JPH0547859B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] This invention converts a long continuous kana character sequence, such as one input as a sentence unit, into kana-kanji characters while appropriately separating them into clause units. This invention relates to a kana-kanji conversion device that can effectively create Japanese sentences.

[Technical background of the invention and its problems]

Conventionally, the unit of character input in kana-kanji conversion devices such as Japanese word processors has generally been limited to a single clause, and even when inputting compound words of nouns, the unit of input has been limited to a few units at most. is the majority. Therefore, when inputting and creating Japanese sentences using this type of device, the operator must always be aware of the units of words or phrases, which places a heavy burden on the operator. Therefore, recently, various studies have been attempted to input a sequence of pronunciations of kana in units of sentences, so-called solid sentences, without limiting the unit of text input, and to perform kana-kanji conversion processing on the solid sentences. These are experimentally possible, for example, by performing clause analysis recursively. However, this requires a considerable amount of processing time and has the drawback of consuming a large amount of buffer memory. Although it has been considered to simplify the algorithm for phrase analysis processing by limiting the processing time and memory amount, it cannot be denied that the accuracy of the conversion processing deteriorates.

For example, if a kana character string is given as ``I want a zandaka'', then ``I want a zandaka//I want a zandaka''.
It is possible to mechanically divide the phrase into phrases such as ``Zandaka wo mo/stop'', but it is also grammatically possible to divide the phrase ``Zandaka wo mo/stop''. At this time, empirically, it is thought that the so-called longest match is correct, but based only on such empirical rules, clause analysis is always performed from the front of the input character string, for example, ``Balance also / If only the conversion result "stop" is extracted, the conversion accuracy will be significantly degraded. Therefore, in the end, a plurality of conversion candidates such as "also/stop balance" and "calculate balance" are extracted, and each of the plurality of conversion candidates is displayed and outputted in order to leave the selection to the operator's support. Efforts are being made.

On the other hand, if we assume that the word "calculation formula" is not registered in the dictionary in the device for the input character string "Keisanshiki", the analysis result is similarly determined by the classification position of the clause. Multiple types occur, such as ``/formula'' and ``calculate/tree.'' Furthermore,
Conversion results such as "hair/heritage/formula" also occur. Therefore, for example, when kana character strings are given consecutively, such as ``Keisanshiki for Zandaka'', at least 2×3 conversion candidates are generated.
Furthermore, since there are words with the same sound as ``stop'' and ``over'' for ``stop'' in the above input string, and ``shiki'' as well, such as ``shiki'' and ``shiki'', we can use longer units. In order to accept input character strings and convert kana to kanji with high accuracy, a considerable amount of memory capacity was required. Furthermore, in the above example, if the word ``rukei = ``exile'''' exists in the dictionary, the candidate ``calculate balance/exile/formula'' also appears. As the length of the input kana character string increases in this way, it is expected that a large capacity memory approximately proportional to the order of powers will be required.

By the way, as an evaluation process to determine the most likely one as the first candidate among various conversion candidates for such an input kana character string, for example, if the input character string has a small number of clauses or words, It is considered that the priorities are determined in order. Specifically, for example, for the input character string "Kou Gakushi Yotoku", "high amount/income" is converted as a more likely conversion result than "term ga/phrase/income" or "term ga/kukoto/phrase". It is something to judge.
As for homophones, it goes without saying that words that are used more frequently are preferentially output as correct conversion results. However, such evaluation processing
For each phrase series obtained by the phrase analysis process, it largely depends on the overall structure of the phrase series from the beginning to the end. For this reason, new problems have naturally arisen, such as the longer the length of the input kana character string, the more calculation time is required for sorting processing for a plurality of clause sequences.

[Purpose of the invention]

The present invention has been made in consideration of these circumstances, and its purpose is to reduce the processing time and memory capacity even for long input character strings such as so-called solid sentences. To provide a kana-to-kanji conversion device which can effectively perform accurate kana-to-kanji conversion processing without causing problems.

