JPH03276252A - Document treatment system and document treatment equipment - Google Patents

Abstract

PURPOSE: To correctly judge the character kind for even a reading string unregistered in a dictionary by converting the string of reading to the string of phoneme so as to judge the character kinds of a reading string based on the converted phoneme. CONSTITUTION: An inputted reading string is stored in a reading string buffer 110, decomposed into the phoneme strings by a decomposing means 120 and stored in a phoneme buffer 140. Then the phoneme string is sent to a judging means 150, which refers to a frequency table 160 to judge the character kind of the phoneme string. A conversion means 170 converts the reading string stored in the reading string buffer 110 to a specific character kinds corresponding to this judging result and sends the conversion result to a character string buffer 180 to display it on an output means 30 such as CRT or LCD, thereby judging a correct character kind concerning a reading string unregistered in a dictionary.

Description

[Detailed description of the invention] A. Industrial application field The present invention has a function of determining the correct character type (character type) of an input string of readings, that is, whether the input string of readings is correctly written in kanji. Has a function that determines whether something should be written in katakana or hiragana, etc.
It is related to document processing devices.

B. Conventional technology In conventional document processing devices, if the kanji or katakana corresponding to the input reading sequence (usually a hiragana sequence) is not registered in the dictionary, the conversion key is pressed several times. Even when pressed, the input string of readings is usually left as it was when input, or the string of input readings is inappropriately divided into smaller pieces and converted into inappropriate kanji combinations (taji). Met. That is, it is difficult to correctly convert words that are not registered in the dictionary (unknown words). In particular, katakana is often used for words and phrases that are increasing day by day, such as place names and people's names, as well as foreign words and scientific and technological terms, so it is often not possible to register all katakana words and phrases in dictionaries. Therefore, in conventional document processing devices, when the input string of readings should be converted into katakana, the input string of readings may remain as it is, or it may only be converted into ``taji''. , I had to perform a special operation for katakana conversion, which is different from the normal conversion key operation.

In addition, it is desirable to determine the correct character type of an input string of pronunciations, not only when converting to katakana, but also when performing appropriate document processing.

Problems to be Solved by the C0 Invention The purpose of the present invention is to provide a document processing method and device having a function of correctly determining the character type even for a sequence of pronunciations that are not registered in a dictionary. Is Rukoto.

Means for Solving Problem D2 The present inventor discovered that hiragana words have a sound that feels like hiragana, and katakana has a sound that feels like katakana. Each character type has its own sound and the way the sounds are connected, which indicates its character. Therefore, when a sequence of pronunciations is converted into a sequence of phonemes, the character of the character can be recognized in the converted phoneme sequence. I discovered that
The present invention has been accomplished based on such findings.

The document processing method according to the present invention attempts to achieve the above object by converting a string of pronunciations into a string of phonemes, and determining the character type of the string of pronunciations based on the converted string of phonemes. .

A first document processing device according to the present invention includes: an input means for inputting a string of pronunciations; a decomposition means for decomposing the input string of pronunciations into a string of phonemes; By providing a determining means for determining the character type of a string of readings, and a converting means for converting the string of readings into a specific character type according to the determining means, it is possible to convert the input string of readings into correct character types. This is what we are trying to convert into the notation.

Further, a second document processing device according to the present invention includes a reading string converting means for converting a character string into a reading string, a decomposing means for decomposing the reading string into a phoneme string, and a reading string converting means for converting a character string into a reading string. a determining means for determining the character type of the string of readings, and a comparing means for comparing the character type indicated by the determining means with the character type of the character string;
By providing an output means for displaying the comparison results,
It attempts to check the correctness of a word or phrase that has already been written in a certain character type.

E. Example Hereinafter, embodiments of the present invention will be described based on the drawings.

FIG. 1 shows an embodiment of a document processing device according to the present invention.

In the figure, an input hand R20 is a keyboard, a handwriting input tablet, or the like for inputting a reading string, and the reading string input manually is stored in a reading string buffer 110. The yomi string stored in the yomi string buffer 110 is decomposed into a phoneme string by the decomposition means 120, and the decomposed phoneme string is stored in the phoneme string buffer 140. The decomposition means 120 refers to the phoneme conversion table 130 and decomposes (converts) the string of pronunciations into a string of phonemes.

