JP2016218848A - Language expression rewriting apparatus, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve rewriting to wide variety of language expressions.SOLUTION: For each of plural language characteristics containing a literary style, a predicate functional expression and a personal pronoun and expressing characteristics corresponding to a character, a setting section 11 stores setting values related to rewriting of language expressions based on the language characteristics, in a setting value database 22. On the basis of the setting value stored in the setting value database 22, a rewriting processing section 13 constituted of plural processing sections 13a to 13l including a literary style conversion section 13a, a predicate functional expression/character word ending conversion section 13e, a personal pronoun conversion section 13h, applies at least one type of rewriting processing to an input sentence 23, from among language expression rewriting processing based on plural types of language characteristics including: processing of converting a type of a literary style; processing of converting a predicate functional expression according to a character; and processing of converting a personal pronoun according to the character.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a language expression rewriting device, a learning device, a method, and a program, and more particularly, to a language expression rewriting device, method, and program for rewriting the language expression of input text.

  Conventionally, there is a method for characterizing a language expression by rewriting a text only for a function word sequence (end sentence expression) at the end of the sentence. For example, a technique has been proposed in which text data to which an author's attribute is assigned is used to extract sentence end expressions that are used in a biased manner for each attribute value of the author and used for characterizing an utterance (Non-Patent Document 1). ).

Chiaki Miyazaki, Toru Hirano, Ryuichiro Higashinaka, Toshiro Makino, Yoshihiro Matsuo, “Conversion of sentence ending expression to give utterance character”, Artificial Intelligence Society Study Group Material (SIG-SLUD-68), pp. 41-46 , 2013.

  However, in the technique of Non-Patent Document 1, since only the sentence end expression is a target of rewriting, for example, the linguistic features of a character with rich individuality such as characters in TV animation and manga are expressed in various variations. There is a problem that this is not enough.

  The present invention has been made to solve the above problem, and an object of the present invention is to provide a language expression rewriting apparatus, method, and program for realizing rewriting to various language expressions.

  In order to achieve the above object, the language expression rewriting device according to the first invention includes a sentence style, a predicate function expression, and a personal pronoun, and each of a plurality of types of linguistic characteristics that express characteristics according to a character. A setting unit for setting a setting value for rewriting a language expression based on the linguistic feature; a process for converting a style type based on the setting value set by the setting unit; and a predicate function expression according to a character At least one type of rewriting processing based on a plurality of types of linguistic features including processing to convert and morpheme that is a personal pronoun into a personal pronoun corresponding to the character. And a rewrite processing unit to be applied.

  According to the language expression rewriting apparatus according to the first invention, the setting unit includes a sentence style, a predicate function expression, and a personal pronoun, and each of a plurality of types of linguistic characteristics that express characteristics according to the character Set the value related to rewriting linguistic expressions based on specific features. Then, the rewrite processing unit converts the style of the style based on the setting value set by the setting unit, the process of converting the predicate function expression according to the character, and the morpheme that is a personal pronoun according to the character. Of the linguistic expression rewriting processes based on a plurality of types of linguistic features including the process of converting to personal pronouns, at least one type of rewriting process is applied to the input sentence.

  A language expression rewriting method according to a second invention is a language expression rewriting method in a language expression rewriting device including a setting unit and a rewriting processing unit, wherein the setting unit includes a sentence style, a predicate function expression, and a personal name. For each of a plurality of types of linguistic features including pronouns and displaying features according to the character, setting values for rewriting the language expression based on the linguistic features are set, and the rewriting processing unit is set by the setting unit Multiple types including processing to convert stylistic types, processing to convert predicate function expressions according to characters, and processing to convert morphemes that are personal pronouns into personal pronouns according to characters Is a method of applying at least one kind of rewriting process to the input sentence among the rewriting processes of the linguistic expression based on the linguistic features.

  In this way, not only the sentence end expression but also various language expression rewriting processes are arbitrarily combined, so that various variations of language expression can be rewritten.

  In the first and second aspects of the invention, the setting unit further includes a discriminative meaningless expression that can discriminate, but has no meaning, a sentence structure, an inflection form, a slang, a dialect or special vocabulary, a specific phoneme, and a character. The setting value is set for each of the plurality of types of linguistic features including, and the rewrite processing unit converts a sentence structure conversion and a utilization form based on the setting value set by the setting unit Processing, processing to convert to sloppy expression, processing to convert a specific vocabulary to dialect or special vocabulary, processing to convert a specific phoneme to a phoneme according to the character, and insert a discriminative meaningless expression according to the character Of the rewriting processing of the language expression based on the plurality of types of linguistic features further including processing, at least one type of rewriting processing can be applied to the input sentence. As a result, rewriting to more various variations of language expressions can be realized.

  In the first and second aspects of the invention, the setting unit sets the setting value for each of the plurality of types of linguistic features further including character types, segmentation, and symbols, and the rewrite processing unit includes: Language expression rewriting processing based on the plurality of types of linguistic features, further including processing for converting the character type, processing for converting to character writing, and processing for inserting symbols based on the setting value set by the setting unit Among these, at least one type of rewrite processing can be applied to the input sentence. As a result, rewriting to more various variations of language expressions can be realized.

  A language expression rewriting program according to the third invention is a program for causing a computer to function as each part of the language expression rewriting apparatus.

  As described above, according to the language expression rewriting device, method, and program of the present invention, not only the sentence end expression but also various language expression rewriting processes can be arbitrarily combined to implement various variations of language expressions. Rewriting to can be realized.

It is a functional block diagram which shows schematic structure of the language expression rewriting apparatus which concerns on this embodiment. It is a figure which shows an example of the correspondence table | surface of function expression and its meaning. It is a figure which shows an example of the function expression list according to a style. It is a figure which shows an example of an utilization table. It is a figure which shows an example of the replacement rule of morpheme strings. It is a figure which shows an example of the connection expression list classified by character. It is a figure which shows an example of the function expression list according to character. It is a figure which shows an example of a collapse utilization rule. It is a figure which shows an example of the personal pronoun list according to character. It is a figure which shows an example of the vocabulary replacement rule classified by character. It is a figure which shows an example of the phoneme replacement rule classified by character. It is a flowchart which shows an example of the language expression rewriting process routine in this embodiment.

