JPH0336660A - Document preparing device - Google Patents

Abstract

PURPOSE:To improve efficiency for preparing a document by storing a pair of word by learning and one of this pair of the words as a determined word, referring this determined word and preparing the document from the reading input of a character string. CONSTITUTION:When the reading of the character string is inputted from a keyboard 4, a CPU 1 executes the program processing of a Japanese syllabary (KANA)/Chinese character (KANJI) converting program and an input/output control program, etc., in a ROM 2. Then, it is decided whether the converted KANJI of the former or latter word out of a pair of the words exists in the determined word buffer of a RAM 3 or not. When such a KANJI exists, the pair of the words based on learning are provided in a word order table and the word order table is referred. Then, the pair of the words including correspondent connection are selected and the document is prepared. Thus, the document can be speedily and exactly prepared.

Description

[Detailed description of the invention] [Objective of the invention] (Industrial application field) The present invention converts pronunciations into words, and when homophones exist as a result of the conversion, priority is given to appropriate words among them. The present invention relates to a document creation device having a function of outputting data.

(Prior art) In Japanese document creation devices, the "reading" of character strings is manually performed using a keyboard or voice, etc., and character strings containing readings and kanji (
The dictionary memory that stores pairs of hiragana and katakana character strings (sometimes hiragana and katakana character strings) is searched for the input ``yomi'', and the character string obtained from the dictionary memory is output to a display device to obtain a kanji-mixed character string. Has a function.

However, as a characteristic of the Japanese language, there are multiple homonyms corresponding to one reading, and multiple homonyms are output as the conversion result. For this reason, the Japanese document creation device is equipped with a function (homonym selection function) that allows the operator to select a desired word from among a plurality of homophones. For example, one representative of the homophones is displayed in a format different from other characters in the character string of the conversion result (inverted display, overlined display, high brightness display, etc.).

Then, when this homophone is indicated with a cursor etc., the homophones are displayed in order by pressing the key ([Next candidate key), and when the desired homophone is displayed, pressing the other key ([ When the user presses the [Select] key, the word is stored in the document string and the selection is confirmed.

In selecting the homophones, if as many words desired by the operator as possible are output as the initial conversion result, the number of presses of the [Next Candidate] key will be reduced and document creation efficiency will be improved. For example, the reading ``Let's'' includes ``specification, use, trial, personal use, cotyledon, branch and leaf, let's''.
There is a homophone. Assuming that homophones are always displayed in this order, an operator who desires "private use" must always press the [Next Candidate] key three times.

In order to solve this problem, Japanese document creation devices have two
Two methods have been adopted.

The first method (short-term learning function) is to display recently selected words in preference to other homophones. For example, if "Private" is selected in the Jingji example, the following "
In response to the pronunciation input of ``Let's do it,'' it is displayed in the order of ``Private use, Specification, Use, Trial, Cotyledon, Branch and Leaves, Let's.'' This takes advantage of the characteristic that the same word is likely to be selected consecutively in a sentence.

The second method (use frequency learning function) is a method in which the selection frequency of each word is calculated as a coefficient, and the word with the highest selection frequency is given priority and output. For example, "Specifications (3), Usage (5)
, trial (2), private (4), cotyledon (O), branch (1), copy (0) J (0 is the selection frequency) , Let's do it.'' Then, when "private" is selected two more times, homonyms will be displayed for the next pronunciation of "private" in the order of "private, use, specification, trial, branch, leaf, cotyledon, sho." be done.

This takes advantage of the fact that there are biases in the words used in the field of the document being created (letters, patent specifications, etc.) or in the environment in which the device is used (law offices, hospitals, etc.). ing. Note that there are also methods that use the above two methods together.

(Problem to be Solved by the Invention) However, the above-mentioned characteristics at the time of document creation cannot always be used.

For example, if you want to obtain the character string "for personal use," when the short-term learning function is activated, you can input the pronunciation of "sho de sho."

■For example, displaying conversion candidates for "use" (an angline indicates that the word is not selected).

■If you display the next option for the previous "Use" and select "Private use", it will become "Private use", and the short-term learning function will work until the next "Use", and it will be replaced with "Private use". .

■Again, press the [Next option] key for "Private use" to display "Use" and select it.