[Summary of the invention]

The present invention includes a dictionary search section in which a plurality of words are registered in the dictionary, and the phrase extraction section includes a partial string starting from a specified character position in a series of input strings inputted via the input device. , the partial strings that match the words registered in the dictionary search section are extracted as clause units, and the total clause sequence extraction section extracts one extracted string for a specified block section in the input string. The phrase unit extraction process by the phrase extraction unit is repeated using the next character position of the phrase unit as a newly specified character position, and the phrase series consisting of a string of phrase units sequentially obtained through this process is extracted from the block. The kana-kanji conversion device is configured to obtain all possible combinations of phrase divisions in an interval, and to use a block extraction section to control the unit of the block on which the total phrase series extraction section acts on the input character string. In particular, the block extraction section calculates the longest length of all N phrase sequences that are continuous from the specified starting point, which is determined by the phrase extraction section through repeated processing, based on a predetermined value N. The above-mentioned kana-kanji conversion process is controlled by determining the character position within the (N-1)th position in all the series that have the same clause end position as the end point of the block. .

〔Effect of the invention〕

Thus, according to the present invention, for example, for the input kana character string "Keisanshiki to find Zandaka" in the above example, the character string is divided into two blocks in advance, such as "Keisanshiki to find Zandaka". Can be divided. Then, for each of these two blocks, we create phrase series candidates such as ``to/stop the balance'' and ``to find the balance,'' and evaluate their priorities to find likely kana-kanji conversion candidates. Therefore, the memory capacity does not increase and the processing time can be shortened. Generally, for example, conventional 2×2×
According to the present invention, internal processing was performed on two types of phrase series, but according to the present invention, processing can be performed on two types of phrase series. Further, according to the present invention,
For a long input kana character string, the kana-kanji conversion results are obtained for each block section divided from the beginning of the sentence, so the conversion results can be displayed and output sequentially in block units. Therefore, the operator can check the conversion results quickly.
Overall, it is possible to significantly reduce the burden on the operator.

[Embodiments of the invention]

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment device. The input device 1 includes a keyboard device, a voice recognition device, a kana character reading device, and the like. The character string (first character string) representing the reading input via this input device 1 is:
For example, after being converted into a kana character code, it is given to the kana-kanji conversion unit 2. Note that the first character string representing the above-mentioned pronunciation is indicated as, for example, hiragana, katakana, romaji, or the like.

The kana-kanji conversion unit 2 is, for example, a block extraction unit 2.
a, a total clause series extraction unit 2b, a clause extraction unit 2c, and a conversion dictionary 3, and for the first character string transferred from the input device 1, displays a corresponding character string mixed with kanji. The second consisting of
I'm looking for a string. The block extraction section 2a
is to obtain a phrase analysis result that is associated with a preset value N, for example N = 4, as a phrase series within N phrases for the first character string, and when there is no corresponding phrase analysis result. , the first input character string is divided into several block sections. Then, the pronunciation strings of each block divided in this way are extracted by the total bunsetsu sequence extraction unit 2.
Furthermore, the kana-kanji conversion results obtained for each block by the total phrase series extraction section 2b, that is, the second character string, are sequentially sent to the output control section 4.

The total phrase series extraction unit 2b divides the first character string into divisible phrase series using the phrase extraction unit 2c, and converts the conversion results of the kana-kanji mixed notation obtained for each of these phrase series. This is output to the block extraction section 2a. The phrase series is calculated for all combinations of input kana character strings that can be divided into phrases, and the candidates are ranked in order of likelihood based on, for example, priority evaluation, and then output in order starting from the first-ranked candidate. be done.
The above priority evaluation is performed by giving priority to phrase series with a small number of phrases, for example, because phrases with a small number of phrases are generally more likely to correspond to the input target sentence containing kanji. .

The phrase extraction unit 2c searches and matches the input character code string with character strings (words) registered in advance in the conversion dictionary 3, and searches for written characters containing kanji that correspond to the first character string. This is to find the second character string. For example, as shown in FIG. 2, the conversion dictionary 3 has an input heading table area 3a,
It includes an output heading table area 3b and a part of speech area 3c,
The first line representing the reading in the input heading table area 3aa above.
A second character string (word) consisting of notation characters including kanji corresponding to this first character string is stored in the output heading table area 3b. The part of speech area 3c stores part of speech information for the first and second character strings.

