JP2659700B2 - Kana-Kanji conversion method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a kana-kanji conversion method capable of creating a Japanese sentence mixed with kana-kana.

[Technical Background of the Invention and Its Problems] The unit of kana-kanji input in a conventional Japanese word processor or the like is generally limited to a single phrase, and the input of a compound word of noun is limited to at most several units. Most of them are. Therefore, when inputting and creating a Japanese sentence using this type of device, it is necessary to input a delimiter key and a conversion key while always being aware of the unit of a word or a phrase, which imposes a heavy burden on the operator. Natsuta Therefore, recently, various researches have been conducted in which a kana-kanji conversion process of kana-kanji is performed on a solid kana sequence, that is, a so-called solid sentence without input unit limitation. However, this requires a considerable amount of processing time by performing the grammatical analysis for the conversion in all cases, and there is a problem that a large amount of buffer memory must be consumed. Further, it has been considered to simplify the algorithm of the character analysis process by limiting the processing time and the amount of memory, but it is unavoidable that the accuracy of the conversion process deteriorates.

In addition, as a method for displaying an input solid sentence, generally, an input line is provided, and the maximum number of characters that can be input to the input line is the maximum number of consecutive input characters. The operator had to make a conversion request. Then, the conversion process is started at this point, and after confirming the result of the conversion, the operator needs to proceed to the input of the next sentence, and kana-kanji conversion is possible for the solid sentence, but the operability is extremely poor. There were many problems.

[Object of the Invention] The present invention has been made in view of such circumstances.
The aim is to eliminate the restriction on the number of consecutive characters when inputting and automatically convert to kanji,
It is an object of the present invention to provide a kana-kanji conversion method that allows a user to input text as read.

[Summary of the Invention] The kana-kanji conversion method according to the present invention converts kana character strings that are continuously input into kana-kanji characters as appropriate. The character part is searched using a dictionary that stores the kana character string corresponding to the independent word in advance, and the kana character string part with a high possibility of the found independent word and the adjunct to be combined with this independent word When a possible kana character string portion is detected in order from the beginning of the input kana character string, the kana character string corresponding to the independent word and the adjunct word is extracted as one block, and the extracted 1 The present invention is characterized in that an input kana character string is automatically converted to a kana-kanji character every time one block of a kana character string is extracted based on a block.

[Effects of the Invention] According to the present invention, since there is no limit on the number of input characters, the user can naturally input kana as read, and furthermore, since there is no need to input an instruction such as a delimiter key, a conversion key, etc., automatic conversion is performed. Its practical advantage as a kana-kanji conversion method is enormous.

[Embodiment 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 apparatus according to an embodiment of the present invention. The input unit 1 includes a keyboard device, a voice recognition device, a kana character reading device, and the like. A character string (a kana character string) representing a reading input through the input unit 1 is provided to the input control unit 2 after being converted into, for example, a kana character code. Note that the kana character string representing the reading is, for example, hiragana, katakana,
It is shown as Roman letters.

The input control unit 2 displays the input kana character on an input display line (not shown), and requests the dictionary search unit 3 to prefetch a dictionary when the input character is a hiragana. When the number of input characters exceeds a certain number, the virtual block extracting unit 6 is requested to determine whether the virtual block can be divided. When the virtual block division is possible, the information of the extracted character block is removed from the input display line (not shown) by the number of characters, and the information of the removed character string is sent to the kana-kanji conversion unit 8. (These will be described later).

The dictionary search unit 3 prepares information of all independent words included in the sentence in the form of a table during the input of the sentence. For each input of the kana character, the input control unit 2 outputs the kana character information. Then, the dictionary storage unit 4 is searched for a corresponding word.

If there is a corresponding word (independent word), the kana character information having the corresponding independent word is entered into the dictionary search table 5.

