JP3029109B2 - Character processing apparatus and method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a character processing apparatus for inputting characters such as kanji while performing kana-kanji conversion and creating and editing a document such as a Japanese document. The present invention relates to a character processing device that can correctly convert the next input when a phrase separation of a kana mixed sentence is changed.

[Prior Art] In a character processing apparatus for inputting Japanese sentences such as a Japanese word processor, as a means of inputting kanji, it is possible to input a kana string from a keyboard or the like and convert the input kana string into kana-kanji characters. Generally done. In particular, in recent years, there are models that provide “solid writing conversion” or the like that allows the operator to continuously input a kana without being aware of a break between phrases. However, such solid writing conversion has insufficient analysis capability of the system, so that a break in a phrase presented by the system may differ from the conversion desired by the operator (erroneous conversion or erroneous division). In such a case, it is a common practice for the operator to operate a so-called break change key (break reduction, break expansion) to designate a desired segmentation.

For example, when the user inputs "I was informed about the problem" as an input reading sequence, the system erroneously said, "I got an agreement / agreement / at a fixed price / for the / problem" ( “/” Indicates a break in a phrase). In such a case, the operator instructs the phrase “relation” to “delimitation expansion”. As a result, the following conversion is correct: "The following agreement has been reached on the matter."

Unless learning is performed for such a break changing operation, the operator is very inconvenient.
For example, in the above example, when the user inputs "I have heard something about the problem" again next time, "I got an agreement / agreement / on the / problem / about / fixed price" again. The operator must make another break change.

For this reason, some models have a so-called break learning function for reflecting the break change performed by the operator next time. The break learning function is performed by increasing or decreasing the frequency of each word.

As described in the previous example, the frequency of “relation” and “relation” = 2 and the frequency of “price” =
4. Assuming that the frequency of “less than or equal to” = 3, the frequency of “relation / list price” = 2 + 4 = 6, the frequency of “relation / less than” = 2 + 3
Since “= 5”, “relation / fixed price” is converted as the first candidate. Change the delimiter here and enter ""
If you change to the following, change the frequency to the “list price” frequency =
3. If the frequency of "less than or equal to" is reversed to 4, the next time you input "Kashite Ika no", the "relation / fixed price"
Frequency = 2 + 3 = 5, frequency of “about / below” = 2 +
Since 4 = 6, the correct conversion “about / less than” is obtained, indicating that the segmentation learning has been performed.

[Problems to be Solved by the Invention] However, in the conventional segmented learning, unexpected erroneous conversion may occur in an unrelated context, and the conversion rate may be worsened.

For example, in the above example, the frequency of “less than or equal to” has been improved,
When inputting and converting, it is converted to "it / related to / below / view / hear" and the like, and the operator's desired "it / related / home / view / hearing" may not be converted.

In other words, the word "relating" is often used as "relating" or "relating", but is not usually used as "relating". Inability to respond to the phenomenon.

In order to solve the above-described problem, a method of storing a local reading for one phrase before and after the change of the delimiter is also conceivable. For example, it memorizes where to divide a sentence for readings such as "kante" and "kantan". Is a method of converting to "seki / ta". However, the problem remains in that case. For example, in the case of input of "accept a problem," it is natural that the input is converted to "accept / problem / recommend / accept" and "accept / problem / accept / accept." Is unnatural. This is different from the "about / less than" case. That is, there are cases where it is better to convert the same "kante" reading to "seki / te" and cases where it is better to convert it to "seki / te".

To summarize the above, it should be converted to "I want to do something" → "I want to do something less", "I want to make something" → "I want to make something", "I want to make something" → "I want to make recommendations". However, in the conventional method of delimiter learning corresponding to the frequency of each word or in a method of storing a local reading of one phrase, a side effect of delimiter learning occurs, resulting in inconvenient kana-kanji conversion.

