JP2001216294A - Kana/kanji conversion method, kana/kanji converter, and recording medium having kana/kanji conversion program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform a suitable Kana (Japanese syllabary)/Kanji (Chinese character) conversion in a small-sized equipment by performing the high-speed processing of writing with a space between words. SOLUTION: This method is composed of a step for extracting a clause shape based on a dictionary, a step for selecting the extracted clause shape by a read length and turning it to a front clause candidate, a step for extracting the clause shape following the front clause candidate based on the dictionary, turning it to a rear clause candidate and extracting the longest two clauses composed of the combination of the front clause candidate and the rear clause candidate, the step for selecting one of the longest two clauses in the case that two or more of the longest two clauses are extracted and the step for performing Kanji conversion for the front clause.

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, a kana-kanji conversion device, and a recording medium on which a kana-kanji conversion program is recorded. The present invention relates to a kana-kanji conversion device that can be used when inputting a.

[0002]

2. Description of the Related Art As a means for inputting Japanese in a computer or a Japanese word processor, a kana-kanji conversion method is used in which a solid kana character string is inputted as a reading and converted into a sentence mixed with kanji kana. . In the kana-kanji conversion process, it is necessary to perform a segmentation process for dividing an input reading character string into words. As a representative method of such segmentation processing, the segmentation method using the two-phrase longest-matching method is described in Reference 1 ("Separation of solid sentences and kana-kanji conversion-segmentation by the two-phrase longest-matching method", Makino, Kizawa, IPSJ Transactions Vol.20 No.4 pp.337-345. (19
79)), and it is known that conversion accuracy is good. Further, a technique in which the technique of Reference 1 is improved to save storage area and speed up is described in Reference 2 (Japanese Patent Laid-Open No. 62-25195).
9).

[0003] Referring to FIG. 14, a description will be given of a dividing method disclosed in Reference 2. In step G1, the brute force phrase candidate extraction means performs phrase conversion on the input character string from the beginning of the sentence to the brute force. The brute-force phrase candidate extraction means extracts only the longest possible phrase for each independent word. Next, in step G2, priority is given to all the phrase candidates obtained in step G1 by the priority order determining means. Specifically, the longest independent word is selected from the longest clause length clauses, and is set as the first-ranked clause. Similarly, the next longest phrase is the second-ranked phrase.

[0004] The first-ranked phrase and the second
The two clause longest matching method is applied to the second clause in steps G3 and G5, and the lengths of the two clauses obtained as a result are stored in variables l and m in steps G4 and G6, respectively. Next, in step G7, the variable 1 and the variable m are compared. If either value is larger, the larger value, that is, the longer one of the two clauses is adopted and stored in the clause area in step G12. If the value of the variable l is equal to the value of the variable m in step G7, another measure (independent word length,
Frequency information) to determine one of the solutions. Further, in step G13, it is determined whether or not all input character strings have been processed, and if there is an unprocessed character string, the process returns to step G1 to repeat the processing.

[0005] According to the method of the above-mentioned document 2, the first phrase (hereinafter referred to as the "previous phrase") for searching for the longest two phrases is subjected to brute force conversion by the brute force phrase candidate extraction means, and priority is determined. A priority is given by means, and the two-phrase longest matching method is applied only to the previous phrase up to a predetermined order. Since the conversion of the second clause (hereinafter referred to as the second clause) following the first clause is attempted after limiting the first clause, the number of searches for the longest two clauses is reduced, and the processing speed is significantly faster than the method of Reference 1. It is said to improve.

[0006]

However, in the method of Reference 2, although the preceding clause to be processed is limited by the priority order, the following clause is located at each position where there is a possibility that a clause is separated in the middle of the preceding clause. In search of the longest two phrases. Therefore, if the preceding phrase is a long phrase containing many adjuncts, there is a large number of character positions for which the subsequent phrase is required, and there is a problem that the effect of speeding up by limiting the preceding phrase cannot be obtained. .

For example, in the case of converting the input "Tomorrow's size", according to Document 2, as a result of phrase extraction from the beginning of the sentence by brute force, the longest "Tomorrow's size" is the first.
The second longest phrase is "Asuka". Here, the delimitation of "from tomorrow" may seem a bit unnatural, but it is interpreted as a noun-equivalent phrase in "from tomorrow", and as an expression expressing the question of colloquial tone in "~ kai". The interpreted delimiter. In the appendix processing for phrase extraction, a connection table is generally used to determine whether or not two adjacent adjuncts can be connected. Therefore, it is necessary to avoid including long phrases due to excessive adjunct connection in the phrase extraction result. I can't.

[0008] According to the method of Document 2, the first-ranked phrase "from tomorrow" can be a boundary of a phrase "from tomorrow""fromtomorrow""fromtomorrow""fromtomorrow""fromtomorrow" A subsequent phrase is extracted for each of the six positions “K” and “Tomorrow”, and the longest two phrases are obtained. The boundary of the phrase in the second-ranked phrase “Asuka” is one of the above six types. Even if the re-calculation is avoided by sharing the processing result of the subsequent phrase starting from the same position on the memory, it is necessary to extract the subsequent phrase in at least six different starting positions. On the other hand, in the method of Reference 1 which does not limit the preceding clause, “Asuka no Kaisai”
Since a post-segment following "A" is also obtained in addition to the street, the post-segment is extracted at seven different start positions. In other words, in the above example, the seven types are limited to only six types, and the effect of speeding up is hardly obtained.

