JP2975613B2 - Kana-Kanji conversion method and device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a kana-kanji conversion method and device for converting an input Japanese sentence of a kana expression into a kanji-kana mixed sentence.

(Prior Art) In Japanese, there are many words (kanji) having the same pronunciation but different expressions, so-called homonyms. For this reason, when converting a certain reading (kana) into a kanji, from the viewpoint of how to present, as a first candidate, a word that the user wants to input among homonyms corresponding to the reading. Development has been done. Heretofore, as this attempt, there has been a method that utilizes a word-word co-occurrence relationship at a surface level or a meaning level. For example, the word "kana" has two notations (kanji), "cry" and "snarling," and "people cry" depending on the subject word.
It is necessary to use them properly, such as "dogs sing". In the conversion method using the co-occurrence relationship at the surface level,
The above relationship is stored in advance as a table of the form "person"... "Crying""dog". Then, at the time of conversion, the most appropriate one of the homonyms is selected using this table. However, in the co-occurrence relation at the surface level, when "Kana ni no genki" or "Akita ken no" is entered, the subject word is different from the previously stored co-occurrence relation table, Cannot convert correctly.

On the other hand, the co-occurrence at the semantic level is one in which at least one of the words is considered as a concept when considering the co-occurrence relationship. For example, when using semantic features,
The meaning information of “human” and “dog” can be expressed as “+ animal, + human” and “animal +, − human”, respectively. Therefore, the above table can be described as follows at a semantic level.

"+ Animals, + People"… "Crying""+ Animals, + People" ... "Crying" Furthermore, "Japanese" and "Akita Inu"
It is a subordinate concept of "people" and "dogs". Therefore, each of the semantic information is also “+ animal, + person” and “+ animal, −
Person. " For this reason, even if the user inputs "Japanese" or "Akita-ken" in kana, it is possible to correctly convert "null" according to the subject.

However, such co-occurrence relationships are extremely diverse, and the number is extremely large. Therefore, it is practically incomprehensible to check in advance the possibility of all combinations of words and prepare complete co-occurrence information. Therefore, in a conventional conversion device using a co-occurrence relationship, when an operator selects a homonym by a key operation, the co-occurrence relationship at the surface level between the selected word and, for example, a preceding word is learned. Thus, examples of co-occurrence relationships were being increased. However, even in this case, learning of the co-occurrence relationship at the semantic level has not been performed. For this reason, even if such a co-occurrence relationship is learned, it is often the case that the conversion rate is not effectively improved.

(Problems to be Solved by the Invention) As described above, in the conventional kana-kanji conversion method and apparatus,
Since the co-occurrence relationship was not learned at the semantic level, while learning the co-occurrence relationship at the surface level,
In many cases, the conversion rate was not improved.

Therefore, when the present invention converts kana into kanji and selects a homonymous word, it searches for words that are co-occurring with the selected word, and not only learns the co-occurrence relationship at the surface level, but also The purpose of the present invention is to learn the co-occurrence relation of the semantic level, mean the result thereof, improve the conversion rate thereafter, and improve the efficiency of the user's text input.

[Configuration of the invention]

(Means for Solving the Problems) A kana-kanji conversion device according to the present invention is a word information storage device that stores word information, and a co-occurrence relationship storage device that stores co-occurrence relationship information of a surface level and a meaning level between words. Based on the reading information input as a conversion target, the word information read from the word information storage means is specified by the co-occurrence relation information read from the co-occurrence relation storage means, and the input reading information is kanji. Kana-kanji conversion means for converting to kana-mixed information, next candidate selection means for selecting another homonym for words having homonyms among words of the kanji-mixed information converted by this conversion means, and next candidate selection In response to the selection of another homonym by the means, it is determined whether or not a new cooccurrence relationship is established between the selected homonym and another word of the kanji-mixed information other than the selected homonym, A co-occurrence learning control unit that newly stores the co-occurrence relationship information at the surface level and the meaning level of the combination of the words that have been determined when the occurrence relationship holds, For a combination of words that are not stored in the co-occurrence relation storage means, by newly storing the co-occurrence relation information at the surface level and the meaning level in the co-occurrence relation storage means,
It is configured as a feature of learning a new co-occurrence relationship.