[Means for Solving the Problems] In order to solve the above problems, a character processing apparatus of the present invention includes an input unit for inputting a reading, and a conversion for converting the reading input from the input unit into a sentence mixed with kanji kana. Means, a delimiter changing means for changing a delimiter position of a phrase of a sentence mixed with a kanji kana converted by the conversion means, a storage means for storing delimiter learning data, and two delimiter positions before and after the delimiter position changed by the delimiter change means. Registration means for registering a reading including a phrase and how to separate each of the two phrases in the reading in the storage means as delimiter learning data; and a reading input from the input means being stored in the storage means. If the reading that includes the delimiter learning data is included, the reading is converted based on the way of delimiting each of the two phrases represented by the delimiter learning data, and the input reading is And control means for controlling the conversion means so as to convert the input reading without depending on the delimiter learning data when the reading does not include the delimiter learning data.

Further, the character processing method of the present invention includes an input step of inputting a reading, a conversion step of converting the reading input in the input step into a sentence mixed with kanji kana, and a conversion of the sentence mixed with kanji kana converted by the conversion step. A delimiter change step of changing a delimiter position of a phrase, a reading including two phrases before and after the delimiter position after the change by the delimiter change process, and a method of delimiting each of the two phrases in the reading are separated as delimiter learning data. A registration step of registering in the learning data storage unit, wherein in the conversion step, when the reading input from the input step includes a reading that matches the separation learning data stored in the separation learning data storage unit, The pronunciation is converted based on the way of delimiting each of the two phrases represented by the delimiter learning data, and the input reading includes a reading that matches the delimiter learning data. If not, and converting the inputted read regardless to the separator learning data.

EXAMPLES Hereinafter, the present invention will be described in detail with reference to the drawings.

 FIG. 1 shows an example of the overall configuration of the present invention.

In the configuration shown in the figure, the CPU is a microprocessor, performs calculations for character processing, performs logical decisions, etc., and is connected to those buses via an address bus AB, a control bus CB, and a data bus DB. Control the components.

The address bus AB transfers an address signal indicating a component to be controlled by the microprocessor CPU.
The control bus CB transfers and applies a control signal of each component to be controlled by the microprocessor CPU. The data bus DB transfers data between the components.

Next, the ROM is a fixed read-only memory.
The control procedure by the microprocessor CPU, which will be described later with reference to FIGS. To 14, and fixed data such as a word dictionary and a grammar dictionary are stored. The word dictionary is a dictionary in which readings, notations, grammar information, and the like are stored correspondingly, and is referred to in kana-kanji conversion and the like. The grammar dictionary stores, for example, connection rules between words necessary for morphological analysis, syntax analysis, and the like.

The RAM is a writable random access memory having a structure of 16 bits per word, and is used for temporarily storing various data from each component. KULDT is delimiter learning data described in detail in FIG. HENTBL is a conversion candidate table detailed in FIG. KBBUF is a keyboard buffer for storing the input reading sequence, and is configured as shown in FIG.

KB is a keyboard. Alphabet key, hiragana key, katakana key and other character symbol input keys, and
Various function keys for instructing various functions to the character processing apparatus, such as a cursor movement key, a kana-kanji conversion key, a delimiter reduction key, and a delimiter expansion key, are provided.

DISK is an external storage for storing document data,
The created document is stored, and the stored document is called up by a keyboard instruction when necessary.

CR is a cursor register. The CPU can read and write the contents of the cursor register. A CRT controller CRTC described later displays a cursor at a position on the display device CRT corresponding to the address stored here.

DBUF is a display buffer memory for storing patterns of data to be displayed. When displaying the contents of the document data, it is performed by developing a pattern on the DBUF.

The CRTC plays a role of displaying the contents stored in the cursor register CR and the buffer DBUF on the display CRT.

The CRT is a display device using a cathode ray tube or the like, and the display pattern of the dot configuration and the display of the cursor on the display device CRT are controlled by a CRT controller. Further, CG is a character generator, and the display device CR
This stores the pattern of characters and symbols displayed on T.