The present invention has been made to solve the above problems, and extracts not only the longest phrase form but also all possible phrase forms for each independent word, and extracts the extracted previous phrase. Are sorted by phrase length to form a phrase block, the phrase blocks that are candidates for the previous phrase are limited to a predetermined order of the phrase length, and the subsequent phrase that follows each phrase block is extracted. The purpose of the present invention is to increase the speed and save memory by applying the two-clause longest matching method. For example, in the above example, the two phrase longest match method is applied to the four phrases “Limiting from tomorrow”, “Regarding from tomorrow”, “Regarding from tomorrow”, and “Regarding from tomorrow”.
Since it is only necessary to try combinations less than the six combinations in the above, it is possible to increase the speed.

As described above, it is general that the conversion accuracy is reduced by increasing the limit of the solution search range. In this case, it is confirmed that the conversion accuracy is not reduced by a conversion experiment using a large number of example sentences. I know from experience. this is,
If a phrase with a large number of adjuncts, such as "Take from tomorrow", is obtained as a candidate for the preceding sentence, a combination of the adjunct parts, for example, reading such as "Kara-no-kakai", "Rano-kai", or "No-kai" Is very unlikely to be included in the independent word dictionary, and the correct answer candidate can be obtained without finding the post-sentence at all the delimiter positions of each attached word.

[0011] As described above, an object of the present invention is to perform faster segmentation processing without lowering the conversion accuracy or increasing the memory capacity than the method disclosed in Document 2, and to increase the memory capacity of a mobile phone or the like. Small, high-speed CP
An object of the present invention is to provide a kana-kanji conversion device suitable for small devices in which U cannot be used.

[0012]

According to the present invention, there is provided a kana-kanji conversion method comprising the steps of: extracting a phrase starting from the beginning of a kana character string or a known phrase break based on a dictionary; A phrase type limitation step of selecting a phrase according to the reading length and setting it as a preceding phrase candidate, and extracting a phrase type following the preceding phrase candidate based on a dictionary as a subsequent phrase candidate, and comprising a combination of a preceding phrase candidate and a subsequent phrase candidate. A longest two clause extraction step of extracting the longest two clauses, and a longest two clause selection step of selecting one of the longest two clauses when two or more longest two clauses are extracted in the longest two clause extraction step. A kanji conversion step of determining a bunsetsu break based on the obtained longest two phrases and performing kanji conversion on the preceding phrase.

It is unlikely that a phrase having a short reading length is included in the longest two phrases. For this reason, by restricting the phrase form that is the preceding phrase candidate in the phrase length limitation step according to the reading length, it is possible to reduce the number of character positions from which the phrase form should be extracted in the maximum two phrase extraction step.
Therefore, the kana character string can be separated and written at high speed.

Further, the kana-kanji conversion method according to the present invention comprises:
The phrase extraction step includes an independent word extraction step of extracting an independent word, an independent word limitation step of selecting the extracted independent word according to an independent word reading length, and an auxiliary word following each selected independent word. Is extracted.

It is unlikely that a phrase composed of independent words having a short reading length is included in the longest two phrases. For this reason, in the independent word limiting step, by limiting the independent words according to the reading length, it is possible to reduce the number of independent words from which the additional words should be extracted in the additional word extracting step. In addition, the phrase forms that are the previous phrase candidates are also reduced, and the character positions from which the phrase forms should be extracted can be reduced in the longest two-phrase extraction step.

Further, the kana-kanji conversion method according to the present invention comprises:
The independent word limitation step finds a maximum reading length of the extracted independent word, and determines an independent word having a reading length whose difference from the maximum reading length is within a predetermined character (preferably within 4 or 5 characters). select.

In comparison with other independent words, it is unlikely that a phrase composed of independent words whose reading length is sufficiently short is included in the longest two phrases. Therefore, by excluding such independent words having a short reading length from the candidates, it is possible to reduce the processing target of the attached word extraction step and the longest two-phrase extraction step without remarkably lowering the conversion accuracy. .

Further, the kana-kanji conversion method according to the present invention comprises:
The above phrase form limitation step selects phrase forms from the longest phrase reading length to a predetermined order (preferably the third or fourth place).

Further, the kana-kanji conversion method according to the present invention comprises:
The phrase form limitation step finds a maximum reading length of the extracted phrase form, and determines a phrase form having a reading length whose difference from this maximum reading length is within a predetermined character (preferably, within 4 or 5 characters). select.

In comparison with other phrase forms, it is unlikely that a phrase form whose reading length is sufficiently short is included in the longest two phrases. For this reason, by excluding such a phrase form having a short reading length from the preceding phrase candidates, it is possible to reduce the processing target of the longest two-phrase extraction step without remarkably lowering the conversion accuracy.

Further, the kana-kanji conversion method according to the present invention comprises:
The phrase form limitation step further includes a step of selecting a phrase form within a predetermined order from the longest independent word reading length.

In comparison with an independent word included in another phrase, it is unlikely that a phrase composed of an independent word having a short reading length is included in the longest two phrases. For this reason, in the phrase form limitation step, it is unlikely that a phrase form composed of an independent word whose reading length is sufficiently short is included in the longest two phrases. For this reason,
By excluding such independent words having a short reading length from the candidates, it is possible to reduce the processing target of the attached word extraction step and the longest two-phrase extraction step without significantly lowering the conversion accuracy.

