JPH05324607A - Method and device for converting japanese syllabary/ chinese character - Google Patents

Abstract

PURPOSE:To attain a meaning analysis, and to improve a clause division recognizing efficiency by operating the analysis of a sentence by using not only conventionally used format information but also meaning information, at the time of researching a clause candidate in a clause division processing. CONSTITUTION:The research of the clause candidate is operated to the clause candidate prepared by a basic analyzing part 62a, by using the format information such as word information 52, grammar information 51, and experience rule by a format information referring part 62b-1 of a clause division processing part 62. Then, the research is further operated to the searched clause candidate, based on the meaning information from a cooccurrence example information 53 by meaning information referring part 62b-2, and a clause dividing position for the conversion candidate is decided. Then, the notation of the decided clause is decided by a notation selecting part 62c, and it can be used as a conversion candidate character column.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a kana-kanji conversion method and device for converting an input reading character string into a kana-kanji mixed character string.

[0002]

2. Description of the Related Art In the conventional kana-kanji conversion method, the conversion result is obtained in the following procedure. First, a phonetic character string to be converted is input, and a conversion instruction key is used to instruct execution of kana-kanji conversion. Upon receiving the conversion instruction, the conversion candidates are displayed to the user. Here, if the intended conversion candidate has not been obtained, the user obtains the intended conversion candidate by displaying the next candidate. If the segment break is different from the intended one, the user re-segments the phrase and then displays the conversion candidates again. In this way, when the intended conversion candidate is displayed, the conversion candidate is confirmed as the conversion result by the confirmation instruction key. The conversion candidate thus determined is output as the conversion result.

Next, the data processing procedure from the above conversion instruction to the display of conversion candidates will be described in more detail.

When a conversion instruction is input, first, a basic analysis of a read character string is performed. In this basic analysis, with respect to the reading to be converted, a dictionary search for a word (stem) based on word information is performed. Then, a combination of all bunsetsu candidates that can be established as a bunsetsu is created from the searched word and the grammatical information.

Next, an optimal combination of clauses is selected from the clauses obtained by the basic analysis. Here, the part-of-speech information obtained from the word information, the priority information of the dependency between the phrases obtained from the grammar information, the rule of thumb for narrowing down the combination candidates, etc. are referred to and the phrase is narrowed down to finally Extract a combination of one clause.

With respect to the phrase combination determined as described above, the notation in each phrase is determined. That is, the kanji notation is determined for each phrase. Here, matching is performed based on the co-occurrence example information, and the notation in which the matching is established is preferentially adopted. At this time, the homophone processing by the co-occurrence example is executed. Then, with respect to the phrase for which the matching based on the co-occurrence example information has not been established, the optimum notation is selected by referring to the word information and the learning information. That is, the most recently used notation is preferentially adopted based on the learning information, or the notation is determined by the order registered in advance in the dictionary. The conversion candidates are determined as described above.

Here, the co-occurrence example information is binary relation information between words having a strong semantic connection. And with this co-occurrence example information, besides simply connecting two words,
It includes examples of co-occurrence in attribute units such as “person” and “flower”, and additional restriction information. As additional restriction information,
For example, there is particle information such as "wo" and "ga", and information about the direction of establishment. Examples of co-occurrence examples are: "Hot-summer", "Thick-book", "Hot-hot water" ... (word connection) "Meet a person (he, she, teacher, lover, etc.)-" (Example of co-occurrence by attribute unit) "Flowers (tulips, chrysanthemums, etc.) bloom" ... (Example of co-occurrence by attribute unit) "Talk-listen", "Medicine-effective", "Twilight-difficult" …
(Particle restriction information) "Home-education", "Education-process" ... (direction restriction information). The co-occurrence example information as described above is semantic information, and by using this, the polysemy of homophones is eliminated.