The character processing device of the present invention comprising such components operates in response to various inputs from the keyboard KB. When an input from the keyboard KB is supplied, first, an interrupt signal is generated by the microprocessor. CPU
The microprocessor CPU reads various control signals stored in the ROM, and performs various controls in accordance with the control signals.

FIG. 2 is a diagram showing a screen configuration of the apparatus of the present invention. In the figure, CRT means a display screen. CM is a cursor,
Next, when a key input is performed, it indicates a position where a character enters. TS is a text screen on which the contents of the stored document are displayed. MS is a monitor line, which is an area where input key data is displayed one by one. When performing Kana-Kanji conversion, etc., a reading column is displayed once on the MS, and after conversion, the conversion result is transferred to the TS. In addition,
The kana-kanji conversion key is indicated by “/”.

FIG. 3 is a view showing a method of operating the decompression and extension according to the present invention. First, the operator inputs "that" in the document, and then tries to input "the following agreement was reached on the problem". And then input the kana / kanji conversion key "/". (A) “I was told about the problem /” was converted to Kana-Kanji characters, and “a fixed price agreement was reached on the problem”
Next enter the text screen. (B) At this time, the operator notices that the portion of "relation price" is not the desired segmentation, and moves the cursor to the position of "relation". (C) Next, a decompression expansion key is input to instruct expansion of the phrase “Seki”. It is re-converted as "About" and displayed. (D) FIG. 4 is a diagram showing an operation method of delimiter reduction according to the present invention. First, the operator inputs "that" in the document, further tries to input "paid for the proposal", reads and inputs the key "I want to do what I want", and then the kana kanji Enter the conversion key "/".
(A) "I want to do what you want /" is converted to Kana-Kanji and becomes "Paid for the proposal" and enters the text screen. (B) At this time, the operator notices that the part of “the corresponding medical department” is not the desired segmentation, and moves the cursor to the “pair”.
Move to the position. (C) Next, a delimiter reduction key is input to instruct to reduce the phrase “to”. It is re-converted to "Consideration" and displayed. (D) FIG. 5 is a diagram showing the configuration of the delimiter learning data KULDT.

The reading is an area for storing readings for two phrases related to the change of the break. The first phrase reading number stores the number of readings of a portion that becomes the first phrase after the delimiter is changed in the reading.
The second phrase reading number is the second of the readings after the break is changed.
Stores the number of readings of the phrase.

For example, consider a case in which the input “Kishite Ikaigaigaiga” is converted to “Relationship / fixed price / agreement” and changed to “Relationship / following / agreement” by changing the break.
At this time, the reading for the two phrases is "I'm gonna do it,"
The reading of the first phrase after the change of the delimiter is “kante”, and the reading of the second phrase is “squid”. Therefore, the registration in the delimiter learning data is as follows: reading = “Kanashi Ikano”, first phrase reading number = 4, second phrase reading number = 3.

As another example, the case where the input “This Heya is very hot” is converted to “Furuno / Yawata / Urgent / Hot”, and modified to “This / Room is / Very / Hot” by changing the separator Think. At this time, the reading of the two phrases before the change of the delimiter is “this heyahata”, and the reading of the two phrases after the change is “this heyahata”, so the longer “this heyahata” Is the reading of the delimiter learning data. The first phrase reading after the delimiter change is “this”, and the second phrase reading is “heyaha”. Therefore, the registration to the delimiter learning data is as follows: reading = “this part”, first phrase reading number = 2, second phrase reading number = 3.

FIG. 6 is a diagram showing an example of a phrase candidate analyzed in the middle of processing when performing kana-kanji conversion.

FIG. 7 illustrates a possible phrase structure for the input “Kanashi Ika no Goai”. The left end is a candidate for the first phrase,
It means that there is a possibility of "government", "relationship" and "relationship". Actually, for example, for "government", "can""kan""ring"
There are homonyms such as, but they are omitted in the figure because they are complicated.

