JPH0246976B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a kanji-kana conversion device that converts a Japanese text containing kanji and kana into a kana character string that is the reading thereof.

Various types of kana-kanji conversion devices have been proposed for converting kana character strings into Japanese text containing kanji and kana, but there is a method that converts Japanese text containing kanji and kana into the kana character string that is its reading. No conversion has been done. A kanji-kana conversion device that converts Japanese sentences containing kanji and kana into kana character strings that are the pronunciation is necessary for, for example, a machine that reads books aloud, and also corrects the word conversion in the kana-kanji conversion device. It is also an effective means in various situations.

The present invention meets the above requirements, and includes:
The purpose of the present invention is to provide a kanji-kana conversion device capable of converting a Japanese sentence containing kanji and kana into a kana character string that is its reading. Therefore, the kanji-kana conversion device of the present invention is a kanji-kana conversion device that converts a kanji-kana mixed sentence into a kana character string that is its reading, and includes an input device 1 capable of inputting a kanji-kana mixed sentence, and a kanji notation or A word dictionary 5 having a plurality of entries in which are written keys consisting of words written in kana or written in kanji and kana, word pronunciations, grammatical information, and frequency of use; and a grammar dictionary 6 in which information on connectivity between parts of speech is written. , an analysis device 3, and a candidate word extraction device 4, the candidate word extraction device 4 searches the word dictionary 5 using the character string on the right side of the analysis position given from the analysis device 3 as a key, and A candidate word that can be grammatically connected to the previous most likely candidate word is selected from among the words obtained as a result of the search with reference to the grammar dictionary 6, and the candidate word obtained as a result of the selection and the pronunciation of the candidate word are included. The analysis device 3 is configured to notify the analysis device 3 of related information, and the analysis device 3 initially sets the analysis position to the left end of the input character string, and uses the input character string continuing to the right of the analysis position to the candidate word extraction device. 4, and when candidate words and related information are notified from the candidate word extraction device 4, the maximum likelihood candidate word is selected from among the candidate words by evaluating the number of characters and frequency of use of the word, and the maximum likelihood candidate word is selected. It is configured to shift the analysis position to the right by the number of characters in the word, and provide the candidate word extraction device 4 with a character string that continues to the right of the analysis position in the input character string, provided that the analysis position is not at the right end of the input character string. It is characterized by the presence of

Hereinafter, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention, and FIG.
The figure is an explanatory diagram of the operation of the embodiment of FIG. 1.

In FIG. 1, 1 is an input device, 2 is an output device, 3 is an analysis device, 4 is a candidate word extraction device, 5 is a word dictionary, and 6 is a grammar dictionary.
The input device 1 is, for example, an optical character reader or a real-time handwritten character input device. The output device 2 is a part that sends the kana character string obtained by the kanji-kana conversion device to another device. The analysis device 3 can be thought of as controlling the whole, and includes, for example, a function of passing a key sentence containing kanji and kana to the candidate word extraction device 4, a function of moving the analysis position of a sentence, a function of selecting a maximum likelihood candidate, It has a function of concatenating the kana with the readings found and passing them to the output device 2. The candidate word extraction device 4 indexes the word dictionary using the kanji/kana mixed sentence sent from the analysis device 3 as a key, extracts matching words, and refers to the grammar dictionary to select candidate words from the single group. The word dictionary 5, which has a retrieval function, has multiple entries, and each entry contains key kanji (including hiragana), reading (katakana), grammatical information, usage frequency, etc. in JIS kanji code. It is written. The above-mentioned grammatical information means, for example, information regarding parts of speech and conjugations. The grammar dictionary 6 shows what types of words can be connected with what words, and various parts of speech are written in the first row and first column of the matrix, and their intersection points are written in the first row and first column of the matrix. Information indicating whether connection is possible is written.

Next, the operation of the embodiment shown in FIG. 1 will be explained with reference to FIG. 2. When converting a sentence containing kanji and kana into a kana character string, the analysis device 3 sets i to 1 and sets the analysis position to the left end of the sentence.