Next, homophone processing by co-occurrence example information will be described. For example, when performing a conversion process on the reading "Atsu / hon", "thick" is selected from the connection information "thick-book" in the co-occurrence example information,
"Hot and hot" etc. are not selected. Therefore, "thick /
The conversion result "book" is immediately obtained. In this way, the conversion efficiency is improved by using the co-occurrence example information.

[0009]

However, in the above-mentioned conventional example, for example, when a conversion instruction is executed by inputting a reading character string of "make a child read a book" (read a book to a child), The phrase segmentation position may be determined, such as "/ read / listen to Japan". This is only formal information such as rules based on grammatical information (such as giving priority to nouns and nouns that do not have particles) and empirical rules (such as giving preference to candidates with a small number of clauses) when deciding punctuation points. It occurs because processing is executed based on. That is, such a wrong conversion candidate is displayed because the delimiter position of the phrase is determined based only on the format information without analyzing the semantic connection between words.

In such a case, the user can obtain a desired conversion result by manually re-segmenting the clause and giving a conversion instruction again. In this way, when the bunsetsu delimiter position is incorrect, the work of re-separating the bunsetsu always occurs, and the operation is very troublesome.

Further, since the homophone selection process in the conventional kana-kanji conversion method is executed after the phrase segmentation process,
As described above, when the phrase segmentation process is different from the user's intention, the homophone selection process is all wasted.

The present invention has been made in view of the above problems, and when narrowing down bunsetsu candidates in bunsetsu segmentation processing, in addition to the format information that has been conventionally used, a sentence analysis using semantic information is performed. It is an object of the present invention to provide a kana-kanji conversion method and device in which semantic analysis is enabled by performing the above and the phrase segmentation recognition rate is improved.

[0013]

A kana-kanji conversion device according to the present invention for achieving the above object has the following configuration. That is, in a kana-kanji conversion device for converting an input phonetic character string into a kana-kanji mixed character string, a semantic information storage means for storing semantic information regarding a semantic connection between words and the reading using the word information and the grammatical information. Generation means for generating bunsetsu candidates from a character string, first narrowing means for narrowing down bunsetsu candidates using format information for the bunsetsu candidates, and the meaning information storage means for the bunsetsu candidates. Based on the phrase candidates selected as a result of narrowing down the phrase candidates by the first narrowing unit and the second narrowing unit, the second narrowing unit narrowing down the phrase candidates using the semantic information. A notation determining means for determining the notation of the conversion candidate.

The Kana-Kanji conversion method according to the present invention for achieving the above object includes the following steps. That is,
In a kana-kanji conversion method for converting an input phonetic character string into a kana-kanji mixed character string, a semantic information storage means for storing semantic information regarding a semantic connection between words and the phonetic character string using word information and grammatical information. A generation process for generating more bunsetsu candidates, a first narrowing process for narrowing down bunsetsu candidates by using format information for the bunsetsu candidates, and the meaning information storage means for the bunsetsu candidates. A conversion candidate based on a second narrowing stroke for narrowing down phrase candidates using the semantic information, and a phrase candidate selected as a result of narrowing down the phrase candidates by the first narrowing stroke and the second narrowing stroke. And a notation determination step for determining the notation of.

The formal information in the present invention is information about a format used when narrowing down bunsetsu candidates, and includes grammatical information (whether or not a grammatical connection between words or bunsetsu is correct and its strength, etc.). ) Or empirical rules (such as giving priority to candidates with a small number of clauses), it refers to information that does not have any semantic elements.

[0016]

In the above structure, first, a plurality of phrase candidates are generated by the generating means. Next, in the first narrowing-down means, the phrase candidates generated by the generating means are narrowed down by a priority rule based on formal information such as word information, grammatical information, and empirical rules. Next, the second narrowing-down means further narrows down the phrase candidates narrowed down by the first narrowing-down means on the basis of the semantic information, and determines a phrase-breaking position to be a conversion candidate. And
The notation determining means determines a notation for the determined clause and sets it as a conversion candidate character string.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A preferred embodiment of the present invention will be described below with reference to the accompanying drawings.