Candidates for the second phrase are “death”, “worker”, and “designation” for the first phrase “government”, and “hand”, “empire”, “price”, “price” Means that there is a possibility of “stomach”, “medical”, “medical” for the first phrase “relating”.

 Hereinafter, the phrase structure is similarly expressed.

FIG. 7 is a diagram showing the structure of a conversion candidate table HENTBL for internally describing the above-mentioned phrase candidates.

The “reading” column describes the reading of each phrase. The column of "dictionary" describes an address in which part of the word in the independent part of the phrase exists in the word dictionary. In the figure, the addresses are enclosed in parentheses. The “number of kana words” describes the number of readings of the kana part of the reading described in the “yomi” excluding the reading of the independent part. For example, in the case of "relating", the number of readings of the sending pseudonym "re" is described. The “next candidate” column is a pointer that indicates a phrase of the next candidate that can be replaced with the phrase. In the phrase candidate diagram shown in FIG. 6, the phrases arranged vertically are linked. For example, “government” is linked to “relation” and “relation”. When there is no next candidate, "-1" is stored so that it can be understood that the link does not continue any more. The "next clause" column is
This is a link to the phrase following the phrase. The phrase candidate diagram in FIG. 6 links a partial phrase if it is at the right position. For example, “death” is linked to “worker” and “designation”. If the next clause does not exist, "-1" is stored so that it can be understood that the link does not continue any more.

FIG. 8 is a diagram showing the configuration of the keyboard buffer KBBUF. Key data input from the keyboard is temporarily stored in this KBBUF. For example, in the case of kana-kanji conversion, the reading sequence is stored in this KBBUF until the conditions for conversion are satisfied, and converted into kanji when the conversion conditions are satisfied, and the buffer is cleared.

Characters are stored in 2 bytes per character using, for example, JIS X 0208 code code. In the figure, "/" means a kana-kanji conversion key, and means that the reading sequence stored up to "/" is converted into kanji.

 The operation of the above embodiment will be described according to a flow.

FIG. 9 is a flow chart of a part for receiving a key input and performing processing.

Step 9-1 is a process for taking in data from the keyboard into the input buffer KBBUF.

In step 9-2, the key contents of the input buffer KBBUF are checked, and the process branches to each process according to the type of the key. If the data of the kana-kanji conversion key is included in the KBBUF, the kana-kanji conversion must be performed, and the process branches to step 9-3. If the head of KBBUF is the delimiter reduction key or delimiter expansion key, the delimiter must be changed, and the process branches to step 9-4. If the contents of the key are other than those described above, normal editing processing is performed, so that the process branches to step 9-5 to perform other processing such as cursor movement, insertion, and deletion, which are found in general word processors. .

In step 9-3, as described in detail in FIG.
Convert the input reading string on F to Kana-Kanji, output it to a document, and display it.

In step 9-4, the delimiter is changed as described in detail in FIG. 14, and the result is output to a document and displayed.

FIG. 10 is a detailed flowchart of step 9-2 “Kana-Kanji conversion”. In step 10-1, KBBU
Performs morphological analysis, syntax analysis, etc. on the input reading sequence on F with reference to a word dictionary, grammar dictionary, etc., and a conversion candidate table
Create HENTBL.

At step 10-2, as detailed in FIG.
Which of the above conversion candidates is to be converted,
Determine candidates.

In step 10-3, the determined first candidate (adopted phrase) is converted into a sentence mixed with kanji and kana one by one.

The converted character string created in step 10-4 is set in the document. Further, the vicinity where the converted character string is set is displayed so that the contents of the change of the document can be understood.

FIG. 11 is a detailed flowchart of step 10-2 "Determine first candidate".