The string of phonemes is sent to the determining means 150, which determines the character type of the string of phonemes by referring to the frequency table 160, and sends the determination result to the converting means 170. Conversion means 170
is the reading sequence stored in the reading sequence buffer 110,
According to the determination result, the character is converted into a specific character type, the conversion result is sent to the character string buffer 180, and the character string buffer 18
The contents of 0 are displayed by output means 30 such as CRT or LCD.

Next, the present embodiment will be described in more detail using an example in which the reading string is "Aiden Difumei".

The reading string "Aiden Tiffany" inputted by the input means 20 is stored in the reading string buffer 1 using a predetermined character code.
10, and the stored string of pronunciations "Aidendeifumei" is decomposed into a string of phonemes by the decomposition means 120.

FIG. 2 shows a phoneme conversion table 130 that is referred to by the decomposition means 120. When the phoneme conversion table 130 is used, the following Table 1 shows the correspondence between the pronunciation string "Aidendeifumei" and the phoneme string.

Table 1 Correspondence between the pronunciation sequence “Aiden teifumei” and its phoneme sequence, that is, the pronunciation sequence “Aiden teifumei” is a means of decomposition! 2
0 makes a sequence of phonemes r AIDENNTEiHUaI J
It is decomposed (converted) into The phoneme sequence r AIDENNTEiHUaI J in the phoneme sequence buffer 140 is the determination means 1
50, and the determining means 150 refers to the frequency table 160 to determine the phoneme sequence r AIDEN.
Determine the character type of NTEiHUaI J.

A frequency table 160 is shown in FIG. All phonemes are arranged on each of the vertical and horizontal axes of the frequency table 160, and at the intersection of a specific vertical axis phoneme row and a specific horizontal axis phoneme column, the vertical axis phoneme column is arranged. Next, how often do phoneme connections in which the phonemes on the horizontal axis are connected occur in words that should be written in katakana, or in words that should be written in hiragana? It shows the relative frequency of each character type. In other words, the frequency table 160 numerically represents the degree of Katakana-likeness or Hiragana-likeness for all combinations (connection forms) of two phonemes.

The frequency table 160 of FIG. 3 will be further explained.

The frequency table 180 is created as follows.

First, the number of occurrences of each hiragana and katakana word in the existing document is counted, and the reading sequence is converted into a phoneme sequence with reference to the phoneme table shown in FIG. Tables 2 and 3 below show examples of the number of occurrences of hiragana words and phrases and the number of occurrences of katakana words in the referenced existing documents, respectively.

Table 3 Examples of the number of occurrences of katakana words Table 2 Examples of the number of occurrences of hiragana words (blank below) From Tables 2 and 3 obtained in this way, the occurrence of phoneme pairs that appear in hiragana words Count the number of times a phoneme pair appears in a katakana phrase. For example, if we calculate the number of times the phoneme pair rAIJ appears in hiragana words from Table 2, we find that rAIJ appears in the second row of Table 2.
The number of times it appears in AAIUJ is 1, and rAIJ in the 3rd line
Since the number of times it appears is two, the number of times it appears is 3 times, which is the sum of these numbers. The results of similar operations performed on a suitable number of existing documents are shown in Tables 4 and 5 below, respectively. That is, Table 4 shows the number of times phoneme pairs appear in hiragana words, and Table 5 shows the number of times phoneme pairs appear in katakana words.

Table 5 Examples of the number of occurrences of phoneme pairs in katakana phrases Table 4 Examples of the number of occurrences of phoneme pairs in hiragana phrases From these tables, for example, the phoneme pair rUaJ (``U
`` first, followed by rB'')
It can be seen that although it appears only once in Hiragana words, it appears 2239 times in Katakana words.

The character type of the phoneme string stored in the phoneme string buffer 140 (
Although the fourth and fifth tables obtained in this way may be used as they are to determine the frequency table 160 (hiragana or katakana), in this embodiment, the frequency table 160 (
FIG. 3) is created and this frequency table 160 is used.

