JPH0638253B2 - Kana-Kanji conversion processor - Google Patents

Description

TECHNICAL FIELD The present invention relates to a kana-kanji conversion processing device, and more particularly to a kana-kanji conversion processing device suitable for improving the reliability of a kana-kanji conversion result.

2. Description of the Related Art Recently, the kana-kanji conversion method has become the mainstream as an input method for Japanese word processors and computers. The full-text character arrangement pen touch tablet and the method of inputting one Kanji with two strokes of kana are still used due to their individual characteristics, but in terms of the ease of use by anyone,
The kana-kanji conversion method is completely leading.

In the Kana-Kanji conversion method, a word dictionary is provided in order to convert sentences input in Kana (Hiragana, Katakana, and Roman characters) into proper Kana-Kana mixed sentences. In addition, as the input method, a method of pressing the Kana-Kanji conversion key for each word,
Method of pressing Kana-Kanji conversion key for each bunsetsu, method of specifying the front and back of Kanji part with keys, system that automatically accepts solid writing, grammatical analysis for each bunsetsu, and system for Kana-Kanji conversion However, the solid input method is the best in terms of speed of input.

However, even if any of the above methods is adopted, it is not possible to obtain a correct kanji-kana mixed sentence in one conversion process due to the characteristics of Japanese, that is, the presence of many homonyms and grammatical complexity. Have difficulty.

For example, consider a case where the example sentence "Shirogaso (white pixel)" is converted into Kana-Kanji.

If the word "white pixel" is in the conversion dictionary, the probability of correct conversion is extremely high in any method of Kana-Kanji conversion. However, a compound word such as "white pixel" means that "black pixel"
Since there are a plurality of "red pixels" and the like and the entire compound word is an astronomical number, the compound word is generally not registered in the word dictionary by itself, but only individual words are registered. Therefore, the system performs grammar analysis and the like, extracts the corresponding word from the word dictionary, selects the word with the highest evaluation value from it, and outputs it as the conversion result.

However, for example, in Kana-Kanji conversion with the phrase specification method,
There is a risk that the display will be erroneous, such as "white glazed". Further, in a system that allows the input of a plurality of clauses, it is possible to display an unnatural conversion result such as “white is sparse” when the word “pixel” is not registered. There is no difference in the difference between "Shiro ga soo" and "Shiro ga sparse", but from the operator's point of view, "Shiro ga soo" is less uncomfortable. Seem.

In this way, despite the high and low reliability between the multiple Kana-Kanji conversion results for the input character string,
When displaying as the conversion result, since the conversion result with high reliability and the conversion result with low reliability were all treated in the same way, "white is sparse"
The conversion result that gives the operator a feeling of strangeness is also displayed. Nowadays, it is becoming more and more common to create documents by handwriting or faster with a kana-kanji conversion processing device, and such an unnatural display of conversion results reduces the psychological burden and fatigue of the operator. Increase and make long-time operation impossible.

It is an object of the present invention to provide a kana-kanji conversion processing device capable of avoiding the display of an unnatural conversion result that gives the operator a feeling of discomfort, in order to solve the above-mentioned problems of the prior art. .

Structure The above-mentioned object of the present invention is to extract a plurality of Kana-Kanji conversion candidates for a character string input by phonetic characters by searching a word dictionary, and determine the maximum likelihood candidate word from them to be the conversion result. A kana-kanji conversion processing device having kana-kanji conversion control means is provided with a means for judging the reliability of the conversion result, and the kana-kanji conversion control means is judged to have low reliability by the reliability judgment means. In this case, the conversion result is converted into kana and displayed, and when it is determined that the reliability is high, the conversion result is displayed as it is. To be achieved.

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

FIG. 1 is a block diagram of a kana-kanji conversion processing device according to an embodiment of the present invention.

In FIG. 1, 1 is an input unit, 2 is a conversion target character string creating unit, 3 is a dictionary search unit, 4 is a word dictionary, 5 is a connection availability determination unit, 6 is a connection availability determination table (connection weight table), 7 Is the evaluation section,
Reference numeral 8 is a conversion result reliability determination unit, and 9 is an output unit.

A conversion target character string creation unit 2 extracts a conversion target character string from a sentence input by a phonetic character (Hiragana, Katakana, Roman character, etc.) by the input unit 1 by a method such as phrase unit or solid writing. From the beginning of the extracted conversion target character string, the dictionary search unit 3 searches the word dictionary 4 to extract a plurality of candidate words corresponding to the reading. The connection availability determination unit 5 refers to the connection availability determination table 6 for the candidate words extracted by the dictionary search unit 3, and determines whether the connection with the immediately preceding word is possible. Next, the evaluation unit 7 evaluates the connectable candidate words using an evaluation formula having parameters such as reading length, connection weight, and appearance frequency as parameters, and sets the word with the highest evaluation value as the conversion result. .

