JPH0697454B2 - Character processor - Google Patents

Description

The present invention relates to a character processing device for converting an input character string into a kanji-kana mixed sentence by kana-kanji conversion and inputting it.

[Prior Art] Currently, a character processing device such as a Japanese word processor generally inputs kanji-kana mixed sentences using kana-kanji conversion.

Kana-Kanji conversion is to convert an input reading string into Kanji by referring to a dictionary. Part-of-speech information such as nouns, sahen nouns, adverbs, adjectives, and adjective verbs is described for each word in the dictionary, and Kana-Kanji conversion analyzes possible reading strings to create possible phrase candidates and combines them. The conversion candidates are determined and presented in order of likelihood. The operator selects a desired candidate from the presented conversion candidates.

For example, for the reading sequence "Shuyoga Hasseishita", "main""mainyo""major""tumor""main""necessary""necessary""tumor""occurred""occurred""occurred" Phrase candidates such as "occurred" and "controlled" are created, and "major occurrence" and "tumor occurrence" are combined and output as conversion candidates in this order.

Although the notation "main" is unfair, in the dictionary, "main" is just a noun, which is not wrong as a morpheme. In addition, the "major occurrence" is not the wrong syntax. In general, “majority” has a higher frequency than “tumor”, so that “majority has occurred” is more likely to be converted as the first candidate.

As it is, the desired candidate is not converted as the first candidate, so a method called example conversion has been proposed in order to increase the conversion rate. The example conversion is a method in which a pair of co-occurring words (examples) is registered in the example dictionary in advance, the example dictionary is referred to during conversion, and the first candidate is changed according to the example in the example dictionary. For example, the pattern "tumor is / occurred" is stored in the example dictionary and "shuyoga hassei"
In response to the input, "tumor has occurred" is converted as the first candidate.

[Problems to be solved by the invention] However, in the example conversion method, in order to correctly convert "tumor is ~" and "tumor ~", it is a perfect conversion unless the words co-occurring with "tumor" are covered. Don't For example, “caused by tumor”, “tumor deteriorates”, “tumor shrinks”, “remove tumor”, “remove tumor”, “remove tumor”, etc. must be registered in the example dictionary. For this reason, a huge memory was required to store the example dictionary. In addition, it has been virtually impossible to cover such examples due to human labor.

Also, when "Shuyouga" was entered in a single phrase, the surrounding situation could not be seen, so the example conversion could not be applied, and an appropriate conversion could not be performed.

[Means (and Action) for Solving Problems] The character processing device of the present invention is configured such that input means for inputting a kana character string, reading of an independent word, notation, word likelihood, and the independent word are An independent word dictionary means that stores the correction value of the word likelihood corresponding to the connected word in association with each other, and the likelihood of each attached word,
An adjunct word dictionary means for storing connection likelihoods between adjunct words and a clause candidate for reading a kana character string input by the input means are created by referring to the independent word dictionary means and the adjunct word dictionary means. By correcting the word likelihood of the independent word in the phrase candidate created by the candidate creating means by the correction value of the word likelihood corresponding to the word to which the independent word is connected, the independent word The likelihood of the part is obtained, the likelihood of the adjunct word is obtained based on the likelihood of each adjunct word in the phrase candidate, and the connection likelihood between the adjunct words, and the likelihood of the independent word part is calculated. Based on the likelihood with the attached word portion, a clause likelihood determining means for determining the clause likelihood of the clause candidate, and based on the clause likelihood determined by the clause likelihood determining means, the candidate creating means Having a phrase candidate output means for outputting the created phrase candidate By correcting the word likelihood of the independent word in the generated bunsetsu candidate by the correction value of the word likelihood corresponding to the word to which the independent word is connected, the likelihood of the independent word part is obtained, and the phrase The likelihood of each adjunct word in the candidates and the likelihood of the adjunct word portion are calculated based on the connection likelihood between each of the adjunct words,
Based on the likelihood of the independent word portion and the likelihood of the adjunct word portion, the clause likelihood of the clause candidate is determined, and based on the determined clause likelihood, the created clause candidate is output. It was done.

EXAMPLES The present invention will be described in detail below with reference to the drawings.

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

In the illustrated configuration, the CPU is a microprocessor, performs arithmetic operations for character processing, logical decisions, etc., and is connected to these buses via an address bus AB, a control bus CB, and a data bus DB. Control elements.

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 the control signals of the respective components to be controlled by the microprocessor CPU. The data bus DB transfers data between the constituent devices.

Next, the ROM is a read-only fixed memory, and stores the procedure of control by the microprocessor CPU described later with reference to FIGS. 12 to 15.