The kana-kanji conversion method of the present invention is the first kana-kanji conversion method, wherein the word information storage means storing word information, and the co-occurrence relation storage means storing co-occurrence relation information of the surface level and the meaning level between words. Based on the reading information input as a conversion target, the word information read from the word information storage means is specified by the co-occurrence relation information read from the co-occurrence relation storage means, and the input reading information is determined. A kana-kanji conversion method for converting to kana-kanji mixed information, wherein, among words of the converted kana-kanji mixed information, a word having a homonymous word is selected when another homonym in the homophone candidate is selected. It is determined whether or not a new co-occurrence relationship is established between the same homophone and the other words in the kanji-mixed information other than the same. If the co-occurrence relationship is established, the combination of the words to be determined is determined. The co-occurrence relation information of the combination surface layer level and meaning level is newly stored in the co-occurrence relation storage means.

(Operation) In the present invention, it is determined whether or not a co-occurrence relationship is newly established between the selected homophone and another word other than the kanji mixed information. When the co-occurrence relation is established, the co-occurrence relation information of the surface level and the meaning level of the combination of the words to be determined is newly stored. That is, a new co-occurrence relationship is learned.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a reading information input unit, 2 is a kana-kanji conversion control unit, 3 is a text editing unit, 4 is a display unit, 5 is a co-occurrence learning control unit, 6 is a word dictionary storage unit, and 7 is a co-occurrence relationship storage. Department.

The kana column input from the reading (kana) information input unit 1 is
Sent to the kana-kanji conversion control unit 2 sequentially. The kana-kanji conversion control section 2 searches the word dictionary storage section 6 and the co-occurrence relation storage section 7 and converts the supplied kana sequence into a sentence mixed with kanji and kana.
The sentence editing unit 3 is supplied with a result of conversion by the kana-kanji conversion control unit 2, a kana column before conversion, a list of other candidates for homonymous words, and the like, temporarily stores them, and outputs and displays them. The power character string and the like are sent to the display unit 4.
The sentence editing unit 3 also receives inputs related to normal editing operations such as cursor movement, character string deletion, and selection of homonyms from the reading information input unit 1 through the kana-kanji conversion control unit 2 and is determined in advance. Performs editing-related operations. The co-occurrence learning control unit 5 checks the co-occurrence relationship between words when the sentence editing unit 3 performs a homonymous word selecting operation, and stores it in the co-occurrence relationship storage unit 7.

FIG. 2 is an example of a format of the word dictionary stored in the word dictionary storage unit 6. FIG. 3 is an example of information in a word dictionary expressed according to the format.
The “reading” portion represents a word reading (eg, “dog”), and the “heading” portion represents a “reading” portion (eg, “dog”).
(Eg, "dog"). The “grammar information” portion stores grammatical information (for example, “noun”) such as the part of speech of this word (for example, “dog”), and the number of the “heading number” portion (for example, “101”) is This is an identification number assigned to each heading. In general, a word represented by a heading generally has a plurality of meanings in many cases. For example, the word "dog" has the meaning of so-called "mammalian canine" and the meaning of "rotator" or "spy". The number in the “concept number” portion is a number assigned to the concept represented by the word of the heading. Second
In FIG. 3 and FIG. 3, the number of concept numbers is three, but the number is not limited to this, and a variable length format in which the number is not fixed may be used.

FIG. 4 is an example of a format of a co-occurrence dictionary stored in the co-occurrence relation storage unit. FIG. 5 is an example of information in the co-occurrence dictionary expressed according to the format. Numbers arranged side by side (for example, “105” and “103”) indicate that they are co-occurring with each other, and these numbers are heading numbers if the type is “0” and concept numbers if the type is “1”. It is. For example, the numbers “105” and “103” are both of type “0”,
It is a heading number, and “105” represents “person” and “103” represents “cry”, which indicates that “person” and “cry” have a co-occurrence relationship.

