JPS59221732A - Kana-kanji conversion processing device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] Technical Field The present invention relates to a kana-kanji conversion processing device, and in particular, to a kana-kanji conversion processing device that has divisions such as two words and clauses, y-, b), b) 1) Yurubeta writing input power, and 4 types of kana-kanji conversion. To improve the validity of candidate word extraction and the speed of confirmation processing by considering the position of each word in a sentence when extracting conversion candidate (hereinafter simply referred to as "candidate") words in a processing device. The present invention relates to a kana-kanji conversion processing device that enables the conversion of kana to kanji.

Prior Art In conventional kana-kanji conversion processing devices, selective extraction,
It is known to use the number of words or characters constituting the selected and extracted word string when determining the selected word string. However, in this device, for example,
Something like "I am" (2 words, 5 characters)
Even in sentences with very little ambiguity, there was a problem in that even ``r watakushi'' was not confirmed until a certain parsing length was obtained. In addition, the so-called evaluation values such as the number of words or the number of characters are not used for selecting and extracting candidate words, but other evaluation values are used for selecting and extracting candidate words, so both evaluation values are calculated. There was also the problem that separate means were required for this purpose.

Furthermore, since the definite unit is a word, particles, auxiliary verbs, etc. are also output as one unit, which poses a problem of heavy burden on the operator.

Purpose The present invention has been made in view of the above circumstances, and its purpose is to eliminate the above-mentioned problems in conventional kana-kanji conversion processing devices and improve the speed of confirmation processing with a simple configuration. Another object of the present invention is to provide a kana-kanji conversion processing device that makes it possible to reduce the burden on an operator.

Configuration The configuration of the present invention will be explained below in terms of 5)-phase by way of examples.

First, the selection and extraction of the 111 most likely candidate words will be explained.

In this embodiment, the following will be described later in the first floor. ! 11 The appearance rate of each word in a sentence (hereinafter simply ``
Z'Il'l')'i
It has the following characteristics. The above probability of occurrence is determined by considering the position of the searched word in the sentence to determine whether or not the word searched by the dictionary search means can be connected with the preceding Peng word, and is combined with other parameters. Together, they form an evaluation value for each word. Maximum likelihood candidate word selection and extraction is performed based on this evaluation value.

FIG. 1 is a block diagram showing an example of the configuration of a maximum likelihood candidate word extraction section of a kana-kanji conversion processing device according to an embodiment of the present invention. In the figure, 1 is the input string buffer, 2 is the word dictionary @
11 is a dictionary search unit that searches for and extracts candidate words; 3 is a word buffer that holds the words extracted by the dictionary unit 2;
Reference numeral 4 denotes a connectability test section that refers to the connectability test section 12 to test whether the word extracted by the dictionary search section 2 and stored in the word buffer 3 can be connected to a preceding word. In addition, 5 is a word bassona in a connectable n which stores words that are determined to be connectable by the connectability testing unit 4.
6 refers to the appearance probability table 13 and adds the appearance probability value (1) to each word in the connectable J and 14 word buffers 5. 7 stores the words with appearance probability values. Qi word buffer, 8 is the total appearance probability, 7th shift; b
9 (output sentence qniji 1) 10 is a buffer for holding candidate words that stores the appearance probability value and a single word with an overall evaluation value, which will be described later.

Figure 2 shows the appearance probability table 13 used by the appearance probability recognition section 6.
This shows an example of the content. In the tree table, the word order, which is a combination of the position of the character string to be extracted and the part of speech and character type of the preceding word, is arranged in rows, and the word order is arranged in columns.

FIG. 3 shows an example of the contents of the word dictionary 11.

The frequency rank is, for example, a numerical value obtained by classifying the frequency of use, and the higher the frequency of use, the larger the numerical value.

The operation of the maximum likelihood candidate word extraction section of this embodiment configured as described above will be described below.

As a first example, in input string buffer 1, from the beginning of the sentence "
Let's consider a case where there is an input saying, "I want this shiyashin."

The dictionary@seasearch section 2 performs Shizuma bisa from the beginning of the above input sentence,
``ChiJ+'9mJ, rkono'' etc. are extracted and this A is stored in the JElt hash sofa 3. Each word stored in the word buffer 3 is tested by a connectability testing section 4 to see if it can be connected. The connectivity test is performed by referring to a table in which the part of speech of the preceding word and the connectivity of the word are expressed as 1' and 'O'. In this example, all of the words listed above can be connected, so all can be connected.
Sent to buffer 5.