The RAM is a writable random access memory having a structure of 1 word and 16 bits, and is used for temporary storage of various data from each component. IBUF is an input buffer for storing key data input by a key, and OBUF is an output buffer for temporarily storing a result of Kana-Kanji conversion. BCTBL is a phrase candidate table that stores phrase candidates that are in the process of conversion. DIC is a dictionary for Kana-Kanji conversion. The FTBL is an adjunct word table in which adjuncts are stored in association with adjunct likelihoods and serial numbers. CTBL is an adjunct word connection table that describes the connection relationship between adjunct words.

KB is a keyboard for instructing various functions for this character processing device such as alphabetic keys, hiragana keys, katakana keys and other character symbol input keys, and conversion keys, next candidate keys, candidate display keys, enter keys, etc. Equipped with various function keys for.

DISK is an external memory for storing document data,
The document created in the text buffer TBUF is saved, and the saved document is called when necessary by the instruction of the keyboard. CR is a cursor register. The CPU can read and write the contents of the cursor register. The CRT controller CRTC, which will be described later, displays the cursor at the position on the display device CRT corresponding to the address stored here.

DBUF is a display buffer memory that stores data to be displayed.

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 the CRT controller. Furthermore, C
G is a character generator, which stores a pattern of characters and symbols to be displayed on the display device CRT.

In the character processing device of the present invention composed of each of the above components, the character processing device operates according to various inputs from the keyboard KB, and when the inputs from the keyboard KB are supplied,
First, an interrupt signal is sent to the microprocessor CPU, the microprocessor CPU reads various control signals stored in the ROM, and various controls are performed in accordance with those control signals.

FIG. 2 is a diagram showing a conversion example by the device of the present invention. In the figure, "" means that the conversion key is pressed there.

If you enter "Saikin", "Recent" will be converted as the first candidate and "Bacteria" will be converted as the second candidate, but if you enter "Sainin", "Bacteria" will be the first candidate. Become. Since "recently" is not converted, there is no second candidate or lower.

If you enter "yes", "yes yesterday" will be converted as the first candidate and "function" will be converted as the second candidate. If "yes" is entered, "function" will be the first candidate. .

If you enter "shuyou", "main" will be converted as the first candidate and "tumor" will be converted as the second candidate, but if you enter "shuyou", "tumor" will be the first candidate. . Since “main” is not converted, there is no second or lower candidate.

If you enter "Kikan", "Term" is converted as the first candidate and "Institution" is converted as the second candidate, but if you enter "Kikan Sankyo", "Kikan" is converted to "Industry". "
Is used as a prefix, and the first candidate becomes the “key”.

If you enter "shichoo", "mayor" will be converted as the first candidate and "city hall" as the second candidate.
If you enter "Shichode", "At City Hall" will be the first candidate. "By mayor" is the second candidate.

FIG. 3 is a diagram showing the configurations of the input buffer IBUF and the output buffer OBUF.

The IBUF and OBUF have the same configuration. The first 2 bytes are the size information of each buffer, and the numerical value obtained by doubling the value obtained by subtracting 1 from the number of characters stored in the buffer is entered. The "" at the end of the input buffer means that the conversion key was pressed there. Each character consists of 2 bytes per character and is stored in JIS X 0208 code or the like.

FIG. 4 is a diagram showing the structure of the dictionary DIC. "reading"
It is composed of "notation", "part of speech", "word likelihood", and "particle connection likelihood".

"Reading" means reading a word, "writing" means writing a word,
The "part of speech" stores the part of speech of the word.

In the "word likelihood", information indicating the likelihood of the word itself such as frequency information is stored as a value of 0-5. A likelihood value of 0 means that it is most likely, and is interpreted as suspicious as the value increases. Likelihood value 6 means that it cannot be considered at all, and therefore does not exist as a word likelihood value. The particle connection likelihood is stored as described in detail in FIG. 5, but when the case particle of a word follows, information indicating how to correct the word likelihood is stored.

In the figure, it means that the word likelihood should be −5 when used as a prefix for “basic”, and the corrected word likelihood of “basic” is 0 when used as a prefix. Becomes That is, it seems more plausible.

For "yesterday", it means that the word likelihood should be +4 when the particle "wo" is connected, and the corrected word likelihood of "yesterday is" is 5. That is, it is more suspicious.

For "recent", it means that the word likelihood should be +5 when the particle "ga" is connected, and the corrected word likelihood of "recent" is 6 and such notation is no longer possible. There is no.

With respect to "city hall", it means that the word likelihood should be -3 when the particle "de" is connected, and the corrected word likelihood of "at the city hall" is 0. That is, it seems more plausible.

