JPH0567076A - Kana/kanji converter - Google Patents

Abstract

PURPOSE:To make it possible to efficiently suppress the output of an unnecessary homonym at the time of performing KANA (Japanese syllabary)/KANJI (Chinese character) conversion processing. CONSTITUTION:Based on reading information to be inputted from an input part 1 inputting the reading information of a Japanese sentence, a word information is retrieved from a dictionary storage part 5 and plural conversion candidates composed of the combinations of the word information and accessory words are extracted from a paragraph group extraction part 3. For these conversion candidates, the existence of synthetic words are detected by using a synthetic words conjunction rules storage part 7 and a synthetic words meanings classification and determination rules storage part 8 and the meaning classification of each existing synthetic word is determined. Based on these determined meaning classifications, the selection priority order of each conversion candidate is determined.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a kana-kanji conversion device used in a Japanese word processor for converting kana-character input into sentence information containing kanji and kana.

[0002]

2. Description of the Related Art Recently, a Japanese word processor has been widely used as an input means for inputting reading information of a Japanese sentence and converting it into sentence information containing kanji and kana.

In such a Japanese word processor, when the reading information of a sentence is input by kana or romaji input from the keyboard, the operation timing of a specific key for instructing these breaks, such as a break of a phrase or a break of a sentence. Alternatively, when punctuation marks are entered during kana input, or when the number of input characters exceeds a predetermined number of characters, conversion processing of kana-kanji mixed notation corresponding to the input kana input is executed, and the user It is supposed that the kana-kanji mixed notation that you want will be obtained.

By the way, since there are various homonyms in Japanese, as a result of kana-kanji conversion processing by a Japanese word processor, a word to be converted is selected from a plurality of conversion candidates. There is. However, such selection of homonyms is very troublesome and sometimes interrupts the thinking of the user who is inputting a sentence, which is a cause of significantly reducing the work efficiency of the sentence input.

Therefore, conventionally, a frequency is assigned in advance to each homonymous phoneme and the one designated as the first candidate is output according to the frequency, or the homophones sequentially selected in the input process. A method of learning a synonym and preferentially outputting it as a first candidate is considered. Such a method greatly contributes to reducing the number of operations required for selecting homonyms when the entire input operation is viewed.

However, there are cases where a plurality of homonyms such as "the reporter returns to work" are input in one input sentence. In this case, in each case,
Since the operation of selecting a homonym is required, the effect of each method described above cannot be expected.

Therefore, assuming that a plurality of homonyms exist in such one input sentence, the semantic classification assigned to each word or the co-occurrence relation of each word is stored in advance. A method of preferentially outputting conversion candidates having this co-occurrence relationship is also considered.

In this method, for example, in response to a reading input such as "kishagakisha", "(person)-
By pre-storing co-occurrence relation data between words and semantic classifications such as (operation), and using this, it is possible to preferentially output "reporter returns to work" from among multiple conversion candidates. I have to.

By the way, it is known that, in general, Japanese can produce an extremely large number of compound words by combining words and affixes, so that the number of co-occurrence relations that can be considered from these is extremely large. Therefore, it is impossible to prepare the co-occurrence relation data for the combination of the word and the compound word in advance in the dictionary as a practical problem, and as long as it does not correspond to the co-occurrence relation data prepared in the dictionary, it remains unchanged. Other unnecessary homonyms were output, and a troublesome operation was required to select these homonyms.

[0010]

As described above, in the conventional kana-kanji conversion device, the co-occurrence relation data prepared in the dictionary is extremely limited. Since a large number of words are output, there is a drawback that a troublesome operation is still required for selecting homonyms.

The present invention has been made in view of the above circumstances, and can effectively suppress the output of unnecessary homonyms when performing kana-kanji conversion processing, and an operation for selecting homonyms. It is an object of the present invention to provide a kana-kanji conversion device which can be easily and minimally.