In the example of the second line, the type of the number “201” is represented as “1”. Therefore, the number "201"
Is a concept number, and it can be understood that the kanji character “meow” is appropriate in the case of “dog” or “cat” due to the co-occurrence relationship between “201” and “102”.

FIG. 6 is an example of a flowchart of the process of the text editing unit 3. The operation based on this flowchart is schematically described as follows. That is, if the character sent to the editing unit 3 is not the code of the next candidate key for outputting another candidate for the homonym, the processing corresponding to the code is performed. In addition, even if the input is the next candidate key, if another candidate of the homonym cannot be selected, that is, if there is no homonym originally, or if the information of the other homonym that has already been determined is lost If so, the process ends. If another candidate can be selected, the selection operation is performed while the key for instructing the display of another candidate such as the next candidate key or the previous candidate key is sent from the kana-kanji conversion control unit. For example, the input is ended by input of a key for fixing a candidate, a key for moving a cursor, or the like. After the selection operation is completed, a search for a word having a co-occurrence relationship with the selected word is performed. If no co-occurring words are found, the process ends.

More specifically, the presence / absence of the next candidate key input is determined (S
1). If the input is not the next candidate key input, other editing processing is performed (S2), and the processing ends. If there is an input, it is determined whether there is a homonym (S3). If not, end. If there is, an operation of selecting a homonym is performed (S4).
Next, the co-occurrence relationship between the selected word and another word is searched (S5). If no co-occurring words are found, the process ends. If found, the co-occurrence dictionary update word (S7) ends.

FIG. 7 is a flowchart showing details of an example of a process of searching for words having a co-occurrence relationship in S5 of FIG. The flowchart of FIG. 7 is schematically described as follows. That is, it is assumed that the word selected by the homonym selecting operation shown in FIG. 6 is set in the variable W appearing in the flowchart. The variable F is a variable for holding whether or not a word having a co-occurrence relationship is found. Here, if the value is 1 at the time when the processing of FIG. 7 is completed, the co-occurrence relationship is found. , And 0 indicates that no item was found. The variables W1 and W2 are used to hold two words having a co-occurrence relationship, and are referred to by the words 1 and 2 in the flowchart of FIG.
Here, we consider W1 as a noun (nominal) and W2 as a verb.

First, the variable F is set to 0. Subsequently, if the selected word W is a noun, it is checked whether the word immediately before W is a continuation-type word or the word immediately after W is a word (however, the word immediately after the word is a period). If so, the selected word W is set in W1, the decree is set in W2, the variable F is set to 1, and the process ends. Also, when W is a verb, contrary to the case where W is a noun, immediately after W is a noun and W is a continuous form, or immediately after W is a punctuation point, If there is, the noun is set to W1, the W itself is set to W2, and finally, the variable F is set to 1 and the processing is ended. Otherwise, the process ends immediately.

More specifically, first, a variable F is set to 0 (S1
1). It is determined whether the selected word W is a noun (S1
2). In the case of a noun, it is determined whether or not the word immediately before the word W is an adjunct word (S13). If it is an adnominal form,
The word W is set to W1, and the last word is set to W2 (S1
4) Then, 1 is set to the variable F (S15), and the processing ends.

In S13, if it is determined that the word immediately before the word W is not a continuous form, it is determined whether the word immediately after the word W is a word (S16). If not, the process ends. If it is a declinable word, it is determined whether or not it is a punctuation mark immediately after that (S17). If it is not a period, the process ends. If it is a period, the word W is set to W1, the word of the direct word is set to W2 (S18), the variable F is set to 1 (S15), and the process ends.

If it is determined in S12 that the word W is not a noun, it is determined whether or not the word W is a verb (S19). If not, the process ends. If not, it is determined whether the word W is a noun immediately after the word W (S20). If it is a noun, it is determined whether the word W is a continuous form (S21). If not, end the process. In the case of the adjoint form, the noun immediately after is set to W1, the word W is set to W2, the variable F is set to 1, and the process ends.

