JPS6210762A - Kana to kanji conversion system - Google Patents

Abstract

PURPOSE:To minimize an analysis error by providing an input means equipped with a homonym selection key by which a user changes a converted notation to a homonym notation and a homonym selection correcting and registering means that stores an operation content corrected by the homonym selection key. CONSTITUTION:At a preceding process part 1, before the character is converted to KANJI (chinese character) by a word dictionary, a preceding process is performed in such a way that: a flat written sentence is sectioned to omit a wasteful process and also, it is classified to a Chinese word and an auxiliary word, or with retrieving a KANA (Japanese syllabary) appeared frequently, it is removed from the converting object, etc. Next, at a candidate extracting part 2, a corresponding work is converted to a KANJI by the control of a conversion control part 5 and a bit of converted information or a bit of removed information is transferred to and stored and managed at a conversion information management part 3. After an eddition at the conversion control part 5, the bits of information is outputted to an output control part 4, and after a feedback process such as a back track, etc., a KANJI and KANA combined sentence is outputted. Thus, the analysis error can be always reduced to the minimum level.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a kana-kanji conversion method, and more particularly to a learning method for a kana-kanji conversion input method.

BACKGROUND ART Conventionally, kana-kanji conversion using input units of single kanji, words, bunsetsu, extended (run) bunsetsu, and sentences has been put into practical use.

In these kana-kanji conversion systems, if the desired reading of an input word cannot be obtained, it is usually possible to select a homophone orthography. Furthermore, some systems learn the operator's selections and give priority to the notation selected before the next time. in this case.

This is not a problem when inputting single kanji or words because the connection between words is not checked, but when inputting a phrase, extended clause, or sentence, the connection between words is generally determined based on the part of speech information of the word. Usually, a method is used to narrow down candidates that can be researched and converted and displayed. In this case, in the case of 0 bunsetsu units, in which the learned candidate is replaced with its part-of-speech information and the next candidate is extracted, this method generally performed the analysis in a good direction; In units.

If the candidates for a word in a sentence change, the structure of the following words will change significantly, and subsequent word-word connection analysis may proceed in a direction that is unsuitable for conversion candidates. This tendency is most evident in words with short pronunciations.

Purpose The purpose of the present invention is to improve such conventional problems and to always minimize parsing errors when learning homophone selection in the process of inputting kana and converting it into sentences containing kanji and kana. To provide a kana-kanji conversion method that can achieve both the candidates desired by the user and the candidates output by the device while suppressing the problems.

Configuration In order to achieve the above object, the kana-kanji conversion method of the present invention inputs the Japanese pronunciation in kana (or romaji) and searches a word dictionary to convert the kana-kanji into a sentence containing kanji and kana. The conversion method includes an input means equipped with a homophone selection key that allows the user to change the converted notation to a homophone notation, and a homophone selection correction registration process that stores the operation content modified by the homophone selection key. and when there is an input with the same pronunciation after registering in the registration means, or when the notation of the memory contents of the registration means can be a conversion candidate.
゛ is characterized in that the notation is the notation of the first candidate for conversion.

The configuration of the present invention will be explained in detail below using an embodiment.6 Fig. 1 is a processing flowchart of a kana-kanji conversion method showing an embodiment of the present invention.

When the initial processing (101) is completed, an input request is displayed on the CRT display device (102), and if the operator has the key strength (103), it is checked whether it is character input or other input (10'4). '), in the latter case, check whether it is a conversion display modification operation or something else (e.g., another word processor function, or another conversion function) (110, 11
2). If it is a termination operation or other operation, perform the corresponding process, and if it is a modification operation, perform the modification process (1
14) If there is a homophone selection operation, the operation contents are stored in the main memory in the form of Table 12 in the homophone selection correction registration process (115, 116).

When it is an input string, preprocess the string (105
), a word candidate string is extracted with reference to the word dictionary 10 and grammar dictionary 11 (106). Refer to the registration table 12 that was registered earlier by giving priority to homophone selection candidates, and if there is a corresponding one, change the first conversion candidate to that candidate and perform the first conversion string editing process to create a conversion candidate. Display as (1
07,108°109). In this way, in the system shown in FIG. 1, by registering and learning the operator's homophone selection and modification operations, the results can be used preferentially to speed up the processing.