Next, a procedure for creating the frequency table 160 from Tables 4 and 5 will be explained.

The total number of occurrences of the phoneme pairs shown in Table 4 is s
If b is the total number of occurrences of the phoneme pairs shown in Table 5, and Sk is the total number of occurrences of the phoneme pairs shown in Table 5, then the value of sh and the value of Sk usually do not match. Therefore, each number in Table 5 is multiplied by Sh/Sk J so that the total number of occurrences of phoneme pairs is the same in Table 4 and Table 5.
Take the values in the table as they are.

The value of the intersection position (number of occurrences) between row i and column j in Table 4 is
lj, and Kij is the value obtained by multiplying the value (number of occurrences) of the intersection of row i and column j in Table 5 by rSh/SkJ, then row i in frequency table 160 in FIG. Taking the value (relative frequency) of the intersection position of and column j as Rij, then Rfj
It is determined from t Hij and Kfj as follows.

(i) When 0 * Hij and 0≠Kij: Rij
Take = RcX Log(Kij/Hij).

Here, Re is a weight constant for actual calculation and is usually 1. Also, it always takes a constant value in one relative frequency table 160. When performing calculations that round down the decimal places, use fl
Increase a to reduce the truncation error. Also, the base of the logarithm is the base of the natural logarithm (e=2.718.,,').

(ii) When Hij=O and Kij=O: Rij
= Take O.

(iii) When Hij≠O and Kij=Q: Take Rfj = -Max.

However, Max is the maximum absolute value that can be handled, and therefore,
r-Max J is the minimum value that can be handled.

In this case, the phoneme pair appears in hiragana words, but never in katakana words. In the frequency table 160 of FIG. 3, cases that appear frequently in katakana words and phrases are represented by large values on the positive side, and cases that appear frequently in hiragana words and phrases are represented by large values on the negative side. , In this case, the value of Rlj is taken as the negative side of the maximum that can be handled (meaning the maximum absolute value).

Values (relative frequencies) f in the frequency table 160 for all adjacent phoneme pairs in 1HUaIJ: When calculated, the results are as shown in Table 6 below.

(Left below) (iv) When Hij=O and Kij≠0: Ri
Take j = + Max.

In this case, the phoneme pair appears in katakana words, but never in hiragana words. Therefore, in this case, the maximum positive value that can be handled is taken as the value of Rij.

Since the frequency table 160 of FIG. 3 is obtained in the above manner, the process of determining the character type of the phoneme string V ATDENNTEiHUaI J with reference to the relative frequency table 160 will now be described. First, the phoneme sequence rAIDENN
TE Table 6 Phoneme sequence rAIDENNTEiHUal
Relative frequency value sequence r of each phoneme pair of J AIDENNTEiHUal It is determined that the character type of J is Katakana. Further, the degree to which the judgment result is expected to be correct is defined as the judgment degree as follows.

The total value has the same sign as the value on the right side of Table 6, that is, the total value of the relative frequency values of the phoneme pairs is +433. Therefore, the sound is the 6th
When applied to tables, the judgment level = 95% (=457/4
81). Therefore, the sequence of phonemes rAIDENNTEi
HUaIJ is determined to be katakana with a determination degree of 95%. The determination result obtained in this way is determined by the determination means 160.
The conversion means 170 converts the string of pronunciations stored in the reading string buffer 110 ``Otenteifai'' into the character type katakana based on the determination result, and sends the converted string to the character string buffer 180. , output means 30
is written as ``identify'' in katakana.

Above, we have explained the process by which the character type for the reading string ``aidendi-fai'' is determined to be katakana and is written in katakana. Explaining the process of determining hiragana.

First, if we put the pronunciation column ``Madorokkoshii'' in the phoneme column, we get something like Table 7.

Table 8 Phoneme sequence r MADOROKKOSI I,
Relative frequency values for each phoneme pair in J Table 7 Correspondence between the pronunciation sequence ``Madorokkoshii'' and its phoneme sequence Thus, the phoneme sequence r MADOROKKOSI I
, J is found, next we need to find the phoneme sequence r MADOROK
Table 8 shows the relative frequency values of all phoneme pairs in KOSI I J.