[0012]

A kana-kanji conversion device according to the present invention includes an input means for inputting reading information of a Japanese sentence, a storage means for storing a plurality of words and a semantic classification corresponding to these words, and an input. Conversion candidate extraction means for searching word information from the storage means based on the reading information input by the means and outputting a plurality of types of conversion candidates composed of combinations of the word information and attached words, at least a first word or meaning It is output from the semantic classification determination information storage means and the conversion candidate extraction means that store, as semantic classification determination information, a rule for determining a semantic classification related to a compound word composed of a classification and a second word or a word belonging to the semantic classification Synthetic word meaning for detecting the existence of a compound word using the semantic classification decision information for each conversion candidate and determining the semantic classification of each existing compound word ; And a kind determining means is configured to determine the selection priority of the conversion candidates based on the determined semantic classification by the compound word semantic classification determining means.

[0013]

As a result, according to the present invention, a conversion candidate consisting of a combination of word information obtained from input reading information and an accessory word is converted even if each word or its semantic classification is not included in the co-occurrence information. The presence of a compound word in a candidate can be used to determine the selection priority of conversion candidates from the semantic class of the compound word by using a rule for determining the semantic class of the compound word. It is possible to effectively suppress the output of necessary homonyms.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a circuit configuration of the embodiment. In the figure, reference numeral 1 denotes an input unit, which is composed of, for example, a keyboard, and reading information for creating a sentence is input by a Kana character string or a Roman character string. The input unit 1 also selects homonyms, registers next candidates, co-occurrence information, and compound word meaning determination rules,
It also has a special key for entering data such as deletion.

The input data from the input unit 1 is sent to the control unit 2. The control unit 2 judges the above input data and gives various control commands to each circuit. For example, when it is judged that the kana-kanji conversion is necessary, the reading information (input from the input unit 1 ( The case of a kana character string will be described below) is given to the phrase series extracting unit 3 and the dictionary searching unit 4, respectively.

The phrase series extraction unit 3 extracts a combination of phrases consisting of words and attached words from the kana character string input from the input unit 1, and searches the dictionary storage unit 5 for word information by the dictionary search unit 4. , The possibility of connecting the part-of-speech of each word and the adjunct word is examined, and the combination of the word and the adjunct word judged to be valid is output as a segment candidate. The dictionary search unit 4 searches the dictionary storage unit 5 and retrieves all the word information for the segment candidates for the input kana character string.

As shown in FIG. 2, the dictionary storage section 5 classifies and registers reading information A, surface character string information B, part-of-speech information C, meaning classification information D, etc. for each of a plurality of words.
Note that E is the storage address information.

The semantic classification information D among the information classified and registered in the dictionary storage unit 5 is used by the phrase candidate extraction unit 3. The semantic classification information D here is information in which semantic attributes of words are assigned. For example, words relating to “relevance” (a), words belonging to “state” (b), words belonging to “plan” (C), (d) for words that belong to "Affair", (e) for words that belong to "Variety",
It is specified that (f) is assigned to a word belonging to “actual”, (g) is assigned to a word belonging to “character”, and (h) is assigned to a word belonging to “influence”.

The output of the phrase sequence extraction unit 3 is the compound word processing unit 6
Sent to. The compound word processing unit 6 refers to the compound word connection rule storage unit 7 to analyze whether each phrase candidate has a compound word. Here, when a compound word is detected as a phrase candidate, the compound word meaning classification determination rule storage unit 8 is searched for a rule that includes a combination of a word and a semantic classification of the constituent elements of the compound word, and the compound word The semantic classification is taken out and this information is added to the phrase candidate.

As shown in FIG. 3, the compound word connection rule storage unit 7 stores the part-of-speech information A of the first word forming the compound word, the part-of-speech information B of the second word forming the compound word, and the compound word. The part-of-speech information C of each is classified and registered. Note that D is storage address information.