Further, the kana-kanji conversion method according to the present invention comprises:
The longest two-phrase selection step determines the syntactic validity between the phrases that are the longest two phrases for the preceding phrase candidates whose independent word usage frequency is up to a predetermined order for each preceding phrase candidate having the same reading length, One of the longest two phrases is selected based on the determination result.

In comparison with other pre-phrase candidates, it is unlikely that syntactic validity of the longest two-phrase phrase including the pre-phrase candidate whose self-sufficient word is sufficiently low is high. For this reason, by excluding the longest two clauses consisting of the preceding clause candidates in which the independent word is not frequently used, the longest two clauses in which the syntactic validity between the clauses should be determined in the longest two clause step are reduced. be able to.

The kana-kanji conversion method according to the present invention comprises:
The longest two-segment selecting step selects the longest two-segment consisting of the preceding clause candidates having a reading length of two or more characters.

It is unlikely that one of the two phrase segments, which has a reading length of one character, is the correct answer. According to this kana-kanji conversion method, in the longest two-phrase selection step, it is possible to preferentially select the longest two-phrase including the preceding phrase candidate having a reading length of two or more characters.

The kana-kanji conversion method according to the present invention comprises:
The longest two-segment selecting step selects the longest two-segment having a smaller sum of the morpheme numbers of the adjunct words of the preceding and succeeding candidate segments.

It is unlikely that the longest two phrases having a large number of morphemes are correct. According to the kana-kanji conversion method, in the longest two-segment selecting step, the longest two-segment having a smaller number of attached word morphemes can be preferentially selected.

Further, the kana-kanji conversion method according to the present invention comprises:
The longest two-phrase selection step selects the longest two-phrase including the preceding phrase candidate ending with a predetermined particle.

For a particular particle, there is a high possibility that the phrase immediately after that particle will be a phrase break. According to this kana-kanji conversion method, it is possible to preferentially select the longest two phrases including the preceding phrase candidate ending with a predetermined particle.

Also, the kana-kanji conversion device according to the present invention
A phrase-form extracting means for extracting a phrase form starting from the beginning of a kana character string or a known phrase break based on a dictionary, and a phrase-form limiting means for selecting the extracted phrase form according to the reading length and setting it as a previous phrase candidate A longest two-phrase extracting means for extracting a phrase form following the preceding phrase candidate based on the dictionary as a later-phrase candidate and extracting a maximum of two phrases consisting of the preceding and succeeding phrase candidates; When the longest two phrases are extracted, a longest two-phrase selecting means for selecting any one of the longest two phrases, a phrase break is determined based on the obtained longest two phrases, and kanji conversion is performed for the preceding phrase. Kanji conversion means.

The recording medium storing the kana-kanji conversion program according to the present invention has a computer-readable kana-kanji conversion program for causing a computer to execute the above steps.

[0033]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 FIG. 1 shows a block diagram of a kana-kanji conversion device according to Embodiment 1 of the present invention. Reference numeral 101 denotes an input unit for inputting a reading character string. Reference numeral 102 denotes kana-kanji conversion control means, which sequentially calls the phrase form extracting means 103, the phrase form sorting means 107, the longest two-phrase extracting means 108, and the longest two-phrase selecting means 109, and separates and inputs the read character string. Create a conversion result. The conversion result is output by the output unit 110. The phrase extractor 103 extracts a phrase from the input reading character string while referring to the independent word dictionary 104, the attached word dictionary 105, and the attached word connection table 106. The phrase type sorting means 107 creates a phrase block by arranging the extracted phrases according to the reading length (the phrase blocks will be described later with reference to FIG. 9). The longest two-phrase extraction unit 108 extracts a subsequent phrase for a phrase block having a predetermined rank or higher among the sorted phrase blocks, and extracts the longest combination of two phrases. The longest two-segment selecting means 109 selects an optimal two-segment delimiting method in accordance with a procedure described later when there are a plurality of two-segment delimiters with the longest reading length.

FIG. 2 shows the kana-kanji conversion control means 102 of FIG.
7 is a flowchart of a segmenting process executed by the user. Hereinafter, each step of FIG. 2 will be described with reference to FIG. 1 and other detailed drawings as appropriate. Input means of FIG.
From 101, it is assumed that a reading character string of "Tomorrow's size" has been input.

First, in step S201, the values of the variables m and s are initialized to 1. The variable m indicates that the phrase to be processed in the two phrase longest matching method is the m-th phrase counted from the beginning of the phrase. The combination of the m-th clause (previous clause) and the (m + 1) -th clause (post-clause) having the longest reading length is obtained, and the segmentation is performed by dividing the m-th clause and the (m + 1) -th clause in the longest combination. This is the longest match method for two phrases. Step from step S201
While counting up the variable m up to S208, the process of obtaining the longest combination of the reading of the m-th clause and the (m + 1) -th clause is repeated until there is no more reading character string to be processed. The variable s is a character position representing an unprocessed part of the read character string, and is a leading character position when extracting the m-th phrase.

In step S202, the phrase form extracting means of FIG.
According to 103, all the clauses starting from the s letter are extracted,
Let it be the m-th phrase candidate Bm (i). Hereinafter, the operation of the phrase extraction unit 103 will be described with reference to FIGS.

FIG. 3 is a flow chart showing a phrase extraction process. In step S301, the independent word dictionary of FIG.
104 is searched, the number of search result words is stored in a variable N, and the content of each word in the search result is stored in an array J [k]. Now, because the input reading character string is “Tomorrow's size” and s = 1,
“Asuka”, “Tomorrow”, “A”, etc. are searched, and J
Stored in [k].