FIG. 2 is an explanatory diagram showing an example of the correction operation in FIG. 1.

Here, when the "reading" is input for the "input original text" and the "converted sentence" is displayed, the operator can correct the spelling that differs from the original text using the homophone selection key (conversion key) as shown in the example. do. For example, the correction operation is "Background" → "Dear sir",
Correct it as follows: "Reporter" → "Your company", "Seiei" → "Kyoei".

Figure 3 shows the result of the correction operation in Figure 2.
It is a figure showing the contents of "Homophone Selection Correction Registration Table" (Third Example).

The table shows (a) the part where the homonym selection correction key operation was performed.
Reading length, (b) reading character string, (c) notation character string, (d
) Registration flag, and (e) notes are stored. Note that the registration flag is set when the item is already registered].

It is 0 if it has not been registered yet. Moreover, even if only the part shown in FIG. 3A is used (i2 embodiment), almost the same effect can be obtained.

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

1 is an input preprocessing section, 2 is a candidate extraction section, 3 is a conversion information management section, 4 is an output control section, and 5 is a conversion control section. When the reading is input, the input preprocessing unit l processes the reading.

Before converting from a word dictionary to kanji, in order to avoid unnecessary processing, we separate solid text and classify it into kanji and adjunct words.
Alternatively, preprocessing such as searching for frequently occurring kana sentences and removing them from conversion targets is performed. Next, the candidate extraction section 2 converts the corresponding word into kanji under the control of the conversion control section 5, and transfers the converted information or excluded information to the conversion information management section 3 for storage management. After being edited by the conversion control unit 5, it is output to the output control unit 4, and after performing feedback processing such as backtracking, a sentence containing kanji and kana is output.

FIG. 5 is a perspective view of the input device of FIG. 4.

As shown in Figure 5, the input device is divided into a display section and a keyboard section, and the keyboard section has function keys,
In addition to character keys, conversion control keys, numeric keys, correction keys, candidate selection keys, delimiter selection keys, etc. Further, the display section displays a text area, a conversion display line, a correction line, a reading input display, and the like.

FIG. 6 is a diagram showing the configuration of the candidate extraction section of FIG. 4.

The candidate extraction section 2 includes a candidate extraction control section 6, a candidate word string storage section 7, a dictionary search section 8, and a conversion dictionary storage section 9. When the kana character string signal is input to the candidate extraction control unit 6,
The control unit 6 searches the conversion dictionary storage unit 9 using the dictionary search unit 8 from the beginning, and extracts all conversion candidates corresponding to the reading.

When one conversion candidate does not exist, the dictionary search section 8 notifies the candidate extraction control section 6 to that effect. The candidate word string is
The candidate word string storage unit 7 stores the candidate word strings.

FIG. 7 is a processing flowchart of a registration method when selecting a homophone correction according to the present invention.

First, when there is a modification request from the operator, it is determined whether the registration conditions are met (201), and if they are,
First, the registration position in the registration table is set (202), and the reading length is registered (203). The second step is to register the reading character string 204), the third step is to register the written character string 205), and the fourth step is to set the registration flag to "register" (206). Note that if some conditions are omitted from the registration table, the processes (203, 206) are omitted. As mentioned above, even if it is deleted, there is a considerable effect.

Also, the sleep conditions in the process (201) are in correction mode.

And when using normal kana skills, refer to a dictionary and think of possible word breaks. Further, to set the registration positions, for example, a fixed length table as shown in FIG. 3 is set, and the registration positions are registered in the order of serial numbers. When the table becomes full, the serial numbers are overwritten and registered again from the beginning. And the reference behavior is
Starting from the most recent registered position, reference is made in reverse order of the serial number (referred to as the ring method).

FIG. 8 is a flowchart of the process of giving priority to homophone selection candidates according to the present invention.