In S20, when it is determined that the part immediately after the word W is not a noun, it is determined whether the part immediately after the word W is a punctuation mark (S2).
3). If it is not a period, the process ends. In the case of a punctuation mark, a noun in front of the word W in the sentence is searched (S24). It is determined whether or not there is a noun (S25). If there is no noun, the process ends. If there is a noun, the noun (the noun in front)
Is set to W1, the word W is set to W2 (S26), and the variable F is set.
Is set to 1 (S15), and the processing ends.

By the processing described in the flowchart of FIG. 7, for example, when the kana input is converted into "dog crying" or "saw dog crying", "crying" is selected by homonym selection. If you change to
It can be seen that "dog" and "crown" are co-occurring.

In this way, when two words having a co-occurrence relationship, that is, word 1 (for example, dog) and word 2 (for example, sounding) are found, the process of updating the co-occurrence dictionary in S7 of FIG. 6 is performed. The details are shown in the flowchart of FIG. FIG. 8 is a flowchart of a process for registering a co-occurrence relation in the co-occurrence relation storage unit. In this flowchart,
The co-occurrence relation at the surface level and the semantic level is registered in the co-occurrence relation storage unit 7, and the processing is terminated. As a result, four types of co-occurrence relation data, namely, heading number of word 1 × heading number of word 2 heading number of word 1 × concept number of word 2 concept number of word 1 × heading number of word 2 The data of concept number × concept number of word 2 is registered. However, there is a possibility that some of the data ~ may not necessarily have a true co-occurrence relationship, but assuming that all the data ~ have a co-occurrence relationship,
All of these data are to be registered. Assuming that the number of concepts of words 1 and 2 is N1 and N2, respectively, the total number of the data is (N1 + 1) × (N2 + 1). For example, when the word 1 is "dog" and the word 2 is "ringing", as can be seen from FIG. 3, since N1 = 1 and N2 = 2,
(N1 + 1) × (N2 + 1) = 6, and six data, that is, data of 101 × 102 101 × 203 201 × 102, 202 × 102 201 × 203, 202 × 203 are obtained.

This will be described in detail with reference to the flowchart of FIG.

Consider two sets of memory arrangements for temporarily storing the index numbers and concept numbers of words 1 and 2 read from the word dictionary storage unit 6, and these are referred to as A1 and A2. First, A1 (0), A2
Register the heading numbers of word 1 and word 2 in (0), change the concept number of word 1 to A1 (1) to A1 (N1), and change the concept number of word 2 to
The data is sequentially stored in A2 (1) to A2 (N2) (S30). Variables I and J appearing in the flowchart are arrays A1 and A2, respectively.
Subscripts to indicate the data inside, variables II and JJ are set to 0 when the values of I and J are each 0, and set to 1 otherwise, and used as values of the type in the format of Fig. 4. Shall be performed. It is assumed that the variable X is a structured type variable having the format shown in FIG. 4 as a data structure as it is. The variables I and II are set to 0 (S31), and subsequently the variables J and JJ are also set to 0 (S32). Next, information to be registered is set in the variable X described above (S33). A1
Since index numbers are registered in (0) and A2 (0), the first set to X is the co-occurrence relationship at the surface level. It is checked whether the data set in X is stored in the co-occurrence relation storage unit (S34). If not registered, it is registered there and the process moves to S36. If it has been registered, the process moves directly to S36. In S36, 1 is added to J, and the value of JJ is set to 1 (S36). Here, the value of the added word of J is checked (S37). If the concept number of the word 2 is equal to or less than the number N2, the process proceeds to S33, where a new value is set to the variable X, and the unregistered co-occurrence relations are sequentially registered. In this way, if it is determined in S37 that the value of J is larger than N2, then 1 is added to I, and 1 is set in II (S38). Repeat until it becomes larger (S39).

By the above operation, as described above, the headword number ×
A maximum of (N1 + 1) × (N2 + 1) data of four types of headword words, headword number × concept number, concept number × headword number, and concept number × concept number are registered.

