JPS589968B2 - Kana-Kanji conversion device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a kana-kanji conversion device, and particularly to a kana-kanji conversion device that accepts a separated or solid sentence in romaji or kana as input, and uses a maximum likelihood evaluation method to convert words (independent words, attached words, etc.) into a dictionary. This invention relates to a kana-kanji conversion device that extracts a word (morpheme), converts it into a kanji code corresponding to the word, and outputs a sentence containing kanji and kana.

Recently, there has been a lot of research into converting sentences input in kana or romaji into sentences mixed with kanji.

Conventionally, the longest match method has generally been used as a conversion method.

FIG. 1 is a diagram illustrating the conventional longest match method.

First, input kana character strings a1, a2 . a3. ...a, then searches the word dictionary.

In the word dictionary, words written in katakana are used as headings, and the corresponding kanji or hirakana and grammatical properties of the words are entered.

For example, a column with the heading "Safe" contains the kanji character "Government" and the fact that the word "Government" is a noun.

In addition, in the column with the heading ``Yotsu'', there is written the word ``Yorotsu'', which has a really strong kanji character, and the fact that this is the conjunctive form of the verb in the five-line conjugation.

First, input kana character string a0. a2. a3. ...Ao is compared with the contents of the word dictionary, and words W1 (al, a2.a3), w2 (al.

a2), W3(al, a2), . . . , Wm(al) are the word candidates to be selected, and among them, the word with the longest kana character string, wl in the example of FIG. 1, is selected.

Next, the remaining kana character strings a4, a5 +..., an
Similarly, the word dictionary is compared with the contents of the word dictionary, and the word corresponding to the longest kana character string is selected.

The same applies below. FIG. 2 illustrates an example of a conventional kana-kanji conversion search method using the longest match method.

This example explains the case of converting the katakana character string ``Seifunomousidenyotsute'' into hirakana mixed with kanji.

In Figure 2, ■ indicates a noun, O indicates an adjunct, and [phase] indicates a verb.

First, the sentence "Seifunomousidenyotsute" and the word dictionary memory are compared using the longest match method.

At first, there is no match, so next we try to match "Seifunomousidenyo", which is one character removed, with the contents of the word dictionary memory.

In this way, when you reach "Safe", it matches, so "
The kanji ``government'' is obtained.

After the kanji ``government'' is obtained, the remaining katakana character strings are compared with the word dictionary memory using the longest match method.

Assume that as a result of repeating such a conversion search, the sentence ``Government's goods are also delivered by cattle'' is obtained.

In Japanese grammar, the verb ``yorite'' does not come next to the noun ``load'', so the connection ``load'' is found to be incorrect.

Therefore, if we choose the number word ``two'' instead of ``load'', this is also a mistake.

Assume that "by" is obtained as a result of multiple trials.

However, since ``de'' is interpreted as a particle, it is impossible for the particle ``ni'' to come next to the particle ``de''.

By performing such grammar checks and conversion searches multiple times, it is possible to obtain the correct hiragana sentence mixed with kanji, ``according to the request of the government.''

Now, looking at the conversion process in Figure 2, after obtaining the kanji for ``Government'', the attached word ``nomo'' was selected.
It can be seen that a considerable number of attempts are repeated in vain.

This is because ``nomo'' has a higher priority than the adjunct ``no'' due to the longest match method.

However, in general, the adjunct ``no'' is used much more frequently than ``nomo,'' and the usage of ``government'' is more likely to occur than the usage of ``government no mo.'' is considered to be high.

Therefore, after selecting the noun "government", the attached word "
If the attached word ``no'' can be selected first instead of ``nomo'', it is possible to improve the kana-kanji conversion speed.

This presentation is based on the above consideration, and the purpose is to efficiently perform kana-kanji conversion by determining words not only by measuring the length of the pronunciation of the word, but also by taking into consideration the frequency of use. Therefore, the kana-kanji conversion device according to the present invention has a word dictionary memory in which words corresponding to kana and the frequency of use of the words are written, and a reading that matches the input kana string of arbitrary length from the beginning of the input kana string. dictionary search means for searching the dictionary for a word having A stack for holding the idioms that have been confirmed, and a maximum likelihood evaluation means for determining the idioms with the highest maximum likelihood evaluation by performing a maximum likelihood evaluation on each word in the stack based on the length of pronunciation and the frequency of use. It is characterized by being prepared.

The present invention will be explained below with reference to the drawings.

FIG. 3 is a diagram explaining the maximum likelihood evaluation method used in the present invention.

In addition to the contents of the conventional word dictionary shown in FIG. 1, the word dictionary shown in FIG. 3 includes usage frequency information for each word.

First, the input kana character string al, C2) C3,...
・Compare the contents of the previous word dictionary and find the words W'(al, C2g C3), w2(a
l, C2), w3(at+a2)s...1wm(al
) as word candidates to be selected.

Next, for each candidate word, evaluation values C1, C2, C3, . . . are given based on the pronunciation length and usage frequency information.

Find cm.

Then, the word with the largest evaluation value is selected.

In the example of FIG. 3, word w2 is selected because C2 has the maximum value, and the remaining kana character strings a3, C4,...t
a, is used for the next word analysis.

Here, when the pronunciation length of a word is li and the frequency of use (for example, the number of past uses) is hi, the evaluation value ci is obtained in the form of C1-hi×32ti, for example.