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

<Configuration of language expression rewriting device>
The language expression rewriting device 10 according to the present embodiment may be configured by a computer including a CPU, a RAM, a language expression rewriting program for executing a language expression rewriting processing routine described later, and a ROM storing various data. it can. The language expression rewriting apparatus 10 is functionally represented by a configuration including a setting unit 11, a basic analysis unit 12, and a rewriting processing unit 13, as shown in FIG.

  The language expression rewriting device 10 receives a Japanese input sentence 23 (text data) as an input, and outputs a rewritten sentence 32 in which the language expression included in the input sentence 23 is rewritten in accordance with a designated setting. . In the present embodiment, the following 12 types of linguistic features that are frequently observed in the utterances of characters with strong personalities such as characters in TV animations and comics can be specified in any combination.

Rewrite item (a) Style, (b) Character type, (c) Split writing, (d) Sentence structure, (e) Predicate function expression / character ending, (f) Practical use, (g) Sword stagnation, (h) Personal pronoun , (I) dialect / special vocabulary, (j) phoneme substitution, (k) discriminative meaningless expression, (l) symbols

  The setting unit 11 reads a setting file 21 in which setting values relating to rewriting of the language expression in the rewriting processing unit 13 described later are described. In the setting file 21, for each of the above rewriting items, the following setting values related to rewriting (values for specifying settings, file names, character strings) are described.

(A) Stylistic conversion setting values (0 = no conversion, 1 = normal body, 2 = respected body (is / mass), 3 = respected body)
(B) Setting value of character type conversion (0 = no conversion, 1 = Hiragana conversion)
(C) Setting value for the split-line conversion (0 = no conversion, 1 = reading mark delimiter, 2 = blank delimiter)
(D) File name of sentence connection conversion character-specific expression 26 (details will be described later) (If no file name is specified, no conversion is assumed)
(E) File name of character-specific function expression list 27 (details will be described later) for predicate function expression / character ending conversion (if no file name is specified, it is regarded as no conversion)
(F) Setting value for utilization type conversion (0 = no conversion, 1 = conversion to broken utilization type)
(G) Set value for stagnation conversion (0 = no conversion, 1 = stagnation)
(H) File name of the personal pronoun list 29 for personal pronoun substitution (details will be described later) (if no file name is specified, it is regarded as no conversion)
(I) File name of character-specific vocabulary replacement rule 30 (details will be described later) for dialect / special vocabulary replacement (if no file name is specified, it is regarded as no conversion)
(J) File name of phoneme replacement character-specific phoneme replacement rule 31 (details will be described later) (if no rule file name is specified, it is regarded as no conversion)
(K) Character string indicating the discriminative meaningless expression to be used (if no discriminative meaningless expression is specified, it is regarded as no conversion)
(L) Character string indicating the symbols to be used (If no symbols are specified, it is considered as no conversion)

  The setting unit 11 stores a setting value for each item described in the read setting file 21 in a setting value database (DB) 22.

  The details of each item will be described together with the description of each processing unit of the corresponding rewrite processing unit 13, and the description here will be omitted.

  The basic analysis unit 12 reads the input sentence 23. The input sentence 23 is text data written in Japanese. For example, any textual Japanese text data such as writing to a blog, SNS (Social Networking Service), speech recognition result of an utterance, or text chat can be applied as the input sentence 23 of the present embodiment. .

  The basic analysis unit 12 applies the read input sentence 23 to the dependency analyzer, and from the output, reads the morpheme boundary, the reading of each morpheme, the part of speech of each morpheme, the use type / use form of the usage word, the phrase boundary, and the phrase head Get information. The basic analysis unit 12 does not necessarily have to acquire information on the morpheme boundary of the input sentence 23, the reading of each morpheme, the part of speech of each morpheme, the utilization type / utilization form of the usage word, the phrase boundary, and the phrase headword. There is no need to use a dependency analyzer. For example, the morpheme boundary, reading, part of speech, and utilization type / utilization form of the usage word may be acquired by the morphological analyzer, and the phrase boundary and the phrase head word may be obtained using individual algorithms.

  Further, the basic analysis unit 12 acquires a semantic label of the functional expression included in the input sentence 23 by character string matching using a correspondence table between the functional expression and its meaning as shown in FIG. 2 (reference document 1). reference). In addition, as a technique for acquiring a semantic label of a functional expression, for example, a technique described in Reference 2 for estimating an appropriate label using machine learning may be used.

References 1: Shun Matsuyoshi, Satoshi Sato, Takehito Utsuro, “Editing Japanese Functional Expression Dictionary”, Natural Language Processing, 14.5, 2007
Reference 2: Kenji Imamura, Atsuko Izumi, Genichiro Kikui, Satoshi Sato, “Semantic Label Tagger for Predicate Functional Expression”, Proc. Of the 17th Annual Conference of the Language Processing Society, 2011

  For example, when the basic analysis unit 12 reads the input sentence 23 “I am cold, I chose warm clothes.” (1) Morphological boundaries, phrase boundaries, phrase headwords, and (2) each morpheme , (3) part of speech of each morpheme, utilization type / utilization form of utilization word, and (4) semantic label of function expression. Hereinafter, these pieces of information acquired by the basic analysis unit 12 are collectively referred to as basic analysis results.

(1) Because {I} _ is / {cold} __, __, {warm} _i / {clothes} _ / {select} _n_.
(2) {I am} _ha / {Samgari} _Na_Node_, / {Atata} _I / {Fusou} _wo / {Era} _n_da_.
(3) {pronoun} _involvement particle / {noun} _auxiliary verb_conjunctive particle_reading point / {adjective stem} _utilization ending: collocation / {noun} _case particle / {verb stem: ba line five steps} _utilization Ending: Consecutive form-sound flight _ auxiliary verb _ punctuation point (4) {*} _ * / {*} _ * _ reason _ * / {*} _ * / {*} _ * / {*} _ * _ done_ *

  In the above basic analysis result, the morpheme boundary is indicated by “_”, the phrase boundary is indicated by “/”, and the phrase main word is indicated by “{}”. It should be noted that where a clause is separated, a morpheme is always separated, so the clause boundary is also a morpheme boundary. In addition, the usage endings of the usage words (verbs, adjectives, auxiliary verbs) are separated from the stem (for example, “select_n_”). Furthermore, in (4), “*” means that it is not a function expression, or even if it is a function expression, a semantic label to be processed in this embodiment has not been assigned.