In the processing according to the flowchart of FIG. 8, the data to are registered as having a co-occurrence relationship. However, as described above, there is a possibility that some of these data are not necessarily in a true co-occurrence relationship. Therefore, data ~
May not be registered as having a co-occurrence relationship, but only those having a true co-occurrence relationship may be registered. That is, when the selected words 1 and 2 have a plurality of concepts (concept numbers), one concept number used in the context is identified, and a more precise co-occurrence between the words 1 and 2 is determined. Relationship data can be obtained and registered. FIG. 9 to FIG. 11 are flowcharts showing such processing.

FIG. 9 is a flowchart of a process for registering only the true co-occurrence signal in the co-occurrence relation storage unit 7. First, the variables M1 and M2
, The heading numbers of words 1 and 2 are stored, and the number of concept numbers of words 1 and 2 is stored in variables N1 and N2 (S41). It is determined whether the value of N1 is 1 (S42). When N1 is 1, the unique concept number is stored in the variable G1 (S43). If N1 is not 1, the operator is caused to select the concept number of the concept of word 1 used in the sentence, and the number is stored in the variable G1 (S4).
Four). The same is true for word 2. That is, it is determined whether the value of N2 is 1 (S45). When N2 is 1, the unique concept number is stored in the variable G2 (S46). If N2 is not 1, the operator is caused to select the concept number of the concept of word 2 used in the sentence, and the number is stored in variable G2 (S4
7). Thus, for each of words 1 and 2,
The heading number and the concept number are determined one by one. That is, for word 1, the heading number is stored in M1, and the concept number is stored in G1. For word 2, the heading number is stored in M2 and the concept number is stored in G2. Based on these, in S48 to S51, four types of co-occurrence relation data
Is registered only if it has not been registered. That is, what is registered in S48 is the co-occurrence relation () at the surface level,
And 94 are co-occurrence relations (,) based on the combination of the surface layer and the semantic level, and in step 95, they are co-occurrence relations of the semantic level ().

FIG. 10A is a flowchart showing the details of S44 in FIG. First, a screen used by the operator to select a concept number is set on the display unit 4 (S61). Variable I is 1
(S62). The I-th concept number of word 1 is set to G (S63). A word having G as the concept number is searched for and set to W (S64). It is determined whether the word set in W is the same as word 1 (S65). If the same, S64
And the above search operation is repeated. If not,
The number I and the character string obtained by replacing the word 1 with W are displayed (S66). By the above operation, for example, "the dog rings"
However, when it is selected as a co-occurrence relationship, when calculating the concept number of the concept of "dog" used there, as shown in Fig. 11, "1. Is displayed. Next, 1 is added to the variable I (S67). Next, it is determined whether I is smaller than N1 (S68). If smaller, the process returns to S64 and the above operation is repeated. Thereby, for example, as can be seen from FIG. 11, "2. The spy rings" is displayed. That is, for each of the N1 concept numbers of word 1, words W other than word 1 having the same concept number are retrieved from dictionary storage unit 6 and sequentially displayed. In S68, when I becomes greater than N1, that is, for example, when the words of the same concept as all the concepts of "dog" have been displayed, the operator, while watching the screen of FIG. , One of the numbers (N) of 1 and 2 is input (S69). It is determined whether the input number (N) is appropriate (S70). If it is appropriate, the Nth concept number of word 1 is set in G1 (S71), and the screen is restored (S72). For example, in the example of FIG.
Is input, the concept number of the concept of “dog” in the case of “dog rings” is determined to be “201”.

The selection of the concept number of word 2 in S47 is performed in substantially the same manner as in the case of word 1 described above. The details are shown in FIG. 10B.

With the above operation, when words 1 and 2 each have a plurality of concept numbers, by allowing the user to select the concept number used in the sentence, more accurate co-occurrence learning can be performed. , The conversion rate afterwards is greatly improved.