In this example, the 32 times difference in usage frequency corresponds to the difference in the reading length of one kana character.

In other words, even if the pronunciation length is one kana character shorter, if the usage frequency is 32 times or more, the word with the higher usage frequency is selected.

Here, the evaluation value compares the possibility that the word will be used.
Other values may be used as long as they can be evaluated.

FIG. 4 is a block diagram of a kana-kanji conversion device according to an embodiment of the present invention.

In Fig. 4, 1 is an input kana buffer, 2 is a word dictionary memory, 3 is a dictionary search device, 4 is a connectivity test device, 5 is a matching word stack, 6 is a word buffer, and 6-1 is a reading and kanji information. holding part, 6-2 is a part holding information indicating the pronunciation length of words, 6-3 is a holding part holding frequency information of words, 7 is an evaluation value calculation circuit, 8 is a word/evaluation value stack, 8-1 is the part that holds reading and kanji information, 8-
2 is a part that holds the evaluation box, 9 is a sorting circuit, 1
0 is the priority stack.

The operation of FIG. 4 is as follows.

First, input kana strings al, a2, a3. ...,
an is put into the input kana buffer 1.

Next, the dictionary search device 3 searches the contents of the word dictionary memory 2 and selects all words having pronunciations that match a kana character string of an arbitrary length from the beginning of the kana character string in the input kana buffer 1.

The words selected by the dictionary search device 3 are sent to the connectivity testing device 4.

Connectivity test device 4 (Detailed operation is provided in Japanese Patent Application No. 54-92270)
If the word sent from the dictionary search device 3 corresponds to the pronunciation of the first part of the input kana character string, the word is put into the matching word stack 5 as is.

Furthermore, if the word sent from the dictionary search device 3 does not correspond to the reading at the beginning of the input kana character string, but is a word that follows a word that has already been searched and confirmed for kana-kanji conversion, The connectivity testing device 4 checks whether or not the words sent from the dictionary search device 3 have grammatical connectivity with the immediately preceding confirmed word, and places only those that are acceptable in the matching word stack 5.

In addition, in FIG. 4, illustration of a buffer that holds words for which kana-kanji conversion has been determined is omitted.

Next, the words in the matching word stack 5 are transferred word by word to the word buffer 6.

The word length information is in the 6-2 part of the word buffer.
The frequency information of each word is held in the -3 part, and these are input to the evaluation value calculation circuit 7, and the evaluation value is calculated using the formula described above, and the word/evaluation value stack 8
Put it in part 8-2.

A portion 6-1 of the word buffer 6 holds word pronunciations and kanji information, and these are input as they are to a portion 8-1 of the word/evaluation value stack 8.

In this way, after obtaining the evaluation values for all the words in the matching word stack 5 and storing the words and evaluation values in the word/evaluation value stack 8, the words in the word/evaluation value stack 8 are sorted by the sorting circuit 9. Take out in order of evaluation value,
Transfer to priority stack 10.

and the highest priority in the priority stack 10,
That is, the word with the highest evaluation value is taken out and placed in a confirmed word storage buffer (not shown).

In this manner, after one word is determined, the input kana character string in the input kana buffer 1 is shifted by the kana characters that match the determined word and are removed from the input kana buffer 1.

Then, the same process as above is repeated for the remaining undefined kana character strings held in the input kana buffer 1 to create a sentence containing kanji.

In the above embodiment, in order to find the word with the maximum likelihood evaluation value, the evaluation values of all the words in the matching word stack 5 are calculated and placed in the word/evaluation value stack 8. For example, word buffer 6
The words in the matched word stack 5 are extracted one by one, and the two words are sequentially compared and the word with the larger evaluation value is held in one of the determined buffers. It is also possible to ask for the meaning word.

Furthermore, it goes without saying that the dictionary search device, connectivity testing device, evaluation value calculation circuit, etc. in Example 11 can also be realized by using a general-purpose processing device and storage device.

As described above, according to the present invention, when converting kana to kanji, words are decided by taking into consideration the pronunciation length and frequency of use, which is better than the conventional simple longest match method. Kana-kanji conversion can be performed efficiently.

[Brief explanation of the drawing]

Figure 1 is a diagram explaining the conventional longest match method, Figure 2 is an example of a kana-kanji conversion search method using the longest match method, and Figure 3 is a diagram explaining the maximum likelihood evaluation method used in the present invention. ,
FIG. 4 is a block diagram of a kana-kanji conversion device according to an embodiment of the present invention. In FIG. 4, 1 is an input buffer, 2 is a word dictionary memory, 3 is a dictionary search device, 4 is a connectivity test device, 5 is a matching word stack, 6 is a word buffer, 7 is an evaluation value calculation circuit, and 8 is a word・Evaluation value stack, 9 is sorting circuit,
10 is a priority stack.

Claims (1)

[Claims] 1 Search the word dictionary memory in which words corresponding to kana and the frequency of use of the words are entered, and the dictionary for words that have pronunciations that match an input kana string of arbitrary length from the beginning of the input kana string. a dictionary search means; a connectivity testing means for fully testing the connectivity between the searched word and the previously searched and determined words; and a stack for holding words whose connectivity has been confirmed by the connectivity testing means. and a maximum likelihood evaluation means for determining the word with the highest maximum likelihood evaluation by performing maximum likelihood evaluation on the pronunciation length and usage frequency of each word in the stack. Device.