JPS6345676A - Kana-kanji conversion device - Google Patents

Abstract

PURPOSE:To perform the KANA (Japanese syllabary)/KANJI (Chinese character) conversion corresponding to input of characters, by carrying out the character input processing through a dictionary retrieving means when characters are short in a memory and performing alternately the input of character strings and the retrieval of dictionary. CONSTITUTION:A dictionary retrieving means 4 reads directly the character strings out of an input character string memory 2 and also performs the character input processing if the characters are short within the memory 2. Then the KANA/KANJI conversion is carried out in response to input of characters by performing alternately the input of characters and the retrieval of dictionary. Thus said conversion is possible in parallel with input of KANA characters to omit such a procedure where the conversion is carried out with push of a conversion key after input of the considerable number of character strings. As a result, an operator can perform the KANA/KANJI conversion with no waiting time. Furthermore the means 4 can be incorporated as a conventional example to realize a KANA/KANJI converting device that can work at a high speed and with high accuracy.

Description

[Detailed description of the invention] [Industrial application field] The present invention converts kana human power into sentences containing kanji and kana.

This invention relates to a kana-to-kanji conversion device, and particularly to a kana-to-kanji conversion device for converting a kana input sentence that is not divided into phrases, following the input without any special conversion start instruction means such as a conversion key.

[Conventional technology]

There are several known methods for converting sentences written in kana (so-called solid input), which have not been divided into individual clauses, into sentences containing kanji and kana.

One of them is the method described in Japanese Patent Application Laid-Open No. 18565/1983, ``Kana-Kanji Conversion Apparatus''. The method shown here is to accurately convert kana human power into sentences containing kanji and kana, and if there is ambiguity in the conversion result, it efficiently extracts and retains multiple candidates, and among them, Get correct conversion results quickly.

Provides an easy way to choose.

In addition, the method described in Nikkei Computer's November 25, 1985 issue, "Technology for full-text kana-kanji conversion method" is as follows:
In parallel with the human power of kana characters, we perform dictionary lookups for independent words that correspond to arbitrary substrings of the kana character strings input so far, and perform simple clause matching, resulting in a constant fi ( When the input of a kana character string (up to 4 clauses) is recognized, formal kana-kanji conversion is automatically started. In this method, conversion is performed without the operator having to press a conversion key to instruct the start of conversion, making it easier for the operator to recognize the conversion.

In addition, JP-A-6O-22226R Kana-Kanji Conversion Device”
describes a method that performs a dictionary search using the longest match every time a character is input, and instantly performs kana-kanji conversion at high speed.

[Problem that the invention seeks to solve]

Among the above-mentioned conventional technologies, the first one requires that a certain amount of kana text be entered first, and then a conversion key be pressed to instruct the start of kana-kanji conversion.

It takes a long time to obtain the conversion result after pressing the conversion key, and even though it is not a phrase-based conversion method, the operator has to be aware of the breaks in the input sentence when pressing the conversion key. There was a problem that it was troublesome.

The second conventional example attempts to solve this problem. However, since the conversion starts after the number of input clauses reaches 4 or more, the problem of time taken to obtain the conversion result is not completely resolved, and the extraction of 4 clauses in preprocessing is not perfect. Instead, it is necessary to make corrections in the subsequent conversion process, and there is a problem that there is overlap in the processing. Furthermore, although all independent words are read from the dictionary before the conversion process, most of them will be unnecessary in the subsequent process, so there is waste in this part as well.

In the last example, the conversion process can be executed at high speed, but dictionary matching is performed exclusively using the longest match, and there is a problem in terms of conversion accuracy compared to the first conventional example.

The present invention improves the problems of conversion accuracy and processing time that have been considered to be solved together in kana-kanji conversion using the prior art, provides a high-speed and high-precision conversion processing means, and also It is an object of the present invention to provide a kana conversion device that can easily extract other candidates even when making corrections.

[Means for solving problems]

In order to achieve the above purpose, the following means are used.