FIG. 4 shows an example of the configuration of the independent word dictionary 104. The configuration includes at least the reading 104a, the headline 104b, and the part of speech 104c, and can be searched by the reading 104a. In order to further improve the conversion accuracy, semantic information, co-occurrence information and the like can be stored in addition to the information shown in FIG.
In general, a method of displaying conversion results of a plurality of independent words (homophonetic words) having the same reading in the order of use frequency in an existing document is used. That is, the frequency of use can be one of the priorities when selecting an independent word. Although FIG. 4 does not show information on the frequency of use of each independent word, the homophones are sorted in advance in the order of frequency of use, and are stored in the array J [k] in the order of frequency of use. Although the above configuration is used in order to save the memory capacity, the configuration may be such that the usage frequency of each independent word is stored in a dictionary, and the search results are sorted by referring to the usage frequency information after the search.

Next, in step S302 of FIG. 3, a loop counter variable i is initialized. Hereinafter, in steps S303 to S308, an attached word process is performed for each word of the independent word search result J [i]. Step S303 is a loop end determination. In step S304, the independent word search result J [i] is substituted for the temporary variable Btmp, and the extraction result buffer is initialized. If the independent word search result J [i] can form a phrase alone, the extraction result buffer is initialized with J [i]. If a clause cannot be composed by itself, the extraction result buffer is initialized to empty. Whether or not a phrase can be composed independently depends on the part of speech of J [i].

In step S305, the auxiliary word dictionary 105 of FIG.
With reference to the attached word connection table 106, the attached word processing for Btmp is performed. FIG. 5 is a configuration example of the attached word dictionary 105.
The configuration includes at least the reading 105a, the heading 105b, the preceding code 105c, and the following code 105d, and is configured to be searchable by the reading 105a. Prefix code 105c and postcode 105d
Is a code for indicating whether or not a connection between the independent word and the attached word and the attached word is possible. In addition, the indirect word does not store the preceding code and the following code, because the independent word is always at the beginning of the clause, the preceding code is unnecessary, and the part-of-speech 104c can substitute the succeeding code. It is.

FIG. 6 shows an example of the structure of the attached word connection table 106.
If the value in the table is 1, connection is possible, and if 0, connection is not possible. In the example shown in the drawing, it is shown that an attached word having a preceding code "10" can be connected to an attached word having a succeeding code "35". For example, it indicates that the attached word “kara” shown in FIG. 5 can be connected to the attached word “ka”. By giving a postfix code corresponding to the part of speech for an independent word, the inflected ending can be included in the attached word dictionary and processed uniformly. It also includes termination flag information indicating whether or not each succeeding code can be a terminating end.
An adjunct with a trailing code whose end flag information is 1 indicates that it can be a clause end.

For example, with respect to the independent word "tomorrow", in step S305 of FIG.
It is obtained from the attached word connection table 106 that the preceding code "30" of "kara" can be connected to the succeeding code "1" corresponding to "noun" which is the part of speech of "". In this way, the connection result "from tomorrow" is obtained from the independent word "tomorrow".
The obtained connection result is stored in a temporary variable Btmp.

In step S306 of FIG. 3, it is determined whether or not the temporary variable Btmp has a newly connected auxiliary word. If not, the process proceeds to step S309, where the variable i is counted up and the same applies to the next independent word. Is performed. Here, since the result "from tomorrow" is obtained, the process proceeds to step S307.

In step S307 of FIG. 3, it is determined whether or not the result stored in the temporary variable Btmp can be the end of a clause. If not, the process proceeds to step S305 to repeat the attached word processing for Btmp. The determination is made based on the subsequent code 105d and the termination flag information in the attached word connection table shown in FIG. Here, since the postfix code 35 of “kara” in “from tomorrow” can be the end of a phrase, the process proceeds to step S308. In step S308, the contents of the temporary variable Btmp are copied to the extraction result buffer, and then the flow advances to step S305 to repeat the attached word processing.

As described above, steps S305 to S308
While repeating the processes up to, the phrase forms derived from the independent word J [i] are sequentially stored in the extraction result buffer. When there are no more new connections, steps S306 to S309
To perform the attached word processing for the next independent word. In the above, for the sake of simplicity, the case where there are a plurality of attached word processing results in step S305 is not described. It is assumed that the processing from step S305 to step S308 is performed in parallel (it may be repeated by backtracking).

FIG. 7 shows an example of a phrase extraction result for the input reading character string "Asukara no Kaisai". Phrases from 701 to 706 are extracted for the independent word "tomorrow". Also, 707 and 708 are extracted for the independent words "Asuka" and "A", respectively.

The description of the phrase extraction processing in step S202 of FIG. 2 has been described above. Returning to FIG. 2, the processing of step S203 and thereafter will be described. In the following, the independent word extraction process will be described with reference to FIG.
The description will be continued on the assumption that the phrase extraction method described in the above is used.

In step S203 of FIG. 2, the phrase extraction result shown in FIG. 7 is sorted by the phrase alignment means 107 of FIG. 1 according to the reading length of the phrase. FIG. 8 is an example of a phrase alignment result. Reference numerals 801 to 807 in FIG. 8 denote phrase blocks for classifying phrases by reading length. The phrase blocks store the phrase reading length and have pointer information to phrase data (701 to 708 in FIG. 7). Since 702 and 707 in FIG. 7 have the same number of reading characters of 3, the phrase blocks in FIG. 8 are both aligned as elements of the phrase block 805. As described above, the phrase block has a function of collecting a plurality of phrases having the same reading length into one and abstracting the processing target of the segmentation.