Therefore, the clause extracting unit 2c searches for a character string (word) registered in advance in the input heading table area 3a of the conversion dictionary 3, using, for example, a known longest forward match method for the given input character string. , the conjugated endings and adjuncts are analyzed, and the part of the input string that matches the first letter of the input string the longest is determined as the phrase extraction result. At this time, the analysis of the conjugated ending is performed based on the part-of-speech information stored in the part-of-speech area 3c. Then, a second character string consisting of written characters mixed with kanji corresponding to the phrase extraction result found through this analysis is read out from the output heading table area 3b and output. In addition, the total clause series extraction unit 2b inputs the unit of input to the clause extraction unit 2c (the clause extraction result) into various combinations that can be divided into clauses in the block section determined for the given input character string. We are trying to find the most probable phrase sequence by varying it.
For such a phrase series (a string of phrase extraction results), the block extracting section 2a obtains a second character string consisting of the mixed kanji notation for each block from the total phrase series extracting section 2b, This is displayed on the display device 6 via the output data memory 5 and the output control section 4.
It is output to.

As shown in FIG. 3a, for example, the output data memory 5 stores the headwords converted by the dictionary search section 2 together with the structure of the combinations of phrases in the block section. FIG. 4 shows an example of the configuration of the output data memory 5. As shown in FIG. The combination table 5a storing the structure of the above combinations in the output data memory 5 is as follows:
Corresponding to the bunsetsu structure of the input character string, the arrangement of each bunsetsu is described by numbering, for example, the row indicates the interpretive type of bunsetsu structure, and the column indicates the individual bunsetsu structure. It is a sequential description of the connections between the clauses in the text. Furthermore, the numerical values shown in the combination table 5a are numbers assigned to bunsetsu, that is, blocks that are a group of headwords with the same corresponding pronunciation, and are shown in the mapping table 5a.
It points to the area with the same number in b. Furthermore, the numerical values shown in the combination table 5a are the numbers assigned to the bunsetsu, that is, the blocks that each group a group of headwords with the same corresponding pronunciation, and refer to the area with the same number in the mapping table 5b. . Further, in the mapping table 5b, pointers indicating each corresponding clause block in the headword table 5c are written. Each phrase series subjected to phrase analysis as shown in FIGS. 3b and 3c is stored in the output data memory 5 with its headword group divided into blocks of the same phrase.

Next, using the specific input example shown in Figure 3a,
An example of the operation of the kana-kanji converter 2 will be explained.

The block extractor 2a obtains a maximum of N clauses as much as possible from the first character position of the input character series. Here, if N is set to 4, for example, Fig. 3a
In the example, the entire input series is first input to the clause extraction unit 2c, and "and" is obtained as the first clause result by the longest match method. Next, the longest matching result is found in the same way, using the point after this bunsetsu break as the starting point (the position of the next bunsetsu start character), to obtain the bunsetsu ``Kongonoto''. Such processing is repeated in order to find the first phrase series candidate ``And/Kongonoto/Ushiha/Kaiteishi'' as shown in item ``A'' in FIG. 3B. Next, in order to obtain a phrase series different from the one obtained in the above item "A", for example, the third phrase result is "Ushiha".
Delete the last character of
In the same way, we obtain the phrase ``Ushi'' as the longest matching result, and find ``HA'' as the next phrase starting with ``HA''. Thereafter, in the same manner, each time a phrase shorter than the third, second, and first phrases is obtained, another phrase series following these phrases is sequentially determined. In this way, all four-clause sequences that can be separated into phrases from the input character string are obtained as shown in FIG. 3b. still,
At this time, a corresponding entry word candidate (second character string) containing kanji is also obtained at the same time.

Next, among these sequences, attention is focused only on the candidate whose total length is the longest (the bunsetsu sequence). This is based on the fact that, as shown above, the sequence with the minimum number of input clauses has a high tendency to match the conversion result intended for input.
This means that the number of clauses in one block should be the minimum, and if the number of clauses in one block is the same,
This means that the block has a longer length.