FIG. 1 is a block diagram showing the schematic arrangement of a document processing apparatus according to this embodiment. In the figure, reference numeral 1 denotes a CPU, which executes various processes of the document processing apparatus.
Reference numeral 2 is a keyboard for inputting reading character strings and various inputs such as conversion instructions and confirmation instructions. 3 is a CRT,
Display the document you are inputting, Kana-Kanji conversion candidates, etc. Four
Is a ROM in which various control programs executed by the CPU 1 are stored. Reference numeral 5 denotes a dictionary, which stores various information referred to when performing kana-kanji conversion processing.
A RAM 6 has a work area for the CPU 1 to execute various processes, a kana-kanji conversion unit 60, and the like. The Kana-Kanji conversion unit 60 also includes an input / output interface unit 61 for Kana-Kanji conversion processing and a conversion engine unit 62 for executing Kana-Kanji conversion. The above-mentioned units are connected to the bus 7 and exchange data with each other.

FIG. 2 shows a kana-kanji conversion unit 60 (input / output interface unit 61 and conversion engine unit 62) and a dictionary 5.
It is a figure showing the detail of.

First, in the input / output interface unit 61, a determination unit 61a determines whether the input from the keyboard 2 is a read character string, a conversion instruction, a confirmation instruction, or the like. A storage unit 61b stores a reading character string input from the keyboard 2 and a conversion candidate character string input from the conversion engine unit. A display unit 61c displays the reading character string, the conversion candidate character string, and the like stored in the storage unit 61b on the CRT 3.

Further, in the conversion engine section 62, 62
Reference numeral a is a basic analysis unit, which obtains all possible combinations of punctuation positions from grammatical information and word information in the dictionary 5. Reference numeral 62b is a phrase segmentation processing unit, which includes grammar information,
A format information reference unit 62b-1 that narrows down phrase break candidates based on format information such as word information, and a semantic information reference unit 62b that narrows down phrase breaks using co-occurrence example information and executes homophone selection processing. -2 is provided. 6
Reference numeral 2c is a notation selecting unit, which determines the notation for a phrase whose notation has not been determined by the phrase segmentation processing unit 62b.

The dictionary 5 includes grammar information 51 and word information 5
2, co-occurrence example information 53 and learning information 54 are stored. The grammatical information 51 includes information on the correctness of the grammatical connection between words and information on the strength of the connection. For example, "noun (no particle) + suffix has strong ties, noun (no particle) + noun has strong ties"
Format information such as is stored. The word information 51 includes part-of-speech information and utilization information of each word. The co-occurrence example information 53 is binary relation information between words having a strong semantic connection, and details thereof have been described in the conventional example. This co-occurrence example information is referred to as semantic information in the phrase segmentation process. A considerable number of such co-occurrence example information is prepared, and hundreds of thousands of information are recorded in this document processing device. The learning information 54 is information for preferentially adopting the most recently used notation.

The operation of the document processing apparatus having the above configuration will be described. FIG. 3 is a flowchart showing the procedure of the kana-kanji conversion operation by the document processing apparatus of this embodiment.
This process is started when there is any key input from the keyboard 2.

In step S11, the determination unit 61a determines whether the key input from the keyboard 2 is a character input. If it is a character input, the process proceeds to step S12. In step S12, the input character is stored in the storage unit 61b as a reading character string, and the process proceeds to step S13. On the other hand, when it is determined that the key input input in step S11 is not character input, step S1
Go to 3.

In step S13, the determination section 61a determines whether or not the key input is a conversion instruction, and if it is not a conversion instruction, the process returns to step S11 to repeat the above processing. If the key input is a conversion instruction, the process proceeds to step S14. In step S14, the phonetic character string stored in the storage unit 61b is transferred to the conversion engine unit 62, and the conversion engine unit 62 executes the conversion process on this phonetic character string. In step S15, the conversion candidate character string obtained by the conversion engine unit 62 is stored in the storage unit 61b and displayed on the CRT 3 by the display unit 61c as a conversion candidate.