Further, the compound word meaning classification determination rule storage unit 8
4, as shown in FIG. 4, meaning classification information A of the first word forming the compound word, surface character string information B of the second word forming the compound word, and meaning classification information of the composed word C
Are classified and registered respectively. Note that D is storage address information.

The outputs of the phrase series extraction unit 3 and the compound word processing unit 6 are sent to the output control unit 9. The output control unit 9 grammatically evaluates all bunsetsu candidates for the input kana character string obtained by the bunsetsu sequence extracting unit 3 and added with the semantic classification information on the compound word by the compound word processing unit 6 and grammar. I try to abandon segment candidates that are unlikely to exist. The output control unit 9 also determines the order of homonyms of each bunsetsu candidate by referring to the semantic classification and the combination information between words in the co-occurrence information storage unit 10. When the combination information stored in the information storage unit 10 is applicable, the priority order is changed.
That is, when the priority is instructed to be high, the output is performed in the first order, and when the priority is instructed to be low, the output is performed last.

The co-occurrence information storage unit 10 stores, as co-occurrence information,
As shown in FIG. 5, the first semantic classification information A, the surface character string information B of the second word that co-occurs with this semantic classification information A, and each pattern information C in this co-occurrence are classified and registered. .. Note that D is storage address information.

Of the classification information here, the case pattern information C is the information attached to words that control the case, such as the Sahen noun, the adjective, and the adjective verb, to which the verb "suru" is connected, and the first meaning classification information. It shows information about which particle the word belonging to A is to be combined with the word shown in the information B or the word belonging to the second meaning classification.

In FIG. 1, 11 is a document storage unit and 12
Indicates the display unit. In this case, the document storage unit 11
The conversion result of the kana-kanji mixture processed by the output control unit 9 is stored. Further, the display unit 12 is adapted to output and display the conversion result of kana-kanji mixture processed by the output control unit 9. Here, the segment candidates having the same homonyms are displayed differently from each other by changing the brightness or color, pulling, or the like. When the homonyms displayed on the display unit 12 are not the desired ones,
By inputting input data such as the next candidate and the recutting of the phrase from the input unit 1, the word is converted into a desired word. Next, the operation of the embodiment configured as described above will be described. Now, when the reading information is input from the input unit 1 as a kana character string, this input kana character string is sent to the phrase series extraction unit 3 and the dictionary search unit 4, respectively.

Then, in the phrase sequence extraction unit 3, the input unit 1
A combination of a clause consisting of a word and an adjunct word is extracted from the kana character string input from, and the word part retrieved from the dictionary storage unit 5 by the dictionary retrieval unit 4 can be connected to the word part of speech and an adjunct word. And outputs the combination of the word and the adjunct word judged to be valid as each bunsetsu candidate.

Then, the output from the phrase sequence extraction unit 3 is
It is sent to the compound word processing unit 6. In this compound word processing unit 6,
For all bunsetsu candidates for the input kana character string, the compound word detection inspection is performed. The compound word detection process and the compound word semantic classification determination process here are executed according to the flowchart shown in FIG. It

In this case, the word information of the phrase candidate is first copied to the phrase candidate register (not shown) (step A).
1). Then, the word pointer is made to point to the word in the first conversion unit of the phrase candidate (step A2).

Next, the information of the word pointed to by the word pointer is copied to a compound word register (not shown) (step A).
3). Then, the process proceeds to step A4, and when the word pointer points to the last word of the phrase candidate register, the process branches to step A14, and otherwise the process proceeds to step A5.

If the word pointer does not point to the last word in the phrase candidate register, the process advances to step A5 to advance the word pointer to the next word. Then, it is determined whether the combination of the part-of-speech information in the compound-word word information stored in the compound-word register and the part-of-speech information in the word information of the word pointed by the word pointer exists in the compound word connection rule storage unit 7. Check (step A6).

Next, the process proceeds to step A7, and step A6
If the applicable rule is found by the inspection of step A8, step A8
If not, the branch process proceeds to step A12.

