JPH0576064B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character processing device for inputting a sentence containing kanji and kana through kana-kanji conversion.

[Prior Art] Currently, character processing devices such as Japanese word processors generally input sentences containing kanji and kana using kana-kanji conversion.

Kana-kanji conversion converts the input pronunciation sequence into kanji by referring to a dictionary. In dictionaries, part-of-speech information such as nouns, sub-nouns, adverbs, adjectives, adjective verbs, etc. are described for each word, and kana-kanji conversion analyzes the pronunciation sequence to create possible clause candidates, and then combines them. Conversion candidates are determined and posted in order of likelihood. The operator selects a desired candidate from among the displayed conversion candidates.

For example, for the pronunciation ``Iinkai ni Katsuta'', ``stomach'', ``committee'', ``committee'', ``committee'', and ``mark'' are used.
"Seal""Fire""Section""Meeting""Meeting""Stomach""Stomach"
"in the stomach""two""two""leaf""measured""figured"
Candidate clauses such as ``Konitsuta'' and ``Buitsuta'' are created.
Combinations of these, ``I planned it to the committee,'' ``I planned it to the committee,'' and ``I consulted the committee,'' are output and displayed as conversion candidates in this order. Among the words "measure,""plan," and "consult,""measure" and "plan" are generally more frequent than "consult," so "I consulted the committee."
is not converted as the first candidate.

Therefore, a method called example conversion has been proposed to increase the conversion rate. Example conversion is a method in which pairs of co-occurring words (examples) are registered in advance in an example dictionary, and the first candidate is changed according to the example in the example dictionary by referring to the example dictionary at the time of conversion. . For example, if you memorize the pattern ``to/consult with a committee'' in the example dictionary, you can use it as ``It's okay.
In response to the input, “I consulted the committee” is the first option.
Allow it to be converted as a candidate.

The above example is an example of an individual usage example that describes the relationship between words, but semantic usage examples that describe the relationship with semantic classification have also been proposed. for example,
In the context of ``It was to ~'', when the ``~'' represents a place, it is most likely to be written as ``Gituta'', and when the ``~'' represents a person, it is most likely to be written as ``Kitata''.
The notation seems plausible. In order to perform this conversion correctly, it is necessary to ``tell/say to <<person>>'' and ``tell/say to <<place>>.
An example of the meaning ``to go'' is registered in advance in the example dictionary. Note that ≪human being≫≪place≫ are semantic classifications, and all words in the dictionary have ≪
It is described whether it is a person, a place, or not. This system allows sentences such as ``I told the mayor'' and ``I went to city hall'' to be correctly translated.

[Problem to be Solved by the Invention] However, in the conventional example conversion technology, when multiple examples are applicable, that is, when examples conflict, which example should be given priority is not specified. Nakatsuta. There is only a guideline to give priority to the individual usage example when there is a slight conflict between the individual usage example and the semantic usage example, but it is not clear which usage example should be given priority in other situations. Therefore, since the first candidate was determined by using an appropriate example, the conversion rate was poor in delicate situations.

[Means for Solving the Problems (and Effects)] In order to solve the above problems, according to the present invention, an input means for inputting a kana character string to a character processing device, a word reading, word dictionary means for storing the notation and meaning classification in association with each other; example dictionary means for storing the notation of a word in association with the notation or meaning classification of the word used in conjunction with the word as a usage example; and said word dictionary. a candidate creation means for creating a candidate for the reading of the kana character string inputted from the input means with reference to the means; a priority storage means for storing the priority of the meaning classification; and the candidate creation means Among the created notation candidates for the same reading, priority is given to candidates that correspond to the example stored in the example dictionary means, and when there are multiple candidates that correspond to the example along with the meaning classification of the word used together, the above-mentioned Display priority determining means for determining the display priority of notation candidates by giving priority to candidates corresponding to usage examples related to high-priority semantic classifications stored in the priority storage means; and the display priority determining means. and a candidate display means for displaying the notation candidates created by the candidate creation means based on the display priority determined by the means, so that the notation candidates corresponding to the same reading correspond to the usage example. Prioritize candidates and apply
When there are multiple candidates that correspond to the semantic classification of words that are used together, the display priority of the notation candidates is determined by giving priority to the candidate that corresponds to the usage example related to the semantic classification with a high priority. The notation candidates are displayed based on the displayed display priority.