In step S204 of FIG. 2, it is determined whether any of the phrases arranged in step S203 reaches the end of the input read character string. If there is a phrase reaching the end, the segmentation processing ends.
Here, the phrase “From tomorrow” (708 in FIG. 7)
Is the longest and has not reached the end of the input read character string, so the process proceeds to step S205.

In step S205 in FIG. 2, the following longest phrase (subsequent phrase) is extracted from the phrase alignment result shown in FIG. 8 by the longest two-phrase extraction means 108 in FIG. Then, in step S206, a combination of phrase blocks having the longest reading length is obtained. It has been empirically known from a conversion experiment using a large number of example sentences that the longest two sentences can be obtained if the succeeding sentences are obtained for the top three or four places in the above-mentioned alignment result. Here, it is assumed that a phrase block of the following phrase is obtained for the phrase block of the preceding phrase up to the fourth place. In obtaining the phrase block of the subsequent phrase, the same method as that of the preceding phrase block is used for each target reading character string described below. FIG. 9 shows a read character string to be subjected to post-segment extraction. Each of the read character strings following the phrase blocks 801, 802, 803, and 804 in FIG. According to the prior art (Reference 2), the number of extraction targets of the later clause is six, but the number is reduced to four, so that the processing can be speeded up.

Further, when the preceding phrase to be processed is limited, the phrase may be limited to a phrase whose difference in the number of characters from the longest phrase block is within a predetermined number of characters. Empirically, it has been found that even if the difference in the number of characters from the longest phrase block is limited to a phrase block of 4 or 5 characters or less, there is no problem in finding the longest two phrases. In the example of FIG. 9, since the difference in the number of characters in the upper four clause blocks is one character at a time,
Although there is no effect of limitation due to the difference in the number of characters, in the example shown in FIG. 10, by limiting the number of characters from the longest phrase block, the extraction target of the subsequent phrase can be limited to three types. Is possible. Of course, a plurality of these limiting methods may be combined, and a method that has the highest limiting effect, that is, a method that minimizes the number of the subsequent phrase extraction targets may be selected.

FIG. 11 shows an example of a method for limiting the preceding phrase block in step S205 in FIG. In step S1101, variables are initialized. The variable k indicates the number of types of independent word length. The variable i is a loop counter indicating that the i-th of the sorted phrase blocks is to be processed. The variable L is the longest read character length of the previous phrase block, and is initialized by the read character length of the first phrase block.

Step S1102 is a determination section for extracting a phrase block having a reading length up to a predetermined order.
As described above, it has been empirically known that if the threshold value Th1 is 3 or 4, the longest two phrases are likely to be obtained. Here, it is assumed that Th1 is 4.

In step S1103, the determination unit is used to limit the phrase whose difference in the number of characters from the longest phrase block is within a predetermined number of characters Th2. It has been empirically found from experiments that it is highly likely that the longest two phrases can be obtained by setting 4 or 5 characters as Th2.

In step S1104, the determination unit is used to limit the number of types of the independent word length when the number is within a predetermined threshold Th3. When finding the longest combination of two clauses for two clauses in which no adjuncts appear, such as compound words, speedup is limited by limiting not only the length of the phrase block but also the length of the independent word. To be peeled off. For example, in response to a reading input of "daigakuseikatsu", step S1
According to the judgment of 102, "Is it a college student?"
Although it is limited to the four preceding phrases "dai", three types of candidates such as "college student", "college", and "dai" are included when focusing on the length of the independent word. It is unlikely that the shortest “table” can be the pre-phrase block of the longest two phrases. Therefore, in step S1104, the number of types of independent word lengths included in the preceding phrase block is limited by Th3 having a value slightly smaller than the threshold value Th1 (here, 2), and the preceding phrase block “daiga” is limited. The speed is increased by not extracting the candidates. Therefore, in step S1104, Th3 having a value slightly smaller than the threshold Th1 (here, 2) is used.
The number of types of independent word length included in the preceding phrase block is limited. Note that, here, it is assumed that in step S203 of FIG. 2, the preceding phrases are sorted in advance for each phrase block based on the independent word length, and that the smaller the i is, the smaller the independent word length Len (i) is. . This concludes the description of the method for limiting the previous phrase block with reference to FIG. 11, and returns to FIG. 2 to describe the processing of step S207 and subsequent steps.

In some cases, there are a plurality of combinations of two phrases having the longest reading length obtained in steps S205 and S206. For example, in the example shown in FIG. 9, there are three longest two phrases, “starting tomorrow / difference”, “starting tomorrow / detailed”, and “starting tomorrow / hold”. In step S207 of FIG. 2, the longest two phrases are selected by the longest two phrases selection means 109 of FIG. 1 in accordance with a procedure described later, and the optimal longest two phrases are selected. Less than,
The operation of the longest two-phrase selection means 109 will be described with reference to FIG.

In step S1201, it is determined whether the longest clause combination is one. If there is one, the following processing is skipped, and the flow advances to step S1213 to determine the m-th delimiter Sep (m). To end the selection process. In the above example, since there are three types, the process proceeds to step S1202.