  As shown in FIG. 1, the rewrite processing unit 13 includes a style conversion unit 13a, a character type conversion unit 13b, a segmentation conversion unit 13c, a sentence structure conversion unit 13d, a predicate function expression / character ending conversion unit 13e, and a utilization type conversion unit 13f. , A word conversion unit 13g, a personal pronoun replacement unit 13h, a dialect / special vocabulary replacement unit 13i, a phoneme replacement unit 13j, a discriminative meaningless expression insertion unit 13k, and a symbol class insertion unit 13l.

  The basic analysis results obtained by analyzing the input sentence 23 by the basic analysis unit 12 are input to the respective processing units of the rewrite processing unit 13. However, when the setting value of the item corresponding to each processing unit stored in the setting value DB 22 indicates that the rewriting process of each processing unit is not applied, the basic analysis result is not input to the processing unit. For example, when the setting value of the item (a) style conversion stored in the setting value DB 22 is 0 = no conversion, the basic analysis result is not input to the style conversion section 13a.

  Hereinafter, the rewriting process of each processing unit to which the basic analysis result is input will be described in detail.

  The style conversion unit 13a uses the setting value DB 22 to set the item (a) style conversion setting values (1 = ordinary body, 2 = respected body), 3 = rewarded body ) Get one). The stylistic conversion unit 13a outputs a sentence obtained by converting the stylization by replacing the functional expression of the predicate of the input sentence 23 with the stylistic indicated by the acquired setting value.

  Specifically, the style conversion unit 13a refers to, for example, a style-specific function expression list 24 as shown in FIG. 3, and the function expression (after the main letter) included in the last sentence of the input sentence 23 (basic analysis result). To obtain the representation after replacement of the morpheme string. In the example of FIG. 3, the style-specific function expression list 24 is divided into “sentence parts of phrase headwords” and “style of phrases”, according to style (regular (datai), respect (mass), honor (body)). The “function expression notation” to be replaced is defined for each “function expression meaning”.

  When the functional expression to be replaced is a verb in the functional expression list 24 according to the style, the “functional expression notation” of the replacement destination indicates the usage form of the verb ending required by the functional expression of the replacement destination. It is defined by a notation including a tag to be represented (hereinafter referred to as [utilized form]. For example, [continuous form]). In this case, the style conversion unit 13a refers to, for example, the utilization table 25 illustrated in FIG. 4 and converts the tag [utilization form] portion into the replacement destination notation. In the utilization table 25 shown in FIG. 4, an empty cell indicates that the stem and the subsequent morpheme are connected without sandwiching the utilization ending. For example, the verb form (A-dan) of the verb used in the first stage is directly connected to the stem “see” and the auxiliary verb “no”, and becomes “do not see”. A cell in which “-” is written indicates a combination that does not exist in Japanese (a combination that does not need to be considered).

  The processing of the stylistic conversion unit 13a will be described in more detail by taking the conversion of the stylization to the basic analysis result of the input sentence 23 “I am cold, so I chose warm clothes.”

  The stylistic conversion unit 13a acquires information on the part of speech of the predicates of the predicate = {verb stem: ba line five steps} and the meaning label of the functional expression of the clause = “complete” from the basic analysis result. Then, the style conversion unit 13a refers to the style-specific function expression list 24 shown in FIG. 3 and is associated with “part of speech of phrase main part” and “meaning of function expression” that match the information acquired from the basic analysis result. Acquire "notation of function expression". When the set value acquired from the set value DB 22 is 2 = respected body (3), or 3 = respect body (is the body), the following processing intermediate result is obtained. In the following, the replaced part is represented by << >>.

Example of honored body:
I'm cold, so I chose warm clothes.
Example of honored body:
I am cold, so I chose warm clothes << [continuous use-sound] >>.

  When the clause whose functional expression is replaced includes the tag [utilization form] as in the above processing intermediate result, the stylistic conversion unit 13a uses the verb utilization form indicated by the tag [utilization form] from the utilization table 25. And the usage form of the verb that is the main part of the phrase is acquired. Then, the style conversion unit 13a replaces the tag [utilization form] included in the processing intermediate result with the utilization form notation acquired from the utilization table 25, as shown below.

Example of honored body:
I'm cold, so I chose warm clothes.
⇒I am cold, so I chose warm clothes.
In the phrase “Selected [Continuous]”, the part of speech of the phrase head is “Verb stem: Ba line 5 Dan”, and the usage form indicated by the tag “Utilization form” is “Consecutive form”. The notation “Bi” is acquired.

Example of honored body:
I am cold, so I chose warm clothes << [continuous use-sound] >>. ⇒I am cold, so I choose warm clothes.
The phrase part of speech in the phrase “Select [Continuous Form-Sound Service]” is “Verb stem: Ba line 5 Dan”, and the utilization form indicated by the tag “Utilization Form” is “Continuous Form (Sound Form)”. Since there is, the usage notation “n” that matches this condition is acquired.

  In addition, if the usage type of the phrase main word is any one of “5 row utilization”, “5 row utilization”, “5 row utilization”, and “5 row utilization”, the replacement destination The first character (the character that follows the ending ending) of the function expression “te” or “ta” is replaced with “de” or “da”, respectively. In the above example, because the phrase headword usage type is “Ba row five-stage usage”, “Selected [continuous form-sound service]” is replaced with “I selected it”.