Then, in step S16, the determination unit 61a
Judges whether or not there is a key input for instructing confirmation from the keyboard 2, and if there is no confirmation instruction, returns to step S14. Further, if there is a confirmation instruction, the process proceeds to step S17 and is output as the conversion result. Although omitted in this figure, if the conversion candidate is not the desired one, step S
At 14, it is assumed that the next candidate is acquired or the segmentation is performed again.

Next, the conversion process in step S14 will be described in more detail with reference to the flowchart of FIG. FIG. 4 is a flow chart showing the procedure of kana-kanji conversion processing by the document processing apparatus of this embodiment.

In step S21, the basic analysis section 62a is referred to by referring to the grammatical information 51 and the word information 52 of the dictionary 5.
Generates bunsetsu candidates. First, a word (stem) is extracted from the reading character string by performing a dictionary search. The part-of-speech of the extracted word is obtained from the word information 52, and from this part-of-speech information and grammatical information 51, the processing of the inflectional ending and the adjunct word is performed.
In this way, all clauses that can be grammatically established are extracted, and combination candidates of a plurality of clauses are generated.

In step S22, the format information reference section 62b-1 in the clause segmentation processing section 62b adds points to each of the candidate combinations of clauses generated in step S21 using the format information to set the priority. calculate. Here, the formal information is information obtained from the word information 52 (information on the part of speech of each word, utilization, etc.), grammar information 5
Information obtained from 1 (information such as correctness of grammatical connection between words, strength of connection, etc.) and empirical rule (information such as giving priority to a candidate having a small number of clauses). Then, in step S23, only bunsetsu candidates having a predetermined priority are extracted based on the calculated priorities, and bunsetsu combination candidates are narrowed down.

In step S24, the semantic information reference unit 62b-2 adds points to each of the candidate combinations of clauses narrowed down in step S23 using the semantic information to calculate the priority. Here, the co-occurrence example information is referred to check whether there is a match, and if there is a match, the priority score is added so as to increase the possibility that the phrase segment candidate will be selected. .. In step S25, the phrase delimiter position in the conversion candidates is determined by adopting the phrase candidate having the highest priority as described above. And step S
In 26, in the clause adopted in step S25,
If there is a match with the co-occurrence example information, the notation of that phrase is determined to execute the homophone selection process. Substantially, at step S25, when a phrase that matches the co-occurrence example information is detected, the notation is determined for this phrase by the co-occurrence example information.

Next, in step S27, the notation selecting unit 62c sets word information and grammatical information for the phrase not determined in step S26 (that is, the phrase whose notation is not determined by the homophone selection process). , Determine the notation by referring to the learning information. Then, in step S28, it is output to the input / output interface unit 61 as a conversion candidate.

Next, the determination of the conversion candidates by the above-mentioned processing will be described by giving a specific example of the reading character string.

For example, as shown in FIG. 5, a case will be described in which a sentence (reading character string) 101, "To read children's books" is input. It is assumed that the document processing apparatus has a setting such as "read a book" as co-occurrence example information. As a result of the narrowing-down process based on the format information from step S21 to step S23 in FIG. 4 described above, the candidate sentences 102 and 103 are extracted as candidates for the phrase combination. In the conventional Kana-Kanji conversion process, bunsetsu candidates are determined only by the formal information, so if there is a rule that prioritizes the connection of "noun (no particle) + noun" in this formal information, "child" and "Japanese" The tie of “” becomes stronger and the phrase candidate 102 is selected. Then, as a result, the conversion candidate sentence 104 is obtained.

