in a sentence corpus;
Accent combination in which the accent of a synthesized word obtained by combining two morphemes adjacent to the insertion position of the character “NO” by the verification phrase generation unit in the morpheme string changes from that of each of the two morphemes An accent combination determination unit that determines whether or not a phenomenon occurs based on the number of appearances for each of the plurality of types of verification phrases;
A morpheme table showing accent information for each morpheme indicating accent information when the morpheme is pronounced alone, and an accent generated in the morpheme constituting the composite word when the accent combination phenomenon occurs in the composite word A storage unit in which an accent combination rule table indicating information indicating a mode of change for each morpheme is stored;
Accent information about each of the plurality of morphemes constituting the morpheme string is acquired from the morpheme table , and the accent combination determination unit of the acquired accent information is determined to cause the accent combination phenomenon. The accent information about the morpheme constituting the compound word is changed according to the information indicating the aspect of change of the accent for the morpheme shown in the combination rule table, and the received Japanese text A result output section for creating and outputting an accent;
Equipped with a,
The accent combination determination unit is configured so that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. And the verification phrase having the maximum number of appearances is generated by inserting the character “NO” at the last boundary among the boundaries in the morpheme string. If not, a determination is made that the accent combining phenomenon does not occur at the insertion position of the character “no” in the verification phrase with the maximum number of appearances.
An accenting apparatus characterized by that. The accent combination determination unit is configured so that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. And the verification phrase having the maximum number of appearances is generated by inserting the character “NO” at the last boundary among the boundaries in the morpheme string. In some cases, whether or not an accent combining phenomenon occurs at the position of the rearmost boundary is determined according to a rule for combining accents for two morphemes adjacent to the rearmost boundary. The accent applying apparatus according to claim 1 , wherein the accent applying apparatus is determined. The accent combination determination unit is configured such that a ratio of the number of appearances with respect to other verification phrases for a verification phrase having the maximum number of appearances among the plurality of types of verification phrases is predetermined in any of the other verification phrases. If it is smaller than the determination threshold, whether or not an accent coupling phenomenon occurs at the position of the rearmost boundary among the boundaries in the morpheme string is adjacent to the rearmost boundary prepared in advance. accentuation device according to claim 1 or 2, wherein the determining in accordance with two binding rules accents for morphemes are. The accent combination determination unit determines that an accent combination phenomenon occurs in all of the boundary positions in the morpheme sequence when the number of appearances for each of the plurality of types of verification phrases is all zero. The accent determination apparatus according to any one of claims 1 to 3 , wherein The accent combination determination unit is a case where the boundary in the morpheme sequence is only two and when it is determined that an accent combination phenomenon does not occur at one of the boundaries, accentuation device according to claim 1, characterized in that to make the determination that the accent coupling phenomenon occurs in any one of the four in the other position. If the boundary in the morpheme sequence is left undecided as to whether an accent coupling phenomenon occurs at the position of the boundary, the verification phrase generation unit A plurality of types of verification phrases are newly generated for the morpheme string by inserting the character “no” into any one of the above,
The example search unit counts the number of appearances for the newly generated verification phrase,
The accent coupling determination unit determines whether an accent coupling phenomenon occurs at the remaining boundary position based on the number of appearances for the newly generated verification phrase.
The accent imparting apparatus according to any one of claims 1 to 5 , wherein When the verification phrase generation unit newly generates the plurality of types of verification phrases, the verification phrase generation unit divides the morpheme string into two at a boundary position where it is already determined that the accent combination phenomenon does not occur in the morpheme string. The accent assignment apparatus according to claim 6 , wherein the verification phrase is newly generated for each of the two divided morpheme strings. Among the plurality of morphemes constituting the morpheme sequence output by the morpheme analysis unit, if the morpheme used in combination with other morphemes is not included in a single lexical meaning, A morpheme correction unit that corrects the morpheme string by treating a composite word obtained by combining the morpheme and the other morpheme as one morpheme,
The verification phrase generation unit generates the verification phrase for a morpheme sequence corrected by the morpheme correction unit.
