JPH02255935A - Kana/kanji conversion system - Google Patents

Abstract

PURPOSE:To produce the KANJI (Chinese character)-KANA (Japanese syllabary) sentences with no mistake by performing the KANA-KANJI conversion for each paragraph and correcting automatically the improper result of conversion if detected by a prescribed number of times into another proper form. CONSTITUTION:A user inputs an unconverted KANA character string to an unconverted character string buffer 1, and this input character string is converted by a KANA/KANJI conversion system 2 and stored in a converted character string buffer 3. If the user is not satisfied with the result of conversion, the converted character string is corrected by a human intervention corrector 4 and the final corrected character string is obtained at a final character string buffer 5. A proper noun adoption/rejection controller 7 decides a case where the value is set with a parameter instruction of the user or with a manual setting operation or a case where the value is set by a proper noun learning computing element 6 with the use of a preference information setter 8. Then the controller 7 controls the system 2 to instructs the adoption or rejection of proper nouns. The element 6 analyzes systematically the information on the buffer 5 and controls the adoption/rejection of the subsequent proper nouns in terms of learning.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a Japanese kana-kanji conversion method, and in particular, to a kana-kanji conversion method that converts kana reading sentences containing proper nouns into sentences containing kanji and kana without errors. Regarding the method.

[Prior Art] Generally, in a kana-kanji conversion processing method for complex phrases on a workstation or word processor, a terminology dictionary is provided that stores conversion information of kanji for hiragana reading.
An algorithm performs automatic clause recognition processing on input Hiragana reading sentences, in which term units that match terms recorded in the term dictionary are regarded as clauses. In this type of kana-kanji conversion processing method, in order to convert proper nouns such as name terms, place name terms, and corporate corporation terms,
It is necessary to additionally register these proper nouns in the term dictionary. Note that Japanese Patent Laid-Open No. 55-43627, for example, discloses a prior art technique for determining the validity of connections between proper nouns in an input Japanese sentence written in kana and converting the proper nouns into kanji.

[Problems to be Solved by the Invention] In the above-mentioned conventional technology, the more proper nouns are added to the terminology dictionary, the more versatile it can be. Or, conversely, a clause consisting of a normal part of speech and a particle may be mistakenly recognized as a proper noun, which has an undesirable effect on the kana-kanji conversion process for a single sentence, resulting in a drop in performance.

For example, in a normal Kana-Kanji conversion processing method for sentences, a method of recognizing phrases using the longest-teaching method is used in the phrase recognition process described above. This method is a method in which when the same pronunciation (Kanayomi) has long and short kanji characters registered in the terminology dictionary, the one with the longest number of kanji characters is adopted, and the proper noun is the one with the longest number of kanji characters. Due to the long word length (number of kanji characters), there is a high possibility that proper nouns will be preferentially adopted over certain kana readings, which is a factor in the performance decline.

Figure 4 shows an example of the conventional normal kana-kanji conversion processing method. 11 is an input Japanese sentence with other readings, and 12 is a kana-kanji conversion unit that converts the input Japanese sentence into kana-kanji by setting phrases using an algorithm. , the conversion unit 12 selects "Ueno" from among the registered characters "Ueno" and "Kami" for the hiragana human power "Ueno" according to the longest-teaching method. Therefore, the character string 33 after conversion incorrectly contains the proper noun "Ueno".
is adopted. Therefore, the user needs to perform manual correction intervention 14 to obtain a correct sentence 15 containing kanji and kana.

On the other hand, in the kana-kanji conversion process of workstations, word processors, etc., depending on the application, more proper nouns such as surnames, place names, and company names may be required than in normal kana-kanji conversion processes. In some cases, a company directory may be created. In that case, a considerable number of proper nouns will be registered in the term dictionary. In this type of kana-kanji conversion process, it is better to use proper nouns than to use common parts of speech for the same kana reading.
The probability of making an error is lower.

Therefore, there are two types of kana-kanji conversion processing: a ``proper noun priority type,'' which prioritizes proper nouns in the terminology dictionary for the same kana-yomi input, and a ``proper noun priority type,'' which prioritizes proper nouns in the terminology dictionary. It is divided into the ``ordinary part-of-speech priority type,'' which prioritizes ordinary names, etc. (including combinations of particles), and it is possible to switch and adopt one of these types depending on the situation such as usage and target. Conceivable. However, with this method of selecting one of these fixed types in advance and then executing the conversion process, sometime during the process (for example, if a proper noun is inappropriately used in a sentence containing kanji and kana that the user expects) When it is found that the compatibility is not good (e.g. mixed with other types), the process must be repeated by changing to a different type, which is cumbersome.