If the rule is found by the check in step A6, the word information of the compound word register is rewritten with the part-of-speech information of the compound word obtained from the rule of the compound word connection rule storage unit 7 (step A8). Then, it is checked whether or not the combination of the synthesized word meaning information stored in the synthesized word register and the word indicated by the word pointer exists in the synthesized word meaning classification determination rule storage unit 8 (step A
9).

Next, the process proceeds to step A10 and step A
If the applicable rule exists in the inspection of step 9, step A
11. If not, the branch process proceeds to step A12 if not.

If the rule is found by the check in step A9, the compound word meaning classification determination rule storage unit 8
The semantic classification information of the synthetic word register is rewritten with the semantic classification information of the synthetic word obtained from the rule (step A11). Then, the process returns to step A4, and the above operation is repeated.

On the other hand, if it is determined in step A7 or step A10 that the corresponding rule does not exist, the process proceeds to step A12. Here, since the end of the compound word has been detected, in the compound word register. The word information of the word immediately before the word pointed by the word pointer is updated with the word information of. Then, after initializing the compound word register (step A13), the process returns to step A3 to repeat the above operation.

After that, if it is determined in step A4 that the word pointer points to the last word in the phrase candidate register, the process proceeds to step A14. At step A14, since the processing of all the phrase candidates has been completed, the word information of the word pointed by the word pointer is updated with the word information in the compound word register. The processing of the compound word as described above is performed for all the phrase candidates, and the output of the compound word processing unit 6 at this time is sent to the output control unit 9.

The output control unit 9 grammatically evaluates all bunsetsu candidates for the input kana character string to which the meaning classification information on the synthesizing word is added in the synthesizing word processing unit 6, and discards the grammatically impossible bunsetsu candidates. To do. Then, referring to the semantic classification and the combination information between words in the co-occurrence information storage unit 10,
The order of homonyms of each phrase candidate is determined, and the result is stored in the document storage unit 11 and displayed on the display unit 12. Next, the above-mentioned operation will be described with reference to specific sentence examples. Now, consider a case where the reading information "Kihonhosakujitai no Hanei" is input and the text information "Reflecting the basic policy itself" is obtained.

Here, when the conventional kana-kanji conversion device is used, "Kihon / Housaku /
Even if the recognition of the break of the word is successfully performed, such as "jitai / no / hanei", due to the error of the homophone,
Occasionally, wrong conversion candidates such as "reflection of basic good harvest situation" and "reflection of basic policy situation" were output. It is considered that this is because the co-occurrence information such as “policy” and “of / reflecting” could not be used successfully because the co-occurrence information was used without analyzing the compound word. On the other hand, the kana-kanji conversion device of the present invention is as described below.

First, when the kana information "Kihonhosakujitai no Hanei" is input from the input unit 1, this reading information is sent to the control unit 2, and then the phrase sequence extraction unit 3 And sent to the dictionary search unit 4.

The phrase series extraction unit 3 extracts a combination of a word and an accessory word from the input kana character string, and the dictionary search unit 4
The segment candidate is output using the part-of-speech of the word obtained by searching for and the connection information of the adjunct. Here, of the combinations of clauses, only the process for “kihon / hosaku / jitai / no / hanei” will be the subject of the following description.

Then, the segment candidate "Kihon / Housaku /
The conversion units that make up "jitai / no / hanei" are "kihon", "hosaku", "jitai", "no", and "hanei". Among these conversion units, The ones that have multiple conversion possibilities are "hosaku" and "jitai". And, of these, there are two possibilities for "hosaku", "policy" and "good harvest", and for "jitai",
It is known from the search result of the dictionary storage unit 5 that there are four possibilities of “situation”, “decline”, “self”, and “letter”. Therefore, in the phrase candidates of the above example sentences,
There are 2 × 4 conversion candidates, and all of these conversion candidates are processed by the compound word processing unit 6. And
The process of selecting the correct "reflection of the basic policy itself" from these candidates will be described with reference to the flowchart of FIG. In this case, the contents of the conversion candidate register at each time point are shown in FIG. 7, and the contents of the word pointer and the compound word register at each time point are shown in FIG.