  In the above description, in order to insert an appropriate usage ending, a tag representing the usage form of the verb usage ending required by the functional expression of the replacement destination is used. However, the present invention is not limited to this method. For example, what kind of effective ending should be inserted when two morphemes are combined may be learned in advance by some machine learning technique and used for inserting the effective ending in the style conversion unit 13a. Good. For example, a model for estimating what kind of ending ending should be between “Select” and “Ta” is learned. If the verb whose stem is “Select” is to be replaced with the form of “stem + utilized ending + auxiliary verb“ ta ””, it is possible to obtain the utilized ending at the replacement destination from the output of the learned model.

  As another method, it is also possible to convert a style using a replacement rule between morpheme sequences that considers what morphemes co-occur before and after the morpheme (sequence) to be replaced. In this case, for example, the style can be converted using a replacement rule between morpheme strings as shown in FIG.

  In this embodiment, the possible setting values are 4 types: 0 = no conversion, 1 = normal body, 2 = respected body, and 3 = respected body. However, a setting value for conversion to another style (for example, “normal body”) may be added.

  The character type conversion unit 13b outputs a sentence in which kanji characters are converted into hiragana according to the item (b) character type conversion setting value (1 = Hiragana conversion) stored in the setting value DB 22. Specifically, as shown below, the character type conversion unit 13b replaces all morpheme readings (portions written in katakana) included in the basic analysis result with hiragana characters.

Hiragana example:
I'm awkward, so I chose a warm fusou.

  In this embodiment, there are two types of setting values that can be taken: 0 = no conversion or 1 = Hiragana conversion. However, an option for “katakana” that converts all characters into katakana is prepared. Also good. Also, specify the rate of character type conversion, such as "50% hiragana" (once every two times), "20% (once every five times) Katakana" You may be able to do it. In addition, the part of speech that is to be subjected to character type conversion may be specified, such as hiraganaizing only the morpheme whose part of speech is “noun”.

  The segmentation conversion unit 13c obtains the setting value (1 = reading point delimiter or 2 = blank delimiter) of the item (c) segmentation conversion from the setting value DB 22. The segmentation conversion unit 13c inserts a designated delimiter character at the clause boundary of the input sentence 23, thereby outputting a sentence in which the segmentation is converted as shown below. Note that if a delimiter (a reading or a blank) is inserted to cause consecutive readings or blanks, no delimiter is inserted. In addition, no delimiter is inserted between two consecutive clauses in a linkage modification relationship, such as between “warm” and “clothes”.

Examples of broken line conversion (punctuation break):
I chose <<, >> warm clothes because I was cold.

  The sentence structure conversion unit 13d acquires the file name of the item-by-character connection expression list 26 for sentence structure conversion from the setting value DB 22. Then, the sentence structure conversion unit 13d divides the input sentence 23 into simple sentences, and converts the sentence structure by inserting connection expressions obtained from the character-specific connection expression list 26 indicated by the acquired file name into the divided portions. Output a sentence.

  Specifically, when the input sentence 23 is a compound sentence, the sentence structure conversion unit 13d, for example, in the character-specific connection expression list 26 shown in FIG. 6, the character-specific connection expression list 26 indicated by the acquired file name. Refer to Then, the sentence structure conversion unit 13d searches the input sentence 23 for a phrase having a semantic label described in the character-specific connection expression list 26, deletes the morpheme to which the corresponding semantic label is assigned, and puts a punctuation mark at the deleted position. Insert. In addition, when the morpheme immediately before the morpheme to which the semantic label is assigned is a usage word (a verb, an adjective, or an auxiliary verb), the sentence structure conversion unit 13d converts the usage word into a final form. As the final form, a “basic form” that is often output as a part of the morphological analysis result may be used, or the utilization table 25 may be used. Then, the sentence structure conversion unit 13d acquires the connection expression corresponding to the semantic label from the character-specific connection expression list 26, and inserts it after the inserted phrase. A reading mark may be inserted together with the connection expression.

  For example, it is assumed that the character-by-character connection representation list 26 of the character A is designated as the file name of the item-by-character connection representation list 26 for sentence structure conversion stored in the setting value DB 22. In the input sentence 23 “I am cold, so I chose warm clothes.” The meaning label “reason” of the morpheme “So” in the phrase “Because it is cold” is listed in the connection representation list 26 for each character. Corresponds to one of the semantic labels. Therefore, “so” and the subsequent punctuation mark “,” are deleted, and a punctuation mark “.” Is inserted at the deleted part. Also, the auxiliary verb “na” before “no” is replaced with the final form “da”. Further, after the inserted punctuation mark “.”, The conjunction “because” associated with the semantic label “reason” is acquired in the character-specific connection expression list 26 of character A and inserted together with the reading mark “,”. As a result, the sentence structure of the input sentence 23 is converted as shown below.

Single culture example:
I'm cold <<. So I chose warm clothes.

  In the above description, a punctuation point is inserted to indicate a sentence boundary. However, any symbol may be used as long as it can indicate a sentence boundary.

  Also, an input sentence 23 in which a plurality of particles representing a suggestion such as “ha” or a particle “ga” representing a main case appears out of the scope of a single culture, that is, out of the rewriting process by the sentence structure conversion unit 13d. Good. For example, the input sentence 23, “I gave him a muffler because he is cold,” “I gave him a muffler, because he was cold.” This is to avoid being overwritten.

  The predicate function expression / character ending conversion unit 13e acquires the file name of the item-specific function expression list 27 for predicate function expression / character ending conversion from the setting value DB 22. An example of the character-specific function expression list 27 is shown in FIG. The structure of the character-specific function expression list 27 is the same as that of the style-specific function expression list 24. The processing of the predicate function expression / character ending conversion unit 13e is the same as the processing of the style conversion unit 13a. However, in the character-specific function expression list 27, predicate function expressions and endings capable of expressing the character individuality can be arbitrarily determined. As in the example of the character-by-character function expression list 27 of the character B shown in FIG. 7, expressions outside the Japanese grammar range (character endings) can be used for character assignment. Thereby, in the predicate function expression / character ending conversion unit 13e, rewriting of a language expression different from that of the style conversion unit 13a can be realized.

  For example, it is assumed that the character-by-character function expression list 27 of the character A is designated as the file name of the item-by-character function expression list 27 for the item (e) predicate function expression / character ending conversion stored in the setting value DB 22. . In this case, the predicate function expression of the input sentence 23 "I chose cold clothes because I am cold" is converted as follows.