Therefore, if it is found that the kana-kanji conversion of some phrases is inappropriate while performing the kana-kanji conversion process, it is possible to dynamically change the phrase recognition process to a different type each time. It is believed that the conversion is easily performed.

Therefore, an object of the present invention is to solve the above-mentioned problems of the prior art, and to make it possible to obtain proper nouns in places that should originally be proper nouns in sentences after kana-kanji conversion, and to make it possible to obtain proper nouns in places where proper nouns should originally be used. In addition to preventing proper nouns from being inappropriately mixed into a single location, the algorithm also performs automatic phrase recognition processing, and when inappropriate kana-kanji conversion is performed for multiple phrases, it can be learned from the next step. The purpose of this invention is to provide a kana-kanji conversion method that can dynamically change the conversion of ``kana'' to an appropriate format.

[Means for Solving the Problems] In order to achieve the above object, the basic concept of the kana-kanji conversion method of the present invention is to perform kana-kanji conversion by automatic phrase recognition processing, while also performing inappropriate kana-kanji conversion processing (each The present invention is characterized in that it is provided with means for dynamically changing to a more appropriate type of kana-kanji conversion method when the process (processing of clauses) has been performed a predetermined number of times.

Specifically, each time an inappropriate bunsetsu kana-kanji conversion process is discovered, it is manually corrected to an appropriate kanji-kana-mixed sentence, and when this manual intervention continues a predetermined number of times (for example, twice), , the selection criteria will be automatically revised so that appropriate parts of speech are preferentially adopted in the subsequent bunsetsu kana-kanji conversion process.

For example, if you are applying a proper noun type that preferentially adopts proper nouns from a terminology dictionary, and the results are appropriate, the proper noun type will continue as is (if a proper noun is used as a candidate for a clause in kana-kanji conversion). However, if the result of the selection is inappropriate twice in a row, it will be changed to the common part-of-speech type, which gives priority to the common part of speech. Conversely, if the common part-of-speech type is applied and the adoption result is appropriate or inappropriate, the form can be continued as is or changed to an appropriate type.

Furthermore, a control means (a proper noun acceptance/rejection controller) is provided for controlling whether a proper noun or a common part of speech is adopted or rejected as a clause candidate in kana-kanji conversion.

This control means allows for parameter indication or manual setting by the user, and the final Kanji-kana mixture obtained by the dynamic learning method described above, i.e., the result of kana-kanji conversion including corrective intervention by the user. Analyze the sentence.

It makes it possible to control the adoption or rejection of proper nouns based on the way proper nouns are used from a statistical perspective.

[Effect] The operation based on the above configuration will be explained.

In the kana-kanji conversion process, when inappropriate processing is performed a predetermined number of times, the learning function dynamically changes to the appropriate conversion method, so the uniqueness of name terms, place name terms, corporate corporation name terms, etc. A noun is not inappropriately mixed into a converted sentence containing kanji and kana, and a deterioration of the conversion function is prevented.

In addition, in the automatic clause recognition process using the algorithm,
By setting a proper noun acceptance/rejection controller that controls the acceptance or rejection of proper nouns, it is possible to adopt a proper noun when the subject is a proper noun, and to discard a proper noun otherwise. processing becomes possible.

The value of this control can be determined either by preferential parameters given from the outside (user) or by manual settings, or by statistically determining the state of existence of proper nouns in sentences containing kanji and kana, which are actually the final decision made by the user. It may be set based on analyzed learning information. In other words, it becomes possible to preferentially adopt or reject proper nouns such as surnames and terms in certain tasks based on instructions from the user.

In addition, if there is no instruction to adopt or reject these proper nouns,
A newly installed proper noun acceptance/rejection learning calculator analyzes the sentences containing kanji and kana, which are finally certified through the user's correction intervention on the results of kana-kanji conversion by the algorithm, and statistically shows the tendency for proper nouns to be adopted more frequently. If indicated,
A learning function is provided to set the value of the controller that determines whether the proper noun term is adopted or not to be adopted, and otherwise set it to be rejected. As a result, the control value is set in a feedback manner, and the user can automatically create the desired sentence using kana-kanji conversion without having to set the state one by one. It becomes possible to set acceptance/rejection in an optimal state.