FIG. 5 is a diagram for explaining the structure of the particle connection likelihood of FIG. 4 in more detail.

The particle connection likelihood is divided in units of 4 bits from the beginning, and each 4 bits stores a correction value of the word likelihood corresponding to a specific particle. For example, the connection likelihood corresponding to “ga” stores a value to be corrected to the word likelihood when the particle “ga” follows the word.

The correction value is stored in the range of -5 to 5. Other values (-8, -7, 7) are invalid values. In particular, the value 0 is stored when the word likelihood should not be corrected.

The likelihood as a prefix component of a compound word and the likelihood as a suffix component of a compound word each store a value to be corrected to the word likelihood when the word is used as a prefix or suffix. For example, “core” when used as “core industry” is used as a prefix, and “period” when used as “sales period” is used as a suffix. In such a case, it is described if it is necessary to correct the word likelihood.

FIG. 6 is a diagram showing the concept of the phrase candidate table BCTBL. The phrase candidate table is a binary phrase representing possible phrase candidates as a result of analysis of input reading. In the figure, a horizontal line means a child pointer, and a vertical line means a younger brother pointer. The younger brother pointer links other bunsetsu candidates (usually shorter ones) starting from a certain reading position, and the child pointer links bunsetsu candidates that follow the bunsetsu. In addition, the words enclosed in parentheses mean homophone relations.

For example, the phrase “(recently, in bacteria)” means a homophonic relationship, which means that there are two possible interpretations, “recently” and “in bacteria”.

In the input reading example "Saiki de wa Saiki Nga ...", there are interpretations such as "(recently bacteria)", "(recently bacteria)", "(recently bacteria)", etc. Linked with a pointer (vertical line).

The phrase following "(Recently Bacteria)" is thought to be "Bacteria" and is linked by a child pointer. "Bacteria"
"Not found" is considered as the clause following "," which is also linked by a child pointer.

When such a phrase candidate table is completed, it is easy to create a phrase candidate string such as "Recently / Bacteria / Not found""Recently / Bacteria / Picture / From honey / No".

FIG. 7 is a diagram showing a specific configuration of the phrase candidate table BCTBL.

The "independent word" is composed of 2 bytes each, and links the independent word (existing in the dictionary) of the bunsetsu candidate.

The "adjunct word string" is composed of 2 bytes, and is an area for specifying the adjunct word string following the independent word of the bunsetsu candidate. The first 1 byte indexes the first character of the auxiliary word string on the input buffer, and the next 1 byte indexes the last character of the auxiliary word string on the input buffer. For example, in the case of the input buffer shown in FIG.
Set 10 in the second byte.

"Bunsetsu Likelihood" stores a value indicating the likelihood of the bunsetsu candidate calculated as shown in FIG.

"Homophone link" links bunsetsu candidates in which the independent word of the bunsetsu candidate is changed to another homophone.

The younger brother link links another phrase candidate starting from the same reading position as the phrase candidate.

The child link links a bunsetsu candidate following the bunsetsu candidate.

For example, the homophone link of the phrase candidate 0 is 1, and the phrase candidate 1 (for bacteria) is linked. Also, phrase candidate 1
It can be seen that the homophone link of is 0 and there are no more homophone links.

The younger brother link of the phrase candidate 0 is 50, and the phrase candidate 50 (recently) is linked. The child link is 100 and the phrase candidate is 100.
Link (recently). Further, the child link of the phrase candidate 100 is 200, and the phrase candidate 200 (not found) is linked. It can be seen that the child link is 0, where the input reading ends.

FIG. 8 is a diagram showing the structure of the attached word table FTBL.

"Yomi" stores the reading of each attached word. The “adjunct word likelihood” stores an adjunct word likelihood indicating the likelihood of each adjunct word. A value of 0 to 3 is stored in the attached word likelihood, and 0 means the most likely, and the larger the value, the more suspicious an attached word. Each attached word is managed by a serial number (how many attached words from the beginning).

For example, the adjunct word of the serial number 10 is “ga”, and the adjunct word likelihood thereof is 0. The attached word of the serial number 50 is "ja", and the attached word likelihood is 3.

FIG. 9 is a diagram showing the structure of the attached word connection table CTBL.

The adjunct word connection table is a table describing the connection relation between adjunct words. The preceding adjuncts are shown in the left part of the table, and the following adjuncts are shown in the upper part. Whether or not the adjunct connection is allowed is shown at the intersection of the adjunct shown on the left and the adjunct shown at the top.