However, in the document processing apparatus, as a result of the narrowing-down process based on the semantic information in steps S24 to S25, the co-occurrence example information "reading a book" matches, so the phrase candidate 103 is selected. Further, the notation is determined in step S26, and the conversion candidate sentence 105 is obtained.

As described above, in the document processing apparatus of this embodiment, the phrase combination candidate is selected using the co-occurrence example information in addition to the format information in the phrase segmentation process. As a result, the phrase break recognition rate is significantly improved compared to the case where the phrase break position is determined only by the format information, so that the conversion efficiency is greatly improved. Also,
For a phrase whose connection has been confirmed by the co-occurrence example information, the notation for that phrase has already been performed, and the phrase segmentation process and the homophone selection process are integrated. By this integration, it is possible to reduce the useless conversion process of the conventional homophone selection process after the erroneous phrase segmentation process.

The user may additionally register the binary relation to the information registered in advance as the co-occurrence example information, or the co-occurrence example information may be added by the learning function. It is also possible.

[0037]

As described above, according to the kana-kanji conversion method and device of the present invention, a sentence is analyzed using semantic information in addition to the format information conventionally used in the phrase segmentation processing. This enables semantic analysis and improves the phrase break recognition rate.

[0038]

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a document processing apparatus according to an embodiment.

FIG. 2 is a block diagram showing a functional configuration of a kana-kanji conversion unit according to the present embodiment.

FIG. 3 is a flowchart showing a procedure of a kana-kanji conversion operation of the present embodiment.

FIG. 4 is a flowchart showing a procedure of kana-kanji conversion processing according to the present embodiment.

FIG. 5 is a diagram showing an input example of a reading character string, a phrase segment candidate, and a conversion result.

[Explanation of symbols]

 1 CPU 2 keyboard 3 CRT 4 ROM 5 dictionary 51 grammar information 52 word information 53 co-occurrence example information 54 learning information 6 RAM 61 input / output interface unit 62 conversion engine unit

Claims (6)

[Claims] 1. A kana-kanji conversion device for converting an input phonetic character string into a kana-kanji mixed character string, using semantic information storage means for storing semantic information regarding a semantic connection between words, and word information and grammatical information. Generating means for generating bunsetsu candidates from the reading character string, first narrowing means for narrowing bunsetsu candidates using format information for the bunsetsu candidates, and the meaning information storing means for the bunsetsu candidates Second narrowing-down means for narrowing down bunsetsu candidates using the semantic information stored in, and a bunsetsu candidate selected as a result of narrowing down bunsetsu candidates by the first narrowing means and the second narrowing means. A kana-kanji conversion device, comprising: a notation determining means for determining a notation of a conversion candidate based on. 2. The semantic information stored in the semantic information storage means is co-occurrence example information representing binary relation information between words having a strong semantic connection. Nokana-Kanji conversion device. 3. The kana-kanji conversion device according to claim 2, wherein the notation determining means preferentially adopts the notation in the co-occurrence example information adopted by the second narrowing means. 4. A kana-kanji conversion method for converting an input phonetic character string into a kana-kanji mixed character string, using semantic information storage means for storing semantic information regarding a semantic connection between words, and word information and grammatical information. Generating process for generating bunsetsu candidates from the reading character string, a first narrowing process for narrowing bunsetsu candidates using format information for the bunsetsu candidates, and the semantic information storage unit for the bunsetsu candidates A second narrowing step for narrowing down the bunsetsu candidates using the semantic information stored in, and a bunsetsu candidate selected as a result of narrowing down the bunsetsu candidates by the first narrowing stroke and the second narrowing stroke. A kana-kanji conversion method comprising: a notation determination process for determining the notation of a conversion candidate based on. 5. The semantic information stored in the semantic information storage means is co-occurrence example information representing binary relation information between words having a strong semantic connection. Nokana-Kanji conversion method. 6. The kana-kanji conversion method according to claim 5, wherein in the notation determining process, the notation in the co-occurrence example information adopted in the second narrowing process is preferentially adopted.