JP4041875B2 - Written word style conversion system and written word style conversion processing program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a text word style conversion system that converts a text string of a text word described in a certain natural language into a text string of text words in another text style described in the same natural language.
[0002]
[Prior art]
A typical example of conversion processing of sentences or sentences related to sentences written in a natural language is machine translation. In machine translation, sentences or sentences written in a natural language of one country are converted into sentences or sentences written in a natural language of another country.
[0003]
In contrast to machine translation, which translates into languages of other countries, systems that convert sentences or sentences between the same natural languages are also being used. For example, it is a system that automatically generates a summary sentence or recommends a sentence.
[0004]
In general, in the process of converting sentences between the same natural language, a large number of conversion rules consisting of pairs of patterns of words, phrases, sentences, etc. before conversion and patterns of words, phrases, sentences, etc. after conversion are prepared. The pattern matching is used to find a pre-conversion pattern that appears in the input sentence, and if there is a corresponding pattern, replace it with a pattern such as a word / phrase / sentence after conversion.
[0005]
In addition, as a sentence conversion process between the same natural languages, summary sentences were automatically generated and sentences were reviewed. For example, a novel written by Ryunosuke Ninagawa was written in the style of Natsume Soseki. There was no such thing as converting a particular individual's style or general style to another particular person's style, such as converting to a novel.
[0006]
[Problems to be solved by the invention]
In conventional sentence or sentence conversion processing in the same natural language, uniform conversion is generally performed according to conversion rules, and the quality of conversion results has not been evaluated. Therefore, whether or not a good conversion is actually performed depends greatly on the quality of the conversion rules prepared in advance, and depending on the conversion rules applied, the conversion result may differ from the target conversion. there were.
[0007]
In addition, in order to improve the conversion accuracy, it is necessary to select only truly valid conversion rules, and selecting a large number of conversion rules has been a very difficult task. For example, a conversion rule includes a rule for converting a character string “A” into a character string “B” and a rule for converting a character string “B” into a character string “A”. When mixed, the conventional technology could not obtain the correct conversion result.
[0008]
For this reason, the conventional technology has a problem that it is difficult to construct a system that converts personal writing styles and difficult sentences into plain sentences that are easy for elementary school students to understand.
[0009]
The present invention solves the above-mentioned problems and does not select carefully selected rules as rules for changing the expression of text words. For example, even when various rules such as those automatically acquired by a computer are used, It aims at providing the system which can be appropriately converted into the style of the object.
[0010]
[Means for Solving the Problems]
In order to solve the above-mentioned problems, the present invention provides a first system described in a natural language in a system that converts a text word described in a certain natural language into a text word having another style described in the same natural language. Is a transformation rule that paraphrases the character string of the above into a second character string having the same meaning, and the computer extracts the description of the same word from a plurality of different dictionary files, matches the language information of the plurality of extracted descriptions, Stores a transformation rule that does not depend on the conversion direction of the target sentence style, which is composed of the first character string and the second character string automatically generated by the synonym or synonym phrase obtained from the matching result. a transformation rule storing means, numerical information for the results obtained by modifying the string representation to assess whether it is the style of interest, function group or subroutines, evaluation method Appearance frequency of conversion candidates in an example in a database in which a set of sentences in a database storing a set of sentences according to a conversion destination is used as evaluation information composed of written rules or information obtained by combining these rules. Or an evaluation information storage means for storing predetermined evaluation information so that the higher the appearance probability is, the higher the evaluation value is given, and when inputting a character string subject to stylistic conversion described in natural language, The transformation processing means transforms the input character string using the transformation rules stored in the transformation rule storage means to generate a plurality of conversion candidates. Next, the plurality of generated conversion candidates are evaluated by the evaluation processing means using the evaluation function or the evaluation rule stored in the evaluation information storage means, and the best expression of the evaluation result is selected, and the selected expression is selected. The result of conversion is output as a sentence word converted to the target style.
[0011]
For example, the scale of evaluation by an evaluation function or evaluation rule includes the appearance frequency or probability of conversion candidates in a large number of examples consisting of plain text sets, and gives a high evaluation to those with a high appearance frequency or appearance probability. By doing so, it is possible to realize a text-to-sentence conversion system that converts a character string of an input difficult sentence into a plain text.