Next, the appearance probability is determined. The probability of appearance, Shinoda, is
The aforementioned appearance probability table 13 is used. As shown in Figure 2,
When a word is at the beginning of a sentence, only the adnominal ``kono'' is given an appearance probability of '1', and other nouns are
The probability of appearance of an "arc" is 'O'. The words with these values added by f are sent to the word buffer 7 which stores words with appearance probabilities.

The appearance probability comprehensive evaluation section 8 extracts the IJI words from the tli word buffer 7 one by one and performs an appearance probability comprehensive evaluation as follows. That is, the appearance probability comprehensive evaluation value (hereinafter simply referred to as "evaluation value") F is expressed as F (Q, f, c) = I +, +
Define it as f+2e. Here, p is the reading length, f is the frequency rank, and e is the value of the appearance probability. This means that the word with the highest value of the above-mentioned evaluation 1st round F is output.

Here, "F child" has I +2 +2 X O= 3"
For "arc", 1+1+2XO=2 for "this",
2 + 3 + 2 X], = 7 is given as a numerical value. As a result, "kono" is sent to the output character string buffer 9, and "child" and "arc" are stored in the candidate word holding buffer 10, so that they can be stored in the bank trough 9, etc. later.

Next, as a second example, consider a case where "Konoshiyashin" is input in a state where "Onna no" has already been extracted. In this case, the operation before the recognition of the appearance probability is the same as in the first example of Section 2, so the explanation will be omitted.

As shown in Figure 2, the probability of appearance of the word following each particle ``i'' is ``1・'' in ``child'' and ``arc'', and ``kono''
is given as 0'.

Therefore, the above 5.[I value F is 1 for “child”, 1 + 2 + 2 X l”
5 “Arc” has ]+] +2 X 1 =
4 “This” has 2 + 3 + 2 X O=
A numerical value such as 5 is given. In this case, "child" and "
This' is the same value, so jS's first 'I-3Q r
child' is sent to the output string buffer 9.

As mentioned above, according to the device of this embodiment, different words are extracted depending on the position of the immediate consent in the sentence, so the validity of selecting and extracting the most likely candidate word is significantly improved.
I will do this.

Note that the contents of the "-2 evaluation value F" are not limited to those shown in the second embodiment, and various other expressions are possible. For example, use frequency directly instead of frequency rank, and write appearance probability table 1'.

It is possible to use other suitable values instead of 0'.

Next, the determination of a word string (hereinafter referred to as a "maximum likelihood candidate word string") consisting of the maximum likelihood candidate words selected and extracted by a method such as ``two series'' will be explained.

The most likely candidate word string is determined by accumulating the number of independent words among each of the first candidate words, and at point 113' when JL reaches a predetermined threshold. In addition, in this embodiment, at this time, the most powerful silver proverb string corresponding to a separately predetermined time, which is different from the above-mentioned threshold value, is determined.

This will be explained in detail below based on examples.

FIG. 4 is a block diagram showing the main parts of a kana-kanji conversion processing device according to an embodiment of the present invention. In the figure, 21 is an input character string holding unit that holds the input character string, 22 is the above-mentioned Nou likely candidate single edict J11J output unit, and 23 is the maximum likelihood candidate extracted by the maximum likelihood candidate word extraction unit 22! 1 if? The word extraction result creation unit that holds f, 24, is the first candidate 'l'
L::d A part-of-speech fW interpretation unit that interprets the part-of-speech of the word extracted by the extraction unit 22, and 25 is an independent word number counter that is incremented by 1 when the result of part-of-speech interpretation by the part-of-speech interpretation unit 24 is an independent word. It is. In this embodiment, the determination process is performed at point 035, when the content of the independent word counter becomes 3', and the value of the independent word counter is reset to 0' again. In addition, 26 is a confirmation processing unit that is activated by the independent word counter 25 and confirms the word string in the unconfirmed area in units of phrases, and 27 is an unconfirmation processing unit that maintains the word strings in the unconfirmed area in units of clauses. 11°C constant word string holding section, 28
is an output section of the display section which outputs the word string that has been determined by the second determination process.

The operation of the apparatus of this embodiment will be explained below.