Example of predicate function expression and character ending conversion:
I'm cold, so I choose warm clothes.

  The utilization form conversion unit 13f replaces the adjectives (stems and utilization endings) with corrupted expressions in accordance with the item (f) utilization form conversion setting values (1 = converted into corrupted utilization forms) stored in the setting value DB 22. By doing so, the sentence which converted the utilization form is output. Specifically, the utilization form conversion unit 13f collates the notation of the adjective stem and utilization ending included in the input sentence 23 with the “input (notation)” portion of the collapse utilization rule 28 as shown in FIG. , Acquire and replace the notation of the stem and inflection ending of the corresponding “output” part.

  For example, in the case of the input sentence 23 “I am cold, I chose warm clothes.” The adjective “warm” part of “Kai” collapses and is one of the “input (notation)” of the utilization rule 28. It corresponds to. Therefore, by replacing the “kai” part with “ke” which is the corresponding “output (notation)”, as shown below, the utilization form of the input sentence 23 is converted into a disrupted utilization form.

Example of conversion to a broken usage:
I was cold, so I chose warm clothes.

  In the present embodiment, for example, the case where the conversion process is performed by a rule-based method using the collapse utilization rule 28 as illustrated in FIG. 8 is described, but the present invention is not limited to this. For example, using a model that has been learned from a corpus (text data that uses corrupted usage) using some kind of machine learning, the broken stems and usage endings that are the replacement destinations of unbroken stems and used endings May be estimated.

  The stagnation conversion unit 13g converts the input sentence 23 into a sentence in which stagnation appears according to the item (g) stagnation conversion setting value (1 = scrambled) stored in the setting value DB 22. Specifically, the sloppy conversion unit 13g hiraganaizes the first character of the reading of the morpheme at the beginning of the input sentence 23 (described in katakana in the example of the basic analysis result of this embodiment) and inserts it before the morpheme at the beginning of the sentence To do. A reading mark may be inserted after the inserted hiragana.

  For example, in the case of the input sentence 23 “I am cold and I chose warm clothes”, the input sentence 23 is converted into a scrambled sentence as shown below.

Example of stagnation:
<< Wa >>I'm cold, so I chose warm clothes.

  In the above, the case where only the morpheme at the beginning of the sentence is the object of stagnation is explained, but the morpheme at the beginning of the section is targeted, “Wow, I am cold, so I chose warm clothes.” You can convert it like this, or even if you convert the target morpheme of each phrase as “Wow, I ’m cold, so I chose warm clothes. Good. In addition, if the notation of the morpheme subject to stagnation is katakana, the characters to be inserted before the sneaking target morpheme may be katakana. Further, for example, hiragana or katakana to be inserted may be inserted twice or more like “Wa, Wa, I ...”.

  The personal pronoun replacement unit 13h obtains the file name of the character-specific personal pronoun list 29 for item (h) personal pronoun replacement from the setting value DB 22. Then, the personal pronoun replacement unit 13h outputs a sentence in which the morphemes of the personal pronouns included in the input sentence 23 are replaced with personal pronouns obtained from the character-specific personal pronoun list 29 indicated by the acquired file name.

  Specifically, the personal pronoun replacement unit 13h refers to, for example, the character-specific personal pronoun list 29 indicated by the acquired file name in the character-specific personal pronoun list 29 as shown in FIG. Then, the personal pronoun replacement unit 13h replaces the morpheme whose notation matches the “input” portion of the character-specific personal pronoun list 29 in the input sentence 23 with the corresponding “output” portion notation.

  For example, it is assumed that the character-specific personal pronoun list 29 of character A is designated as the file name of the character-specific personal pronoun list 29 for item (h) personal pronoun replacement stored in the setting value DB 22. In this case, the above-mentioned input sentence 23 “I chose cold clothes because I am cold” is replaced with the personal pronoun as follows.

Example of personal pronoun substitution:
Since << I >> was cold, I chose warm clothes.

  The dialect / special vocabulary replacement unit 13i acquires the file name of the character-specific vocabulary replacement rule 30 for item (i) dialect / special vocabulary replacement from the setting value DB 22. Then, the dialect / special vocabulary replacement unit 13i outputs a sentence in which a specific morpheme included in the input sentence 23 is replaced with a dialect or special vocabulary according to the character-specific vocabulary replacement rule 30 indicated by the acquired file name.

  Specifically, the dialect / special vocabulary replacement unit 13i refers to, for example, the character-specific vocabulary replacement rule 30 indicated by the acquired file name among the character-specific vocabulary replacement rules 30 as shown in FIG. Then, the dialect / special vocabulary replacing unit 13i replaces the morpheme whose notation matches the “input” portion of the character-specific vocabulary replacement rule 30 in the input sentence 23 with the notation of the corresponding “output” portion. In the character-specific vocabulary replacement rule 30, a rule for converting a specific replacement source vocabulary into a vocabulary like a target character is defined.

  Also, the dialect / special vocabulary replacement unit 13i, when the replacement destination vocabulary is a utilization word (verb, adjective, auxiliary verb), the utilization ending is adapted to the utilization form of the replacement vocabulary (the “input” part is applicable). May be adjusted. Regarding the use of auxiliary verbs, those using the verb type can be obtained by referring to the verb usage table, and those using the adjective type can be obtained by referring to the adjective usage table.

  For example, it is assumed that the character-by-character vocabulary replacement rule 30 for character A is specified as the file name of the character-specific vocabulary replacement rule 30 for item (i) dialect / special vocabulary replacement stored in the setting value DB 22. In this case, in the input sentence 23 “I chose cold clothes because I am cold”, a specific vocabulary is replaced with a dialect or a special vocabulary as follows.

Examples of dialect / special vocabulary substitution:
I chose cold clothes because I was cold.

  Note that the expression replaced by the dialect / special vocabulary replacing unit 13i may be protected from being subjected to further rewriting processing in other rewriting processing.

  The phoneme replacement unit 13j acquires the file name of the character-specific phoneme replacement rule 31 for item (j) phoneme replacement from the setting value DB 22. Then, the phoneme replacement unit 13j outputs a sentence in which a specific character included in the input sentence 23 is replaced with a replacement destination character according to the character-specific phoneme replacement rule 31 indicated by the acquired file name.

  Specifically, the phoneme replacement unit 13j refers to, for example, the character-specific phoneme replacement rule 31 indicated by the acquired file name among the character-specific phoneme replacement rules 31 as illustrated in FIG. Then, the phoneme replacement unit 13j replaces the character whose notation matches the “input” portion of the character-specific phoneme replacement rule 31 in the input sentence 23 with the corresponding “output” portion notation. In the present embodiment, the phoneme replacement is recognized and replaced in units of characters. For example, character substitution from “na” to “nya” and “no” to “nyo” is a phonetic substitution from “na” to “nya” and “no” to “nyo” (“n”). Phoneme replacement from “ny” to “ny”).

  For example, it is assumed that the character-by-character phoneme replacement rule 30 for the character A is designated as the file name of the item-by-character phoneme replacement rule 31 for phoneme replacement stored in the setting value DB 22. In this case, the above-mentioned input sentence 23 “I chose cold clothes because I am cold” is replaced with a specific character as follows.

Examples of character substitution that captures phonemic substitution:
I got cold and chose warm clothes.

  In addition, it is not limited to replacing all characters that match the “input” portion of the character-specific phoneme replacement rule 31, and whether or not to perform character replacement depends on the part of speech of the morpheme to which the character belongs, It may be determined by using the condition (constraint) as the appearance position at. For example, restrictions such as “do not replace if the part of speech of a morpheme to which a certain character belongs are nouns”, “do not replace if a certain character is located at the beginning of the morpheme” can be provided. In addition, there may be provided a constraint that “a plurality of substitutions should not be performed within the same morpheme” or a constraint that “two or more consecutive characters should not be replaced”.

  The problem here is that as the number of constraints increases, it becomes difficult to manually define rules that take into account all combinations of constraints. Therefore, the replacement source (input) character, the replacement destination (output) character, the part of speech of the morpheme to which the replacement source character belongs, the appearance position of the replacement source character, the number of replacements already performed in the morpheme to which the replacement source character belongs, A model may be learned by some machine learning using the number of consecutive character replacements up to the replacement source character as a feature amount, and character replacement may be performed using this model.

  The discriminative meaningless expression insertion unit 13k acquires a character string indicating the item (k) discriminative meaningless expression stored in the setting value DB 22 and inserts it at the end of the input sentence 23, thereby discriminating meaningless expression. The statement with is inserted. In this embodiment, the discriminative meaningless expression refers to an expression that has no meaning as Japanese but assists the discrimination of characters.

  For example, when “Pyon!” Is specified as a character string indicating a discriminative meaningless expression, the input sentence 23 “I chose cold clothes because I am cold” is as follows. A discriminative meaningless expression is inserted in

Example of discriminative meaningless insertion:
I chose cold clothes because it was cold. << Pyon! >>

  In this embodiment, a discriminative meaningless expression is inserted after the ending point of the sentence. If there is no punctuation, the punctuation is inserted after the punctuation, and a discriminative meaningless expression follows the punctuation.

  For example, it is assumed that there are three types of characters: a character that looks like a frog (frog), a round character like a ball (ball), and a character that has a lot of thorns (thorn). Because these three use similar language expressions, even if it is difficult to communicate linguistic differences, the frog's utterance, not the ball or thorns, is inserted by inserting the discriminative meaningless expression "Pyon!" It can be impressed strongly to the reader or the listener.

  In the above, the case where a discriminative meaningless expression is inserted at the end of the input sentence 23 has been described. However, “Pyon! I am cold, so I chose warm clothes.” A discriminative meaningless expression may be inserted in other places.

  The symbol insertion unit 13l acquires a character string indicating the item (l) symbol stored in the setting value DB 22, and inserts it at the end of the input sentence 23, thereby outputting a sentence in which the symbol is inserted. The processing of the symbol insertion unit 13l is the same as that of the discrimination meaningless expression insertion unit 13k except that the character string to be inserted is not a discrimination meaningless expression but a symbol class. In the present embodiment, the symbols indicate symbols such as ★ (star) and ♪ (note), and emoticons such as (* ^ o ^ *) and (> _ <;). To do.

  For example, when “(* ^ o ^ *)” is designated as a character string indicating symbols, the above-mentioned input sentence 23 “I chose cold clothes because I am cold” Symbols are inserted as follows.

Example of symbol insertion:
I chose cold clothes because it was cold. << (* ^ o ^ *) >>

  As described above, the rewrite processing for the input sentence 23 (basic analysis result) is performed in each of the processing units 13a to 13l of the rewrite processing unit 13 that is set to perform the rewrite processing, and the rewriting is performed from each of the processing units 13a to 13l. A statement 32 is output.

  Note that the rewriting processes of the processing units 13a to 13l of the rewriting processing unit 13 can be executed in any combination. For example, when the result rewritten by a certain processing unit is further rewritten by another processing unit, the output of a certain processing unit is again passed to the basic analysis unit 12, and the newly acquired basic analysis result can be passed to the next processing unit. That's fine. Alternatively, the basic analysis result of the input sentence 23 may be rewritten based on rewriting by a certain processing unit and then passed to the next processing unit. For example, when the result rewritten by the style conversion unit 13a is passed to another processing unit, the following basic analysis result can be passed to the next processing unit.

(1) Because {I} _ is / {cold} ___, / {warm} _i / {clothes} _ / {selection} _ << and >> _ << done >> _.
(2) {I am} _Ha / {Samgari} _Na_Node_, / {Attaka} _I / {Fuse} _wo / {Era} _ << dummy >> _ << dummy >> _.
(3) {pronoun} _involvement particle / {noun} _auxiliary verb_conjunctive particle_reading point / {adjective stem} _utilization ending: collocation / {noun} _case particle / {verb stem: ba line five steps} _ <<Utilization ending: continuous form >> _ << dummy >> _ Place (4) {*} _ * / {*} _ * _ reason _ * / {*} _ * / {*} _ * / {*} _ * _Done_ *

  << >> indicates information about the rewritten morpheme. In the above example, information that has no opportunity to be used by other processing units is “dummy”, but correct information may be given. Note that (2) morpheme reading is used for hiragana conversion in the character type conversion unit 13b. However, since “bi_ha” is already hiragana, there is no problem with “dummy”. In the present embodiment, since there is no processing that uses the part of speech of the functional expression, there is no problem even if the part of speech of “ta” is “dummy”.

  By arbitrarily combining the rewriting processes of the processing units 13a to 13l of the rewriting processing unit 13, the input sentence 23 “I chose cold clothes because it is cold,” for example, “(I) << >> is cold <<, so >>, warm << Ke >> chose clothes << ze! >> "and" (II) << That's why it becomes possible to rewrite the warm fussy >> as if it were <<, Elder >>>>. When the rewriting process of the plurality of processing units 13 a to 13 l is applied, the final output of the processing unit is output as the rewritten sentence 32. The example of (I) is an example in which (h) personal pronoun, (c) split writing, (f) utilization form, and (e) predicate function expression / rewriting processing for character ending items are combined. In addition, the example of (II) combines (b) character type, (g) sloppy, (h) personal pronoun, (c) split writing, and (e) rewrite processing for predicate function expression / character ending items. It is an example.

In the present embodiment, when a plurality of rewriting processes are applied, for example, the order of the rewriting processes to be applied can be determined in consideration of the following points.
(D) The sentence structure conversion should be applied before the other 11 types of rewrite processing.
(G) Conversion to slogan and (b) character type conversion are (d) sentence structure, (i) dialect / special vocabulary, (a) style, (e) predicate function expression, character ending, and ( h) It should be applied after personal pronouns. Here, either (g) conversion to stagnation or (b) conversion of character type may be applied first.
(I) Dialect / special vocabulary replacement should be applied after (d) sentence structure and before (a) sentence style, (e) predicate function expression / character ending conversion.
In this embodiment, based on the above three points, (d) sentence structure, (i) dialect / special vocabulary substitution, (a) style, (e) predicate function expression / character ending, (h) personal pronoun, (b Rewrite processing shall be applied in the order of :) character type, (f) inflection, (g) sloppy, (j) phoneme substitution, (c) segmentation, (k) discriminative meaningless expression, (l) symbols. .

  Note that * in each list or each rule shown in FIGS. 3 to 11 represents an arbitrary character string. In addition, “example” or “remarks” in each list or each rule shown in FIGS. 3 to 11 is an application example that assists in explanation of each list or each rule, and is an item of each list or each rule. There is no need to specify.

<Operation of language expression rewriting device>
Next, the operation of the language expression rewriting device 10 according to the present embodiment will be described. When a setting file 21 in which setting values relating to rewriting of a language expression according to a target character and an input sentence 23 to be rewritten to a language expression according to a character are input to the language expression rewriting device 10, In the language expression rewriting processing device 10, a language expression rewriting processing routine shown in FIG. 12 is executed.

  In step S <b> 11, the setting unit 11 reads the setting file 21 and stores setting values for each item described in the read setting file 21 in the setting value DB 22.

  Next, in step S12, the basic analysis unit 12 reads the input sentence 23 that has been input, analyzes the input sentence 23 that has been read, reads the morpheme boundary, the reading of each morpheme, the part of speech of each morpheme, Acquires information on the usage form, phrase boundaries, phrase headings, and semantic labels of functional expressions. The basic analysis unit 12 outputs the acquired information to the rewrite processing unit 13 as a basic analysis result.

  Next, in step S13, as described in detail above, the rewriting process for the input sentence 23 (basic analysis result) is performed in each of the processing units 13a to 13l of the rewriting processing unit 13. Next, in step S14, the rewrite processing unit 13 outputs the rewritten sentence 32 that is the result of the rewrite process in step S13, and the language expression rewrite process routine ends.

  As described above, according to the language expression rewriting processing device according to the present embodiment, (a) style, (b) character type, (c) split writing, (d) sentence structure, (e) predicate function expression / character Ending, (f) inflection, (g) stagnation, (h) personal pronoun, (i) dialect / special vocabulary, (j) phoneme substitution, (k) discriminative meaningless expression, and (l) 12 Rewrite processing is performed by arbitrarily combining types of rewrite items. For this reason, not only sentence end expressions but also various language expressions can be rewritten in any combination, and various variations of language expressions can be rewritten.

  In addition, since the rewriting target is not limited to the sentence end expression, it can be converted into a language expression that seems to be a character with a strong individuality such as a TV animation or a cartoon character, which is difficult to realize with the technique of Non-Patent Document 1. Become.

  For example, if the input sentence is "I am cold, I chose warm clothes." For example, in the conventional technology that only converts the sentence ending expression, for example, "I am cold, so I chose warm clothes! "become that way. On the other hand, as in this embodiment, by rewriting various language expressions in any combination, for example, “(I) I am cold. So I chose warm clothes. "(II) Ah, I'm sad. That's why I've chosen a warm fusou.

  When this embodiment is applied to a system that interacts with people (dialogue system), the person who operates the system (the person who designs the characters of the system) has a setting file for each character, several expression lists, and conversion. By preparing the rules, it is possible to easily create utterances having various character characteristics, and to greatly reduce the cost required for increasing the number of characters in the dialogue system.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the rewrite processing method in each of the processing units 13a to 13l of the rewrite processing unit 13 is not limited to the method described above, and other methods may be applied. In addition, the rewrite items are not limited to the above (a) to (l), and may be rewrite items related to linguistic features in which the features of the character appear in the utterance according to the character.