[Example] 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 that performs arithmetic operations for character processing, logical judgments, etc. Control components.

Address bus AB transfers address signals indicating the components to be controlled by the microprocessor CPU. The control bus CB transfers and applies control signals for each component to be controlled by the microprocessor CPU. The data bus DB transfers data between each component.

Next, the ROM is a read-only fixed memory, and stores the control procedure by the microprocessor CPU, which will be described later with reference to FIGS. 10 to 15.

Further, the RAM is a writable random access memory having a configuration of 1 word and 16 bits, and is used for temporary storage of various data from each component.
IBUF is an input buffer that stores input key data, and OBUF is an output buffer that temporarily stores the results of kana-kanji conversion.
DIC is a dictionary for performing kana-kanji conversion.
YDIC is an example dictionary that describes the co-occurrence relationships between the words ``violation and unclassified''. PRTBL is a semantic classification priority table that describes the priority relationship between semantic classifications. BCTBL is a clause candidate table that stores clause candidates that are currently being converted.

The KB is a keyboard and is equipped with character symbol input keys such as alphanumeric keys, hirakana keys, and katakana keys, as well as various function keys such as conversion keys for instructing various functions to the character processing device. There is.

DISK is an external storage unit for storing document data, and stores documents created on a text buffer, and the stored documents can be called up when necessary by instructions from the keyboard.

CR is the cursor register. By CPU,
Can read and write the contents of the cursor register. A CRT controller CRTC, which will be described later, displays a cursor at a position on the display device CRT corresponding to the address stored here.

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

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

Further, a CRT is a display device using a cathode ray tube or the like, and a CRT controller controls the dot-configured display pattern and cursor display on the display device CRT.

Furthermore, CG is a character generator that stores patterns of characters and symbols to be displayed on the display device CRT.

The character processing device of the present invention, which is composed of each of these components, operates in response to various inputs from the keyboard KB, and when input from the keyboard KB is supplied, an interrupt signal is first sent to the microprocessor. The microprocessor CPU reads various control signals stored in the ROM, and performs various controls in accordance with these control signals.

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

If you enter ``You should never own a cat,'' the first option will be converted to ``You should never own a cat.'' This is because the usage example ``to keep <<animal>>'' is registered in the usage dictionary, and furthermore, ``cat'' is described as <<animal>> as a semantic classification. Without such a mechanism for example conversion, there would be homonyms for ``Katsuwa'' such as ``Buiteha'', ``Katsuteha'', and ``Karittewa'', so it would not necessarily be converted correctly. Put it away.

Next, consider the case where the user inputs "Shishiyamo should not be used in the past." Regarding ``shishiyamo'', ``willow fish'' is registered in the dictionary, and two types of semantic categories are described: <<animal>><<food>. Considering the conventions of Part 1 mentioned above, if it were interpreted as <<animal>>, it should be translated into "Do not keep willow fish." However, in reality, it is correctly translated as "Don't buy willow leaf fish."
This is because the example ``buying <<food>>'' has been registered, and ``willow fish'' was correctly interpreted as <<food>>.

Figure 3 shows input buffer IBUF and output buffer
FIG. 3 is a diagram showing the configuration of an OBUF.

Both IBUF and OBUF have the same configuration. The first two bytes are size information for each buffer, and contain a value obtained by subtracting 1 from the number of characters stored in the buffer and multiplying by 2. The "" at the end of the input buffer means that the conversion key was pressed there. Each character consists of 1 character and 2 bytes,
Stored in JIS X 0208 code, etc.

FIG. 4 is a diagram showing the configuration of the dictionary DIC.
"Reading", "Orthography", "Part of speech", "Word likelihood", "Semantic classification"
It consists of fields.

"Yomi" stores the pronunciation of the word, "notation" stores the word notation, and "part of speech" stores the part of speech of the word.

"Word Likelihood" stores information indicating the likelihood of the word itself, such as frequency information, as a value of 1 to 5.
A likelihood value of 5 means the most likely, and the smaller the value, the more doubtful it is. A likelihood value of 0 means that it is completely unthinkable, so it does not exist as a word likelihood value.

In the "semantic classification" field, a plurality of semantic classifications of the word are generally described, such as <<organization>><<animal>><<food>. However, semantic classification is only described when the word is a noun.

FIG. 5 is a diagram showing the structure of the example dictionary YDIC. It consists of fields of "first word", "second word", and "particle".