First, a word candidate comparison position is set to the top position of the candidate sequence table (301), and it is determined whether the reference conditions of the registration table are satisfied (302). If the list is complete, the registration table is searched (303), and it is determined whether a corresponding item exists (304).

If there is a corresponding one, the notation of the first candidate for conversion display is set as the notation of the corresponding candidate (305). On the other hand, if the reference conditions for the registration table are not in place, or if there is no matching item in the registration table, it is determined whether there are any candidates to compare (306), and if there are no candidates, the process is terminated, and if there are The candidate comparison position is set to the next position in the candidate sequence table (307). Note that the reference conditions for the process (302) are input mode and normal kana-kanji conversion. Further, the candidate sequence table is shown in FIG. 11, which will be described later.

FIG. 9 is a processing flowchart of the eighth registration table search.

When searching the registration table by giving priority to homophone selection candidates,
First, set the reference position to the latest registered position (401), determine whether the reading lengths match (402), and if they do not match, determine whether there is still something to refer to (402).
07), Set the reference position to the previous latest registered position (
408). If the reading lengths match, it is first determined whether or not the reading character strings match (403). If there is no match, determine whether there is anything else to refer to, and
A reference position is set to the previous position (407, 408).

If they match, it is determined whether the registration flag is in the registered state (404), and if it is in the registered state, it is determined whether the written character strings match (405). If the written characters match, the table position where the corresponding homophone candidate is located is set (406
). Furthermore, if the registration flag is not in the registered state or if the written character strings do not match, a state is set in which there is no corresponding homophone candidate (409). In the second embodiment, since there is no reading length and registration flag in the registration table, (4
02) (404) is deleted. In the above process, checking whether it is applicable by "reading length" is as follows:
This is effective when there is sufficient space in the main memory and processing speed is a priority. In addition, the "registration flag" is set in the registered state at the time of registration, and when there is a reconversion with the same reading, the "registration flag" is set to the canceled state and the registration candidate is invalidated. When a homophone selection is made, it is registered in a new position.

FIG. 10 shows the cancellation of homophone selection correction registration according to the present invention (
3 is a processing flowchart for the third embodiment).

First, it is determined whether the reference conditions of the registration table are satisfied (501), and if they are, the reference position is set to the latest registered position (502). Next, it is determined whether the reading lengths match (503), and if they match, it is determined whether the reading character strings match (504). If both match, set the registration flag of the reference position to "Cancel" (5
05). Also, if the reading length or reading strings do not match.

Check whether there is anything else to refer to (506);
If so, set the reference position to the previous latest registered position (
507), and then move on to comparing and determining the reading length again (503).
). If there is nothing to refer to, the process ends.

FIG. 11 is an explanatory diagram showing an example of a word extraction candidate string in the present invention.

In the process (106) in FIG. 1, a word candidate string is extracted by referring to the word dictionary 10 and grammar dictionary 11 for the input character string that has been preprocessed.

As shown in Figure 11, for example, in response to "Haikei...", the background and greetings are extracted, and "Kishiyamasu..."
For example, reporter, train, returning to work, your company, Ki, yellow, and vessel are extracted.

FIG. 12 is an explanatory diagram showing an example of canceling homophone selection correction registration contents in the present invention.

In the first registration, as shown in Figure 12(a), the notation "
``Reporter'' was corrected by the operator to ``Your company.''
For serial number l, reading length 3, reading character string ``Kishiya'', notation character string ``your company'', and registration flag 1 were registered. next,"
In order to convert or not convert "Kishiya" again, the registration flag is changed from 1 to 0, as shown in FIG. 12(b). Next, as shown in FIG. 12(c), the operator
As a result of selecting a homophone for ``Kishiya'', register the reading length 3, the reading character string ``Kishiya'', the notation character string ``Reporter'', and the registration flag 1 in 1 serial number 2, and change ``Your company'' to ``Kishiya''. ``Reporter'' to indicate that it has been corrected.

FIG. 13 is a processing flowchart for canceling homophone selection correction registration contents according to the second embodiment of the present invention.