However, among the results obtained as shown in Figure 3b above, the longest phrase sequence is the items "A" and "U".
It is. Next, we find points where these phrase series have phrase breaks in common. In this example, "and/~" and "~wa/~" are found as the common phrase break positions of each phrase series. The block extracting unit 2a determines these two positions as delimiters of blocks, and defines the first block section as "and" and the second block section as "congo no toshiwa." Then, the character strings in these sections are sequentially analyzed by the total clause string analysis section 2b, and the conversion results are sent to the output control section 4. As a result, the first block section becomes a candidate for only "and", and this information is first sent to the output control section 4. Thereafter, the second block section is analyzed.

Here, the total bunsetsu sequence analysis unit 2b obtains a character sequence corresponding to the pronunciation of a given block section by a so-called round-robin method, but in reality, it is calculated as shown in FIG.
Since the bunsetsu series has already been found as shown in , it is sufficient to select only those that correspond to the range of the specified interval. As a priority evaluation, for example, if only candidates with the minimum number of clauses are selected, the analysis result will be as shown in the item "" in FIG. 3d. Of course, other clause candidate sequences may also be provided to the output data memory 5. For example, “Future/
It is also possible to add ``Question/Magazine wa'' to the output result.

Next, the block extracting unit 2a calculates the block unit in the same way for the remaining part of the input character sequence for which no block has been determined yet, that is, the character string "Kiteishige...". The conversion result shown in the item "" in FIG. 3d is obtained. Subsequently, the conversion result ``limited'' is obtained as shown in the item ``'' in Figure 3d, and the conversion for the entire input series is completed.

Here, when N=4, the reason why the length of up to three phrases is set as a block section is as follows. For example, if N is set to 3 and there are a maximum of 3 clauses, that is, N clauses are a block section, then in a case like the above example of "Kiteishi...", "Kiteishi/Genkai/Hatsunika/" is calculated as the longest number of clauses. However, this is called "Hatsunika" (= first shipment).
If this is done, a clause such as the next character string ``Gitsu~'' will not exist in the search dictionary, and further analysis will be impossible, resulting in incorrect segmentation of the clause. In other words, the fourth syllable is ``Kagitsute''.
It is only when this exists that the validity up to the third clause will be highly evaluated. By the way, if N is set to 4 as in this example, the end of the fourth bunsetsu becomes uncertain as a block break, but if the end of the fourth bunsetsu is further back, it is a bunsetsu break until just before that. It can be regarded as certain. The setting of the block section described above takes advantage of this property. Thereafter, the conversion results obtained as described above are sent to the output control section 4 for each block, and the output control section 4 sequentially stacks these data in the output data memory 5 and outputs them to the display device 6. and is displayed.

This result display device 6 receives the above-mentioned conversion processing result and displays "(and)", "(future and)" (cow is).
”,
The conversion results will be displayed in the following order: ``(Revised) (Limit) (From)'' and ``(Limited)''. Here, the input device 1 is provided with selection keys 1a and 1b for selecting a target entry word. By selecting the selection keys 1a and 1b, for example,
When the cursor is placed on the part "(in the future)" and the first selection key 1a is inputted, the phrase string of the second candidate is changed to the structure of the first candidate in FIG. The structure, ie, "(future) (investment)" is output and displayed. The second selection key 1b is used, for example, to change (revise) in the block part of "(revise) (resources) (development)" to (seafloor) in the same phrase candidate. be.

When the first selection key 1a is input, the display device 6 displays the next candidate clause structure shown in the combination table 5a instead of the currently displayed clause structure. After that, if the first selection key 1a is inputted one after another, the structure of the phrase to be displayed will be switched to the next candidate in order, and after all the candidates have been displayed, the structure of the first candidate will be displayed again. The display is controlled. Also, when the second selection key 1b is inputted, the headword of the phrase at the cursor position on the screen at that time is replaced with other homophones existing in the same phrase block as the headword in the headword table. 5c and output it for display. This display switching is performed by advancing the value of the pointer in the mapping table 5b. When the headword to be displayed reaches the end of the clause block in the headword table 5c, the headword to be displayed is controlled to be returned to the beginning of the block.