(1) A keyboard for inputting characters including kana, a memory for storing input character strings, and a dictionary that stores information such as word spellings and parts of speech using word pronunciation as an indexing means,
dictionary search means for cutting out a partial string from the input string memory and searching for a word with a corresponding reading in the dictionary;
A means for performing kana-kanji conversion including a connection test with preceding and succeeding words based on a searched word, a means for storing kana-kanji conversion candidates of degree generated by the conversion process, and a means for storing conversion candidates in order of likelihood. and means for selecting the correct word from the displayed candidates, and the dictionary search means directly reads the character string in the input character string memory, and also reads out the character string when there are insufficient characters in the memory. Perform input processing.

A means of executing kana-kanji conversion following character input by sequentially and alternately processing character string input and dictionary search.

(2) Additionally, a backspace key is provided to correct previous input errors.

A means for deleting the last character in the input string memory that has already been input in response to input of the back key.

Correspondingly, a means for deleting conversion candidates including the deleted input characters from the conversion storage means, and a means for deleting a new conversion from the end of each conversion candidate remaining in the conversion candidate storage means or from the beginning position of the conversion candidate string. A means is provided to initiate the creation of candidates.

(3) Furthermore, at the start of the process in which multiple candidates are generated as described above, a means for storing and managing the start of the conversion process for each candidate, and an executable A means for sequentially fetching processes and executing them, and a storage management method as described above, in which when the process being executed waits for input of an input kana character string that would interrupt dictionary search, the above-mentioned storage management is used to notify that this process is waiting for input. A means is provided for registering the processing in the processing means, extracting another executable processing that is waiting for execution, and restarting the execution.

[Effect]

The first means (1) allows the conversion process to be performed in parallel with the input of kana characters.

Therefore, the first conventional example (Japanese Unexamined Patent Publication No. 6O-1895
65), there is no need to take steps such as inputting a certain string of characters and then pressing a conversion key to execute the conversion, and the conversion process can be executed without making the operator wait. Furthermore, the present means can be incorporated as a part of the conventional example, thereby realizing a highly accurate and high-speed kana-kanji conversion device.

The second means (2) describes internal processing means for the function of the back key, and is realized by operating the respective means described here in conjunction. Furthermore, by starting reconversion from a possible position, it is possible to correspond to conversion for new input characters after deletion.

The third means (3) is to increase the processing efficiency and speed by expanding the scope of parallel processing and performing more thorough parallel processing.

Means (1) and (3) enable faster and more immediate conversion processing than the second conventional example.

〔Example〕

Hereinafter, the present invention will be explained in detail using examples.

FIG. 1 shows an embodiment of a kana-kanji conversion device based on the present invention. 1 is an input keyboard, 2 is an input string memory for storing input character strings, 3 is a conversion dictionary, 4 is a dictionary search section, and 5 performs kana-kanji conversion based on the word obtained from the dictionary search section. a conversion unit; 6, a storage unit for conversion candidates; 7, a display selection control unit that controls display and selection of conversion candidates; 8;
is a display unit that displays conversion candidates on the screen in an easy-to-see format.

FIG. 2 shows the kana-kanji conversion section 5 of FIG. 7 is a flowchart summarizing the processing in the conversion candidate storage section 61 and the display selection control section 7. FIG.

When the conversion process starts, first, in step 201, a pointer i indicating the phrase conversion start position in the input character string memory 2 is set.
is set to an initial value of 0 and positioned at the tip R of the input character string (a state in which no characters have been input yet). Next step 20
In step 2, it is checked whether the position is a break between phrases, that is, at the end of a phrase. The beginning of the input string is considered to be the end of the clause, so in this case, proceed to the next step 203.
Proceed to clause conversion.

In the bunsetsu conversion, as you can see from the configuration diagram in Figure 1, characters such as kana are input from the keyboard.

These are sequentially converted into morphemes that make up a clause, where a clause consists of one independent word, followed by one or more omitted clitics, and one or more omitted words after it. Learn about possible suffixes followed by one or more adjuncts that can be omitted. Compound words are divided into blocks containing one independent word, and each block is considered a clause. For example, for "industrial application fields".