In this embodiment, when the user (operator) selects a concept after each word, the word having the same signal as the concept number is searched from the dictionary and displayed. However, the word information shown in FIG. In the above, it is also possible to add a description in a natural language, such as that described in an ordinary Japanese dictionary, to each concept number, and to display the description to select the concept number.

In order to obtain co-occurrence relation data in which more appropriate conversion is performed, the format of the co-occurrence dictionary shown in FIG. 4 can be extended to include a term of frequency information. For this purpose, the operations in S34 and S35 in FIG. 8 may be changed as follows. That is, first, when the co-occurrence relation information set in the variable X has already been registered, the value of the frequency information is incremented by 1 within the range of the maximum value of the frequency information. The initial value is registered as 1. Kana-Kanji conversion control unit 2
In, if only data whose frequency information is equal to or more than a predetermined value is used at the time of conversion, conversion with higher accuracy can be performed. If a co-occurrence relation that has not been used for conversion for a certain period of time or whose value has not been updated is deleted from the co-occurrence relation storage unit 7 as in normal word learning, the area of the co-occurrence dictionary can be deleted. , And faster conversion is possible.

〔The invention's effect〕

As described above, according to the present invention, when a homonymous word is selected, a word having a co-occurrence relation with the word is obtained, and in addition to the co-occurrence relation at the surface level, the co-occurrence at the semantic level is determined. Since the relationship is also learned, it is possible to improve the subsequent appropriate kana-kanji conversion rate and improve the efficiency of the operator's text input.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, FIG. 2 is an explanatory diagram showing an example of a format of a word dictionary stored in a word dictionary storage unit, and FIG. 3 is a word dictionary expressed in accordance with the above-mentioned format. Chart showing an example of the information in the fourth
FIG. 5 is an explanatory diagram showing an example of a format of a co-occurrence dictionary stored in the co-occurrence relation storage unit. FIG. 5 is a table showing an example of information in the co-occurrence dictionary expressed according to the above format. 8 is a flowchart showing an example of the processing of the text editing unit, FIGS. 9 and 10 are flowcharts showing an example of the co-occurrence dictionary update processing, and FIG. 11 is a screen displayed to the user in the co-occurrence dictionary update. FIG. 4 is an explanatory diagram showing one configuration example of FIG. 1 ... Reading information input unit, 2 ... Kana-Kanji conversion control unit, 3
... A sentence editing unit, 4 a display unit, 5 a co-occurrence learning control unit, 6 a word dictionary storage unit.

Continuation of front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 17/21

Claims (2)

(57) [Claims] 1. Word information storage means for storing word information, co-occurrence relation storage means for storing co-occurrence relation information of surface level and meaning level between words, and reading information inputted as a conversion target Kana-Kanji conversion means for specifying word information read from the word information storage means by co-occurrence relation information read from the co-occurrence relation storage means, and converting the input reading information into Kanji mixed information. Of the words of kanji mixed information converted by
A next candidate selecting means for selecting another homonym for a word having a homonym, and excluding the selected homonym in response to selection of another homonym by the next candidate selecting means It is determined whether or not a new co-occurrence relationship is established with other words that contain kanji information. A co-occurrence learning control means for newly storing the occurrence relation information in the co-occurrence relation storage means. A kana-kanji conversion apparatus characterized in that a new co-occurrence relation is learned by storing the co-occurrence relation information in the co-occurrence relation storage means. 2. Word information storage means for storing word information,
Using co-occurrence relation storage means storing the co-occurrence relation information of the surface level and the meaning level between words, based on the read information input as a conversion target, the word information read from the word information storage means, A kana-kanji conversion method for specifying input reading information to kanji-mixed information by specifying the co-occurrence relation information read from the co-occurrence relation storage means, wherein, among the words of the converted kanji-mixed information, If another homonym is selected from homonym candidates for a certain word, whether or not a new co-occurrence relationship is established between the selected homonym and another word other than the kanji mixture information Is determined, and when the co-occurrence relation is established, the co-occurrence relation information of the surface level and the semantic level of the combination of the words to be determined is newly stored in the co-occurrence relation storage means. Characteristic kana-kanji conversion method.