First, in step A1, the phrase series extraction unit 3
The word information related to the conversion candidate "basic / policy / self / no / reflection /" output from is copied to the conversion candidate register (time T1). Next, in step A2, the word pointer is set to the first word "basic" of the conversion candidate "basic / policy / self / no / reflection" (time T2). Continued Step A
At 3, the word information of "basic", which is the word pointed to by the word pointer, is copied to the compound word register (time T3). Then, in step A4, it is determined whether or not the word pointer points to the last word. However, since it does not point to the last word here, the process proceeds to step A5.

At step A5, the word pointer is advanced to the next word "policy" (time point T4). And step A
In step 6, it is checked whether or not a combination of the part-of-speech information “noun” of the compound word register and the part-of-speech information “noun” of the word “policy” indicated by the word pointer exists in the compound word connection rule storage unit 7.

In this case, in the next step A7, it is judged whether or not the corresponding rule exists in the compound word connection rule storage unit 7 shown in FIG. 3. However, since the corresponding rule exists in the address P11 here, the process goes to step A8. Execute forward branch processing.

At step A8, since the part-of-speech information C of the compound word at the address P11 in the compound-word connection rule storage unit 7 is "noun", the part-of-speech information of the compound word register is rewritten to "noun" (time T5). In this example, the part of speech is the same as before rewriting.

Next, in step A9, the combination of the synthesized word meaning information (a) in the synthesized word register and the word "policy" indicated by the word pointer is stored in the synthesized word meaning classification determination rule storage unit 8
To see if it exists.

In this case, in the next step A10, it is judged whether or not the corresponding rule exists in the compound word meaning classification determination rule storage unit 8 shown in FIG. The branch process of proceeding to A11 is executed.

At step A11, since the semantic classification C of the compound word at the address P11 of the synthetic word semantic classification determination rule storage unit 8 is (c), the semantic classification of the synthetic word register is rewritten with this information, and the process proceeds to step A4. Return (time point T6).

In step A4, it is again determined whether or not the word pointer points to the last word. Again, since the word pointer has not reached the last word, step A
Go to 5.

At step A5, the word pointer is advanced to "itself" (time point T7). Then, in step A6,
It is checked whether or not there is a combination of the part-of-speech information "noun" in the compound word register and the part-of-speech information "noun" of the word "itself" pointed to by the word pointer in the compound word connection rule storage unit 7. In this case, even in the next step A7 inspection, the previous step A7
Since it is determined that the corresponding rule exists in the same manner as, the process proceeds to step A8.

In step A8 as well, the same processing as in the previous step A8 is performed, and the part-of-speech information in the compound word register is rewritten to "noun". (Time point T8) Also in this example, the part of speech is the same as before rewriting.

Next, in step A9, it is determined whether or not the combination of the synthesized word meaning information (c) in the synthesized word register and the word "itself" pointed to by the word pointer exists in the synthesized word meaning classification determination rule memory word 8. Find out.

In this case, in step A10, it is determined whether or not the corresponding rule exists in the compound word meaning classification determination storage unit 8 shown in FIG. 4, but since the corresponding rule exists in the address P21 here, the process proceeds to step A11. Execute the process.

At step A11, since the semantic classification C of the compound word at the address P21 of the synthetic word semantic classification determination rule storage unit 8 is (c), the semantic classification of the compound word register is rewritten with this information, and step A4 is again performed. Return to (time point T
9).

In step A4, it is again determined whether the word pointer points to the last word. Again, since the word pointer has not reached the last word, step A
Go to 5.