In steps S1202 and S1203,
If there is a syntactically valid phrase combination, this is selected. As shown in FIG. 8, one phrase block may include a plurality of phrases. In such a case, it is computationally expensive to verify syntactic validity of all clauses and combinations of clauses. Therefore, the syntactic validity is determined only for a predetermined number of clauses in each clause block. For example, in FIG. 12, if M = 1, the syntactic validity is determined only for the top clause.

Here, the preceding phrase ends with the case particle and ends with the latter phrase, which is noun, and the latter phrase is noun, such as "Kasugai / difference from tomorrow" and "Kasuga / Dessai". The phrase “from / tomorrow” whose noun is a noun is determined to be highly grammatically valid, and the combination “from / tomorrow / hold” is selected.

In step S1204, it is determined whether or not the longest clause combination is one. If there is one, the following processing is skipped, and the flow advances to step S1213 to determine the m-th segment Sep (m). To end the selection process. Here, the combination “from / tomorrow” is selected in step S1203, and the process advances to step S1213. Hereinafter, the processing of steps 1205 to 1212 will be described using another input example different from the above.

If the input character string is "something better", there are three longest two phrases, "name / similar", "what / going", and "something / good". "First name /
It is unlikely that a combination such as "similar" in which one phrase reads one character will be a correct answer. Therefore, if there is a combination in which one character is read in step S1205, the combination is excluded from the candidates in step S1206. However, it is assumed that other candidates remain as a result of the exclusion here. Although omitted in FIG. 12 to avoid complicating the chart, the exclusion process in step 1206 is exceptionally performed when there are only two combinations of a phrase combination where the first phrase is a single character and a phrase combination where the second phrase is a single character. Shall not be performed.

In general, the way of delimiting the number of attached words is often correct. For example, if "Yofu no Kaguga" is entered, "Western style / furniture"
The tool can be divided in two ways. Step S120
In step 8, “Western style / furniture” with a small number of attached words is selected.

In some cases, priorities cannot be determined in each of the above processes. For example, if you enter "Pankakara", "Bread or cormorant""Bread / buy"
There are two ways of dividing. In these examples, the number of attached words is "bread + from + or cormorant +" and "bread + / buy + from +". However, in this case, it is empirically clear that there is a high possibility that a segment break immediately follows "". Therefore, in step S1210, it is determined whether or not the process ends with a specific particle, such as "", and prioritization is performed.

Further, in step S1212, as a default process in the case where the priority cannot be determined by the above process, a candidate having a short independent word length of the subsequent sentence is selected.

In the above, each processing from step S1202 to step S1212 is applied in a predetermined order, and when there is only one candidate, the following processing is skipped and step S1 is executed.
Although described as an exclusive process that proceeds to step 213, step S12
02 to the step S1212
In steps S1203, S1206, S1208, S1210, and S1212), only the points may be given as appropriate, and the process that selects the combination with the highest score in step S1213 may be performed. With this configuration, evaluation is performed in a plurality of stages, so that the possibility of selecting a more appropriate candidate can be increased.

For example, in step S1203, +100 points are given to a syntactically valid phrase block combination. In step S1206, -100 points are given to a combination in which one reading of the phrase block is one character. In step S1208, the score obtained by multiplying the number of attached word morphemes by 10 is reduced for each phrase combination. In step S1210, a score corresponding to the particle is added to a phrase ending with a specific particle. For example,
+10 points are given to a phrase ending with. In step S1212, the independent word length of the subsequent sentence is added as a score.

By adding the points as described above and selecting the phrase block combination having the highest score in step S1213, the same effect as the exclusive processing described above can be obtained. Furthermore, since a comprehensive judgment is made using a plurality of selection measures, there is a higher possibility that a correct combination of phrase blocks will be selected.

Thus, the detailed description of the phrase candidate selection processing in step S207 of FIG. 2 is completed. Returning to FIG. 2, in step S208, the selected segment is set to Sep (m).
The character position next to the m-th clause stored in Sep (m) is substituted for the variable s, the variable m is counted up, and the processing from step S202 to step S208 is repeated. Finally, the end determination of step S204 becomes YES, and the sharing process ends.

The result of division by the kana-kanji conversion control means 102 in FIG. 1 is shaped and output by the output means 110. The output unit 110 controls the priority of the candidates in each of the segmented phrases. Also, a configuration may be adopted in which a learning dictionary for storing candidates selected by the user in the past is provided, and the candidate selected by the user in the past is preferentially output by referring to the learning dictionary at the time of output.

As described above, according to the present invention, all phrase forms including the same independent word are extracted in the phrase extraction step, and the extracted phrase forms are aligned in the phrase alignment step to form a phrase block. Since the phrase block is limited by the reading length to be the preceding phrase block, the succeeding subsequent phrase is extracted, and the longest candidate in two phrases is selected, so that the processing is faster than the conventional technique (Reference 2). Becomes possible.

Embodiment 2 In the first embodiment, an example has been described in which the adjunct processing is performed on all the words in the independent word search results in the phrase extraction processing (FIG. 3). An example in which the speed is increased by limiting the independent words to be executed will be described.

FIG. 13 shows a flowchart in the case of performing the attached word processing by limiting the independent word search result. Steps
In S1301, in addition to the same processing as in step S301 in FIG. 3, the length of the longest independent word is stored in a variable Lmax. Here, it is assumed that the independent word search result J [k] is sorted in advance by the length of the read character string. Normally, since the dictionary search is repeated to search for partial matching character strings while shortening the input reading character string by one character, the above-described premise that the reading character strings are sorted in the length order requires special processing. Not something.