A pair of words that co-occur is described in "first word" and "second word." If it is not an individual example but a semantic example, the semantic classification is written instead of the word.

"Particle" describes a particle that connects a pair of words.

In other words, the example shown in the figure is "to the committee/
"Consult", "Keep an animal", "buy food"
becomes.

FIG. 6 is a diagram showing the structure of semantic classification priorities that describe the priority relationships between semantic classifications. "Semantic classification"
It consists of a "Priority" field.

"Semantic classification" stores the semantic classification code,
"Priority" stores the priority for semantic classification. Information indicating the likelihood of the meaning classification is stored as a value of 1 to 5 as the priority. A likelihood value of 5 means the most likely, and the smaller the value, the more doubtful it is.

FIG. 7 is a diagram showing the concept of the phrase candidate table BCTBL. The phrase candidate table is a binary tree representation of possible phrase candidates as a result of analyzing the input reading. In the figure, horizontal lines mean child pointers, and vertical lines mean younger brother pointers. The younger brother pointer links other clause candidates (usually shorter candidates) starting from a certain reading position, and the child pointer links the clause candidates that follow that clause.

The first part of the input pronunciation sequence ``Shishiyamo wo wa wa no wa ni ni'' has interpretations such as ``Yanagiba fish'' and ``dead,'' and these are linked in order with younger brother pointers (vertical lines).

The phrase that follows ``Yanagiha fish'' is thought to be ``Don't keep them,'' and is linked with a child pointer.

Once a phrase candidate table like this is created,
It is easy to create phrase candidate sequences such as "You must not keep willow fish,""You must not buy willow fish,""You cannot keep willow fish," and "You cannot buy willow fish without permission."

FIG. 8 is a diagram showing the specific structure of the phrase candidate table BCTBL.

"Independent word" stores a pointer to the top position in the dictionary where the independent word of the bunsetsu candidate exists.

The "adjunct word string" consists of 2 bytes, and is an area for specifying the adjunct word string that follows the independent word of the bunsetsu candidate. The first 1 byte indexes the first character of the attached word string on the input buffer, and the next 1 byte indexes the last character of the attached word string on the input buffer. For example, in the case of the input buffer shown in Fig. 3,
To express "Do not touch", set 8 in the first byte and 20 in the second byte. In the figure, "φ" means that there is no attached word string.

"Little brother link" links another clause candidate that starts from the same reading position as the clause candidate.

A "child link" links a clause candidate that follows the clause candidate.

For example, the younger brother link of phrase candidate 0 is 50, and phrase candidate 50 (dead) is linked. There are 200 child links, and 200 phrase candidates (don't keep them) are linked. Furthermore, the child link of clause candidate 200 is 0,
It can be seen that the input reading ends there.

FIG. 9 is a diagram showing an example of calculating sentence likelihood. Sentence likelihood expresses the likelihood of a sentence that is a clause candidate sequence, and the larger the value, the more likely it is.

The sentence likelihood is calculated by adding the sum of clause likelihoods, the sum of inter-clause likelihoods, and the sum of example likelihoods.

The clause likelihood expresses the likelihood of each clause candidate, and the larger the value, the more likely it is. The clause likelihood is the dictionary DIC of the independent words of the clause.
The word likelihood described in is used.

The inter-clause likelihood expresses the likelihood of the connection between adjacent clause candidates, and in this example, the value -
Fixed at 20. If the sentence is composed of n clauses, there are (n-1) between clauses, so the sum of inter-clause likelihoods is always -20(n-1).

Example likelihood is applied between each clause. It is added to the sentence likelihood for each example. The calculation formula for example likelihood is different depending on whether it is an individual example or a semantic example. When a semantic example is applied, the example likelihood is calculated from the priority value of the semantic classification used,
The number of clauses to be skipped when the example is applied is reduced by five times. When an individual example is applied, the example likelihood is the value 20 minus five times the number of clauses to skip when applying the example.

According to Figure 9, the sentence likelihood of ``You must not/keep willow leaf fish'' is -10, and the sentence likelihood of ``You must not /buy willow leaf fish'' is -8. It can be seen that ``Don't be angry'' is converted to the first place.

The reason why it is converted correctly in this way is that when applying the example ``to/keep willow-leaf fish'', ``willow-leaf fish'' is interpreted as <<animal>>, so the usage likelihood = 2, whereas the usage example ``to/keep willow-leaf fish'' This is because when applying "buy", "willow fish" is interpreted as <<food>>, so the example likelihood=4. In other words, ``willow fish'' is more likely to be interpreted as <<food>>, so it is correctly converted.