Step 11-1 is a variable initialization process. The current phrase pointer, which is a pointer pointing to a phrase in the conversion candidate table HETBL, is initialized to 1 so as to point to the first candidate of the first phrase. Next, a counter i indicating the order of the adopted phrase is initialized to one. Finally, a pointer indicating which part of the reading on the input reading buffer KBBUF is currently being processed is set to one.
To indicate the beginning of the reading.

In step 11-2, it is determined whether the processing of all the clauses has been completed. Specifically, the current phrase pointer is -1.
Is determined. When the processing of all the clauses has been completed (when the current clause pointer = -1), the process returns.

In step 11-3, a delimiter learning search is performed as described in detail in FIG. 12, and a search is made as to whether the currently processed reading is registered in the delimiter learning data. If there is, it is output as adoption break learning.

At step 11-4, it is determined whether or not adoption break learning has been output. If so, the process proceeds to step 11-5, but if not, the process branches to step 11-8.

In step 11-5, a conversion candidate search is performed as described in detail in FIG. 13, and a search is made to determine whether there is a conversion candidate that matches the adopted break learning output in step 11-3. If so, a first adopted conversion candidate and a second conversion candidate are output. At this time, the number of adopted phrases = 2 is set.

At step 11-6, it is determined whether or not a matching conversion candidate is found, and if found, step 1-7 is performed.
If not, the flow branches to step 11-8.

In step 11-7, the first adopted conversion candidate and the second conversion candidate output in step 11-5 are referred to as an i-th clause and an i + th conversion candidate.
Adopted as one phrase and branches to steps 11-10.

In step 11-8, since there is no matching segmentation learning, the adopted phrase is determined according to the ordinary two-phrase longest matching method.

 In step 11-9, the number of employed phrases is set to one.

In step 11-10, the value of the number of employed phrases is added to i.

In step 11-11, the current phrase pointer is updated, and
Point to the next clause after the adoption clause. Specifically, the value of the column of “next clause” of the conversion candidate table HENTBL is substituted. When the conversion candidate is adopted in step 11-5, the "next clause" of the second adopted conversion candidate is substituted. Also, the input reading pointer is added and updated by the number of readings of the adopted phrase. When a conversion candidate is adopted in step 11-5, the sum of the number of readings of the first adopted conversion candidate and the second adopted conversion candidate is added.

FIG. 12 is a flowchart showing the details of step 11-3 “separation learning search”.

In step 12-1, variables are initialized. In other words, the value of the “kugaku pointer”, which is the pattern indicating the segmented learning data currently being processed, is initialized to one. In addition, the value of the variable “maximum number of readings”, which means the largest number of readings among the matching delimiter learning data, is initialized to zero.

At step 11-2, it is determined whether or not there is no more break learning data to be processed. Specifically, it is determined whether or not the value of the ward pointer has become -1. If the segmented learning data ends (when the ward pointer = -1), the process branches to step 12-8.

In step 12-3, the reading of the segmented learning data currently being processed (pointed to by the ku school pointer) is compared with the current input reading (the reading on the KBBUF and beyond the position indicated by the input reading pointer).

At step 12-4, it is determined whether or not the comparison at step 12-3 matches. If not, step
Branch to 12-7.

At step 12-5, it is checked whether or not the number of readings of the matching segmented learning data exceeds the maximum number of readings. If not, the flow branches to step 12-7.

In step 12-6, the found segmented learning data is used as the currently adopted segmented learning data.

In step 12-7, in order to process the next section learning data, the section pointer is updated by adding +1 to indicate the next section learning.

FIG. 13 is a detailed flowchart of step 11-5 “Conversion candidate search”.

In step 13-1, the variable i indicating the phrase currently being processed among the phrases on the conversion candidate table HENTBL is initialized to the value of the current phrase pointer.

In step 13-2, it is determined whether or not the next candidate of the current phrase is already over. Specifically, the value of i is-
It is determined based on whether it is 1. If the next candidate ends (i
If -1), the flow branches to step 3-14.