[0012]
In addition, for example, the evaluation scale by the evaluation function or the evaluation rule includes the appearance frequency or appearance probability of conversion candidates in a large number of examples made up of a specific individual's sentence set, and is high for those with a high appearance frequency or appearance probability. By giving the evaluation, it is possible to realize a text-to-sentence style conversion system that converts the expression of the input character string into a specific personal style.
[0013]
Although it is not always necessary to use the frequency of appearance or the probability of appearance as an evaluation scale, it is possible to evaluate whether or not the text style is suitable for the purpose after transformation of the written word by some kind of evaluation scale. Since the transformation is selected as the conversion result, it is sufficient that the information prepared in advance as the transformation rules of the style is basically synonymous with the character string, and a variety of transformation rules can be used without careful selection. There is no need to consider the direction of the deformation rule such as the deformation source and the deformation destination. That is, for example, in a deformation rule, a rule that transforms a character string “A” into a character string “B”, and a rule that transforms a character string “B” into a character string “A”. Even if both are mixed, the final conversion result can be obtained by the evaluation. Therefore, it is easy to create a transformation rule, and it is possible to use a transformation rule prepared for a certain style conversion for another style conversion.
[0014]
The above means can be realized by a computer and a software program that is installed and executed on the computer, and the program is a computer-readable portable medium memory, semiconductor memory, hard disk, or other suitable recording medium. Can be stored.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 shows a system configuration example of the present invention. In the figure, reference numeral 1 denotes a text-to-sentence conversion processing device realized by a computer including a CPU and a memory, and includes a deformation processing unit 11, an evaluation processing unit 12, a deformation rule storage unit 14, and an evaluation information storage unit 15.
[0016]
The conversion target sentence 10 is a natural language sentence that becomes an input in the present system. Hereinafter, the conversion target sentence 10 is not necessarily limited to one sentence, but may be a sentence, a phrase, a clause, or the like, although not particularly specified. The conversion result sentence 13 is an output of the present system, and is the conversion target sentence 10 having the same type of natural language and a different style from the original style.
[0017]
The module of the text / word style conversion processing device 1 basically includes a deformation processing unit 11 and an evaluation processing unit 12. The deformation processing unit 11 is a module that acquires conversion candidates using the deformation rules stored in the deformation rule storage unit 14. The evaluation processing unit 12 evaluates whether or not the conversion candidate is good, that is, whether the sentence is suitable for the purpose, using an evaluation scale (evaluation function or the like) stored in the evaluation information storage unit 15 in advance, and has the highest evaluation. This module selects conversion candidates.
[0018]
When the conversion target sentence 10 is input, the transformation processing unit 11 uses the transformation rules to list transformation candidates, and the evaluation processing unit 12 checks the validity of the transformed text and finds that it is the most appropriate. The determined one is selected and the result is output as a conversion result sentence 13.
[0019]
The transformation rules stored in the transformation rule storage unit 14 are synonyms obtained from the result of matching the extracted language explanation information of a plurality of explanation texts by extracting explanation texts of the same word from a plurality of different dictionary files. It consists of the first character string and the second character string automatically generated by a word or synonym phrase. This transformation rule does not depend on the direction of the target style. Evaluation information of the evaluation function for evaluating the conversion candidates (rating scale) prepares what is appropriate for each issue to be handled.
[0020]
The evaluation information may be numerical information for evaluation, or may be procedural information such as a function group or a subroutine group. Also, it may be a rule (rule) describing the evaluation method. It can also be realized by a combination of these. A typical example of the evaluation scale used in the evaluation processing unit 12 is an appearance frequency (or appearance probability) in a large amount of language data composed of a sentence set of a target style.
[0021]
For example, in the present system that changes the style of a sentence word, it is assumed that all deformation rules used by the deformation processing unit 11 satisfy synonyms. In this case, if the data whose style is to be converted is transformed so that the probability of occurrence (occurrence) of the data is higher in the text set of the desired style, the text word is very close to the target style.
[0022]
This will be explained with a simpler example. For example, it is assumed that the input sentence has a so-called “Datone” style, and there are many character strings “~”. Suppose the transformation rule has a rule that transforms “is” into “is”. If the target style is “Naruto”, a database storing a large amount of text data of the target style is prepared and used for evaluation. Count the number of occurrences of “is” and “is” in the database. If there are more occurrences of “is”, the evaluation of “is” is higher than “is”. By this evaluation, the style of “Datone” is automatically converted to the style of “Datone”.