A part-of-speech interpretation unit 24 performs part-of-speech interpretation for each of the l words obtained by the maximum likelihood candidate word extraction unit 22 described above. As a result of the part-of-speech interpretation, for words that are determined to be independent words (indicative, idiom), the independent word number counter 25 counts them from 1 to 1.

When the content of the independent word number counter 25 reaches 3', the determination processing section 26 is activated and determines the j-first phrase among at least three phrases including the three independent words.

For example, consider the following sentences.

In this example sentence, the confirmation condition is met when three of the underlined independent words are extracted, and as a result, 1 The phrase ``,'qL element wa'' containing the second independent word ``heimoto'' is defined as 111'.

In this case, the backtrack range due to the inability to extract words after the first phrase is set after the first phrase, that is, 1 m after the second phrase.

In addition, by exploring Kasa's method of displaying the display on the output section 28, for example, displaying the confirmed part in the display character string in a fixed display, and displaying the unconfirmed part in a blinking display, the operator can be informed of the current information. It is not possible to inform MtJ7 of the situation.

FIG. 5 shows another embodiment of the present invention, and in the figure, symbols ], 9, 11 to 13 indicate the same components as shown in FIG. Further, 30 indicates a kana-kanji conversion control unit, and 40 indicates a general meeting 5''j' fullfi M word buffer. , a connection possibility test means, an appearance probability; a B-determined means, and an appearance probability comprehensive evaluation means.

The operation of the device of this embodiment is as follows: First, a dictionary search is performed on the input character string using the word dictionary 11, and when one word is found, a connectability test is performed based on the word. Perform this with reference to Test Table 12.

Furthermore, the appearance probabilities of connectable words are determined using the appearance probability table 13, and a comprehensive evaluation is performed using the appearance probabilities, reading lengths, and frequency ranks. The results are stored in the comprehensively evaluated qt word buffer 40 together with the contents of the word dictionary 11.

The above processing is performed until no matching word is found from the same position in the input character string, and the above verification, lif value, and storage procedures are repeated each time a word is found. As a result, from among the single πt1 in the comprehensively evaluated word buffer 40, the one with the largest Jf value is selected and sent to the output character string buffer 9.

The present invention should not be limited to the above-described embodiments, but may be modified in various ways, and the present invention includes such modifications. As mentioned above, the present invention takes into consideration the position of each word in the sentence when extracting candidate words, thereby increasing the speed of the confirmation process by a factor of 1.7. This has the remarkable effect of realizing a kana-kanji conversion processing unit that can improve the performance of the kana-kanji conversion processing unit.

[Brief explanation of drawings]

FIG. 1 is a diagram showing an example of the configuration of the maximum likelihood candidate word extraction unit, FIG. 2 is a diagram showing an example of the contents of an appearance probability table, FIG. 3 is a diagram showing an example of the contents of a word dictionary, and FIG. FIG. 5 is block I71 showing an embodiment of the present invention. Human string buffer, 2: 7) Kenkyobu, 3; 1
11゜Edict buffer, 4 = Connectability test unit, 5: Connectable word buffer, 6: Appearance probability recognition unit, 7: Word buffer, 8: Appearance probability total evaluation unit, 9: Output string buffer,
10: Buffer for holding candidate words, 11: Word dictionary, 12
: Connectability test table, 13: Appearance confirmation table 1, 21: Human string creation section, 22: First candidate word extraction section, 23: Uji extraction result holding section, 24: Product fli'1fli: Interpretation section , 25
: Independent word counter, 26: Determined prescription ■ part, 27: Undetermined word string creation part, 28: Output part, 30: Kana-kanji conversion part, 40: Comprehensive ii'I' valued single 55 kunofa C

Claims (2)

[Claims] (1) In a kana-kanji conversion processing device, means for giving an evaluation value unique to each searched word, means for determining a maximum likelihood candidate word based on the evaluation value, and means for determining a maximum likelihood candidate word based on the evaluation value. Provided is an independent word number counter that selects an independent word if it is an independent word, and a means for comparing the value of the independent word number counter with a predetermined threshold, and when the content of the independent word number counter exceeds the threshold, 1. A kana-kanji conversion processing device comprising: determining at least the first -- clause of an undetermined maximum likelihood candidate word string. (2) The kana-kanji conversion processing device according to claim 1, wherein the evaluation value is an evaluation value that is calculated by multiplying the probability of appearance of the word depending on its position in a sentence.