This phenomenon occurs, which actually worsens operational efficiency.

In the case of the usage frequency learning function, priority is always given to the most frequently used item. Therefore, if the usage frequency of "private use" is higher than that of "use", the conversion result is always "private use and private use". Furthermore, if the frequency of use of "use" exceeds that of "private use," it will always become "use by use," and the desired conversion result of "use by private use" cannot be obtained from the beginning.

In view of the above problems, the present invention learns and memorizes the relationship between the previous word and the following word (referred to as "co-occurrence relationship"), and uses the learned co-occurrence relationship to select from homophones. The purpose of this invention is to provide a Japanese document creation device that gives priority to what is desired.

[Structure of the invention] (Means for solving the problem) In order to achieve the above object, the document creation device of the present invention has the following features:
a storage means for storing pairs of words; (b) At the time of converting a reading into a word, after the first word is output as a conversion result, a plurality of homophones are registered for the reading input after this word. When a synonym is output, it is checked whether a pair of the first word and any one of the homophones is registered in the storage means, and the first word is selected from among the homophones. and a processing unit that outputs words registered in the storage means as conversion results preferentially.

(Operation) With the above configuration, when a certain word is selected, the co-occurrence relationship between this word and other words (which may come before or after it) is learned, and this learning result is used during subsequent conversion. This can be used to prioritize output of homonyms, improving document creation efficiency.

(Example) FIG. 1 is a diagram showing the system configuration of a Japanese document creation device that implements the present invention.

A ROM 2, a RAM 3, a keyboard 4, a display device 5, and an external storage device (for example, a floppy disk device) 6 are connected under the control of the CPUI.

A controller and a buffer memory (for example, a video RAM for storing display images in the display device 5) are connected to the keyboard 4, display device 5, and external storage device 6 to control data storage and data input/output. However, since this is a well-known matter, it is omitted from the drawing. On the keyboard 4, there is a [character] key for manually reading "yomi".
[Convert] key to start the conversion of , [Next candidate key] to instruct the display of conversion candidates (homonyms) in sequence, [Select] key to instruct to confirm the selection of the conversion candidate. [Cursor] key to move the cursor within the display screen. ] key is provided.

The ROM 2 contains functions necessary for the Japanese document creation device such as a program for kana-kanji conversion related to the present invention, a program for co-occurrence relationship learning, a data input/output control program, a screen editor program, etc. Other programs to be executed by the terminal equipment 4.5.6 are also stored. Parts of the program stored in the ROM 2 that are particularly relevant to the present invention will be explained later using flowcharts. Further, the ROM 2 stores a dictionary in which pairs of "yomi" and words corresponding to the "yomi" (character strings containing kanji, katakana words, etc.) are stored.

Under the control of the Kana-Kanji conversion program, the CPU
By searching this dictionary for ``Yomi,'' the word corresponding to the yomi can be read from the dictionary. In addition, each word is assigned a number (called a "dictionary number"), and the CPU
In addition to searching for words by "reading", l- can also read words from the dictionary by dictionary number. Furthermore, R
The OM2 also stores character patterns for displaying created documents and character strings as conversion results on the display device 5.

RAM3 stores various buffers used for kana-kanji conversion (a manual buffer that stores the input pronunciation, a stack that stores homonyms, a bunsetsu buffer that stores fixed phrases, etc.), and the characters that make up the created document. A document buffer for storing character codes (including codes for control characters indicating line breaks, etc.) and a co-occurrence table for storing co-occurrence relationships are provided.

FIG. 2 shows the internal structure of the co-occurrence table used in this embodiment. A co-occurrence table is a table that stores reference relationships between words (such as a relationship where the connection between "kanji" and "conversion" is prioritized). One entry in the table contains a pair of ``words that come before'' and ``words that come after'' and a code (
types of particles, conjunctions (directly connected like in ``kanji conversion''), conjugation endings when the previous word is a pragmatic, etc.)
is memorized. The "code indicating the type of connection (grammatical connection between words)" is, for example, [0 = conjunction (connections such as "community", "ten", "donation", running" + "include")]
[1="'s"], [2="is"], [3="to"]
, [4= wo”]...[8=adjunct form], [9=
It is set like [unnatural form]. “The word that comes before” and “
The following words are each stored in the form of a dictionary number.

