JPS63213061A - System for classifying declensional kana ending - Google Patents

Abstract

PURPOSE:To eliminate the need of registering all decleneional KANA (Japanese syllabary) endings into a dictionary by classifying and registering the relations between the index in the dictionary and the decleneional KANA endings of those words to be retrieved when the KANJI (Chinese characters)-KANA words are retrieved. CONSTITUTION:When no coincident index is not detected out of a dictionary when the KANJI-KANA words are detected, the relation between the index read out of the dictionary and including an extracted KANJI and the declension al ending of a word to be retrieved is classified. In this case, a simple (A) type where the KANA following the KANJI can be omitted is defined together with a KANJI word ending feed (B) type where where the extra declensional KANA endings can be secured, a continuous word ending addition (R) type where the KANJI has a continuous word ending and can be used as a noun, and declensional KANA ending coincidence enable (O<+>) type where the index in the dictionary contains HIRAGANA (cursive form of Japanese syllabary) equivalent one or more characters in the form of a word having the omission of a declensional KANA ending. Thus it is not required to register all declension al KANA endings into a dictionary.

Description

[Detailed Description of the Invention] [Summary] This invention, when searching to see if a certain kanji-kana mixed word is registered as a word in a dictionary, uses the heading found in the dictionary and the name of the word to be searched for. In order to solve the problem of not being able to search for a word when the words are different from each other, the word is classified into types that can match okurigana, simple types, and kanji endings, based on the relationship between the heading in the dictionary and the okurikana in the word. By doing so, it is possible to search for a desired word even if no heading with the same okurikana is found in the dictionary.

[Industrial application field]

The present invention classifies these types based on the relationship between the heading in the dictionary and the okurikana in the word to be searched, so that the desired word can be searched even if there is no matching word in the dictionary. This paper relates to the constructed okurikana classification system.

[Problems to be solved by the prior art and the invention] In the field of natural language processing, particularly in machine translation, natural language interfaces (for example, database searches using natural language, etc.), it is necessary to make computers understand sentences. In the first stage, a dictionary is used to break down sentences into clauses and at the same time extract the meanings of words.

Conventionally, when searching a dictionary using a computer, there has been a problem that if the okurikana of a certain word is different from that of the heading in the dictionary, the word cannot be retrieved. In order to be able to force a word to be looked up, it is necessary to register in advance all possible okurikana for the word to be looked up as a heading in the dictionary, which increases the capacity of the dictionary and makes it impractical. In addition, there was a problem that not all okurigana could be registered in the dictionary. For this reason, instead of pre-registering all possible okurikana of a word in a dictionary, we classify the relationship between the heading in the dictionary and the okurikana of a word, even if there is no match in the dictionary. It is desired to be able to search for a desired word even if the user is searching for a desired word.

[Means for solving problems]

In order to solve the above-mentioned problems, the present invention classifies the words into at least okuri-kana matching type, simple type, kanji ending-adjusting type, etc. based on the relationship between the heading in the dictionary and the okurikana in the word. Even if the same okurikana is not found in the dictionary, it is possible to search for a desired word.

FIG. 1 shows a basic configuration diagram of the present invention. A search unit 1 in the figure searches a dictionary 6 for a heading that matches a character string that is input data.

The start position detection unit 2 detects, from an input character string, a start position of a character string to which it is estimated that the classification method according to the present invention can be applied.

The kanji extractor 3 extracts only kanji from the character string detected by the start position detector 2.

The collation unit 4 finds out from the dictionary 6 all headings that contain the kanji extracted by the kanji extraction unit 3.

The classification processing unit 5 classifies the okurigana of the entry word based on the relationship between the character string detected by the start position detection unit 2 and the entry in the dictionary 6 found by the matching unit 4. .

[Effect]

Next, the operation will be explained.

In FIG. 1, a search unit 1 searches a dictionary 6 for a matching heading for character string input data, and if a matching heading is found, the heading data is sequentially output from the beginning as output data. On the other hand, if no match is found, the start position detection unit 2 finds the start position of the character string that is presumed to be unsearchable due to the difference in okurigana. Starting from this found position, the kanji extracting unit 3 extracts only kanji included in a character string of a length estimated to constitute a word. The collation unit 4 collates the heading containing the extracted kanji with those registered in advance in the dictionary 6 and reads it out. The classification processing unit 5 performs classification based on the relationship between the heading containing the above-mentioned kanji read from the dictionary 6 and the okurikana in the word.

As described above, for words that include character strings that do not match the headings in the dictionary 6, the headings that include the extracted kanji parts are read out from the dictionary 6, and classification is performed based on the relationship between the okurikana characters. This makes it possible to search for words even if they are not registered in the dictionary 6.

〔Example〕

Next, the configuration and operation of one embodiment of the present invention will be explained in detail using FIGS. 2 to 4.