In step 13-3, it is compared whether or not the number of readings of the conversion candidate indicated by i matches the number of readings of the first phrase in the adoption break learning.

In step 13-4, it is determined how the reading numbers match, and if there is a match, the process proceeds to step 13-6 to enter the check of the second phrase. If they do not match, the flow branches to step 13-5, where the value of i is updated to indicate the next candidate of i (the "next candidate" column of HENTBL is substituted), and the process returns to step 13-2.

In step 13-6, the conversion candidate indicated by i is set as the first adopted conversion candidate.

In step 13-7, the value of j is initialized to indicate the next clause of the conversion candidate of i. ("Next sentence" of HENTBL
Substitute the field of In step 13-8, it is determined whether the next candidate for j is already over. Specifically, it is determined whether the value of j is -1. If the next candidate ends (j = -1)
In the case of, the process branches to steps 12-14.

In step 13-9, it is compared whether the number of readings of the conversion candidate indicated by j matches the number of readings of the second phrase in the adoption break learning.

It is determined in step 13-10 how the reading numbers match, and if they match, the process proceeds to step 13-12.
If not, the process branches to step 13-11, where the value of j is set to j
Is updated to indicate the next candidate (substitute the "next candidate" column of HENTBL), and the process returns to step 13-8.

In step 13-12, the conversion candidate indicated by j is set as a second adopted conversion candidate.

In step 13-13, the number of employed phrases is substituted for 2, and the routine returns.

In steps 13-14, the number of employed phrases is set to 0, and the routine returns.

FIG. 14 is a flowchart showing the details of step 9-4 "Change break".

In step 14-1, the process of decompressing or decompressing is executed according to whether the input key is a decompression / reduction key or a decompression / expansion key. This processing is actually realized in a Japanese word processor or the like, and is a well-known technique, and will not be described.

In step 14-2, is substituted for L ₁ Retrieves the number readings readings 2 clause minutes before delimiting changes.

In step 14-3, and assigned to the L ₂ taken out number reading readings 2 clause content after separator changes.

In step 14-4, it compares the values of L ₁ and L ₂ obtained above, if, when the L _1> L _2, i.e., the longer the number reading 2 clause minutes before delimiting change, step 14 Branch to -5. If L ₁ ≦ L _2, i.e., the longer the number reading 2 clause content after delimiting change, the process branches to step 14-6.

In step 14-5, the readings of the two phrases before the change of the delimiter are defined as "readings" for registering the delimiter learning data.

In step 14-6, the readings of the two phrases after the change of the delimiter are defined as "readings" for registering the delimiter learning data.

In step 14-7, the number of readings of the first phrase after the change of the delimiter is set as the "first phrase reading number" for registering the delimiter learning data.

In step 14-8, the number of readings of the second phrase after the change of the delimiter is set as the "number of second phrase readings" for registering the delimiter learning data.

At step 14-9, the delimiter learning data is registered as set above.

In step 14-10, search is made for delimiter learning data inconsistent with the registered delimiter learning data, and if inconsistent delimiter learning data is found, it is deleted. The contradictory delimiter learning data is the following learning data.

If there is a "reading" that matches the registered delimiter learning data "reading", or the delimiter learning matches the input reading longer than the registered delimited learning data "reading", and the registered delimiter learning data does not match the segmentation,
That is, segmentation learning in which "the number of first segment readings" or "the number of second segment readings" do not match This inconsistency segmentation learning data deletion process prevents a case where a learning effect cannot be obtained.

[Other Embodiments] In the above description, the case where the break change is explicitly specified by the operator as the timing at which the break learning data is registered, that is, the case where the break reduction key and the break expansion key are operated has been described. However, when the operator does not know the method of changing the delimiter and the conversion result of the erroneous division is presented, it is possible that the input reading sequence may be finely divided next time and converted. is there. An embodiment in which the segmentation learning is performed in the above case will be described below.