[0023]
Here, by changing the corpus for examining the appearance frequency (or appearance probability), the results of various style conversions can be obtained. For example, when the input data is a legal sentence, if a set of plain sentences is given as a corpus, it can be expected that a difficult sentence related to the law will be transformed into a plain sentence.
[0024]
Also, if you put a novel sentence written by someone as input data and put a Shakespeare novel as a corpus, a novel with a Shakespeare style will be completed. In the same way, it is possible to convert the novel of Ryunosuke Akutagawa into the style of Natsume Soseki.
[0025]
The deformation rules used in this system are automatically generated by a computer, for example, as follows. First, language information having a semantic correspondence described in a plurality of the same languages is extracted. Specifically, by preparing a plurality of different dictionary file, we extract a description of the same word from them. Next, a plurality of extracted language information is matched, and a synonym or synonym phrase is extracted from the result. From the extracted synonym or synonym phrase, a deformation rule that automatically translates the first character string into a synonymous second character string is automatically generated.
[0026]
FIG. 2 is a processing flowchart of the deformation processing unit 11. First, in step S10, the transformation processing unit 11 inputs the conversion target sentence 10 designated as the style conversion target. The input method does not matter, such as input from a keyboard, input from a file, input from an application program.
[0027]
In step S11, a deformation rule necessary for conversion is read from the deformation rule storage unit 14. If it has already been read, reading here is not necessary. In step S12, the input conversion target sentence 10 is transformed using a transformation rule. The transformed expression candidate is delivered to the evaluation processing unit 12. At this time, the transformed expression candidates may be passed to the evaluation processing unit 12 one by one, or when there are a plurality of candidates, they may be passed together.
[0028]
In step S <b> 13, the evaluation processing unit 12 receives a candidate for expression after transforming the conversion target sentence 10 from the transformation processing unit 11, and uses the evaluation information stored in the evaluation information storage unit 15 to obtain a target sentence style. Evaluate whether the expression is suitable for This evaluation information may be an evaluation function called from the evaluation processing unit 12, or may be a parameter used by the evaluation function. For each candidate for expression, the evaluation result is calculated as a numerical value (evaluation value). In step S14, the expression after modification having the best evaluation result is selected, and the modified expression is output as the conversion result sentence 13.
[0029]
Hereinafter, specific application examples to various text-to-sentence conversion processing systems will be described.
[0030]
(A) Application Example to Difficult Sentence Conversion System FIG. 3 shows an application example to the difficult sentence conversion system. The difficult sentence conversion system shown in FIG. 3 performs processing for rewriting legal sentences into plain sentences and rewriting difficult newspaper articles into easy sentences for elementary school students.
[0031]
For example, as in the example of FIG. 3, it is assumed that a sentence “I exempt the minister” is input as the conversion target sentence 10. The transformation processing unit 11 paraphrases the conversion target sentence 10 into different expressions having the same meaning using the transformation rules prepared in advance in the transformation rule storage unit 14. Here, as a transformation rule,
“Relieve” → “Stop”
...
If there is a rule, the transformation processing unit 11 applies the transformation rule to the conversion target sentence 10 to generate a sentence “stop the minister” from a sentence “dismiss the minister”. In addition, there are various transformation rules, and many transformed sentences are generated as candidates. These sentences are passed to the evaluation processing unit 12. Note that the conversion target sentence 10 that has not been transformed is also passed to the evaluation processing unit 12 as one of the candidates.
[0032]
The evaluation processing unit 12 evaluates the sentence deformed by the deformation processing unit 11 using evaluation information (evaluation function) prepared in advance in the evaluation information storage unit 15. Here, the evaluation scale gives a high evaluation to the conversion that increases the appearance frequency or the appearance probability in a sentence set for a low age group such as for elementary school students. Therefore, in this example, the evaluation processing unit 12 obtains the appearance frequencies of “dismiss the minister” and “stop the minister” in the sentence set for the younger age group within a predetermined range. A simple method is to count how many times the character string in the small area range including the deformed part appears in the language data. If the appearance frequency is higher for “quit the minister”, this expression is easier to express for younger age groups. As a result, this modification is accepted, and the conversion result sentence 13 is output “Stop the Minister”. Note that the appearance (occurrence) probability may be calculated instead of the appearance frequency.
[0033]
Further, the evaluation scale is not limited to the appearance frequency and the appearance probability in a predetermined sentence set, and any other scale can be used. For example, it is possible to preliminarily set evaluation points for word connections and grammatical phrases obtained from the result of parsing, and to use them for evaluation. Another possible method is to learn some measure of simplicity from the converted text set and evaluate it accordingly.
[0034]
In addition, in the sentence set for the younger age group, instead of counting the appearance frequency of conversion candidates each time it is evaluated, the appearance frequency (appearance) for each sentence, phrase, phrase, word, etc. that appears in the sentence set in advance. It goes without saying that an appearance frequency (appearance probability) may be obtained by preparing a table for which the probability) has been obtained and searching the table.
[0035]
(B) Application Example to Personal Style Conversion System FIG. 4 shows an application example to the personal style conversion system. In the personal style conversion system shown in FIG. 4, for example, a novel written by Ryunosuke Ayukawa is rewritten into a novel written by Natsume Soseki, or a novel of an unknown writer is rewritten as a novel written by Shakespeare.
[0036]
For example, as in the example of (1) in FIG. 4, it is assumed that a sentence “I have exempted the minister” has been input as the conversion target sentence 10. The transformation processing unit 11 paraphrases the conversion target sentence 10 into different synonymous expressions using the transformation rules prepared in advance in the transformation rule storage unit 14. Here, as a transformation rule to the style of a person who frequently uses the expression “such as”
“To do” → “To do”
...
The transformation processing unit 11 applies the transformation rule to the conversion target sentence 10, so that the sentence “I exempted the minister” was used to “dismiss the minister”. Is generated. In addition, there are various transformation rules, and many transformed sentences are generated as candidates. These sentences are passed to the evaluation processing unit 12. Note that the conversion target sentence 10 that has not been transformed is also passed to the evaluation processing unit 12 as one of the candidates.
[0037]
The evaluation processing unit 12 evaluates the sentence deformed by the deformation processing unit 11 using evaluation information (evaluation function) prepared in advance in the evaluation rule storage unit 15. Here, the evaluation scale gives a high evaluation to an expression having a high appearance frequency or appearance probability in a sentence set of a specific individual that is the purpose of conversion. Therefore, in this example, the evaluation processing unit 12 obtains the appearance frequency of the sentence “I have exempted the minister” or the sentence “I have exempted the minister” in the sentence set of the specific individual. . Note that the appearance frequency does not necessarily have to be the number of appearances of the entire sentence, and it may be the number of times a character string in a small area range including the deformed portion appears in the sentence set. When the appearance frequency of the sentence “I did something like exempting the minister” is high, the evaluation processing unit 12 outputs “I did something like exempting the minister” as the conversion result sentence 13.
[0038]
Also, in the case of a change to the style of a person who frequently uses “will”, for example, as a transformation rule,
"I think" → "I will"
...
As a measure of evaluation, an expression that makes the appearance frequency or the probability of occurrence in a sentence set of a specific individual who frequently uses the word “probably” be used.
[0039]
When the transformation processing unit 11 inputs the conversion target sentence 10 “I think that the Minister is exempted” as shown in (2) of FIG. 4, the transformation processing unit 11 applies the transformation rule to this input, Will be transformed. If the evaluation by the evaluation processing unit 12 shows that the evaluation value of the expression “I will exempt the minister” is the highest, the evaluation processing unit 12 outputs this sentence as a conversion result sentence 13.
[0040]
Note that the evaluation scale is not limited to the appearance frequency and the appearance probability in a predetermined sentence set, and any other scale can be used, as in the system example described above.
[0041]
As described above, the application example to the difficult sentence conversion system and the application example to the personal style conversion system have been described. However, the present system can be applied in the same manner as long as it converts some styles of sentence words. In this case, each transformation rule can be shared. For example, even if there is a transformation rule in which a rule that transforms the style of the author A into the style of the author B and a rule that transforms the style of the author B into the style of the author A, there is only a change in the evaluation information. Using the same transformation rule, conversion from the style of author A to the style of author B, and conversely, conversion from the style of author B to the style of author A can be realized. This is a significant difference from the prior art. Of course, it goes without saying that the transformation rules may be selected and used as rules suitable for conversion of each style.