In step S1302 in FIG. 13, the loop counter i is initialized to 1 as in step S302 in FIG. 3, and in step S1303, the end of the loop processing is determined as in step S303 in FIG.

Step S1310 in FIG. 13 is a processing step that does not exist in the flowchart in FIG. 3, and is the length Len (J [i]) of the independent word currently being processed and the length Lmax of the longest independent word. The end of the loop processing is determined based on the difference in the number of characters from. Here, it is assumed that 4 is given as the threshold Th. For example, if the input character string is "Futago Tamagawa Karen no Kaze", "Futako Tamagawaen", "Twins", "Lid", etc. are obtained as an independent word dictionary search result. Is very unlikely to be the longest match for "twin", "lid", etc. Therefore, in step S1310, the “independent” word extraction process ends without performing the auxiliary word processing for “twin”, “lid”, etc., in which the difference in the number of characters read from the longest independent word “Futakotamagawaen” is 4 or more characters. . When an independent word with a long reading is input as described above, short-term independent words that become a partial character string are excluded from adjunct word processing and the phrase extraction process ends, so speed up the process. Can be. The processing from step S1304 to step S1309 in FIG. 13 is the same as the processing from step S304 to step S309 in FIG.

As understood from the above description, the kana-kanji conversion method according to the present invention can be realized by, for example, the following configuration. (1) an input step of inputting a reading character string; and (2) extracting all phrase forms including the same independent word from the input reading character string while referring to the independent word dictionary, the auxiliary word dictionary, and the auxiliary word connection table. A phrase shape extraction step, (3) a phrase shape sorting step of sorting the phrase shapes extracted in the phrase shape extraction step in the phrase reading length order to form a phrase block for each phrase reading length, and (4) a phrase block Is defined as a pre-clause block by a predetermined condition, a post-clause following each pre-clause block is extracted, and a longest two-clause extraction step of obtaining a combination of the longest clause blocks in the two clauses; When a plurality of combinations of phrase blocks are extracted in the extraction step, a maximum order of two phrase selection steps of performing a predetermined ranking and selecting an appropriate two phrases, and (6) outputting a conversion result Each step of the output step that.

Further, with respect to the above basic configuration, various applications and modifications as described below can be realized. (2-1) The phrase extraction step includes an adjunct word processing target limiting step of restricting an independent word to be an adjunct word processing target according to the length of a reading character string of the independent word. (2-2) The attached word processing target limitation step limits the independent word whose difference from the reading length of the longest independent word in the independent word search result is within a predetermined number of characters. (2-3) The attached word processing target limiting step limits independent words whose difference from the reading length of the longest independent word among independent word search results is within 4 to 5 characters.

(4-1) The longest two phrase extraction step includes a processing target phrase block limitation step of limiting the preceding phrase block from the one having the longer reading length to a predetermined order. (4-2) The longest two-phrase extraction step comprises reading the preceding phrase block from the first to the third or fourth, whichever is longer in length.
And a process target clause block limiting step for limiting to the order. (4-3) The longest two-phrase extraction step includes a processing-target-phrase-block limiting step of limiting a previous phrase block to be processed based on a difference in reading length from a phrase block having the longest reading length. (4-4) The processing target phrase block limiting step restricts the preceding phrase block in which the difference in reading length from the phrase block having the longest reading length is within 4 to 5 characters. (4-5) The longest two-phrase extraction step includes a processing-target-phrase-block-limiting step of limiting a previous phrase block to be a processing target limited by the number of types of independent word reading lengths.

(5-1) The longest two-phrase selection step is a step in which, when there are a plurality of combinations of two phrase blocks having the longest read character length, the degree of connection between the phrases up to a predetermined order in each phrase block. And selecting a syntactically valid phrase block combination to select a syntactically valid phrase block combination. (5-2) The longest two phrase selection step includes a one-character phrase block deletion step of deleting a combination including a phrase block whose reading character length is one character from candidates of the preceding phrase block.
(5-3) The longest two-phrase selection step includes a phrase block morpheme number minimum phrase block combination selection step of selecting a phrase block combination that minimizes the sum of the adjunct morpheme numbers. (5-4) The longest two phrase selection step includes a specific particle phrase block combination selection step of selecting a phrase block combination including a phrase block ending with a specific particle. (5-5) The longest two-phrase selection step is a combination including a phrase block in which the degree of connection between phrases up to a predetermined order in each phrase block is determined, and the read character length is one character. No, the total number of adjunct morphemes, and whether or not the phrase block ends with a specific particle are comprehensively determined, and an appropriate phrase block combination is selected.

The kana-kanji conversion method according to the present invention
The present invention can be realized in a computer for executing the above steps, or in an electronic device incorporating a microprocessor. For example, by incorporating the microprocessor into a mobile communication device (mobile phone, mobile information terminal, etc.) capable of large-capacity wireless data communication, kana-kanji conversion can be performed at a low-cost and small communication terminal. .

The kana-kanji conversion method according to the present invention
It can be provided by a computer-readable storage medium as a computer program. Such storage media include recording media for optical reading means (CD-ROM, DVD, etc.), recording media for magnetic reading means (floppy disk, hard disk, etc.), semiconductor memory, and the like.

[0081]

According to the present invention, it is possible to reduce the number of objects for extracting a maximum of two phrases by limiting the phrase forms which are the preceding phrase candidates according to the reading length. Therefore, high-speed kanji conversion processing can be realized, or desired kanji conversion can be realized using cheaper or smaller hardware.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of a kana-kanji conversion device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing an example of a segmentation process executed by the kana-kanji conversion control means 102 of FIG.