In addition to the document buffer, a confirmed word buffer for selected and confirmed words (including uniquely confirmed words with no homophones) is provided in the RAM 3. The internal structure of this confirmed word buffer is shown in FIG. In this embodiment, the confirmed word buffer has a capacity to store approximately one page's worth of confirmed words.

As the reading skill and conversion progresses, a ``code indicating unselected'' and ``code indicating the type of clause link (as described above)'' will be registered for each phrase obtained as a result of the conversion. Also,
In the final clause of a sentence (the clause before a control code such as a full stop or line break), instead of the "code indicating the type of clause connection",
A "code indicating that it is a break in a sentence" is stored.

Then, when the candidate word is selected and confirmed, the "code indicating unselected" corresponding to this word is replaced with the dictionary number of the selected candidate. Also, “reading”
As a result of conversion from - For words that are determined at will, the dictionary number of the word is stored from the beginning instead of the "code indicating unselected".

The co-occurrence relationship learning program will be explained below with reference to FIGS. 4 and 5.

Now, as shown in FIG. 5(a), it is assumed that an unselected sentence "I give 1 star to (I)" is displayed on the screen of the display device 5.The operator selects the cursor ] key and move the cursor to "feeling", "凰1"
The word becomes the selection target (step 10). In addition, at the position of an unselected word in the document buffer, a ``code indicating unselected'' and ``address of the buffer where homophones are stacked'' are stored, and when the [Select] key is pressed, If ``feeling'' is the beginning of an unselected word from the beginning of the displayed character string, press the [Select] key.
"Feeling" becomes the selection target and the cursor jumps to that digit position.

Here, each time the [Next Candidate] key is pressed (step 12
), the stacked homophones are read out, and for example, ``Feel → Kanji → Secretary → Auditor → Kanji → Kanji →...
Conversion candidates are switched and displayed in the order of "...→Feeling→Feeling→..." (Step 14).

Assuming that "Kanji" is the desired word, press the [Select] key while "Kanji" is displayed (display state shown in Figure 5 (b)) to confirm the word "Kanji". (Step 16). In this case, the character code of "Kanji" is
They are stored at the same digit position in the document buffer instead of the code indicating "unselected" and the address of the buffer where the homophone is stacked (step 18). Here, the confirmed word buffer is accessed, and confirmed words within four clauses before and after the clause in which the currently selected word exists are extracted (step 20). Note that the extraction of confirmed words is performed within a range of one sentence (a range separated by clauses having the above-mentioned "code indicating that it is a break in a sentence"). Then, the pair of the "currently selected word" and the extracted "confirmed word" and the "code indicating a link" attached to the earlier word in this pair are stored in the co-occurrence table. (step 22). Subsequently, the document buffer is searched to see if there is an unselected phrase after the currently selected word (step 24). If there are still unselected phrases, the words in those phrases become new selection targets (step 26), and the above-described process is repeated. Furthermore, if there are no unselected phrases, the process returns to the normal "reading" manual mode.

In the state shown in FIG. 5(b), when "Kanji" is selected, the pair is not registered because the later clause is not yet determined, but the earlier clause ([this] [in the device, ko) is not registered. Since these are fixed, [the pair of dictionary numbers for "ko" and "kanji" and the "code" for "particle"] and ["device" and "kanji"]
The pair of dictionary numbers and the code for ``particle'' are registered in the co-occurrence table.Similar co-occurrence relationships are subsequently registered in the selection process for ``eligibility'' and ``return.''

Here, if you change "Return" to "Convert" and select it,
In particular, [the pair of dictionary codes for "kanji" and "conversion" and the code for the particle "no" attached to "kanji"] will be registered, so in the next conversion, "kanji no" - There is no longer a pair that cannot exist in the Japanese word ``return''.

Here, a method of using the co-occurrence table in kana-kanji conversion will be explained with reference to FIG.