The analysis device 3 passes the sentence to the right of the analysis position to the candidate word extraction device 4. The candidate word extraction device 4 extracts all matching words from the word dictionary 5 using the given sentence as a key. This extraction method will be explained. Let the input character string be a ₁ ...a _o , and let the leftmost input character of the partial input character string that has not been converted into a word string be a _j . In this case, candidate word extraction device 4
extracts from the word dictionary 5 all words with the heading a _j ...a _j+k (1≦j≦n, 1≦j+k≦n).

For example, if there is a kanji-kana character string "Japanese sentence" to the right of the analysis position, the words "Japan", "Japanese", "Japanese sentence" and their related information (i.e., reading, grammatical information, usage frequency, etc.) are taken out from the word dictionary 5. After all matching words have been extracted, the candidate word extraction device 4 refers to the grammar dictionary 6 and sets {candidate}i the set of only words that can be grammatically connected to the maximum likelihood candidate (i-1). . Note that sentence-initial grammar rules are applied to words at the beginning of sentences. The candidate word extraction device 4 passes the candidate words and related information (ie, pronunciation, grammatical information, usage frequency, etc.) to the analysis device 3. If {candidate} i exists, the analysis device 3 evaluates the number of characters in the word (in this case, the number of kanji characters) and the frequency of use, selects the maximum likelihood candidate (i), and selects the maximum likelihood candidate (i). Shift the analysis position to the right by the number of characters, and check whether the analysis position is at the right end of the sentence. If the analysis position does not exist at the right end of the sentence, the analysis device 3 updates i to i+1 and passes the character string to the right of the analysis position to the candidate word extraction device 4. This causes the above operation to be repeated. If the analysis position is at the right end of the sentence, it is considered to be the end 1. At end 1, the analysis device 3 connects the readings of the maximum likelihood candidate (1) to the maximum likelihood candidate (i) and passes this to the output device 2.

{Candidate} When notified that i does not exist, the analysis device 3 checks whether or not i is 1, and if i is not 1, sets i to i-1 and sets the analysis position to the maximum likelihood candidate ( Shift it to the left by the number of characters in i),
{Candidate} The maximum likelihood candidate (i) is removed from i, and the process illustrated and explained is repeated for the obtained set of candidate words. If end 2 is reached, it means that the kana character string has not been found.

As is clear from the above description, according to the present invention, it is possible to obtain a kanji-kana conversion device that converts a sentence containing kanji and kana into a kanji character string with the correct reading.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of one embodiment of the present invention, and FIG.
The figure is an explanatory diagram of the embodiment of FIG. 1. 1... Input device, 2... Output device, 3... Analysis device,
4... Candidate word extraction device, 5... Word dictionary, 6... Grammar dictionary.

Claims (1)

[Scope of Claims] 1. A kanji-kana conversion device that converts a kanji-kana mixed sentence into a kana character string that is its reading, comprising an input device 1 capable of inputting a kanji-kana mixed sentence, and a kanji notation, a kana notation, or a kanji-kana character string. A word dictionary 5 that has a plurality of entries in which keys consisting of words written in mixed spellings, word pronunciations, grammatical information, and frequency of use are written; a grammar dictionary 6 in which information on connectivity between parts of speech is written; and an analysis device 3. , a candidate word extraction device 4, the candidate word extraction device 4 searches the word dictionary 5 using the character string on the right side of the analysis position given from the analysis device 3 as a key, and refers to the grammar dictionary 6. From among the words obtained as a result of the search, a candidate word that can be grammatically connected to the previous most likely candidate word is selected, and the candidate word obtained as a result of the selection and related information including the candidate word's pronunciation are analyzed as described above. The analysis device 3 initially sets the analysis position to the left end of the input character string and provides the input character string continuing to the right of the analysis position to the candidate word extraction device 4, and When candidate words and related information are notified from the candidate word extraction device 4, the most likely candidate word is selected by evaluating the number of characters and frequency of use of the word from among the candidate words, and only the number of characters of the most likely candidate word is analyzed. The candidate word extraction device 4 is configured to shift the position to the right and provide the candidate word extraction device 4 with a character string that continues to the right of the analysis position in the input character string on the condition that the analysis position is not at the right end of the input character string. Kanji-kana conversion device.