The accent imparting device according to any one of claims 1 to 7 , wherein When the morpheme that is an affix is included in a plurality of morphemes that constitute the morpheme sequence output by the morpheme analysis unit, the morpheme correction unit includes the morpheme that is the affix and the morpheme 9. The accent assigning apparatus according to claim 8 , wherein the morpheme string is corrected by treating a composite word obtained by combining a morpheme which is a stem of the affix as a single morpheme. . An accenting method performed by an accenting apparatus,
The morpheme analysis unit provided in the accent assigning device generates a morpheme sequence divided into a plurality of morphemes by performing morpheme analysis on the input Japanese text,
The generated morpheme string is generated by the verification phrase generator included in the accent assigning device inserting the character “no” into any one of the adjacent morpheme boundaries in the generated morpheme string. Generate multiple verification phrases for
The example search unit provided in the accenting device counts the number of appearances of the generated plurality of types of verification phrases that appear in a sentence corpus,
Two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when the accent combination determination unit provided in the accent assignment device generates the plurality of types of verification phrases Is determined based on the number of appearances for each of the plurality of types of verification phrases, whether or not an accent combining phenomenon occurs in which the accent of the synthesized word obtained by combining the two is different from that of each of the two morphemes,
The result output unit provided in the accent assigning apparatus acquires accent information for each of the plurality of morphemes constituting the generated morpheme string from the morpheme table , and the accent information among the acquired accent information The input is performed by changing the accent information for the morpheme constituting the compound word determined to cause the accent combining phenomenon according to the information indicating the aspect of the accent change for the morpheme shown in the combining rule table. and outputs to create accents for has been Japanese text,
The morpheme table is a table showing accent information representing an accent when a morpheme is singly generated for each morpheme, and is stored in a storage unit provided in the accent assigning device,
The combination rule table is a table that shows information for each morpheme that represents an aspect of an accent change that occurs in a morpheme constituting the compound word when the accent combining phenomenon occurs in the compound word. Stored in the storage unit of the accenting device,
In the determination by the accent combination determination unit, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is any of the other verification phrases. And the verification phrase having the maximum number of appearances is generated by inserting the character “NO” at the last boundary among the boundaries in the morpheme string. If it is not, it is determined that the accent combining phenomenon does not occur at the insertion position of the character “of” in the verification phrase having the maximum number of appearances.
An accenting method characterized by that. Perform morpheme analysis on the input Japanese text to generate a morpheme string divided into multiple morphemes,
Generating a plurality of types of verification phrases for the generated morpheme sequence by inserting the character “no” into any one of the adjacent morpheme boundaries in the generated morpheme sequence;
Counting the number of times the generated multiple types of verification phrases appear in the sentence corpus,
An accent of a composite word obtained by combining two morphemes adjacent to the insertion position in the generated morpheme string for the character “no” inserted when the plural types of verification phrases are generated is the two Whether or not an accent coupling phenomenon that changes from that of each morpheme is determined based on the number of appearances for each of the plurality of types of verification phrases,
Accent information for each of the plurality of morphemes constituting the generated morpheme sequence is acquired from a morpheme table , and a composite word determined to cause the accent combining phenomenon in the acquired accent information is configured The accent information for the morpheme is changed in accordance with the information indicating the mode of change of the accent for the morpheme shown in the combination rule table, and the accent for the input Japanese text is created and output To
Let the computer execute the process ,
The morpheme table is a table showing accent information representing an accent when a morpheme is singly generated for each morpheme, and is stored in a storage unit provided in the computer,
The combination rule table is a table that shows information for each morpheme that represents an aspect of an accent change that occurs in a morpheme constituting the compound word when the accent combining phenomenon occurs in the compound word. Store it in the storage section of the computer,
In determining whether or not the accent coupling phenomenon occurs, the ratio of the number of appearances to the other verification phrases with respect to the verification phrase having the maximum number of appearances among the plurality of types of verification phrases is the other verification. In any of the phrases, the verification phrase having the maximum number of appearances is greater than a predetermined determination threshold, and the character “NO” is inserted at the last boundary among the boundaries in the morpheme string. If it is not generated in the verification phrase that the number of appearances is the maximum, it is determined that the accent combining phenomenon does not occur at the insertion position of the character “of” in the verification phrase.
program.

Images