A ``yomi'' is inputted from the keyboard 3 (for example, ``Kanji no Koritsuki de sosai ni yoi henkan''), and the ``yomi'' is stacked in the input buffer (step 30). When the [conversion] key is pressed, the kana-kanji conversion program is activated, and the conversion of the "yomi" stored in the input buffer is instructed (step 32). Note that the conversion program is also started when it is determined that a phrase will definitely be cut off by inputting punctuation marks, line breaks, tabs, etc. First, the Kana-Kanji conversion program extracts phrases from the reading and outputs conversion candidates one by one. For example, a word is extracted by matching a word registered in a word dictionary with a pronunciation character string, and a particle extraction is performed using an adjunct word dictionary that stores particles, etc., and an attempt is made to cut a clause. Put the homophones of ' into the stack. In the process of loading onto the stack, the short-term learning function and selection frequency learning function described above are used to prioritize each homophone. Then, there is a method that uses a grammar dictionary word dictionary that stores grammatical connection relationships to check the connection relationship with the next character string, discarding grammatically incorrect conversion candidates, and redoing bunsetsu cutting. Are known. Here [Kanji auxiliary verb “no”]
[The adnominal form of the ten auxiliary verb "da" = 2] [the auxiliary verb "sousei"] [good] [henkan] are extracted (step 34). Here, it is assumed that the short-term learning function and the selection frequency learning function described above are activated for the word ``kanji'' in the first clause, and ``kanji'' is displayed as the first candidate (step 36). In this case, the "previous word" portion of the co-occurrence table is searched using the dictionary number corresponding to "kanji", and it is checked whether the same word is registered (step 38). Here, "Kanji" is registered, so next, in the co-occurrence table, the "Yomi" is registered with the one registered in the "Connection type" column that is stored corresponding to the dictionary number of "Kanji". It is checked whether the connecting part in
. In this example, "no" matches, so step 4
Proceed to step 2, read the dictionary number ("conversion") of the "following word" stored in correspondence with "kanji" in the co-occurrence table, and read out the dictionary number ("conversion") of the "word that comes after" that corresponds to the "kanji" in the co-occurrence table. Checks each homophone group on the stack as a conversion candidate for "1 [good] [henkan]) of [Sousai Jusuke-dou]". Here, "conversion" is the priority candidate (conversion) because a match can be found with "conversion" among the phrase candidates for [henkan] (■return, ■conversion, ■henkan).
(the target of the first display of this clause) (step 44)
, in step 46, conversion result candidates are displayed. If each of the conditions is not satisfied in steps 38, 40, and 42, the process proceeds to step 46, and the highest priority word among the homonyms stacked on the stack is displayed as a conversion result candidate, as in normal kana-kanji conversion.

On the other hand, in step 36, it is assumed that the short-term learning function and the selection frequency learning function described above are activated for the word "kanji" in the first clause, and for example, "feeling" is displayed as the first candidate.

In this case, since "feeling" is not stored in the co-occurrence table, candidates similar to conventional conversion results are displayed. However, after step 46, in the homophone selection mode,
When the [Next candidate] key is pressed (step 48) and "feeling" is changed to another homophone and displayed (step 50), the process returns to step 38, the co-occurrence table is referred to again, and the subsequent Conversion candidates for phrases are also changed according to co-occurrence relationships. When "feeling" is changed to "kanji", "conversion" is displayed as a candidate corresponding to the reading of "henkan".

The reason for storing the "type of linkage" information in the co-occurrence table is as follows. If you memorize the co-occurrence relationship of only the words "reporter" - "return" (note: the stem is registered for the terminology), it is correct in the case of "kishakaeru" → "journalist return". Although the conversion result is obtained, a co-occurrence also occurs in the case of ``Kishadekaeru'', resulting in a semantically incorrect conversion as ``I'm going home as a reporter.'' If we add the ``bridging information''``ga'' to the pair ``reporter'' - ``F return (ru)'', we can create a co-occurrence relationship due to the existence of the particle ``de'' in the case of ``kisha de kaeru.'' is not referenced, so unnecessary co-occurrences such as ``I'm going home as a reporter'' do not occur.

In co-occurrence registration (co-occurrence learning), co-occurrence relationships are registered in the co-occurrence table each time a conversion candidate is selected, but it is not possible to have an unlimited capacity for the co-occurrence table. It's fleeting. Therefore, it is necessary to update the contents of the co-occurrence table by any suitable method. "Updating the contents of the co-occurrence table" includes ■ Fast in-fast out (FI
FO) method of evicting the oldest pair and registering a new one.