r industrial” (independent word 1. suffix 1. attached word 1), “
``Use'' and ``Field'' (1 independent word for each) are each one clause. This method of converting kana to kanji in phrase units is as follows: NHK Technical Research Vol. 25, No. 5 (May 1970)
The method shown in ``Kana-Kanji Conversion Using a Computer'' is well known, and that method can basically be applied here as well. When one complete clause is extracted here, the likelihood of the conversion candidate is determined in step 204, and whether it should be retained in the conversion candidate storage unit 6 or discarded ("
Decide whether to prune (also called "pruning"). The likelihood of certainty is Proceedings of the 19th National Conference of the Information Processing Society of Japan in 1978, 5
E-4r Kana-Kanji Conversion System for Solid Text, 1981 Information Processing Society of Japan Computational Linguistics Study Group Materials 25-6 r
As described in J and others regarding clause structure analysis algorithms using tabular format and their efficiency, it is much better to break down the input string into a string of longer clauses, or in other words, into a string of fewer clauses. Decide to increase the degree. However, adnominals such as ``kono'' and ``so'' and formal nouns such as ``koto'' and ``mono'' are often used attached to other clauses, so they are not independent clauses like nouns and verbs. 1. When counting the number of bunsetsu without doing so, it should be a value smaller than 1. In this way, the number of phrases is calculated by weighting the part of speech, frequency of appearance, etc., and it is assumed that the smaller the number, the higher the likelihood. Specifically, the weight is 1° for nouns, verbs, adjectives, adjective 2, etc., and o for formal nouns, complementary Rh verbs, adjectives, etc.
1. Prefixes and suffixes are treated as semi-independent words and given a weight of 0.5.

Pruning calculates the likelihood from the beginning of the character string to the end of the clause currently being judged, and sets it as the likelihood at the character position at the end of the clause. If the degree is also constant, it is compared with that value, and if the value exceeds a certain tolerance value, pruning is performed. The allowable value is, for example, the minimum value of the number of weighted clauses at the same character position +1.

Each phrase subjected to the technique cutting in step 204 is stored in the conversion candidate storage unit 6 in step 205. In addition, in step 206, the phrase string of the candidate with the highest likelihood among the phrases obtained so far is displayed on the screen of the display unit 8 (actually, for each phrase processed, the most likely phrase string is displayed. (Additional candidates with high likelihood are displayed).

In step 207, 1 is added to i to advance the pointer to the next input character position. In step 208, it is determined whether there is still a human-written character at the character position of pointer i, and if so, the process returns to step 202.

Note that the input of one character is performed in parallel during phrase conversion in step 203. If input continues, the input path cannot be reached in step 208, and the loop in FIG. 2 is repeated. Furthermore, even if character input has been completed, if there are unprocessed characters remaining in the input string, the loop is repeated.

Repeat the above loop, capture all input characters,
Also, when the conversion process for it is completed.

The process proceeds to step 209, where selection input is accepted from the operator, and the operator selects the desired part from among the candidates for each part in order and finalizes the result. Details of the 0 selection operation and processing will not be described here. For example, it can be carried out by the method shown in Japanese Patent Application Laid-Open No. 60-189565.

FIG. 3 shows a specific example of data formed in the conversion candidate storage unit 6 according to the processing described above.

The input character string taken as an example is ``Sukaku Kai Sekiji Yoha'', which is shown at the top of the figure.The bottom of the figure shows the data of multiple conversion candidates created as a result of the conversion process. There is. A network-like data structure is adopted to prevent duplication of conversion candidate data, thereby saving storage capacity.

FIG. 4 shows the processing flow of phrase conversion in step 203 of FIG. 2. Prefix processing (step 311), independent word processing (step 312), suffix processing (step 313),
It consists of adjunct word processing (step 314). As is clear from the figure, only independent word processing is essential, and other processing can be omitted. This is a diagrammatic version of the definition of the clause mentioned above, and is actually a prefix,
For each suffix and adjunct word process, the process is performed for each combination of cases where the process is performed and cases where the process is not performed, and an attempt is made to extract all possible bunsetsu conversion candidates. In addition, although there are cases in which there are not only one adjunct word but several adjunct words, all processing is attempted for these words as well. In this sense, FIG. 4 differs from the usual flowchart representation and is intended to assist in explanation.

In each of the processes shown in FIG. 4, a word whose pronunciation starts with the kana character pointed to by pointer i is retrieved from the input string memory 2 while being checked against the dictionary, and if there are not enough characters in the input string memory 2. When a character is input from the keyboard 1, a matching word is retrieved by checking with the dictionary, and a connection check with the preceding word is performed.

