JPS61221868A - Kana-to-kanji converting system - Google Patents

Abstract

PURPOSE:To convert accurately a KANA (Japanese syllabary) character string into a mixture sentence of KANA and KANJI (Chinese characters), by selecting the independent words with priority. CONSTITUTION:A CPU 1 allocates the different weight values to an independent word and an affix and also calculates the sum total of the weight values of each group extracted as the word candidates. Then a group having a small sum total of weight values is selected with priority in case the weight value lower than the affix is allocated to the independent word. While a group having a large sum total weight values is selected with priority if the weight value higher than the affix is allocated to the independent work. The different weight values are allocated between the independent words and other additional words including affixes, numerals, etc. to increase the probability for selection of independent words. Thus an input KANA character string can be converted quickly into a desired KANA-KANJI mixture sentence without dividing the KANA character string into words of a small number of characters.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a kana-to-kanji conversion method for converting a text input in kana characters into a mixed kana-to-kanji text.

(Prior art) When converting a kana character string input from a keyboard or voice recognition device into a sentence containing kanji, there is a method that determines the kana string that matches the maximum number of characters as a headword and converts it into the corresponding kanji. , the so-called longest match method is used.

This method has the disadvantage that a headword that is probabilistically correct is semantically incorrect, and it takes a long time to select the optimal headword. .

In order to solve this problem, among the headwords stored in the kana-kanji conversion dictionary, attached words such as affixes, numerals, and time are assigned higher weight values than independent words, and these attached words are A kana-to-kanji conversion method has been proposed that reduces ambiguity by preferentially selecting words (Japanese Patent Laid-Open No. 5
9-121529).

According to this method, the same sounds as the prefix or suffix, ``ko'', ``
If you try to convert a word containing "san", for example the kana character string "Kotishisan" (fixed asset), into a sentence containing kana and kanji, the prefix "ko" and the suffix "r-san" will be preferentially selected. There is a problem in that the sentence ``Ko-stop-san'' is mixed with kana and kanji, making it difficult for the independent word ``fixed assets'' to be selected.

(Objective) In view of these problems, the present invention provides a kana-kanji conversion method that can select independent words preferentially and thereby accurately convert a kana character string into a sentence containing kana and kanji. With the goal.

(Structure) Therefore, details of the present invention will be described below based on illustrated embodiments.

FIG. 1 shows an embodiment of a device used in the present invention, and reference numeral 1 in the figure indicates a CPU that controls the entire device based on data input from an input section 2 such as a keyboard.
The input kana character string and control code are output to a kana-kanji converter 3, which will be described later. 3
is the above-mentioned kana-kanji converter, and the code 4 in the figure is 1
In the grammar analysis section, the data bus 6 is input by the kana character string input section 5.
The independent word search unit 7 uses the kana characters tada imported from
, are output to the adjunct word search unit 8 and affix search unit 9, while the headwords selected by each search unit and weight values, that is, a weight value of “2” for independent words and a weight value of “2” for affixes, are output. It is configured to add a weight value of "3" and a weight value of "0" to attached words other than affixes, and to store them in the word candidate storage unit 10 as word candidate data. The independent word search section 7 includes an independent word expansion section 7a and an independent word dictionary index section 7b.
, an independent word dictionary 7c, an attached word search section 8 consists of an attached word index section 8a and an attached word dictionary 8b, and an affix search section 9 consists of an affix index section 9a and an affix dictionary 9b. . Reference numeral 11 denotes a clause composing unit which reads word candidate data stored in the word candidate storage unit 10 and outputs the kana-kanji character string output unit 12 in descending order of the total weight value.
The data is configured to be output to the data bus 6 via the data bus 6.

In addition, the reference numerals in the figure are: 13 is a kana-kanji candidate selection section, 14 is a kana-kanji candidate display section, 15 is a kana-kanji candidate confirmation section, and 16 is a kana-kanji candidate selection section.
indicates the display section, respectively.

In the device configured as described above, when a kana character string ``rkoteishi sanwohi 薖tsu kasuru 1'' is inputted from the input section 2, and then a kana-kanji conversion execution command is inputted, this kana character string is inputted to the kana character string input section 5. 0 to the grammar analysis unit 4 via
The independent word search unit 7. It is directed to the adjunct word search section 8 and the affix search section 9, and searches for headwords in each of the dictionaries 7c, 8b, and 9b from the beginning of the kana character string.