In step A5, the word pointer is advanced to the word "no" (time point T10). And step A6
Then, it is checked whether or not a combination of the part-of-speech information "noun" of the compound word register and the part-of-speech information "particle" of the word "no" pointed by the word pointer exists in the compound word connection rule storage unit 7. In this case, in the inspection of step A7, it is determined that the corresponding rule does not exist in the compound word connection rule storage unit 7 of FIG. 3, so the process proceeds to step A12.

In step A12, since the end of the compound word is detected, the word information of the word "itself" immediately before the word indicated by the word pointer is updated with the word information of the compound word register (time point T11). ). And step A13
Then, the compound word register is initialized and the process returns to step A3 (time point T12). At step A3, the word information of the word "NO" pointed by the word pointer is copied to the compound word register (time point T13).

Then, again in step A4, it is judged whether or not the word pointer points to the last word. However, since the word pointer does not point to the last word also here, step A5
Go to.

At the next step A5, the word pointer is advanced to the next word "reflection" (time point T14). Then, in step A6, the part-of-speech information “particle” of the compound word register and the part-of-speech information “noun” of the word “reflection” pointed by the word pointer.
It is checked whether or not the combination of is present in the compound word connection rule storage unit 7. In this case as well, it is determined in step A7 whether or not the corresponding rule exists, but since there is no corresponding rule here, the process proceeds to step A12.

At step A12, the word information of the word "itself" immediately before the word indicated by the word pointer is updated with the word information of the compound word register (time point T15). here,
This corresponds to the case where a compound word starting from the word "no" was not detected, but since only one word "no" was processed, the contents of the conversion candidate register are the same as before this rewriting. There is. Then, in step A13, the compound word register is initialized, and the process returns to step A3 (time point T1.
6). In step A3, the word information of the word "reflection" pointed by the word pointer is copied to the compound word register (time point T17).

Then, again in step A4, it is judged whether or not the word pointer points to the last word. Here, since the word pointer points to the last word "reflection" of the conversion candidate register, the process goes to step A14. move on.

At step A14, since the processing has been completed for all the segment candidates, the word information of the word "reflection" pointed to by the word Poitan is updated with the word information in the compound word register (time point T18). Again, since the word "reflection" information itself is stored in the compound word register as the word information of the compound word, the contents of the conversion candidate register before and after this rewriting are the same, but in general, This process is necessary because the last word of the conversion candidate may be a constituent of the compound word.

By the series of processing as described above, the conversion candidate “reflecting the basic policy itself” is replaced with “basic: policy: self”
Is detected to be a compound word, and its semantic classification is found to be (c).

The processing of such a compound word is performed for all bunsetsu candidates, but the above-mentioned "/ Kihon /
FIG. 9 shows the processing results for the eight conversion candidates for "hosaku / jitai / no / hanei /".
In this case, in FIG. 9, the symbol “/” represents a break in the phrase, and the symbol “+” represents the connection of the annex in the phrase. Further, the symbol “:” indicates the connection between the constituent elements of the compound word detected by the above processing.

The output of the compound word processing unit 6 thus obtained is sent to the output control unit 9. The output control unit 9 refers to the combination of the semantic classification, the word surface character string information and each pattern information of the co-occurrence information storage unit 10 shown in FIG. change.

For example, in the candidate number 3 for "/ kihon / hosaku / jitai / no / hanei /" shown in FIG.
It has been obtained that the above-mentioned compound word "basic: policy: itself" is detected, and its semantic classification information is (c).
Then, this compound word is connected to the word "reflection" by the case pattern "/", and by referring to the content of the address P11 of the co-occurrence information storage unit 10 shown in FIG. The output priority of can be increased. In other words, in the conventional method, since the word “policy” is not adjacent to the particle “of”, the information that the word “policy” co-occurs with the word “reflection” in each pattern “/” cannot be properly used. That point can be improved by the present invention.
The result is stored in the document storage unit 11 and can be displayed on the display unit 12.