The basic structure of prefix processing, suffix processing, and adjunct word processing, whose flowchart is shown in FIG. 5, is the same as that of independent Jt1 processing, so the following description will be made using independent word processing as a representative.

In step 321, a request is first made to the dictionary search unit 4 to read words that match the partial character string in the human character string memory 2 from the dictionary 3. At this time, all words having the same first character position (the same value of pointer i) on the input character string memory 2 are extracted. In the example in Figure 2, "nest", "vinegar", "number"
"Suck", "Mathematics", etc. are extracted as conversion candidates from the beginning. Next, in step 322, the words extracted as conversion candidates are extracted one by one, and in step 323, the connection with the preceding form J4 (word) is checked. Do the following. In the above example, the extraction start point is the beginning of the input string, that is, the beginning of the clause, so all independent words satisfy the connection condition,
Words such as ``nest,''``vinegar,''``number,''``suck,'' and ``mathematics'' pass the check in step 324 and are recognized as part of the phrase candidates. Then, in step 325, after the existing part (
In this case, it is connected to the beginning). In the general position example,
In Figure 3, there is a part where the particle ``ga'' (corresponding to an adjunct) is cut out, but the particle ``ga'' can be a case particle or a conjunctive particle, and each can be used to express a nominal expression such as a noun or a verb. Since it can be connected to predicates, as shown in Figure 1, the case particle ``ga'' is cut out for ``number,'' and the conjunctive particle ``ga'' is cut out and connected to ``suck.'' When the connection check is completed for all the conversion candidates extracted in this way (step 326), the independent word processing ends.

The processing contents of prefix processing, suffix processing, and attached word processing are almost the same.

The processing flow of the dictionary search unit 4 in step 321 of FIG. 5 is shown in FIG. Initialize to the value of i. Next, in step 402, the pointer k on the input string memory 2 is
Check if the input character is already registered at position e,
If there is, the process moves down; if there are no characters to input, it waits for characters to be input from the keyboard (step 403).

When a character is input, it is read and registered at the position of the pointer in the input character string memory 2 (step 404).

In step 405, the dictionary is searched using the character string in the interval between i and k as the reading. When a matching word is obtained as a result of the search (step 406), it is registered in the dictionary search word buffer (step 407). It is a memory for storage. Next, in step 408, the pointer k is advanced by one and the process returns to the beginning of the loop. In step 406, if no matching word is found in the dictionary, the loop is exited and the process ends.

Further, the words registered in the dictionary search word buffer are passed to step 322 in FIG.

Note that the process ends when no matching word is found in the dictionary in step 406, but in this processing flow, processing is performed from short readings to long readings, so words with matching readings are processed. Even if there happens to be no
If you lengthen the reading further, you may be able to find a matching word. For example, when the character string "Suugaku" is in the input character string memory 2, there are "number" and "suck" as words for the pronunciation "Suugaku", but there is no word for "Suuga". However, if we extend the pronunciation further and say ``suukaku,'' we get the word ``mathematics.''

In order to deal with such a situation, the structure of the index section of the dictionary may be made into a tree structure of reading order as shown in FIG. 7, for example. In the above example, the reading ``su'' is pronounced ``su''.
You can see that there are words such as ``Vinegar'', but for ``Su'' there is an r number'' and ``J to suck''.

There is no corresponding word for ``suuga'', but if you read further, you can easily determine that there is a matching word. Also, "<" in "suugaku"
When you reach that point, you can immediately see that there are no matching words beyond that point. Taking advantage of this point, when searching for "mathematics", it is detected that there are no longer words, and the flow in Figure 6 is configured to immediately exit the loop without going through the loop in Figure 6. You can also change the .

In addition, when obtaining words with multiple reading lengths corresponding to a given character string as shown in Figure 6, it is necessary to store each word obtained as a result of the search in a buffer (dictionary search word buffer). Control can be performed such that each time a word is obtained, the word is passed to the next step (in this case, step 322 in FIG. 5). In this way, there is no need to wait for other processes such as connection check until all words are obtained from the dictionary, and the processes can be executed in parallel, thereby increasing processing efficiency.