The dictionary storage unit 4 stores kana character string information corresponding to an independent word. For example, a read-only memory (ROM)
And so on. In the present embodiment, a reverse lookup structure is adopted so that a reference is prepared for each input read. this is,
For example, if a kana character string such as “Joho” is entered and “Ji” is entered, search for “Ji” is performed (S
1) When "cho" is entered, search for "jo" and "cho" (S2), and when "u" is entered, search for "jo" and "u" (S3), and "ho" When entered, search for "joho", "hoho" and "ho" (S4), and when "uu" is entered, search for "joho", "ho" and "u" (S5) are the same. Since the head of the character string searched in the step is different, the dictionary search becomes complicated. Therefore, if the dictionary is structured so that it can be referred to as "U", "Uho", or "Uhojoji" when "U" is input (S5), the word group that starts with the input character can be used. This means that it is efficient because only the search needs to be performed.
Also, words having the same reading are stored in a group, and the corresponding information is represented by a word related to the first word of the group.

The dictionary search result table 5 stores whether or not there is an independent word corresponding to the input kana character string (or a part thereof) in the dictionary storage unit 4. FIG. 2 shows a storage column of the dictionary search result cables. The symbol “•” indicates that there is a corresponding independent word in the dictionary. The line number indicates the start of reading and the column number indicates the end.
For example, "joho" is represented at the intersection of row number -1 and column number -5.

The independent word recognition unit 7 searches the kana character every time the input is performed as described above, and sets the presence / absence information of the corresponding independent word in the dictionary search result table. A portion having a high possibility of an independent word is detected, and for the detected independent word, the virtual block extraction unit 6 uses knowledge information such as the case particle, the strength of the connection of the particle, the reading length and the number of phonetic reading elements. , Is divided into virtual blocks at locations where there is no error. In other words, for example, instead of directly inputting a long solid sentence, such as “Joho Shokaikai is now a genuine thing,” into the kana-kanji conversion unit 8 that performs strict grammatical analysis, simple After the information has been divided into virtual blocks in advance as in “Joho / Shikaiga / Genjitsu / Now,” using the information, the kana character string for each virtual block that has been shortened to some extent is converted to a kana-kanji conversion unit. 8 is input. By doing so, the processing of the kana-kanji conversion unit can be reduced by simple preprocessing (virtual block extraction) without the operator making a conversion request each time.

In other words, the virtual block extraction unit 6 does not divide the kana character string into virtual blocks that are rather short, so that the kana kanji conversion unit in the subsequent stage is not strictly divided into correct phrases. Even if this virtual block is incorrect, it is correctly re-cut and converted by the strict analysis of the kana-kanji conversion unit.

However, in the present embodiment, when dividing into virtual blocks, if, for example, "johono" (of information) is cut where it should not be cut like "joho" and "uno", Since the Kana-Kanji conversion unit 8 does not have to re-cut, the device is devised in this regard. In other words, the virtual block extraction unit 6 according to the present embodiment is realized based on the idea of taking the safe side, that is, it does not cut where it should not cut, but it does not cut where it should not. This devised point will be described in detail below.

FIG. 3 is an operation flow of the independent word recognition unit 7. When a character other than kana (numeric characters, Roman characters, etc.) is input together with a kana character string from the input control unit 2, the independent word recognition unit 7 sets a virtual part-of-speech mark as a character that may be sent to the kana-kanji conversion unit ( 3 (FIGS. 11 and 12). When a certain number or more of kana character strings are received, they are divided into phonetic reading elements by referring to a sound element dictionary stored in advance (FIGS. 3, 13 and 14).
(See the sound reading element in FIG. 4). As the sound reading elements, for example, those disclosed in Japanese Patent Publication No. 58-4378 are used. That is, as partially described in FIG. 4, a, a, c ... an, in, un ...
…｝ ｛Gya, Ca, Show ……｝ ｛Show, Butterfly ……｝
Is defined as a sound reading element. By the way, since there are many independent words in the dictionary storage unit 4 having one or two readings (the number of characters), if all of them are recognized as independent words, erroneous parts can be recognized as independent words. The nature becomes high. Therefore, in this embodiment, an independent word having a high possibility of being an independent word (an independent word to be converted into kanji in a kana character string) is extracted according to the following rules. (Fig. 3 15) Number of readings is 3 or more Number of reading elements is 2 or more Reading start point and end point are not in the middle of the phonetic reading element Applying the rule to, for example, "jojo shorigagaigagenji" , "Joho", "sho", "shakai", and "genjitsu" are candidates for independent words, and other independent words are excluded from the candidates. When the independent words are continuous, the recognition number is obtained from the number of the phonetic reading elements as follows. (Fig. 3) This recognition number is only used as a judgment material in the division in the subsequent virtual block extraction unit 6, and may be an approximate number, and it is necessary to accurately represent the number of independent words. There is no.