The conversion result reliability determination unit 8 determines the reliability of the conversion result. For example, when inputting the above-mentioned sentence "Shirogaso" in a solid form, only the words that have the reading "so" in the subsequent character string "so" after the analysis has proceeded to "white / ga" Is obtained. The evaluation unit 7 evaluates a plurality of words having the reading of “so” and takes the one having the highest evaluation value among them as the conversion result. However, the immediately preceding words “ga” and “so” are Since the reading length is short, the connection weight is not large, and the frequency of occurrence of "so" is not so large, all of these evaluation values are small.

If the conversion result is Kanji, the evaluation value is compared with the threshold value provided in the conversion result reliability determination unit 8. If the conversion value is smaller than the threshold value, the conversion result is unlikely to be accurate and the operator feels uncomfortable. Since it is given, it is converted into kana and output from the output unit 9. The threshold value is determined by experimentally obtaining an optimum value according to the format of the evaluation formula.

The algorithm in this case is shown in FIG. When the conversion result is a Chinese character, the conversion result reliability determination unit 8 determines whether the evaluation value of the conversion result is equal to or more than a threshold value (201, 20).
2). If it is equal to or more than the threshold value, even if the conversion result is incorrect, the operator does not feel a sense of incongruity. Therefore, the conversion result is sent as it is as the notation information to the output unit 9 (202, 204). . If the conversion result is less than the threshold value, the conversion result has a very high probability of being erroneous, and it makes the operator feel uncomfortable. Therefore, the conversion result is converted into kana (reading) and is output as notation information in the output unit. 9 (202, 203). In addition,
When the conversion result is Kana, that is, when it is not converted, since it is originally Kana, it is meaningless to determine by the conversion result reliability determination unit 8 whether or not to convert to Kana, so the conversion result is output as the notation information as it is. Send to Part 9 (201,
204).

FIG. 3 is a flow chart showing a specific operation example of the conversion result reliability determination unit 8 in FIG.

In this example, if the reading length is short, the reliability is judged based on the reading length based on the empirical knowledge that the reliability of the conversion result is low, and the unreliable words are output in Kana notation. It is configured to do. That is, the conversion result reliability determination unit 8 determines whether or not the reading length of the word is one character when the conversion result is a Chinese character, and if it is two or more characters, the notation information as it is. To the output unit 9 (301, 3
02, 305). If it is 1 character, further check the reading length of the immediately preceding word, and if it is longer than 2 characters, display it as it is (301, 302, 303, 305), 2
When the number of characters is less than or equal to the number of characters, it is converted to kana (reading) and sent as notation information to the output unit 9 (303, 304). For example, in the above example of “shiroga so”, it is assumed that “sparse” is sent to the conversion result reliability determination unit 8 as the conversion result for “so”. In this case, since the reading length of "sparse" is 1, the reading length of the immediately preceding word is determined. The reading length of the immediately preceding word "ga" is 2 or less, so "sparse" is converted to kana and "so" is converted.
Is the notation information.

The reading length may be entered in the word dictionary, or a reading length counter may be provided without entering it in the word dictionary.

The above example is an example in which the conversion result is evaluated using only the length of the word "reading" as a parameter, but the reliability can be more accurately determined by using a plurality of parameters. An example will be described below.

That is, the reading weight, the frequency rank, the category of the word, and the connection weight based on the appearance position are used as parameters.

FIG. 4 is a diagram showing the contents of the word dictionary 4 in the case of having the frequency rank information. In this case, the word dictionary 4 is
It consists of "reading", "notation", "frequency rank", "words", and "other information".

The item of “frequency rank” is a numerical value obtained by classifying the frequency of use of words, and the higher the frequency of use, the larger the value.

FIG. 5 is a diagram showing connection weights, and displays connection weights according to word categories and appearance positions. In this table, the categories in which the position of the conversion target word and the character type of the part of speech of the preceding word are combined are arranged in rows, and the types are arranged in columns.

For example, when the conversion target character "Kono Shiro ga soo" is created by the conversion target character creation unit 2, the operation for obtaining the evaluation value using FIGS. 4 and 5 is as follows.

The dictionary search unit 4 performs a dictionary search from the beginning of the conversion target character string, extracts “child”, “arc”, “this”, etc. and sends them to the connection verification unit 5. The connection verification unit 5 verifies whether the extracted words can be connected, and sends all connectable words to the evaluation unit 7.