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

FIG. 10 is a flowchart of the part in which key input is received and processed.

Step 10-1 is a process for reading data from the keyboard. At step 10-2, the type of key taken in is determined, and the process branches to a processing routine for each key.

If it is a conversion key, the process branches to step 10-3, and in step 10-3, conversion processing for kana-kanji conversion is performed as detailed in FIG. 11. If any other key is pressed, the process branches to step 10-4, and other processing such as insertion, deletion, etc. performed in a normal character processing device is performed. Then step
Branch to 10-1.

FIG. 11 is a detailed flowchart of the "conversion process" in step 10-3.

In step 11-1, the phrase candidate creation process detailed in FIG. 12 is performed, and the phrase candidate table is created.
Create BCTBL.

In step 11-2, a first candidate determination process, detailed in FIG. 13, is performed.

In step 11-3, a conversion result is created and output based on the determined first candidate.

FIG. 12 is a detailed flowchart of the ``phrase candidate creation process'' in step 11-1.

In step 12-1, the input buffer index i and phrase candidate table index j are set to 0.
Initialize to .

In step 12-2, a dictionary is searched to find word candidates based on the pronunciation in the input buffer indicated by i.

In step 12-3, morphological analysis processing is performed to analyze adjunct word strings connected to the found word candidates. As a result, phrase candidates are obtained.

The clause candidates obtained in step 12-4 are stored in the clause candidate table. j+1 when storing
Store in the th entry. Also, set the necessary information. For example, for a clause candidate that has this clause candidate as a child or younger brother, a child link and a younger brother link are set. After storing, count up the value of j.

In step 12-5, an unterminated phrase candidate, that is, a phrase candidate whose child link has not yet been determined, is found from the phrase candidate table and its next reading position is assigned to i.

In step 12-6, it is determined whether the child links of all clause candidates have been determined, and if there are any child links that have not been determined, the process branches to step 12-2.
Otherwise, return.

FIG. 13 is a detailed flowchart of the "first candidate determination process" in step 11-2.

In step 13-1, the maximum sentence likelihood is initialized to the minimum value allowed for processing, for example, -32767.

In step 13-2, one clause candidate string is extracted from the clause candidate table.

In step 13-3, as detailed in FIG. 14, an example applicable to the extracted sentence candidate string is searched, and a "sum of example likelihoods" which is a sum of example likelihoods is calculated.

In step 13-4, the sentence likelihood of the phrase candidate sequence is calculated as shown in FIG.

In step 13-5, it is determined whether the calculated sentence likelihood is more likely than the maximum likelihood sentence likelihood, specifically whether it is larger, and if it is larger, in step 13-6 the maximum likelihood sentence likelihood is compared to the calculated sentence likelihood. Update every time. Further, in step 13-7, the current phrase candidate string is stored as a phrase candidate string corresponding to the maximum sentence likelihood.

In step 13-8, it is determined whether another phrase candidate column can be retrieved from the phrase candidate table, and
If it can be taken out, the process branches to step 13-2. Otherwise, return.

As a result, the clause candidate sequence that is the basis for calculating the maximum likelihood sentence likelihood is determined as the first candidate.

FIG. 14 is a detailed flowchart of step 13-3, ``Calculation of example likelihood sum''.

In step 14-1, the example likelihood sum is first set to 0.
Initialize to .

In step 14-2, one clause is extracted from the clause candidate string as the clause of interest. In the following processing, the likelihood of examples related to this phrase of interest will be determined.

In step 14-3, the maximum likelihood example likelihood is initialized to zero.

In step 14-4, a phrase that pairs with the phrase of interest is extracted from the phrase candidates.

In step 14-5, if a paired clause of the target clause cannot be extracted, processing of any further target clauses is given up and the process branches to step 14-8.

In step 14-6, as detailed in FIG. 15, it is checked whether the example between the target phrase and the paired phrase is applied, and the example likelihood is set depending on the application status.

In step 14-7, it is checked whether the obtained example likelihood is more likely (i.e., larger) than the maximum likelihood example likelihood, and if it is more likely (i.e., larger), the value of the example likelihood is calculated as Assign each time. Thereafter, the process branches to step 14-4, and the example likelihood is calculated for another pair.