FIG. 15 is an example of the operator input described above. (A) shows a state in which “that” has already been input on the text screen, and the operator is further trying to input “the following agreement has been obtained regarding the problem”. However, since the input was repeated last time and the input was incorrectly converted, this time, the input unit was finely divided, and the conversion was started by inputting "//". Then, as shown in (b), it is correctly converted to "about a problem". Next, as shown in (c), "I got a squid /"
Enter As a result, as shown in (d), it is converted to "the following agreement has been obtained". By performing this series of operations, the segmentation learning is performed for the part "about / below", and the next time, if "input is completed" is correctly input "The following agreement has been reached on the matter."

The flow chart of “Kana-Kanji conversion” shown in FIG. 10 is changed as shown in FIG.

In step 16-1, morphological analysis, syntax analysis, and the like are performed on the input reading sequence on the KBBUF with reference to a word dictionary, a grammar dictionary, and the like, to create a conversion candidate table HENTBL.

In step 16-2, as described in detail in FIG.
Which of the above conversion candidates is to be converted,
Determine candidates.

In step 16-3, it is determined whether or not the current conversion is immediately after the conversion of the previous phrase. If it is immediately after the conversion, the process proceeds to step 16-4, and the registration processing of the segmented learning data is performed. If it is not immediately after, the process branches to step 16-9 to skip the registration process of the delimiter learning data.

In step 16-4, the stored reading of the phrase of the previous conversion stored is taken out and merged with the reading of the head phrase to be converted this time, and is set as the "reading" for registering the delimiter learning data.

In step 16-5, the number of readings of the phrase converted last time is taken out and set as "the first number of readings of the phrase" for registering the segmented learning data.

In step 16-6, the number of readings of the first phrase to be converted this time is taken out and set as the "second phrase reading number" for registering the delimiter learning data.

In step 16-7, the delimiter learning data is registered as set above.

In step 16-8, search is made for the delimiter learning data inconsistent with the registered delimiter learning data, and if contradictory departure learning data is found, it is deleted.

In step 16-10, the determined first candidate (adopted phrase) is converted into a sentence mixed with kanji and kana one by one.

The converted character string created in step 16-11 is set in the document. Further, the vicinity where the converted character string is set is displayed so that the contents of the change of the document can be understood.

[Effects of the Invention] As described above, according to the present invention, a reading including two phrases before and after a changed delimiter position and a method of delimiting each of the two phrases in the read are registered as delimiter learning data. In addition, if the input reading includes a reading that matches the delimiter learning data, the reading is converted based on the delimiter of each of the two phrases represented by the delimiter learning data,
When the input reading does not include a reading that matches the delimiter learning data, the input reading is converted without relying on the delimiter learning data. Since the readings of the two phrases before and after the delimited position can be correctly converted thereafter, and there is little side effect of causing erroneous conversion for the readings other than the reading of the two phrases before and after the delimited position, There is an effect that the conversion rate of the entire sentence can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram of the overall configuration of the present invention. FIG. 2 is a diagram showing a screen configuration of a CRT. FIG. 3 is a diagram showing a method of operating the decompression and expansion in the present invention. FIG. 5 shows an example of a phrase candidate analyzed in the middle of processing when performing kana-kanji conversion. FIG. 7 shows an example of a phrase candidate. FIG. 8 shows the structure of the conversion buffer table HENTBL. FIG. 8 shows the structure of the keyboard buffer KBBUF. FIGS. 9 to 14 are flowcharts showing the operation of the character processing apparatus of the present invention. FIG. 16 is a flowchart showing a display example. FIG. 16 is a flowchart showing a control procedure of the operation of the present invention. DISK... External storage CPU... Microprocessor ROM... Read-only memory RAM... Random access memory KBBUF. …… Separated learning data HE NTBL …… Conversion candidate table

Claims (12)