FIG. 2 shows the operation in the classification processing section 5. ■ in the figure is
This shows a state in which it is determined whether or not the type is a type that can match forwarded kana (hereinafter referred to as 0゛ type). This 0゛ type is registered in the dictionary as the heading "□●□●...□●○".
Do you use too many parts of your mouth in your sentences, or do you use ・O?
Represents something that exists as a word in which part of is taken into the previous mouth. For example, in the dictionary 6 shown in Figure 3 (2〉(al), the entry ``necessarily'' consists of ``mouth O'', the word ``necessarily'' consists of ``mouth 00','' and others"
One kana has been sent too many times. This “ra” is “
・Since there is no question as to whether it exists or not, 'must be' means 0.
It corresponds to type 1. If YES, it is classified as type 09. In the case of NO, execute the process (■) in the figure.

■ in the figure indicates a state in which it is determined whether the ending of the heading is a kana. This means to determine whether or not it is a simple type (hereinafter referred to as type A). It is in the form of "mouth...□●○", meaning that it is determined whether or not the word appears as a word with the "・○" part omitted in the sentence. For example, the heading "author" in the dictionary 6 shown using (b) in FIG. Since "shi" is omitted, this corresponds to type A. If YES, it is classified as type A. In the case of No, execute ■ in the figure.

In the figure, ■ indicates a state in which it is determined whether or not the kanji is a word-ending type (hereinafter referred to as type B). This means determining whether the heading in the dictionary is in the form of “★□” and in the sentence, the final sound of “mouth” appears as an extra word. are doing. For example, as shown in FIG.
is composed of "mouth", and the word "angoru" is a word with an extra sound "o" in the entry in Dictionary 6, and the conjugated ending "ru" (in the example, the conjugated ending is treated as a separate word). ) is added, so this corresponds to type B. YES
In this case, it is classified as type B.

In the case of No, execute ■ in the figure.

In the figure, ■ indicates a state in which it is determined whether or not the adjunctive form is the suffix addition type (hereinafter referred to as the R type). This is because the dictionary only registers verbs with the heading "★□", and if there is a conjunctive ending, it is morphologically tense.

It is treated as a noun in subsequent analyses. For example, the third
Figure (c) As shown in the figure (dl), the entry "ma" in the dictionary 6 consists of "mouth", which is registered as a verb, and the noun "ma" (for example, " If it appears as a type (e.g. "between 1"), it is classified as the relevant R type.YE
In the case of S, it is classified as R type. If No, it is classified as simple type without kana (hereinafter referred to as 〇-type). This 〇−
The pattern is registered in the dictionary as "□●□●...mouth.", and the "•" part in the word can be increased or decreased arbitrarily. For example, as shown in Figure 4 (E) and (F), the dictionary entry "Kansora" consists of "roro", and the word "Kamusora" has an increased "mu" part. Yes, it is classified as type ○-.

In addition, as shown in FIG. 4(f), there is an R-0-complex type. This is a combination of R type and O-type.

As mentioned above, it is now possible to classify the word as described above based on the relationship between the heading in the dictionary and the okurikana between the word, and even if there is no heading registered in advance in the dictionary, it is possible to search for that word. It can be performed.

In FIGS. 3 and 4, the words shown in the left column indicate words that cannot be analyzed using the conventional method, and the center column shows words that can be correctly decomposed and are registered in advance in the dictionary 6. Show what is being done.

In the figure, "1" represents a break registered as another word in the dictionary 6. Incidentally, the right column shows examples of failure patterns in which the conventional dictionary 6 was used. ■ in the diagram represents an incorrect word,
◎ indicates that the correct word was drawn by chance, and ■ indicates that it was processed as an unregistered word.

〔Effect of the invention〕

As explained above, according to the present invention, a configuration is adopted in which classification is performed based on the relationship between the heading in the dictionary and the okurikana in the word, such as the okuri-kana matching type, the simple type, and the kanji-ending type. Therefore, even if a heading containing the same okurikana is not found in the dictionary, classification can be performed based on the headings and words in the dictionary, and a desired word can be searched.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the principle configuration of the present invention, FIG. 2 is a flow chart explaining the operation of the present invention, and FIGS. 3 and 4 are diagrams explaining the classification of the present invention. In the figure, 1 is a search section, 2 is a start position detection section, 3 is a kanji extraction section, 4 is a collation section, 5 is a classification processing section, and 6 is a dictionary.

Claims (1)

[Claims] In the okurikana classification method for classifying the okurikana of words included in input kana-mixed sentences, the headings in the dictionary are □●□●...□●○ (□ is a single character) Kanji, ● represents 0 or more flat names, ○ represents one character flat name, the same applies hereafter)
It is registered in the form of , and as words in the sentence, there is a type (1) in which part of □ is sent too much, or a part of ●○ is incorporated into the immediately preceding □, and a dictionary. The headings inside are registered in the form □...□●○, and the words in the sentence are in the simple form (with the ●○ part omitted).
2), the heading in the dictionary is registered in the form of ★□ (★ is any character of 0 or more), and the word in the sentence is a kanji ending with the last sound of □ added. An okurikana classification system characterized in that it is configured to classify at least into okurikana type (3).