Recognition number = continuous number of phonetic reading elements / 2 + 1 For example, the number of recognized independent words is calculated as 4 (exactly 3 but can be approximate number). It is sent to the extraction unit 6. When the virtual block extractor 6 receives the independent word and the number of recognized independent words that are likely to be true in the kana character string from the independent word recognizer 7, the virtual block extractor 6 extracts the virtual block as described above. The Kana-Kanji conversion unit 8 refers to the dictionary search result table 5 for the reading sequence of the virtual blocks divided by the virtual block extraction unit 6 via the input control unit 2, and refers to the dictionary search result table 5 for all the independent words and the phrases of the attached words. A candidate string is obtained, and a unit of conversion to kanji is determined from the combination of the phrase candidate strings. In this embodiment, if the number of kana characters in the virtual block is large, the end point of three segments with high certainty is set as the division point from the combination in which four segments are the longest. Then, in order to perform correct kanji conversion as one sentence on all the phrase candidate strings of the conversion block thus determined, the order is determined by priority control. The output unit 9 converts the candidate sequence of the rank determined by the kana-kanji conversion unit 8 into kanji and outputs it. If it is determined whether or not there is a common connection point in a phrase to be converted into kanji for a plurality of candidate strings, the converted kanji is combined in that unit. If there is no candidate sentence, candidate sentences that have been converted to kanji in accordance with a combination instruction of all candidate strings are output in order.

Next, FIG. 5 is a flowchart showing the operation of the virtual block extracting unit, FIG. 6 is a diagram for explaining the dividing process, and FIG. 7 is a diagram showing an attached word table in the virtual block extracting unit.

The virtual block extracting unit 6 will be described in detail with reference to FIGS. 5, 6, and 7. FIG. The virtual block extracting unit 6 converts the kana character string (in which the recognized independent word information is added to the corresponding kana character string) of the format shown in FIG. The attached words are analyzed as shown in FIG. 6 (b) with reference to the attached word table of FIG. 7, and the division points are determined as shown in FIG. 6 (c). Virtual block extraction unit 6
The basic algorithm of “is independent word (A) + attached word + independent word (B)” for the input kana character string (including independent word information)
When this condition is satisfied in the pattern match of "+ appended word", the dividing point is set in principle before the independent word (B) (FIGS. 11, 12, 13, 14) (FIG. 21). Also, in the present embodiment, even if the pattern match is established, if the following conditions are satisfied, an error occurs, that is, the middle of a character string that is to be a unit for conversion to kanji is set as a division point. Because there is a possibility, it is not recognized as a division point.

When the auxiliary word of the independent word (A) is equal to the auxiliary word of the independent word (B). However, "to" (parallel particles) is an exception. (Figs. 5, 19, 20) For example, in the sentence "Exercise is Study", no particles are continuous. Only in the case of "exercise and study" is the division point before "study and".

If the attached word after the independent word (A) is a single letter, there is a possibility that it will be erroneous (for example, although the independent word recognition unit 7 recognizes that "I have a tooth ..."・
・ ”And based on“ me /
If it is divided into "", it will not be possible to correct it as in the Kana-Kanji conversion unit 8), so the following verification will be performed. If it is a Chinese word, it is recognized as a division point, but if not, the process proceeds to the next verification, where the combination of the auxiliary word after the independent word (A) and the head of the independent word (B) is If it is an adjunct to the character, it is not recognized as a division point, otherwise it is recognized (this will result in the above example not being split at that point) (FIGS. 15, 15, 16, 17). , 18) When the pattern match is not established, the kana-kanji conversion unit 8 determines that there is no malfunction of the kana-kanji conversion for a kana character string having a certain number or more of independent words, and in this embodiment, N = Relax the pattern match condition when four or more independent words are recognized That the third division point before the independent words (C) of the Kana character string (FIG. 5 22, 24, 32
3,24,25). Also, MaX is provided as a limit on the number of input kana characters, and when the number of kana characters reaches this MaX value, the division point is set before the independent word after MaX '(FIGS. 26 and 27). For example, it is assumed that MaX = 40 and MaX '= 32.

As shown in FIG. 5, the virtual block extraction unit 6 reliably extracts a virtual block to be converted to kanji at an appropriate position even when readings are continuously input, and the kana-kanji conversion unit 8 executes this virtual block conversion. Operation starts for the block. Further, according to the present invention, when a division point having a high possibility is extracted at the time when two or more phrases are input, the kana-kanji conversion unit 8 can be operated even with one phrase, and the kana-kanji conversion unit can be operated. No. 8 can reduce the number of combination candidates in the same manner as in the conventional conversion in units of one clause.

Next, the operation of the kana-kanji conversion unit 8 will be described. FIG. 8 is a diagram showing the configuration of the kana-kanji conversion unit 8, FIG. 9 is a diagram showing a priority rule and a deduction rule, and FIG. 10 is a diagram showing how a kana-kanji conversion unit divides a virtual block into segments and gives a score. is there.

As shown in FIG. 8, the kana-kanji conversion unit 8 includes a block extraction unit 8a, a total phrase sequence extraction unit 8b, a phrase extraction unit 8c, and a priority evaluation unit 8d. The kana-kanji conversion unit 8 obtains a character string mixed with the kanji corresponding to the virtual block transferred from the input control unit 2 and outputs this to the output unit 9.
Has given to. When the sent virtual block is a kana character string having a particularly long length (step 27 in FIG. 5), the block extracting unit 8a analyzes the virtual block and sets a predetermined numerical value N, for example, N = 4, the three phrases from the beginning of the longest one of the four phrases are set as the dividing points, and the divided kana character strings are passed to the phrase extracting unit 8c as one unit (block) to be converted into kanji characters. For example, if the sent virtual block has 30 kana characters and the longest 4 phrases are 20 characters, the 3 phrases (for example, 15 characters) from the beginning of the 20 characters are extracted as a unit (block) to be converted into kanji characters. Department
Pass to 8c (in this case, the remaining 15 characters are combined with the next incoming virtual block). If the virtual block sent to the block extracting unit 8a is a block divided before the independent word (B) or (C) (FIGS. 21, 25 in FIG. 5),
It is passed as it is to the phrase extraction unit 8c as a unit (block) composed of phrases or two phrases. The phrase extraction unit 8c refers to the dictionary search result table 5 for the sent block and finds the corresponding independent word part for all combinations.
That is, since all the independent word information of the input kana character string is stored in the dictionary search result table, all independent words in the block are identified accordingly. Then, the phrase extraction unit analyzes the adjunct word connected to the independent word identified in the block with reference to the grammar dictionary stored in advance,
It analyzes the inflection endings of independent words and determines the names of these parts of speech. (This part-of-speech name determination is used when applying the rules in FIG. 9.) Thus, as shown on the left side of FIG. 10, the kana character string (in which the structures of the independent words and attached words in the block are clarified) Blocks). (Although not shown, each of these is given a part of speech name). Upon receiving these blocks from the phrase extraction unit 8c, the total phrase series extraction unit 8b calculates a score for each of the blocks with reference to the priority evaluation unit 8d. The number of basic priority points is one independent word and 0.5 attached word (prefix / suffix), and the total score (basic priority point) is obtained for each kana character string in each block.
The priority evaluation section 8d stores the priority rule and the deduction rule shown in FIG. 9, and applies the rule to the combination of the parts of speech constituting each block to make the final rule as shown on the right side of FIG. Calculate a good score. In FIG. 9, the priority rule is applied to those words which are more likely to be independent words to be converted into kanji, and -1
Points are given. If it is unlikely, the deduction rule will be applied and one point will be given. Therefore, the lower the score, the higher the possibility of the kanji conversion. (In FIG. 9, φ
Indicates that there is no attached word. In addition, as for the evaluation of the priority, a rule having a small number of phrases (the number of independent words + the number of attached words) generally has a higher tendency to correspond to a kanji mixed sentence to be input. The block may prioritize a character sequence with a small number of phrases. (Additional points and deduction points may be weighted in accordance with each rule.) Thus, the total phrase sequence extraction unit 8b sequentially selects the blocks (kana character strings) shown in FIG. (At this time, it is also sent with the part of speech information). The block extracting unit 8a converts the sent kana character string to kanji with reference to a kanji dictionary stored in advance. (The kana character part with the part-of-speech mark of the independent word is converted into the corresponding kanji.) The kanji mixed kana sentence subjected to this conversion processing is sequentially sent to the output unit 9 and displayed via a CRT display (not shown). . Here, the kana-kanji conversion unit 8 may send the candidate character string having a high score to the output unit 9 as it is, and the output unit 9 may convert this kana character string into kanji characters. At this time, if a plurality of candidate strings with high scores are sent, it is checked whether or not there is a common connection point between them (for example, in the case of “jojo-sho-ri” and “joho-sho-ri”, “jojo-sho-ri” The common conversion unit and the non-common conversion unit may be separately subjected to kanji conversion, homophone processing, and the like.

As described above, according to this method, it is possible to realize a kanji conversion method in which it is not necessary to limit the number of characters to be continuously input. Further, when a certain number of kana character strings are input, the characters are automatically converted to kanji, so that a long sentence is not collectively instructed to be converted, the response speed is high, and the efficiency of the operator's operation can be improved.

In other words, when creating a Japanese sentence, the operator simply inputs the sentence as he sees it, and in natural units,
Kana-Kanji conversion is started, and a sentence mixed with Kana-Kanji can be obtained while input is being continued, so that highly operable text input can be realized. Therefore, the burden on the operator can be greatly reduced, and its practical advantage is enormous.

The present invention is not limited to the above embodiment.
For example, if the independent word analysis time is short, the independent word recognizing unit may extract the independent word after analyzing the utilization error and ending the additional word analysis. This means that the virtual block extraction unit can perform division without error in shorter units. In short, the present invention can be variously modified and implemented without departing from the gist thereof.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram showing one embodiment of the present invention, FIG. 2 is a table showing a test result of a dictionary, FIG. 3 is an operation flowchart of an independent word recognition unit, and FIG. Figure showing
FIG. 5 is a flowchart showing the operation of the virtual block extracting unit, FIG. 6 is a diagram showing a state of extracting a virtual block, FIG. 7 is a diagram showing an attached word table, FIG. FIG. 9 is a diagram showing a priority and deduction rule table, and FIG. 10 is a diagram showing a state of phrase extraction and priority evaluation. REFERENCE SIGNS LIST 1 input unit 2 input control unit 3 dictionary check unit 4 dictionary storage unit 5 dictionary check result table 6 virtual block extraction unit 7 independent word recognition unit 8 kana-kanji conversion unit 9 output unit 8a: block extraction unit, 8b: total phrase sequence extraction unit 8c: phrase extraction unit, 8d: priority evaluation unit

 ──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Hiromi Saito 1 Toshiba-cho, Komukai, Koyuki-ku, Kawasaki-shi Toshiba Research Institute, Inc. (56) References JP-A-58-80771 (JP, A) JP-A-58- 114224 (JP, A)

Claims (1)

(57) [Claims] 1. A kana-kanji conversion method for converting kana-kanji character strings that are continuously input into kana-kanji characters as appropriate using a dictionary that stores kana-character strings that can become independent words from an input kana character string, A portion of the first kana character string that can be a retrieved independent word, a portion of a second kana character string that can be an adjunct that joins the first kana character string portion, and the second kana character string A part of a third kana character string which can be an independent word following the character string and a part of a fourth kana character string which can be an adjunct connecting the third kana character string are input. When the kana character string is sequentially detected from the head, the first and second kana character string sets and the third and fourth kana character string sets are extracted. And a second set of kana character strings and the third and fourth set of kana character strings Kana-Kanji conversion characterized by automatically converting the input kana character string corresponding to the first and second kana character string sets to kana-kanji each time the division point is obtained. Method.