The value of the intersection is -1, 0, 1, 2, or 3, and -1 means that they cannot be connected. 0 to 3 means that connection is possible, and the value indicates the likelihood between attached words (likelihood of connection). 0 is a plausible connection, and the larger the value, the more suspicious the connection.

For example, the intersection of the left part 50 and the upper part 130 may be read to see whether the adjunct connection “Ja / Nai” is recognized. Since a value of 0 is obtained, it can be seen that connection can be performed with an inter-word likelihood of 0.

Further, when the intersection of the left part 50 and the upper part 180 is read, it is -1, so it can be seen that the connection "ja / ha" is not recognized.

In addition, when the intersection of the left part 150 and the upper part 180 is read, it is 3. Therefore, it can be seen that the connection “no / ha” is recognized with the inter-adjunct likelihood 3.

FIG. 10 is a diagram showing a calculation example of the phrase likelihood. The phrase likelihood expresses the likelihood of each phrase candidate, and the smaller the value, the more likely it is. Phrase likelihood is calculated by adding α times the independent word likelihood and β times the adjunct likelihood. Here, the independent word likelihood is obtained by adding the particle likelihood to the word likelihood according to the analysis of the adjunct word string of the bunsetsu (what particle is following). The adjunct likelihood is calculated by calculating the sum of the individual likelihoods of individual adjuncts of the adjuncts of the adjunct word sequence that follows the independent word of the bunsetsu and the sum of the likelihoods between the adjuncts. The smaller the clause likelihood, the more likely it is. The values of α and β are α = 2 and β = 5 in this example.
Is.

According to Fig. 10, the phrase likelihood of "function" is 4,
The phrase likelihood of "Yesterday" is (the connection likelihood of "Function / Wo" is +
Since it is 10 (since it is 4), it can be seen that “function” is converted to the first place. Also, the phrase likelihood of "yesterday's ha" is 17 because the adjoining word likelihood of "no / ha" is 3, which means that it is more difficult to convert than a normal phrase.

FIG. 11 is a diagram showing an example of sentence likelihood calculation. The sentence likelihood expresses the likelihood of a sentence that is a phrase candidate sequence,
The smaller the value, the more likely it is. The sentence likelihood is calculated by adding (the number of phrases-1) × inter-phrase likelihood to the sum of the phrase likelihoods. The inter-section likelihood is 20 here.

According to Fig. 11, "recently / bacteria / not found"
The sentence likelihood of is 50 and the sentence likelihood of "recently / with bacteria / from nectar / not" is 80, so "recently / with bacteria / not found" is converted to the first place. I understand.

Also, since the sentence likelihood of "yesterday / enhancement" is 32 and the sentence likelihood of "function / enhancement" is 26, it can be seen that "function / enhancement" is converted to the first place. .

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

FIG. 12 is a flow chart of a part for receiving a key input and performing a process.

Step 12-1 is a process for fetching data from the keyboard. The type of the key fetched in step 12-2 is determined, and the process branches to each key processing routine.

If it is a conversion key, the process branches to step 12-3, and in step 12-3, conversion processing of kana-kanji conversion is performed as described in detail in FIG. If it is any other key, the process branches to step 12-4 to perform other processes such as insertion and deletion which are carried out in a normal character processing device.

In step 12-5, the conversion result or the processing result is displayed. After that, the process branches to step 12-1.

FIG. 13 is a detailed flowchart of the “conversion process” in step 12-3.

In step 13-1, the input reading string is analyzed by performing dictionary search, morphological analysis, syntactic analysis, etc., and bunsetsu candidates are created. In step 13-2, the likelihood of each phrase candidate is calculated, which phrase is most likely to be converted is determined, and the phrase is determined as the first candidate. In step 13-3, create a conversion result based on the determined first candidate,
Output.

FIG. 14 is a detailed flowchart of the “creation candidate creation” in step 13-1.

In step 14-1, the input buffer index i and the phrase candidate table index j are initialized to 0.

In step 14-2, the dictionary is searched for the word candidate based on the reading in the input buffer indicated by i.

In step 14-3, the attached word string connected to the found word candidate is analyzed. As a result, bunsetsu candidates are obtained.

The phrase candidates obtained in step 14-4 are stored in the phrase candidate table. When storing, it is stored in the (j + 1) th entry. Also, set the necessary information. For example,
The phrase likelihood of this phrase candidate is calculated as shown in FIG.
Store. Alternatively, a child link or a younger brother link is set for a phrase candidate whose child or brother is this phrase candidate. After storing, the value of j is incremented.

In step 14-5, a bunsetsu candidate which has not been terminated, that is, a bunsetsu candidate whose child link is not determined yet is found from the bunsetsu candidate table, and the next reading position is assigned to i.