(57) [Claims] An input means for inputting a reading, a converting means for converting a reading input from the input means into a sentence mixed with kanji kana, and a delimiter position of a phrase of the sentence mixed with kanji kana converted by the converting means. Delimiter changing means to be changed; storage means for storing delimiter learning data;
Registration means for registering in the storage means a reading including a phrase and how to separate each of the two phrases in the reading as delimiter learning data; and a reading input from the input means being stored in the storage means. If the reading includes a reading that matches the delimiter learning data, the reading is converted based on how to separate each of the two phrases represented by the delimiter learning data, and the input reading includes a reading that matches the delimiter learning data. A character processing device comprising: a control unit that controls the conversion unit so that the input reading is converted without using the delimiter learning data when there is no reading. 2. The method according to claim 1, wherein when the input reading includes a reading that matches a plurality of delimiter learning data, the control unit determines the reading based on how to separate each of the two phrases represented by the longest matching delimiter learning data. 2. The character processing device according to claim 1, wherein said conversion means is controlled so as to perform conversion. 3. The method according to claim 2, wherein the registering means registers, as the reading in the segmented learning data, a longer one of two phrases before and after the changed segmentation position. 2. The character processing device according to 1. 4. The character processing apparatus according to claim 1, wherein when registering the delimiter learning data, the registration means deletes other delimiter learning data inconsistent with the delimiter learning data from the storage means. . 5. The method according to claim 1, wherein the registering means registers the number of characters defining the length of each of the two clauses as the way of dividing each of the two clauses before and after the delimiter position in the delimiter learning data. The character processing device according to claim 1. 6. When a new reading is input by the input means and conversion is performed by the conversion means following the conversion by the conversion means, the last sentence of the conversion result and the conversion result of the subsequent conversion result are obtained. A second registration unit for registering in the storage unit the reading of two phrases including the first phrase and the method of delimiting each of the two phrases in the reading as delimiter learning data; The character processing device according to claim 1. 7. An input step of inputting a reading, a conversion step of converting the reading input in the input step into a sentence mixed with kanji kana, and a delimiter position of a phrase of the sentence mixed with kanji kana converted by the conversion step. A delimiter change data to be changed, a reading including two phrases before and after the delimiter position after the change by the delimiter change process, and a method of delimiting each of the two phrases in the reading are stored in the delimiter learning data storage unit as delimiter learning data. A registration step of registering. In the converting step, when the reading input from the input step includes a reading that matches the delimiter learning data stored in the delimiter learning data storage unit, the reading is performed by the delimiter learning. If the input reading does not include a reading that matches the delimiter learning data, the conversion is performed based on the method of delimiting each of the two phrases represented by the data. A character processing method characterized in that input reading is converted without depending on the delimiter learning data. 8. In the converting step, when the input reading includes a reading that matches a plurality of segment learning data, the reading indicates the longest matching segment learning data.
8. The character processing method according to claim 7, wherein the conversion is performed on the basis of how each of the clauses is separated. 9. In the registering step, as a reading in the delimiter learning data, a longer one of two readings before and after the delimiter position before and after the change is registered. 7. The character processing method according to item 7. 10. The registration step according to claim 7, wherein when registering the delimiter learning data, other departure learning data inconsistent with the delimiter learning data is deleted from the delimiter learning data storage unit. Character processing method. 11. The method according to claim 7, wherein, in the registering step, the number of characters defining the length of each of the two clauses is registered as a way of dividing each of the two clauses before and after the break position in the break learning data. Character processing method. 12. After the conversion in the conversion step, when a new reading is input in the input step and the conversion is performed in the conversion step, the last sentence of the conversion result and the conversion of the subsequent conversion result are obtained. A second registration step of registering in the delimiter learning data storage unit the two phrase readings including the first phrase and the method of delimiting each of the two phrases in the reading as delimiter learning data. 8. The character processing method according to claim 7, wherein: