JPH0736878A - Homonym selecting device - Google Patents

Abstract

PURPOSE:To provide a homonym selecting device capable of improving the conversion efficiency of 'KANA' (Japanese syllabary)/'KANJI' (Chinese character) conversion. CONSTITUTION:This selecting device is characterized by including a 'KANJI' conversion means (1 to 4) for converting an inputted 'HIRAGANA' (cursive form of Japanese syllabary) string into a 'KANA'/'KANJI' string including 'KANJI' conversion candidates homonymous with the 'HIRAGANA' string to be converted into 'HIRAGANA' and 'KANJI', a storage means 5 for storing 'KANJI' whose 'KANJI' conversion is determined out of the 'KANA'/'KANJI' string converted by the means (1 to 4) and a selecting means (6, 7) for selecting the conversion order of the 'KANJI' conversion candidates corresponding to the 'HIRAGANA' string to be converted so that the meaning of the 'KANJI' stored in the means 5 is matched with that of a context.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a homophone / synonym selection device, and more particularly to a homophone / synonym selection device capable of improving conversion efficiency of kana-kanji conversion.

[0002]

2. Description of the Related Art Conventionally, when selecting a homonym, a frequently used word described in a dictionary in advance is used. If an independent word A is produced using a co-occurrence combination dictionary created manually, B Kanji conversion was performed using a one-to-one dictionary such as selecting a candidate having a high description frequency.

[0003]

Since the conventional method of selecting a candidate with a high frequency does not consider the context of the context, “road construction” and “public announcement of land price” are applied to the same reading “Koji”. It is not possible to deal with cases where the conversion result must be changed in the context such as ". When the one-to-one correspondence is stored, “road”-“construction”,
It suffices to memorize the combination of "land price"-"public notice", but this method only finds the relation to one independent word, and selects conversion candidates considering multiple independent words. I can't.

An object of the present invention is to provide a homonym synonym selection device capable of improving the conversion efficiency of kana-kanji conversion.

[0005]

[Means for Solving the Problems] Kana-Kanji conversion means for converting an input hiragana string into a kana-kanji string containing hiragana and kanji conversion candidates that have the same phonetic differences as the kana-kanji conversion means and kana-kanji conversion means A kana-kanji conversion candidate is stored in the kana-kanji string, and the conversion order of the kanji conversion candidates corresponding to the hiragana to be converted so that the kanji stored in the memory and the meaning of the context match Selection means for selecting.

[0006]

[Function] A kana-kanji character converted by the kana-kanji conversion means into a kana-kanji string containing hiragana and a kanji conversion candidate having the same phonetic meaning as the kana-kanji conversion means to be converted into hiragana and kanji. Since the kanji whose conversion to kanji has been confirmed is stored in the sequence, the selecting means selects the conversion order of the kanji conversion candidates corresponding to the hiragana to be converted so that the meaning matches the kanji stored in the storage means and the context. , Kanji conversion candidates that are semantically similar to already entered words (sentences) can be preferentially selected to improve the Kanji conversion efficiency.

[0007]

1 is a block diagram of an embodiment of a homonym synonym selection device of the present invention. In FIG. 1, reference numeral 1 denotes an input buffer for storing a previously input HIRANA string, 2 denotes Kana-Kanji for the HIRANA string stored in the input buffer 1,
That is, the kana-kanji conversion unit 3 which converts a kana-kanji string including a kana-kanji conversion candidate having the same phonetic meaning as that of hiragana and kanji and outputs the converted kana-kanji string is used for the conversion of the kanji conversion part 2. Suru-Kana-Kanji conversion dictionary, 4 is a conversion candidate buffer for storing Kanji conversion candidates converted by the Kana-Kanji conversion unit 2, 5 is a definite sentence buffer for storing a Kana-Kanji string whose conversion result is definite, and 6 is a definite sentence buffer 5. A candidate rearranging unit that obtains the meaning vector of the sentence input up to that time from the independent words of the Kana-Kana character string and changes the order of the conversion candidates stored in the conversion candidate buffer 4 using the result, 7 will be described later. It is a co-occurrence table in which a meaning vector is stored.

The input buffer 1, the kana-kanji conversion unit 2, the kana-kanji conversion dictionary 3 and the conversion candidate buffer 4 are kana-kanji conversion means, the fixed sentence buffer 5 is a storage means, and the candidate rearrangement unit 6 and the co-occurrence table 7 are provided. And constitute selection means.

Here, the semantic information of the words or sentences input so far is expressed as a vector (semantic vector), and the conversion result is determined (or the order of the conversion results is changed according to the semantic information). ), The conversion result that is semantically similar to the word (sentence) already input, that is, the meaning of matching the context is prioritized.

FIG. 2 is a co-occurrence table showing the order in which the meanings of the Kanji conversion candidates match, that is, the ease of connection between each word. FIGS. It is a figure explaining the example in the case of performing.

In FIG. 2, the "teacher" in the second row and the "lecture" in the fourth column are 7, and the connection is deeper than that of "teacher" in the second row and "protest" in the fifth row. Is shown.
As an example, "He did the decision."
When a sentence "is input", there are two types of conversion results of the character string "Kougi", "protest" and "lecture". In this case, since independent words such as “he” and “decision” whose kanji conversion has been confirmed have already been input, the order of kanji conversion candidates is determined from these words. First, the semantic vector of "he" in Fig. 2 (the element in the row direction of "he" in the co-occurrence table) is "8, 7, 2, 4, 3, 2," and similarly for "decision". The meaning vector (row-wise element of "decision" in co-occurrence table) is "2, 5, 7, 3, 7, 1". When the current meaning vector is calculated by averaging these two vectors, it becomes (5, 6, 4, 3, 5, 1) ((8 + 2) when the average value of the corresponding numerical values is calculated.
÷ 2 = 5, (7 + 5) ÷ 2 = 6, and so on, except for the decimal point. (See FIG. 3). Among the elements of this vector, looking at the places corresponding to “lecture” and “protest”, “lecture” is 3 and “protest” is 5 (see FIG. 3). In other words, if the meaning vector is calculated from the input up to that point, “protest” is more “lecture” than the current input “Kogi”.
It means that the connection is deeper, and "protest" is selected and output as the first place as the conversion result.

As another example, in the case of "He received the teacher's kogi", there are two types of conversion results of the character string "kogi", "protest" and "lecture". In this case, as in the above example, the meaning vector of "he" (the element in the row direction of "he" of the co-occurrence table) in FIG.
Similarly, the meaning vector of "teacher" (the element in the row direction of "teacher" in the co-occurrence table) is "7,
9, 4, 7, 5, 1 ". When the current meaning vector is calculated by averaging these two vectors,
(7,8,3,5,4,1) ((8 + 7) / 2 = 7, (7+
9) ÷ 2 = 8. Similarly, refer to FIG. 4). Among the elements of this vector, looking at the places corresponding to "lecture" and "protest", "lecture" is 5 and "protest" is 4 (see Fig. 4). In other words, if the meaning vector is calculated from the input up to that point, it means that "lecture" is more connected than "protest" to the current input "Kogi", and outputs "lecture" as the conversion result. .

The co-occurrence table can be created by learning from a large amount of data. When automatically created from a large amount of data, it is possible to create a high-quality co-occurrence table without variations caused by the creator, and as a result, it is expected that the kana-kanji conversion efficiency is improved.

The current meaning vector (5, 6, 4,
When replacing Kanji conversion candidates from 3, 5, 1), in the description of the first example, the size of the element was compared (Protest: 5>
“Lecture: 3) may use the distance between the current meaning vector and the meaning vector of the candidate word. In the case of“ protest ”, the meaning vector of the word (protest of the co-occurrence table in FIG. 2). Element in the row direction) is (2,4,6,2,8,1), and the Euclidean distance D to the current semantic vector is D (protest) = (5-2) 2+ (6 -4) 2 + (4-6)
2 + (3-2) 2 + (5-8) 2 + (1-1) 2 = 27 Similarly, in the case of "lecture", D (lecture) = (5-3) 2 + (6-5 ) 2 + (4-3)
2+ (3-9) 2+ (5-5) 2+ (1-3) 2 = 46. In this case, since it is appropriate to select the Kanji conversion candidate from the one having the shortest distance, the “protest” having the shortest distance is selected as the first place.

[0015]

The kana-kanji conversion means converts the input hiragana string into a kana-kanji string containing hiragana and kanji conversion candidates having the same phonetic meaning as kana and kanji, and the storage means is converted by the kana-kanji conversion means. The kanji whose kanji conversion has been confirmed is stored in the Takana kanji string, and the selection means selects the conversion order of the kanji conversion candidates corresponding to the hiragana to be converted so that the meaning matches the kanji stored in the storage means and the context. Therefore, the kanji conversion candidates that are semantically similar to the already input word (sentence) are preferentially selected, and the kanji conversion efficiency can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a homonym synonym selection device of the present invention.

FIG. 2 is a diagram showing a co-occurrence table.

FIG. 3 is a diagram illustrating an example in which kanji conversion is performed based on the connection between words.

FIG. 4 is a diagram illustrating another example of a case where Kanji conversion is performed from the connection between words.

[Explanation of symbols]

 1 Input buffer 2 Kana-Kanji conversion unit 3 Kana-Kanji conversion dictionary 4 Conversion candidate buffer 5 Fixed sentence buffer 6 Candidate rearrangement unit 7 Co-occurrence table

Claims (2)

[Claims] 1. A kana-kanji conversion means for converting an input hiragana string into a kana-kanji string that includes kana-kanji conversion candidates having the same phonetic meaning as hiragana and kanji, and kana-kanji converted by the kana-kanji conversion means. A storage means for storing a kanji character for which a kanji conversion has been confirmed among the columns and a conversion order of kanji conversion candidates corresponding to the hiragana to be converted so that the kanji characters stored in the storage means have the same meaning in terms of context. A homonym synonym selection device comprising: selection means. 2. The homonym synonym selection device according to claim 1, wherein the selection means includes a co-occurrence table that defines an order in which the meanings of the Kanji conversion candidates match.