The evaluation unit 7 evaluates each sent word as follows. That is, the evaluation value F is defined as F (l, f, e) = l + f + 2e (1). Here, 1 is the reading length, f is the frequency rank shown in FIG. 5, and e is the connection weight. As shown in FIG. 5, when the word is at the beginning of a sentence, the connection weight "1 '" is given only to the adnominal "kono", and the connection of other nouns "child" and "ko". Has a weight of 0 '. Further, the reading length of "child" is 1 and the frequency rank is 3, and the reading length of "Ko" is 1 and the frequency rank is 1. Therefore, these evaluation values F are “child” ... 1 + 3 + 2 × 0 = 4 “arc” ... 1 + 1 + 2 × 0 = 2 “this” ... 2 + 5 + 2 × 1 = 9.

As a result, “this” having the highest evaluation value is selected and sent to the conversion result reliability determination unit 8. The conversion result reliability determination unit 8 compares with a threshold value suitable for this evaluation method, for example, threshold value = 4, and if larger than that, sends it to the output unit 9 as it is.

Next, as a conversion candidate word for "so",
It is assumed that "group", "suit", and "sparse" have been sent. The evaluation value F for these is as follows from FIGS. 4 and 5.

"Elementary" ... 1 + 2 + 2x0 = 3 "Group" ... 1 + 3 + 2x0 = 4 "Sparse" ... 1 + 1 + 2x0 = 2 "Action" ... 1 + 2 + 2x0 = 3 Although selected, the evaluation value of this “pair” is not larger than the threshold value 4 of the conversion result reliability determination unit 8, so it is converted to kana and output as “so”.

Therefore, the display "This white is a pair" which gives the operator a feeling of strangeness is avoided, and a relatively natural display "This white is a pair" is obtained.

It should be noted that the above-described evaluation methods are merely examples of the evaluation method used in the present invention, and it is obvious that other evaluation methods may be used. However, it goes without saying that the threshold value provided in the conversion result reliability determination unit 8 differs depending on each evaluation method.

Regarding the display method, as shown in FIG. 6, when the conversion result sent as the most appropriate Kana-Kanji conversion is highly reliable, the notation information of the word that is the conversion result is output as it is. If the reliability of the conversion result is low, the word notation information may be converted to katakana and sent to the output unit 9 (602, 603).

In this case, the sense of incongruity is less than that of the Chinese character notation, and it becomes easier for the operator to determine that the reliability of the conversion result is low, and the response to it is speeded up. For example,
Responses such as changing the input method thereafter and registering a word will be speeded up.

In this way, the reliability of the conversion result that has been evaluated and sent as the most appropriate kana-kanji conversion is judged, and if the conversion result is low, it is converted to kana and displayed. It is possible to avoid the display of natural kana-kanji mixed.

Further, when combined with an editing function capable of re-conversion processing based on the output display, there is an effect that re-inputting can be reduced. Further, there is an effect that the operator can determine that the necessary word should be registered based on the display.

Effect As described above, according to the kana-kanji conversion processing device of the present invention, the reliability of the conversion result is determined, and the conversion result having low reliability is converted to kana and output, thereby giving the operator a feeling of strangeness. It is possible to avoid such an unnatural display of the conversion result and reduce the operator's fatigue level.

[Brief description of drawings]

FIG. 1 is a block diagram of a kana-kanji conversion processing device according to an embodiment of the present invention, FIG. 2 is a flow chart showing the operation of the conversion result reliability judgment unit shown in FIG. 1, and FIG. Flowcharts showing specific examples of operations, FIGS. 4 and 5 are data of a conversion result reliability determination unit according to another embodiment of the present invention, and FIG. 6 is a flowchart showing an output mode according to another embodiment of the present invention. Is. 1: input unit, 2: conversion target character string creating unit, 3: dictionary search unit, 4: word dictionary, 5: connection availability determination unit, 6: connection availability determination table, 7: evaluation unit, 8: conversion result reliability Judgment part, 9:
Output section.

Claims (1)

[Claims] 1. A kana-kanji conversion control in which a plurality of kana-kanji conversion candidates for a character string input by phonetic characters are extracted by searching a word dictionary, and the maximum likelihood candidate word is determined from these to be the conversion result. A kana-kanji conversion processing device having means is provided with means for judging the reliability of the conversion result, and the kana-kanji conversion control means is provided with the above-mentioned means when the reliability judgment means judges that the reliability is low. A kana-kanji conversion processing device, characterized in that the conversion result is converted into kana and displayed, and when it is determined that the reliability is high, the conversion result is displayed as it is.