[0042]
【The invention's effect】
As described above, according to the present invention, it is possible to automatically perform style conversion for various purposes. The transformation rules for character strings for stylistic conversion do not necessarily have to be transformation rules with directionality that can be converted into the target style, and it is sufficient if at least the synonym is satisfied. It is possible to easily collect and accumulate rules.
[Brief description of the drawings]
FIG. 1 is a diagram showing a system configuration example of the present invention.
FIG. 2 is a processing flowchart of sentence word style conversion.
FIG. 3 is a diagram illustrating an application example to a difficult sentence conversion system.
FIG. 4 is a diagram illustrating an application example to a personal style conversion system.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Text word style conversion processing apparatus 10 Conversion object sentence 11 Deformation process part 12 Evaluation process part 13 Conversion result sentence 14 Deformation rule memory | storage part 15 Evaluation information storage part

Claims (3)

A system for converting a sentence word described in a natural language into a sentence word with another style described in the same natural language,
A modification rule for paraphrasing a first character string described in the natural language into a second character string having the same meaning, and a plurality of extracted explanations of the same word extracted from a plurality of different dictionary files by a computer The linguistic information of the description sentence is matched, and conversion of the target sentence style consisting of the first character string and the second character string automatically generated by the synonym or synonym phrase obtained from the result of the matching Deformation rule storage means for storing deformation rules independent of direction;
Evaluation information consisting of numerical information for evaluating whether the expression of the result of transforming the character string has the target style, functions or subroutines, rules describing the evaluation method, or information combining these The evaluation scale that prescribes the evaluation information gives a higher evaluation value as the appearance frequency or appearance probability of the conversion candidate in the example in the database storing the set of sentences by the conversion destination style is higher. Evaluation information storage means for storing evaluation information determined in advance,
An input means for inputting a character string to be converted in a natural language;
A transformation processing means for transforming the input character string using a transformation rule stored in the transformation rule storage means, and generating a plurality of conversion candidates;
An evaluation value based on the evaluation scale is calculated using evaluation information stored in the evaluation information storage unit for a plurality of conversion candidates generated by the deformation processing unit, and each conversion candidate is represented by the calculated evaluation value. Evaluation processing means that evaluates and selects the expression with the highest evaluation value;
An output unit that outputs the conversion result of the selected expression as a sentence word converted into a target sentence style. When evaluating the expression of each conversion candidate, the evaluation processing means determines how many times the character string in the area range including the deformed portion of the character string deformed by the conversion candidate is included in the sentence set in the database. 2. The sentence-to-sentence conversion according to claim 1 , wherein an appearance frequency indicating whether it has appeared is used as an evaluation scale, and a higher evaluation value is given to the higher appearance frequency to evaluate the candidate expression for conversion. system. Ru sentences word written in natural language Oh, a system for converting the text word by another style described in the same natural language, a program for realizing a computer,
A modification rule that paraphrases the first character string described in the natural language into a second character string having the same meaning, and extracts a plurality of explanatory sentences of the same word extracted from a plurality of different dictionary files. Depends on the direction of conversion of the target style consisting of the first character string and the second character string automatically generated by the synonym or synonym phrase obtained from the result of matching A deformation rule storage means for storing a deformation rule that is not to be stored;
Evaluation information consisting of numerical information for evaluating whether the expression of the result of transforming the character string has the target style, functions or subroutines, rules describing the evaluation method, or information combining these The evaluation scale that prescribes the evaluation information gives a higher evaluation value as the appearance frequency or appearance probability of the conversion candidate in the example in the database storing the set of sentences by the conversion destination style is higher. Evaluation information storage means for storing evaluation information determined in advance,
An input means for inputting a character string to be converted in a natural language;
The character string the input deformation with the rules of deformation stored in the transformation rule storing means, and deformation processing means for generating a plurality of candidate conversion,
An evaluation value based on the evaluation scale is calculated using evaluation information stored in the evaluation information storage unit for a plurality of conversion candidates generated by the deformation processing unit , and each conversion candidate is represented by the calculated evaluation value. Evaluation processing means that evaluates and selects the expression with the highest evaluation value;
As an output means for outputting the conversion result of the selected expression as a sentence word converted to a target sentence style,
A text-to-sentence conversion program for making a computer function.

Images