(Left below) Here, the total value of the values in the right column of Table 8, that is, the relative frequency values of phoneme pairs, is -38. Therefore, the sequence of phonemes r M
The character type of ADOROKKOS I I J is determined to be Hiragana. Also, the judgment degree is 64% (=85/13
2). In this way, according to this embodiment, it is possible to convert words (reading sequences) that are not registered in the dictionary into the correct character type.

FIG. 4 shows an embodiment other than the above, and in this embodiment, the character type of the word in the sentence after Kana-Kan conversion or the sentence translated by an automatic translator ( verification).

In FIG. 4, a sentence memory 210 stores sentences that have already undergone kana-kan conversion or sentences that have been translated by an automatic translator. The character strings in the sentences in the sentence memory 210 are extracted by character type by the character string extraction means 220, the extracted character strings are stored in the character string buffer 230, and the character types of the extracted character strings are stored in the character type buffer 240. be done. The character string stored in the character string buffer 230 is converted into a reading string by the reading string conversion means 250. At this time, the pronunciation conversion means 250 refers to the pronunciation table 260 that shows the correspondence between character strings and pronunciation sequences.

The pronunciation sequence obtained by the pronunciation conversion means 250 is decomposed into a sequence of phonemes by the decomposition means 120, and the decomposition means 120 decomposes the sequence of phonemes into a sequence of phonemes, as in the embodiment of FIG. (conversion), and the determining means 150 uses the relative frequency table 160t! : Reference and determine the character type of the phoneme string,
The determination result is sent to comparison means 170.

The comparison means 170 compares the judgment result with the character type buffer 2.
40, the comparison result is sent to the output buffer 280, and the output means displays the comparison result. The display method at this time is, for example, if the determination result and the character type information stored in the character type buffer 240 are different, the character string is highlighted, and at the same time, the character type indicated by the determination result is displayed. .

Note that the correspondence between pronunciations and phonemes is not limited to that shown in the phoneme conversion table shown in FIG. 4, but may be, for example, the correspondence shown in the phoneme conversion table shown in FIG. 5.

In addition, in the above embodiment, the frequency table 160 shows the appearance frequency of all connection forms of phonemes, but the frequency table is used to determine the character type, and the connection of some phonemes, which are the bold elements above. It may also indicate the appearance frequency of only the form. Also, when determining the character type,
Rather than constantly checking the frequency of appearance of all connected forms of the phoneme string to be determined, the determination of the character type may be completed immediately when a phoneme string that is extremely characteristic of a certain character type is recognized. Furthermore, the present invention is applicable not only to Japanese, but also to document processing for foreign languages that have different character types.

F8 Effects of the Invention As described above, according to the present invention, it is possible to determine the correct character type for a sequence of pronunciations that are not registered in a dictionary.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a document processing device according to the present invention, FIG. 2 is a diagram showing a phoneme conversion table of the embodiment, and FIG. 3 is a relative frequency table of the embodiment. FIG. 4 is a block diagram showing the configuration of an embodiment other than the above, and FIG. 5 is a chart showing a phoneme conversion table other than the above.

Claims (5)

[Claims] (1) A document processing method including a process of converting a string of readings into a string of phonemes and determining the character type of the string of readings based on the converted string of phonemes. (2) an input means for inputting a string of readings; a decomposition means for breaking down the string of readings into a string of phonemes; a determining means for determining the character type of the string of readings based on the string of phonemes; A document processing device comprising: a conversion means for converting a string of characters into a character type indicated by a determination means; and an output means for displaying a conversion result. (3) The document processing apparatus according to claim 2, wherein the determining means determines the character type based on the connection form of at least a portion of each phoneme forming a string of phonemes. (4) The document processing device according to claim 2 or 3, wherein the determining means refers to a frequency table representing a relationship between a phoneme connection form and the frequency with which the connection form appears in character types. (5) reading string conversion means for converting a character string into a reading string;
a decomposition means for decomposing a string of pronunciations into a string of phonemes; a determining means for determining the character type of the string of pronunciations based on the string of phonemes;
A document processing device comprising: a comparison means for comparing the character type indicated by the determination means with a character type of the character string; and an output means for displaying the comparison result.