[Example] An embodiment of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

In the figure, 1 is a character string buffer before conversion, 2 is a kana-kanji converter system, 3 is a character string buffer after conversion, 4 is a human intervention corrector, 5 is a final character string buffer, and 6 is a proper noun acceptance/rejection learning calculator. , 7 is a proper noun adoption/rejection controller, 8 is a priority information setter, and 9 is an initializer.

The user inputs the pre-converted kana character string into the pre-conversion character string buffer 1, and obtains the converted character string in the post-conversion character string buffer 3 by the kana-kanji conversion system 2, but if the user is dissatisfied with this conversion result, A final character string is obtained in a final character string buffer 5 through correction by a manual intervention corrector 4. The proper noun acceptance/rejection controller 7 has a value set by the user's parameter instruction or manual setting by the priority information setting device 8, and a value is set by the proper noun learning calculator 6, and the kana-kanji conversion system 2 and instructs whether or not to use proper nouns. The proper noun acceptance/rejection learning calculator 6 statistically analyzes the information in the final character string buffer 5 and controls the acceptance/rejection of subsequent proper nouns in a learning manner. 9 is the initializer for 6 and 7.

FIG. 2 is a flowchart for explaining the proper noun learning function within the proper noun learning calculator 6 used in one embodiment of the present invention.

In the figure, 61 is a proper noun learning acceptance/rejection register (counter) which forms the main part of the proper noun learning calculator 6;
It consists of n bits from 2'' to 2''. The numerical symbols t, ~t5, etc. of the register 61 at each position on the left side of the figure are as follows:
The values at each corresponding step of the flow on the right are shown.

In the figure, first, t and t of the register 61 are all set to O as initial values. If there is no correction operation such as homophone selection or reconversion (step 21)
Since this is when the conversion result desired by the user has been obtained, the conversion result is adopted as is and taken out as an output, and the operation is immediately transferred to the proper noun acceptance/rejection controller 7.

In step 21, when correction operations such as homophone selection and reconversion are performed by the user's manual intervention (by the manual intervention corrector), the correction results are extracted as output, and
Check whether the corrected clause is a clause that includes a proper noun before or after the correction (step 22);
If it is determined by the check that the phrase does not include a proper noun, the process immediately moves to the next phrase (step 26), but if it is determined that the phrase does include a proper noun,
The register 61 is shifted to the left by 1 bit, and the free rightmost bit is set to 0 (step 23).
)1 Next, check whether the correction result is adopted as a proper noun (step 24).If it is not adopted as a proper noun, the value of register 61 is not changed, but if it is adopted as a proper noun, the value of register 61 is changed. Set the rightmost bit to 1 (change from O to l) (step 25) 6 In other words, the bit value of register 61 is set to I when a proper noun is adopted.
I I II, it On is set when a proper noun is not adopted.

On the other hand, the proper noun adoption/rejection controller 7 first checks whether a proper noun is included in the clause as the first candidate (step 31), and if it is not included, moves to the next process (step 36).
, when it is included, it is checked whether the current type is the proper noun preferential adoption type (step 32).It is optional whether the controller 7 is first set to the proper noun preferential adoption type or the other type. This setting can be made by the user by referring to the priority information setting unit 8. No matter which type is initially set, through repeated learning, the type adapted to the purpose and field of the target text will be adopted, and appropriate kana-kanji conversion processing will be performed. go. If the proper noun type is adopted in step 32, a proper noun is selected immediately (step 34), but if the proper noun type is not adopted, the register 61 is checked (step 33),
If both the 2nd and 21st digits are on (bit Jjj), a proper noun is selected (step 34).

In other words, in general, when manual intervention for adoption or rejection of a proper noun occurs i times in a row, the acceptance or rejection mode in the subsequent kana-kanji conversion changes in a learning manner. If i=2, 2', 2@bits store the state mode, and i
=3, then 22.212 fi bits store the state mode. In the embodiment shown in FIG. 2, if i=2 and a proper noun is adopted due to two consecutive manual interventions (at step 24), a proper noun will be selected from then on (step 34). ), step 33, bit 212
If both ° are not on, kana-kanji conversion is performed for the second and subsequent candidates (other than proper nouns) (35).

FIG. 3 shows that by controlling the adoption or rejection of proper nouns according to the present invention,
In the same figure, which shows the overall flowchart when the desired kana-kanji conversion is performed, in step 41, the phrase candidates to be recognized (for the input "Ueno", the proper noun "Ueno" or the common noun on the particle combination r)
is extracted, and in step 42, it is checked whether there is a proper noun in the extracted clause candidate. If there is no proper noun, the process immediately moves to step 45, where the longest clause is set using the teaching method, but if there is a proper noun, step Moving on to 43.