FIG. 3 is a flowchart illustrating an example of a phrase extraction process;

FIG. 4 is a diagram showing a configuration example of an independent word dictionary 104 of FIG. 1;

FIG. 5 is a diagram showing an example of a configuration of an attached word dictionary 105 in FIG. 1;

FIG. 6 is a diagram showing a configuration example of an attached word connection table 106 in FIG. 1;

FIG. 7 is a diagram illustrating an example of a phrase extraction result for an input reading character string “Asukara no Kaisai”;

8 is a diagram showing an example of a result of the alignment of the clauses in FIG. 7;

FIG. 9 is a diagram showing an example of a read character string to be subjected to post-segment extraction.

FIG. 10 is a diagram showing another example of a read character string to be subjected to post-segment extraction.

FIG. 11 is a diagram showing an example of a method for limiting a previous phrase block in step S205 of FIG. 2;

FIG. 12 is a flowchart showing an example of the operation of the longest two-phrase selection means 109 of FIG. 1;

FIG. 13 is a diagram showing an example of an operation of the kana-kanji conversion device according to the second embodiment of the present invention, and is a flowchart showing an operation of a phrase form extraction process for performing an adjunct word process by limiting an independent word search result; It is.

FIG. 14 is a flowchart showing an operation of a conventional kana-kanji conversion process.

 ────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Yasuhiro Takayama 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F-term (reference) 5B009 MB13

Claims (15)

[Claims] 1. A phrase extraction step for extracting a phrase starting from the beginning of a kana character string or a known phrase delimiter based on a dictionary, and selecting the extracted phrase according to the reading length to obtain a preceding phrase candidate. A phrase form limitation step, a phrase form following the previous phrase candidate is extracted based on the dictionary as a later phrase candidate, and a longest two phrase extraction step of extracting the longest two phrases consisting of a combination of the previous phrase candidate and the later phrase candidate; When two or more longest two clauses are extracted in the longest two clause extraction step, a longest two clause selection step of selecting any one of the longest two clauses, and a segment break is determined based on the obtained longest two clauses, A kana-kanji conversion method comprising a kanji conversion step for performing kanji conversion. 2. The phrase extraction step includes: an independent word extraction step of extracting an independent word; an independent word limitation step of selecting the extracted independent word according to an independent word reading length;
2. The kana-kanji conversion method according to claim 1, further comprising an attached word extracting step of extracting an attached word following each selected independent word. 3. The independent word limiting step determines a maximum reading length of the extracted independent word, and selects an independent word having a reading length whose difference from the maximum reading length is within a predetermined character. The kana-kanji conversion method described in. 4. The kana-kanji conversion method according to claim 3, wherein the independent word limitation step selects an independent word having a reading length whose difference from the maximum reading length is within 4 or 5 characters. 5. The kana-kanji conversion method according to claim 1, wherein the phrase form limitation step selects a phrase form from a long phrase reading length to a predetermined rank. 6. The kana-kanji conversion method according to claim 5, wherein the phrase form limitation step selects a phrase form from the longest phrase reading length to the third or fourth place. 7. The phrase form limiting step finds a maximum reading length of the extracted phrase form, and selects a phrase form having a reading length whose difference from the maximum reading length is within a predetermined character. The kana-kanji conversion method described in. 8. The kana-kanji conversion method according to claim 7, wherein the phrase form limitation step selects a phrase form having a reading length whose difference from the maximum reading length is within 4 or 5 characters. 9. The kana-kanji conversion method according to claim 5, wherein the phrase form limitation step further comprises a step of selecting a phrase form within a predetermined rank from the longest independent word reading length. 10. The longest two-phrase selection step includes, for each of the preceding phrase candidates having the same reading length, syntactical syntactics between the phrases having the longest two phrases for the preceding phrase candidates whose priority based on independent words is up to a predetermined order. The kana-kanji conversion method according to claim 1, wherein validity is determined, and any one of the longest two phrases is selected based on the determination result. 11. The kana-kanji conversion method according to claim 1, wherein the longest two-phrase selection step selects the longest two-phrase consisting of preceding phrase candidates having a reading length of two or more characters. 12. The kana-kanji conversion method according to claim 1, wherein the longest two-segment selecting step preferentially selects the longest two-segment having a smaller sum of adjunct morpheme numbers of the preceding and succeeding phrase candidates. 13. The kana-kanji conversion method according to claim 1, wherein the longest two-phrase selection step selects a longest two-phrase including a preceding phrase candidate ending with a predetermined particle. 14. A phrase extraction means for extracting a phrase starting from the beginning of a kana character string or a known phrase delimiter based on a dictionary, and selecting the extracted phrase according to the reading length to obtain a preceding phrase candidate. A phrase form limiting means, a phrase form following the preceding phrase candidate is extracted based on the dictionary as a later phrase candidate, and a longest two phrase extraction means for extracting a maximum of two phrases consisting of the preceding and succeeding phrase candidates; When two or more longest bunsetsu are extracted in the bunsetsu extraction step, a longest two bunsetsu selecting means for selecting any one of the longest two bunsetsu, a bunsetsu break is determined based on the obtained longest two bunsetsu, A kana-kanji conversion device comprising kanji conversion means for performing kanji conversion on a phrase. 15. A computer-readable recording medium on which a kana-kanji conversion program for causing a computer to execute the steps according to claim 1 is recorded.