The present invention is not limited to the above-mentioned embodiments, and can be carried out by appropriately modifying it within the scope of the invention.
For example, in the above-described embodiment, the part-of-speech information of each word is used in the process of detecting a compound word, but this process may be performed using syntactic information other than the part-of-speech information. In the above, the determination process of the semantic classification of the compound word was performed by the combination of the semantic classification information and the surface character information of the word. However, this is performed by the combination of the semantic classification information or the rule including the syntax information. You can go. Further, a rule for inheriting the semantic classification information of the constituent elements of the synthetic word to the semantic classification information of the synthetic word may be included in the synthetic word semantic classification rule. Further, in the above-described embodiment, as the co-occurrence information, a combination of the semantic classification information, the surface character information, and each pattern information is used.
You may express and use it by the combination information of semantic classification information and syntax information. Further, as the semantic classification information, the co-occurrence information described efficiently may be used by using the information having the hierarchical property.

[0069]

According to the present invention, with respect to a conversion candidate consisting of a combination of word information obtained from input reading information and an accessory word, even if each word or its semantic classification is not included in the co-occurrence information, the conversion candidate. Using the rules for determining the meaning classification of the compound word due to the presence of the compound word in the selection word, the selection priority of conversion candidates can be determined from the meaning classification of the compound word, and unnecessary homophones in kana-kanji conversion The output of can be effectively suppressed. This means that, in general, in Japanese, an extremely large number of compound words can be created by appropriately connecting multiple words, so it is realistic to prepare co-occurrence information for all compound words in advance. Although it was impossible, it is effective in solving such problems, and the number of times the user can select synonyms can be greatly reduced, and smooth Japanese input can be realized. it can.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of an embodiment of the present invention.

FIG. 2 is a block diagram showing a dictionary storage unit used in the embodiment of FIG.

FIG. 3 is a block diagram showing a compound word connection rule storage unit used in the embodiment of FIG.

FIG. 4 is a configuration diagram showing a compound word meaning classification determination rule storage unit used in the embodiment of FIG.

5 is a block diagram showing a co-occurrence information storage unit used in the embodiment of FIG.

6 is a flowchart for explaining the operation of the embodiment of FIG.

7 is a diagram showing the contents of a conversion candidate register at each time point for explaining the operation of the embodiment of FIG.

8 is a diagram showing the contents of a word pointer and a compound word register at each time point for explaining the operation of the embodiment of FIG.

9 is a diagram showing the result of the compound word processing for each conversion candidate in the embodiment of FIG.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Control part, 3 ... Phrase sequence extraction part, 4 ... Dictionary search part, 5 ... Dictionary storage part, 6 ... Synthetic word processing part, 7 ... Synthetic word connection rule storage part, 8 ... Synthetic word meaning Classification determination rule storage unit, 9 ... Output control unit, 10 ... Co-occurrence information storage unit, 11 ... Document storage unit, 12 ... Display unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Tatsuya Uehara 1 Komukai Toshiba-cho, Sachi-ku, Kawasaki-shi, Kanagawa Stock company Toshiba Research Institute

Claims (1)

[Claims] 1. An input unit for inputting reading information of a Japanese sentence, a storage unit for storing a plurality of words and a semantic classification corresponding to these words, and a reading unit based on the reading information input by the input unit. Conversion candidate extraction means for searching the storage means for word information and outputting a plurality of types of conversion candidates composed of a combination of the word information and adjunct words; and at least a first word or semantic classification and a second word or Semantic classification determination information storage means that stores, as semantic classification determination information, a rule for determining a semantic classification related to a compound word composed of words belonging to the semantic classification; and for each conversion candidate output from the conversion candidate extraction means, And a composite word meaning classification determining unit that detects the presence of the composite word based on the meaning classification determination information and determines the meaning classification of each existing composite word. Then, the kana-kanji conversion device is characterized in that the selection priority order of the conversion candidates is determined based on the semantic classification determined by the composite word semantic classification determination means.