By the way, independent words often have a longer pronunciation than prefixes, suffixes, or adjuncts. In this case, it is often useless to perform processes such as connection checks for everything from short readings to long readings. Practically, you can limit it to words obtained by longest match or next longest match (a word whose length matches the pronunciation of the word with the longest match).

There is almost no problem with the accuracy of kana-kanji conversion. Therefore, in this case, it is also possible to first increase the reading length one after another until the longest matching word is obtained, and then search for the next longest matching word after the longest matching word is obtained.

The above has described an example in which characters are input without errors, but when humans are inputting characters from a keyboard, they sometimes make input errors. However, input errors are often noticed on the spot, and keyboards are usually provided with a back key to cancel the previous input character in order to correct them. Next, the operation regarding this backward key will be explained.

FIG. 8 shows a flowchart of the backward key process. When the backward key is pressed, first, in step 341, the last character in the input character string memory 2 is deleted. Next, in step 342, the dictionary access that is being executed by the dictionary search unit 4 and includes that character is canceled, and in step 343
Now, conversion candidates for morphemes (words) containing deleted characters in the conversion candidate storage unit 6 are deleted from the network. Finally (step 344), each morpheme on the network (
The conversion process is attempted again from the end of the string (word) and the beginning of the converted string. In this case, the morpheme (word) that has already been extracted
There is no need to perform the process again for , and it is sufficient to perform the conversion process, that is, dictionary search, on the character string that includes the newly input characters starting from the part of the deleted character. If the length is known, there is no need to perform reconversion processing from the end of all morphemes (words) on the network and the beginning of the conversion candidate string, and the maximum pronunciation of the word in the dictionary is calculated from the end of the input string. It is sufficient to perform re-conversion processing up to the essential point.

This concludes the description of the first embodiment.

By the way, as is clear from FIG. 3, when an input character string to be converted is generally given, there are a plurality of corresponding kana-kanji conversion candidates. Not just homonyms,
There may be multiple candidates for how to divide the input character string into words or for the part of speech. At this time, the processing of each part in FIG. 1 shown in the first embodiment is sufficiently fast, and the necessary processing is executed every time one character is input, and the processing is completed before the next input. There is no problem if this is the case, but in reality, the processing of each part in FIG. 1 requires a finite execution time, so there is a possibility that the processing speed will not be able to keep up with the input of characters from the keyboard. This occurs, for example, when a skilled operator inputs characters at a very high speed, or when a large number of conversion candidates exist in parallel.

A second embodiment that can efficiently perform kana-kanji conversion even in such a situation will be described next.

FIG. 9 shows the configuration of this embodiment.
8 is the same as in FIG.

is a parallel processing control unit that sequentially switches and executes the processes of the plurality of candidates existing in parallel.

The operation of the kana-kanji conversion device in this embodiment will be explained using the specific example shown in FIG. This example shows a state in which "suukakuka" has been input in the process that ultimately results in the result shown in FIG.

In the ``suugaku'' part, conversion candidates have become clear, and processing for ``ka'' and subsequent parts has started. "mosquito"
For ',' we use the adverb 'ka' or the suffix 'ka'.
, and these are already registered on the network. Furthermore, since both can end the clause at this position, processing of the next clause starts from the trailing end. It also indicates that processing of a phrase consisting of "ka" and the input character string following it has also started.

Let us consider a case where a situation like that shown in FIG. 10 occurs when the keyboard input waiting state is entered in steps 402 and 403 of FIG. 6 during execution of kana-kanji conversion in the embodiment of FIG. 1. At this time, there would be no problem if the subparticle ``ka'' and the suffix ``ka'' and ``ka'' were extracted, and then the processing of the clause consisting of ``ka'' and the input character string that followed it was started, but this was reversed. ``KA'' or ``KA'' Before ``KA'' is extracted, it enters the input waiting state during the processing of the clause for ``KA'' and the input string that follows, and ``KA'' or ``KA''``KA'' is ” may be delayed even though it can be executed. At times like this,
If you create a configuration that detects when one clause process is waiting for input, and uses that free time to retrieve and execute another executable process, you can determine which process is executed first. Deaf too.

At that time, it becomes possible to execute all executable processes.

The following are candidates for parallel processing that utilizes the waiting state for keyboard input.

(1) Processing of a plurality of phrases with different phrase start points on the input character string memory 2, and subsequent processing.