■ A method in which each entry in the co-occurrence table is provided with a storage section for the frequency of use, the frequency of use is updated every time a pair in the entry is used, and the least frequently used ones are purged.

■ A method of setting a usage flag for each entry in the co-occurrence table, setting the flag when a pair within an entry is used, and periodically turning off the flags of all entries, and at a certain point, expelling those whose flags are off. .

For example, a method of evicting address pairs from the cache directory can be applied.

The co-occurrence table can be saved externally with the created document when it is saved to external memory, and can be reused for adding or changing documents.The co-occurrence table can be saved as a single file in external memory, and other It may also be possible to load it into the device and set the same environment.

Further, in the above embodiment, a batch selection method is used in which homophones existing in multiple locations in a document can be selected using the [Select] key independently of reading input. As a homophone selection method, there is a sequential selection method in addition to a -bloc selection method. The sequential selection method means that after a reading is input and converted, the next candidate for the conversion result can be displayed and selected until the next reading is input, and when the next reading is input, the previous one This is a method in which the converted results are automatically determined. When applying the present invention to a sequential selection method, the selection results (dictionary numbers of confirmed words and transition information) are buffered until a break in the sentence (control code input such as a comma or line break) is reached. Every time a word is determined, pairs with previously buffered words are learned and stored in the co-occurrence table. Even at the time of conversion, the conversion results are retained up to the end of the sentence, and when a certain pronunciation is input and multiple homophones are converted and output, the pair is shared with the buffered word that has already been converted. It is only necessary to check the origination table and output the registered homophones preferentially from among the homonyms.

In the above embodiment, the homonyms are displayed in one place in order, but each homonym is numbered and a plurality of homonyms are displayed together, and the number input or cursor instruction is required. It is also possible to select homophones from among these. In this case, the word with the highest priority due to co-occurrence may be displayed at the beginning.

[Effects of the Invention] According to the present invention, even in the case of "Let's do it" taken as an example earlier, the co-occurrence relationship between "private" and "use" is learned with one selection. Undesired conversion candidates are not output, and the conversion result (candidate) is “for personal use”.
can be obtained accurately. Therefore, unnecessary homophone selection operations (pressing the [Next Candidate] key) are eliminated, and operability is improved.

[Brief explanation of drawings]

Fig. 1 is a system block diagram of a Japanese document creation device implementing the present invention, Fig. 2 is a diagram showing the internal structure of a co-occurrence table provided in RAM, and Fig. 3 is a diagram showing a fixed word buffer provided in RAM. Figure 4 is a flowchart showing the operation of the co-occurrence relationship learning program, Figure 5 is a diagram showing an example of selecting homophones, and Figure 6 is a conversion using the learned co-occurrence relationships. It is a flowchart showing the operation. 2... ROM3 that stores the co-occurrence relationship learning program.
・RAM with co-occurrence table (7317)
Agent Patent attorney Norichika Kensuke Yamashita → Figure 2 Figure 3 Affected word buffer Figure 4 Figure 5 (a) Figure 5 (b) → Translation step 46

Claims (2)

[Claims] (1) An input unit for inputting reading information of a character string; a dictionary storage unit storing a plurality of homophones having this reading in correspondence with the reading information; a processing unit that reads words from a dictionary storage unit and performs conversion; A document creation device that creates a document while uniquely determining a word by having a selection input means for giving an instruction for selection, comprising a storage means for storing a pair of two words, and the processing unit ( a) When a certain word is selected by the selection input means, a pair of this word and a word existing before or after this word is registered in the storage means; (b) When converting the reading into a word; In , after a first word is output as a conversion result, when multiple homonyms are output for the reading input after this word, the first word and the homonyms are It is characterized by checking whether a pair with any of the words is registered in the storage means, and preferentially outputting the words registered in the storage means from among the homophones as a conversion result. document creation device. (2) The storage means stores information on the grammatical connection of both words together with the pair of words, and the processing device stores information on the grammatical connection of both words when registering the word pair at the time of word selection. Information on connections is stored in the storage means, and at the time of conversion, a pair of the first word and any one of the homophones and a grammatical connection between the two words are registered in the storage means. 2. The document creation device according to claim 1, wherein when the word is a homophone, the word among the homophones is preferentially outputted as a conversion result.