FIG. 5 shows a schematic control flow of the kana-kanji conversion process in this device. As shown in this figure, the input codes input through the various keys of the input device 1 are constantly checked, and if the input code is a kana character code that corresponds to the Japanese sentence 〓yomi〓, it is sequentially checked. stored in the stack. If the input code indicates a conversion request, the input character code string stored in the stack is subjected to kana-kanji conversion processing, for example, in accordance with the processing flow shown in FIG. still,
FIG. 6 illustrates the process described above. If the input device 1 has a conversion request key, this conversion request is generated by the operator inputting a character string of an appropriate length and then pressing the conversion request key. Regardless of whether the input device 1 has the conversion request key or not, it detects that a character string of a predetermined length is stored in the stack, or the input code is a code indicating a punctuation mark. It is desirable to detect this and automatically generate the conversion request.

Further, when the input code corresponds to the first or second selection key 1a, 1b, the homophone selection process is performed, for example, along the process flow shown in FIG. (correction, insertion, deletion, etc.), editing processing is performed on sentences that have already been converted and displayed.

By the way, Japanese word processors generally
There are methods in which homophones are selected sequentially for each conversion result, and methods in which they are selected all at once after inputting, for example, one page of character strings, but the process according to the present invention described above can be applied to both. Needless to say.

As explained above, according to this device, even if the input kana character string is relatively long, when it is analyzed and converted into a character string containing kanji to create a Japanese sentence, the resulting A large number of clause sequence candidates can be efficiently found without consuming a huge amount of memory and a lot of processing time. Therefore, tremendous practical effects can be achieved, such as improving processing speed, reducing device costs, and greatly reducing the burden on the operator.

Note that the present invention is not limited to the above embodiments. For example, the clause extraction unit 2c may simultaneously obtain all clauses whose clauses have a common starting position from the input character string. For example, for the input string "And this..." mentioned above, "And (and), Soshi (prevention)",
They say "shite (shite)", "te (te)", "kongo no to (future), kongo (future), kongo (future), kon (now)", etc. The conversion result may be obtained as follows. Among these, the phrases "future, future, future" have the same independent word, and only the way of taking adjunct word strings differs, so in such a case, for the word "future"
You only need to search the dictionary once. and,
It is also possible to create a phrase series each time from among the phrase candidates obtained sequentially, and to perform the block division evaluation process when up to N phrases have been obtained. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of a device showing an embodiment of the present invention, FIG. 2 is a diagram showing the configuration of a memory of a conversion dictionary,
FIG. 3 is a diagram showing an example of an input character string and a phrase series obtained by dividing the phrase into phrases, FIG. 4 is a diagram showing an example of the configuration of an output data memory, and FIGS. 5 to 7 are flowcharts of conversion processing in the embodiment device. FIG. 1... Input device, 2... Kana-kanji conversion section, 3... Conversion dictionary, 4... Output control section, 5... Output data memory,
6... Display control unit, 1a... First selection key, 1b...
Second selection key, 2a...Block extractor, 2b...
Total clause series extraction unit, 2c... Clause extraction unit, 3a... Input heading table area, 3b... Output heading table area.

Claims (1)

[Claims] 1. An input section for inputting a kana character string, a phrase extraction section that identifies partial character strings in the kana character string as phrase units by comparing them with a word dictionary, and a phrase extraction section that Among the clause unit series in the identified kana character string, select the longest clause series with a predetermined integer N of 2 or more, and select the longest clause sequence in which the clause endings are the same except for the Nth clause ending in this series. a block extractor that sets a certain position as the end of the block;
A kana-kanji conversion device comprising: an output section that sequentially displays homophone candidates converted into kanji for each block extracted by the block extraction section. 2. The kana-kanji conversion device according to claim 1, wherein the phrase extraction unit obtains a kanji conversion result for the entire character string for each extracted block. 3. The block extraction unit obtains a kanji conversion result for the entire input kana character string by repeatedly extracting blocks from the beginning of the remaining portions of the input kana character string that have not been extracted as blocks. A kana-kanji conversion device according to claim 1.