(2) Multiple processes with different connections before and after, such as prefixes, independent words, suffixes, and attached words, and multiple processes with polysemy, such as parts of speech and conjugations.

(3) Multiple processes for words with the same starting point but different ending points, or different lengths.

Regarding these, when ambiguity occurs, the execution start of each process is registered to the row processing control unit 9, and when each process waits for keyboard input while accessing the dictionary, the parallel processing control unit 9 checks whether the process is already registered or not. Select the executable one from among them and launch it.

Note that the point at which polysemy occurs is, for example, when pointer i is moved in Figure 2 and clause conversion is started while checking the presence or absence of the end of a clause, and in Figure 4, when clause conversion is started, prefix processing, suffix processing, and adjunct word processing are performed. This includes branching parts regarding whether or not to execute a process.

〔Effect of the invention〕

As described above, according to the present invention, in a device that inputs a kana character string that is not divided into phrases and converts it into kana-kanji, it is possible to perform kana-kanji conversion without pressing a special conversion key. Moreover, since the conversion is performed following the input, it becomes easy to check the conversion result, and a kana-kanji conversion device with good operability can be realized.

Further, by incorporating the present invention as a part of the conventional example (Japanese Patent Laid-Open No. 60-189565), a highly accurate and high-speed kana-kanji conversion device can be realized.

Note that the present invention is configured to sequentially convert kana character strings that are input in sequence, and is suitable for devices such as interactive voice input devices where kana character strings are input sequentially. However, even 3 is not effective.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a first embodiment of the present invention, FIG. 2 is an overall flow of kana-kanji conversion processing, and FIG. 3 is a specific example of data formed in the conversion candidate storage unit 6 by this processing. be. 4 shows a processing flow of phrase conversion, FIG. 5 shows a processing flow of independent word processing, and FIG. 6 shows a processing flow of the dictionary search unit 4. FIG. 7 shows the configuration of the index section of the dictionary used in the process flow of FIG. 6. FIG. 8 is a flowchart of back key processing. FIG. 9 shows the configuration of the second embodiment, and FIG. 10 shows a specific example for explaining its operation. 1...Keyboard, 2...Manual character string memory, 3.
...Dictionary, 4.ff? F search unit, 5... Kana-kanji conversion unit, 6... Conversion candidate storage unit, 7... Display selection control unit, 8... Display unit, 9... Parallel processing control unit.

Claims (1)

[Scope of Claims] 1. A keyboard for inputting characters including kana, a memory for storing input character strings, and a dictionary storing information such as word pronunciations and parts of speech using word pronunciations as index means; Dictionary search means for cutting out a partial string from the input string memory and searching for a word with a corresponding pronunciation in the dictionary, and kana-kanji conversion including a connection test with preceding and succeeding words based on the searched word. The method includes means for performing processing, means for storing a plurality of kana-kanji conversion candidates generated by the conversion process, means for displaying the conversion candidates in order of likelihood, and means for selecting a correct word from the displayed candidates. , the dictionary search means directly reads the character string in the input character string memory, performs character input processing when there is a shortage of characters in the memory, and processes character string input and dictionary search sequentially and alternately. By this,
A kana-kanji conversion device that performs kana-kanji conversion following input of characters. 2. The kana-kanji conversion device according to item 1, wherein the dictionary has an index section with a tree structure in reading order. 3. A means for having a backward key in a part of the keyboard for canceling the previous character input, and deleting the last character in the input character string memory that has already been input in response to the input of the backward key, and the deleted character. Means for deleting conversion candidates containing from the conversion storage means described in paragraph 1, and starting creation of new conversion candidates from the end of each conversion candidate remaining in the conversion candidate storage means or from the beginning position of the conversion candidate string 2. The kana-kanji conversion device according to claim 1, having means for converting kana to kanji. 4. The conversion means includes means for storing and managing the start of the conversion process for each candidate at the start of a process in which multiple candidates are generated, and sequentially extracting and processing executable processes stored in the storage management means. and when the process being executed is waiting for input of characters during dictionary search,
2. The method according to claim 1 or 2, further comprising means for registering in the storage management means that the process is in an input waiting state, and retrieving another executable process in the execution waiting state to resume execution. Kana-kanji conversion device installed.