In step 14-8, since the processing of the target phrase is completed, processing moves on to the next target phrase, but it is determined whether there are any unprocessed phrases remaining. If there are none left, the process returns as is, but if there are, the process branches to step 14-9, and the previously determined maximum likelihood example is added to the sum of example likelihoods. Thereafter, the process branches to step 14-2, and the next phrase of interest is extracted.

FIG. 15 is a detailed flowchart of step 14-6, ``setting of example likelihood value.''

In step 15-1, it is determined whether there is an example applied between the two phrases of interest and a paired phrase, and if so, whether it is an individual example or a semantic example.

If there is no applied example, the process branches to step 15-2, assigns 0 to the example likelihood, and returns.

If the individual example is applicable, the process branches to step 15-3, and in step 15-3, 20 is assigned to the example likelihood. Thereafter, the process branches to step 15-5.

If the semantic example is applicable, the process branches to step 15-4, and the priority of the semantic classification used for the semantic example is substituted as the example likelihood. Thereafter, the process branches to step 15-5.

In step 15-5, the distance between two clauses, that is, the number of clauses to be skipped, is determined, and the value is multiplied by 5 and added to the example likelihood. If the two clauses are adjacent, the clause to be skipped is 0, so 0 is added to the example likelihood.

In step 15-6, it is determined whether the value of the example likelihood calculated above has become negative. If it is negative, it is corrected to 0 at plug 15-7, and the process returns. If it is not negative, return without changing the value.

[Other Examples] In the above explanation, the example dictionary was explained as a pair of two co-occurring words (or categories), but it can be processed in the same way even if it is a triplet or more generally a set of n words. can do.

In addition, the calculation of the likelihood is merely an example; if the priority of the semantic classification of the applied semantic example is considered in the process of likelihood calculation, even if another calculation method is used. This does not particularly impair the spirit of the present invention.

[Effects of the Invention] As is clear from the above explanation, according to the present invention, when a plurality of semantic examples are applicable to the reading sequence input by the operator, the priority of the semantic classification used in the semantic examples is used. Since it is possible to judge which semantic example should be applied based on the given situation, more appropriate conversion can be performed. This makes it possible to realize a comfortable character processing device with a high conversion rate.

[Brief explanation of the drawing]

Figure 1 is a block diagram of the overall configuration of the present invention, Figure 2 is a block diagram of the overall configuration of the present invention.
The figure shows an example of Kana-Kanji conversion in the present invention, and Figure 3 shows the input buffer in the present invention.
Diagram showing the configuration of IBUF and output buffer OBUF,
Figure 4 shows the DIC dictionary for kana-kanji conversion in the present invention.
FIG. 5 is a diagram showing the configuration of the example dictionary YDIC in the present invention, and FIG. 6 is a diagram showing the configuration of the semantic classification priority table PRTBL that describes the priority relationship of semantic classification in the present invention. 7 is a diagram conceptually showing the storage contents of the phrase candidate table BCTBL in the present invention, FIG. 8 is a diagram showing the structure of the phrase candidate table BCTBL in the present invention, and FIG. 10 to 15 are flowcharts showing the operation of the character processing device of the present invention. DISK...external storage, CPU...microprocessor, ROM...read-only memory, RAM...
…Random access memory, IBUF…Input buffer, OBUF…Output buffer, DIC…Kana-kanji conversion dictionary, YDIC…Example dictionary, PRTBL…
...Semantic classification priority table, BCTBL...Phrase candidate table.

Claims (1)

[Scope of Claims] 1. An input means for inputting a kana character string, a word dictionary means that stores the pronunciation of a word, its notation, and a meaning classification in association with each other; example dictionary means for storing as usage examples the corresponding representations or semantic classifications of words; and the word dictionary means to create candidates for the reading of the kana character string input from the input means. a candidate creation means; a priority storage means for storing priorities of semantic classification; and among the notation candidates for the same pronunciation created by the candidate creation means, priority is given to candidates corresponding to the usage examples stored in the example dictionary means. However, when there are multiple candidates corresponding to the usage example along with the semantic classification of the word used together, priority is given to the candidate corresponding to the usage example related to the meaning classification with a high priority stored in the priority storage means. a display priority determining means for determining the display priority of the notation candidates; and a notation candidate created by the candidate creation means based on the display priority determined by the display priority determining means. 1. A character processing device comprising: candidate display means for displaying candidates.