In the second embodiment, since the reading length is not registered in the registration table, the reading length comparison process of the process (503) in FIG. 10 is not necessary. That is, first, it is determined whether the reference conditions for the registration table are in place (601), and if so, the reference position is set to the latest registered position (602).
It is determined whether or not the reading character strings match (603), and if they match, the reading at the reference position is set to a "cancellation" code (604).

Then I checked to see if there was anything else to refer to.
If not, the process ends (605). In addition, if there is something to be referenced, the reference position is set to the previous latest registered position (606), and the repeated reading character string comparison process is performed (
603).

In addition, when performing the cancellation process, it is sufficient to delete the corrected registration with the same reading as the registered one. Furthermore, the "cancellation" code may be, for example, a "space J code."

In this manner, in the first embodiment of the present invention, when the operator uses the homophone selection key to modify the notation converted and displayed by the device for the input reading, the operation content is memorized, and from then on, When the same reading is input and the notation of the stored modification content can be a conversion candidate, that notation is displayed as the first conversion candidate notation displayed by the device. Also,
In the second embodiment, the "pronunciation" and "notation" are stored in a registration table that stores the contents modified using the homophone selection key. In addition, in the third embodiment, "
In addition to "pronunciation" and "notation,""length of pronunciation" and "registration flag" are stored.

Effects As explained above, according to the present invention, the analysis of the device always proceeds in a predetermined direction for the input of homophones and the same pronunciation, so that analysis errors can always be minimized. , it is possible to make the conversion candidates desired by the user compatible with the conversion candidates output by the device, and it is possible to configure an optimal device as a whole.

[Brief explanation of drawings]

Fig. 1 is a processing flowchart of a kana-kanji conversion method showing an embodiment of the present invention, Fig. 2 is a diagram showing an example of a correction operation in the invention, and Fig. 3 is a homophone that stores the correction operation in Fig. 2. FIG. 4 is a diagram for explaining the contents of the selection correction registration table; FIG. 4 is a configuration diagram of a kana-kanji conversion input device showing an embodiment of the present invention;
4 is a perspective view of the device shown in FIG. 4, FIG. 6 is a block diagram of the candidate extraction unit shown in FIG. FIG. 9 is a processing flowchart of giving priority to homophone selection candidates according to the present invention; FIG. 9 is a processing flowchart of registration table search according to the present invention;
Figure 0 is a processing flowchart for canceling homophone selection, selection, correction, and registration contents according to the third embodiment of the present invention, Figure 11 is a diagram showing an example of a word extraction candidate string in the present invention, and Figure 12 is a flowchart of the homophone selection correction registration content cancellation process according to the third embodiment of the present invention. A diagram showing an example of canceling selective correction registration contents,
FIG. 13 is a processing flowchart for canceling homophone selection correction registration contents according to the second embodiment of the present invention. 1: Input preprocessing unit, 2: Candidate extraction unit, 3: Conversion information management unit, 4: Output control unit, 5: Conversion control unit, 6: Candidate extraction unit, 7: Candidate word string storage unit, 8: Dictionary search Section 9: Conversion dictionary storage section.

Claims (3)

[Claims] (1) In the kana-kanji conversion method, which converts the Japanese pronunciation into a sentence containing kanji and kana by inputting the Japanese pronunciation in kana (or romaji) and searching a word dictionary, the user can write the converted notation as a homophone. an input means equipped with a homophone selection key for changing to a homonym selection key, and a homonym selection correction registration means for storing the operation content modified by the homonym selection key, and after registering in the registration means, inputting the same reading When the notation of the memory contents of the registration means can be a conversion candidate,
A kana-kanji conversion method characterized in that the notation is used as a first candidate for conversion. (2) The above-mentioned homophone selection correction registration means has “
The reading and notation are memorized, and when a word-by-word conversion candidate having the same pronunciation appears after being memorized, the notation of the conversion candidate is replaced with the memorized notation and displayed. A kana-kanji conversion method according to claim 1. (3) The above-mentioned homophone selection correction registration means has “
2. The kana-kanji conversion method according to claim 1, wherein in addition to the reading and notation, a reading length and a registration flag are stored.