  In the above embodiment, when a plurality of rewriting processes are applied in combination, the case where they are applied in a predetermined order has been described. However, the order of application can be changed by specifying the order of the rewriting processes to be applied. May be. In this case, information for specifying the order of the rewriting process is also described in the setting file, and each processing unit may be caused to function based on this information. For example, in the case where (j) phoneme replacement and (b) character type of hiragana are applied to the input sentence 23 of “Nohara no Hana”, the character-specific phoneme replacement rule 31 of character A shown in FIG. 11 is applied. think of. (J) Phoneme replacement → (b) When rewriting processing is applied in the order of character types, it is rewritten as “Noharanya Hana”. On the other hand, when the rewriting process is applied in the order of (b) character type → (j) phoneme replacement, it is rewritten as “Nyohara Nyonya”. In this way, by changing the order of the rewriting process, it is possible to express various variations in which the strength of the character personality is different.

  Further, (b) character type, (c) split writing, and (l) symbols are rewrite items that are effective when the rewritten text 32 to be output is text data. That is, when applied to an utterance of a dialogue system or the like whose final output is reading by speech synthesis, the character imparting effect is not exhibited. Therefore, when the output of the rewritten sentence 32 is only read out by speech synthesis, the character type conversion unit 13b, the segmentation conversion unit 13c, and the symbol insertion unit 13l may be omitted from the configuration of the rewrite processing unit 13. Or, according to the output form of the rewritten sentence 32, the setting values of the rewriting items of (b) character type, (c) split writing, and (l) symbols may be changed. For example, when the output of the rewritten sentence 32 is only read out by speech synthesis, a setting value indicating that the rewriting process of (b) character type, (c) segmentation, and (l) symbols is applied to the setting file. When the setting value is stored in the setting value DB 22, the setting value is changed to a setting value indicating that the rewriting process is not applied and stored.

  Further, in the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium or provided via a network. It is also possible to do.

DESCRIPTION OF SYMBOLS 10 Language expression rewriting apparatus 11 Setting part 12 Basic analysis part 13 Rewriting process part 13a Stylistic conversion part 13b Character type conversion part 13c Separation writing part 13d Sentence structure conversion part 13e Predicate function expression and character ending conversion part 13f Utilization form conversion part 13g Conversion unit 13h Personal pronoun replacement unit 13i Dialect / special vocabulary replacement unit 13j Phoneme replacement unit 13k Discriminant meaningless expression insertion unit 13l Symbol class insertion unit 21 Setting file 22 Setting value database 23 Input sentence 24 Stylistic function expression list 25 Usage table 26 Character-specific connection expression list 27 Character-specific function expression list 28 Collapse utilization rule 29 Character-specific personal pronoun list 30 Character-specific vocabulary replacement rule 31 Character-specific phoneme replacement rule 32 Rewritten sentence

Claims (7)

A setting unit that sets a setting value related to rewriting of the language expression based on the linguistic feature for each of a plurality of types of linguistic features that include a stylistic style, a predicate function expression, and a personal pronoun; ,
Based on the setting value set by the setting unit, the process of converting the type of style, the process of converting the predicate function expression according to the character, and converting the morpheme that is a personal pronoun into the personal pronoun according to the character A rewriting processing unit that applies at least one type of rewriting processing to an input sentence among rewriting processing of language expressions based on a plurality of types of linguistic features including processing;
Language expression rewriting device including For each of the plurality of types of linguistic features, the setting unit further includes a sentence structure, an inflection, a slang, a dialect or special vocabulary, a specific phoneme, and a distinct meaningless expression that can distinguish a character but has no meaning. Set the setting value,
The rewrite processing unit converts a sentence structure, converts a utilization form, converts to a sloppy expression, converts a specific vocabulary to a dialect or special vocabulary based on the setting value set by the setting unit A language expression rewriting process based on the plurality of types of linguistic features, further comprising: a process of converting, a process of converting a specific phoneme into a phoneme corresponding to the character, and a process of inserting a discriminative meaningless expression corresponding to the character. The language expression rewriting device according to claim 1, wherein at least one type of rewriting processing is applied to the input sentence. The setting unit sets the setting value for each of the plurality of types of linguistic features further including a character type, a division, and symbols,
The rewriting processing unit includes the plurality of types of linguistic features further including a process of converting the character type, a process of converting to a split type, and a process of inserting symbols based on the setting value set by the setting unit. The language expression rewriting device according to claim 1 or 2, wherein at least one type of rewriting process is applied to the input sentence among the rewriting processes of the linguistic expression based on the language expression. A language expression rewriting method in a language expression rewriting device including a setting unit and a rewriting processing unit,
For each of a plurality of types of linguistic features that include a style, a predicate functional expression, and a personal pronoun, and a feature corresponding to a character appears, a setting value for rewriting a linguistic expression based on the linguistic feature Set,
Based on the setting value set by the setting unit, the rewriting processing unit converts the type of stylistic, the processing for converting the predicate function expression according to the character, and the morpheme that is a personal pronoun according to the character. A language expression rewriting method that applies at least one type of rewriting process to an input sentence among rewriting processes of a language expression based on a plurality of types of linguistic features including a process of converting to a personal pronoun. For each of the plurality of types of linguistic features, the setting unit further includes a sentence structure, an inflection, a slang, a dialect or special vocabulary, a specific phoneme, and a distinct meaningless expression that can distinguish a character but has no meaning. Set the setting value,
The rewrite processing unit converts a sentence structure, converts a utilization form, converts to a sloppy expression, converts a specific vocabulary to a dialect or special vocabulary based on the setting value set by the setting unit A language expression rewriting process based on the plurality of types of linguistic features, further comprising: a process of converting, a process of converting a specific phoneme into a phoneme corresponding to the character, and a process of inserting a discriminative meaningless expression corresponding to the character. The language expression rewriting method according to claim 4, wherein at least one type of rewriting processing is applied to the input sentence. The setting unit sets the setting value for each of the plurality of types of linguistic features further including a character type, a division, and symbols,
The rewrite processing unit includes the plurality of types of linguistic features further including a process of converting the character type, a process of converting a division, and a process of inserting symbols based on the setting value set by the setting unit. 6. The language expression rewriting method according to claim 4, wherein at least one type of rewriting process is applied to the input sentence among the rewriting processes of the linguistic expression based on the language expression.   The language expression rewriting program for functioning a computer as each part of the language expression rewriting apparatus of any one of Claims 1-3.

Images