In step 43, it is checked whether the proper noun is not adopted, and if NO, the process immediately moves to step 45;
If S, the proper noun "Ueno" is removed from the clause candidates, and the common noun/particle combination "on r" is to be adopted (step 44).

In addition, in the example shown in FIG. 2 above, a case was shown in which i = 2 and correction was made to adopt a proper noun through two consecutive manual interventions, but i may be set to an appropriate value of 2 or more. You can do it like this. Setting i=1 is not appropriate because the molds are changed every time and the system may vibrate and the operation may become unstable.

Furthermore, in the above embodiment, two formats are used: a proper noun priority type and a proper noun non-priority type.
Unless there is a change in the type, other proper nouns (for example, ``Nakano'') will also be given priority.

The user can set the value of the controller 7 using the priority information setting device 8 at any time, and this value can be fixed at the user's will by parameters or manual settings. For purposes such as creating a directory, it is possible to always preferentially use first and last names, which are proper nouns, and to eliminate the inclusion of proper nouns when creating regular sentences.

[Effects of the Invention] As described in detail above, according to the kana-kanji conversion method of the present invention, when an inappropriate conversion result is found a predetermined number of times when each clause of a sentence is converted into kana-kanji using a certain type, There is a learning function that automatically changes this to another more appropriate type and performs the subsequent kana-kanji conversion process, so it can quickly and flexibly respond to the usage field and type of input Japanese text, and always perform appropriate kana-kanji conversion. This has the effect of generating error-free sentences containing kanji and kana. In particular, whether or not to adopt proper noun information such as surname terms, place name terms, corporate and corporate name terms, etc. that have been adopted in order to improve the versatility of the terminology dictionary is determined by preferential designation or by the kanji-kana characters that are the result of the conversion process. This is determined by analysis of mixed sentences and statistical learning functions, and for input Japanese sentences that require a lot of proper noun information such as a list of people, proper nouns are given priority, and normal Japanese sentences are prioritized. By adopting a proper noun non-priority or proper noun exclusion type for , it is possible to prevent proper nouns from being accidentally mixed into output sentences containing kanji and kana.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, and FIG. FIG. 3 is a flowchart showing the state of kana-kanji conversion according to an embodiment of the present invention, and FIG. 4 is a diagram showing an example of a conventional kana-kanji conversion processing method. 1...Character string buffer before conversion, 2...
Kana-kanji conversion system, 3... Post-conversion character string buffer, 4... Manual intervention corrector, 5...
- Final string buffer. 6... Proper noun acceptance/rejection learning calculator, 7...
・Proper noun adoption/rejection controller, 8...Priority information setting device, 9...A2 Diagram Diagram Diagram

Claims (1)

[Scope of Claims] 1. A kana device that has a means for inputting the reading of a Japanese sentence and a term dictionary that holds conversion information between the reading and kanji, and converts the input reading of the Japanese sentence into a sentence containing kanji and kana. In a kana-kanji conversion method that is equipped with a kana-kanji conversion means, if a certain format performs kana-kanji conversion by clause setting processing and a predetermined number of inappropriate kana-kanji conversions occur, another appropriate format of kana-kanji is converted. A kana-kanji conversion method characterized by changing to conversion. 2. It has a means for inputting the pronunciation of a Japanese sentence, a term dictionary that holds conversion information between the pronunciation and kanji, and a kana-kanji conversion means for converting the input pronunciation of the Japanese sentence into a sentence containing kanji and kana. In the kana-kanji conversion method, a controller is provided to control whether or not proper nouns are adopted by the kana-kanji conversion means, and the error controller is a proper noun adoption type control when a Japanese sentence mainly consisting of proper nouns is input. When a Japanese sentence that does not consist of proper nouns is input, control is performed using a control format that does not include proper nouns, and furthermore, the control format of the controller is controlled using statistical learning. A computing device is provided which statistically analyzes a sentence containing kanji and kana that has been finally determined through correction intervention, and based on the analysis result, determines a control format that is suitable for the final sentence containing kanji and kana. A kana-kanji conversion method characterized by being configured to select. 3. The kana-kanji conversion system according to claim 2, further comprising means for setting the control format of the controller using externally given parameters or manually, in addition to the arithmetic unit.