In step 14-6, it is judged whether or not the child links of all the bunsetsu candidates have been decided, and if there are some that have not been decided, the process branches to step 14-2. Otherwise it returns.

FIG. 15 is a detailed flowchart of the “first candidate determination” in step 13-2.

In step 15-1, the maximum likelihood sentence likelihood is initialized to the maximum value allowed in processing.

In step 15-2, one phrase candidate string is extracted from the phrase candidate table.

In step 15-2, the sentence likelihood of the extracted phrase candidate sequence is calculated as shown in FIG.

In step 15-4, it is determined whether the calculated sentence likelihood is more likely than the maximum likelihood sentence likelihood or specifically smaller. If it is smaller, the sentence likelihood calculated in step 15-5 is the maximum likelihood sentence likelihood. Update every time.

In step 15-6, it is determined whether another phrase candidate string can be extracted from the phrase candidate table. If it can be extracted, the process branches to step 15-2. If not, return.

As a result, the phrase candidate sequence from which the maximum likelihood sentence likelihood is calculated is determined as the first candidate.

[Other Embodiments] In the above description, a field is previously defined for each case particle as a storage format of particle connection information so as to have likelihood information. However, in this method, 0 is stored in most cases. Therefore, the connection likelihood may be stored together with the necessary particles.

Further, since the likelihood calculation is merely an example, another calculation method does not particularly impair the gist of the present invention.

As described above, according to the character processing device of the present invention,
As the likelihood of an independent word, the word likelihood of the independent word is corrected according to the word to which the independent word is connected, and the likelihood of the attached part is calculated by the likelihood of each attached word and the connection likelihood between the attached words. Since the likelihood of the bunsetsu for outputting the bunsetsu candidate is determined based on the likelihood of the independent word and the likelihood of the attached part, the independent word is connected to a specific particle, for example. Or when connecting to other independent words as a prefix, by correcting the word likelihood of the independent words and considering the likelihood of the appendages There is an effect that a high bunsetsu candidate is output as a high likelihood part candidate.

[Brief description of drawings]

FIG. 1 is a block diagram of the overall configuration of the present invention, FIG. 2 is a diagram showing an example of output of Kana-Kanji conversion in the present invention, and FIG. 3 is a diagram showing configurations of an input buffer and an output buffer in the present invention. FIG. 4 is a diagram showing a configuration of a kana-kanji conversion dictionary according to the present invention, FIG. 5 is a diagram showing a detailed configuration of a particle connection likelihood, and FIG. 6 is a storage content of a phrase candidate table according to the present invention. FIG. 7 is a conceptual diagram, FIG. 7 is a diagram showing a configuration of a clause candidate table in the present invention, FIG. 8 is a diagram showing a configuration of an adjunct word table in the present invention, and FIG. 9 is an adjunct word in the present invention. FIG. 10 is a diagram showing a configuration of a connection table, FIG. 10 is a diagram showing a calculation example of phrase likelihood in the present invention, FIG. 11 is a diagram showing a calculation example of sentence likelihood in the present invention, and FIGS. FIG. 15 is a flowchart showing the operation of the character processing device of the present invention. DISK …… External memory CPU …… Microprocessor ROM …… Read-only memory RAM …… Random access memory IBUF …… Input buffer OBUF …… Output buffer BCTBL …… Branch candidate table DIC …… Kana-Kanji conversion dictionary FTBL …… Attached Word table CTBL …… Adjective word connection table

Claims (1)

[Claims] 1. An input means for inputting a kana character string, a reading of an independent word, a notation, a word likelihood, and a correction value of the word likelihood according to a word to which the independent word connects are associated with each other. The independent word dictionary means stored as an adjunct word, the adjunct word dictionary means that stores the likelihood of each adjunct word, and the connection likelihood between the adjunct words; and the independent word dictionary means and the adjunct word dictionary means, A candidate creating means for creating a bunsetsu candidate that reads the kana character string input by the input means, and a word likelihood of an independent word in the bunsetsu candidate created by the candidate creating means to a word to which the independent word is connected. By correcting with a correction value of the corresponding word likelihood, the likelihood of the independent word portion is obtained, and based on the likelihood of each adjunct word in the phrase candidate and the connection likelihood between the adjunct words. Based on the likelihood of the independent word part and the likelihood of the adjunct word part, the likelihood of the adjunct word part is obtained. And a phrase likelihood determining unit that determines the phrase likelihood of the phrase candidate, and a phrase candidate that outputs the phrase candidate created by the candidate creating unit based on the phrase likelihood determined by the phrase likelihood determining unit. A character processing device comprising: an output unit.