In this way, the headwords of each search section 7, 8, and 9 that matched with the maximum consecutive characters in the kana character string, that is, ■ From the independent word search section 7, "Kotei" (fixed), "Shisan" (assets).

"Hyo Power" (Evaluation) ■From the attached word search section 8, "wo", "suru" Extract.

When the first extraction is completed in this way, similar analysis is performed while shifting the kana character string by one digit and cutting out independent words starting with the second character "te". "Tiku" (stop), "Hyouriki" (evaluation) ■ Extract "wo" and "suru" from the adjunct word search section 8 ■ "ko" from the affix search section 9.

When this second extraction is completed, the same analysis is performed by shifting the kana character string by one digit and cutting out independent words with the third character "i" as the beginning. (Kote), “Ishi” (Stone), “Hyouka”
(Evaluation) ■ “wo” and “suru” are extracted from the adjunct word search unit 8. ■ “san” is extracted from the affix search unit 9.

All the words extracted in this manner are stored in the word candidate storage section 10 with the weight values assigned to the search section 7.8.9.

That is, in the first group, the weight value is 2 for each of "kotei", "shisan", and "hyoka" extracted from the independent word search section 7, and the weight value is 2 for each of "wo" extracted from the adjunct word search section 8. A weight value of 0 is added to each of "Suru" and "Suru". In addition, in the second group, a weight value of 2 is given to each of "tik" and "hyouka", a weight value of 3 is given to "ko", and a weight value of 3 is given to each of "wo" and "suru". weight value 0
is added. Furthermore, in the third group, the weight value is 2 for "Kote", "Ishi", and "Hi1 Uka".
However, a weight value of 3 is assigned to "san", and a weight value of O is assigned to "wo" and "suru".

In this way, each group stored in the word candidate storage unit 1O is tested for the possibility of connection with the headword contained in each group using the part-of-speech information of the headword and the word connection test table. , groups containing headwords that cannot be connected are excluded.

At the end of this test, the total weight value of each group, that is, ■1st group (Kotei (2) + Shisan (2) + Hyouriki (2)) +
(W (0) + Suru (0)) = 6 ■ 2nd group (Tik (2) + Hyouriki (2)) + (K (3))
+ (Wo (0) + Suru (0) ) = 7 ■ 3rd group (Kote (2) + Ishi (2) + Hyoriki (2)) + (Sun (3)) 10 (Wo (0) + Suru (0))=9 is calculated.

Based on this calculation result, the CPU selects the first group whose total weight value is close to the weight value of the independent word, that is, the smallest group, as the first candidate, and displays it on the kana-kanji candidate display section 14. output to section 16.

As a result, "Evaluate fixed assets" is displayed on the display section 16, and since this sentence is the intended sentence containing kana and kanji, it is confirmed by the kana and kanji candidate confirmation section 15.

Note that if the kana-kanji mixed sentence displayed as the first candidate is different from the intended one, the kana-kanji candidate selection section 1
3, the desired kanji-containing sentences can be displayed by sequentially selecting them.

In this example, since the weight value of independent words is smaller than that of affixes, the group with the smallest total weight value is output preferentially when selecting a group. If the total weight value is also set high, the same effect can be obtained by preferentially selecting the one with the largest total weight value.

Also, in this example, the affixes are equally;
, weight values are assigned, but the target kanji can be selected more accurately by differentiating the weight values between the affixes. Further, ambiguity at the time of kanji conversion can be reduced by giving a weight of 1 to attached words and independent words in consideration of their frequency of use.

(Effects) As explained above, according to the present invention, different weight values are assigned between independent words and other affixes and adjuncts such as numerals to increase the probability that independent words will be selected. So,
To quickly convert an input kana character string into a desired sentence containing kanji without cutting it into words with a small number of characters.

[Brief explanation of drawings]

The figure is a configuration diagram showing one embodiment of a device used in the present invention.

Claims (1)

[Claims] In addition to assigning different weight values to independent words and affixes, the total weight value of each group extracted as word candidates is calculated, and when the independent word is assigned a lower weight value than the affix, the weight value is calculated. A kana-to-kanji conversion method that preferentially selects a group with a small total weight value, and a group with a large weight value when an independent word is assigned a higher weight value than an affix.