JPS6118066A - Word extracting system - Google Patents

Abstract

PURPOSE:To minimize the number of character strings to be retrieved in a dictionary retrieval mode and to increase the word extracting speed, by providing input character strings, an attribute deciding part for head KANJI (Chinese characters) sound, character strings to be retrieved, etc. CONSTITUTION:The KANA (Japanese syllabary) character strings corresponding to a Japanese word sentence supplied from a KANA character input part 1 of a word extracting system are stored temporarily to an input character string memory 2. A dividing part 3 for KANJI refers to a KANJI sound table 4 to divide KANJI and KANA in each minimum unit. While an attribute deciding part 5 for input character strings decides the attribute based on the arrangement of KANJI sounds and KANA characters. An attribute deciding part 6 for head KANJI sound decides the attribute of the KANJI sound based on whether an independent word having the same reading as the head KANJI sound exists or not. A deciding part 7 for character string to be retrieved sets the minimum number of character strings to be retrieved based on the information on decisions of both parts 5 and 6. The character strings to be retrieved are added to a dictionary retrieving part 8, and a word dictionary 9 is retrieved to store words to a candidate word memory part 10. This increases the word extracting speed.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a kana-kanji conversion processing device, and more particularly to a tK4FJ extraction method in a kana-kanji conversion processing device that can be applied to a Japanese document creation device, a computer system, etc. .

The conventional kana-kanji conversion processing device uses phonetic characters C hiragana and kakkana. A word dictionary for kana-kanji conversion is provided in order to convert sentences entered in Roman letters) into appropriate sentences containing kanji and kana. This word dictionary search is performed by cutting out the Qt word from the input kana character string, creating the searched character string, and matching the searched character string with the index character string in the 111-word dictionary. However, Japanese grammar is wrong.
Since K1 has many homophones, a dictionary search extracts a plurality of 711 candidate words.

In order to select one of the plurality of candidate song words and use it as a conversion result, the following processing has conventionally been performed. That is, for each of the extracted candidate words, the possibility of connection with the previous converted word (conversion result) is determined, and 4 of the possible connectable words are read and appear! tri If
, the depth of connection, etc. are evaluated as right/parameters, and the candidate fit word with the highest evaluation is output as a conversion result.

By the way, flt"l!1., +1' In order to facilitate word dictionary searches and reduce erroneous analysis, the input character string is pre-processed using kanji preprocessing.

Before the kanji, there is a 1-character kanji sound whose length is one mountain character when written in kana, ('A + 2-character kanji sound whose length is 2 characters,
Samurai) There are three kanji sounds that are three letters. For example, (■)1
The Kanji sound is the most Kanania “A”, I “I”, meaning 9
place, medical, different...'', etc. (■Two-character kanji sound...ai ``love, hello, sadness...'', aku ``
Evil, grip...'', etc.■3 Kanji sounds...Shu ``Collect, collect 1 week, Shu, #.

Xi, Shu's 1st lap, Shu...'', Show “Sang, small, provincial 9 wins, small, commercial, proof.

Erase, correct...” etc. It is.

By the way, in the above two-character kanji sounds and three-character kanji sounds, the kana that occupy the second and third characters are limited to the following 18 types]I
,ru.

“I, tsu, ki, wa, chi, tsu, ya, yu, yo, yu.

Metsu, yaku, koku, yoku, yutsu, yun, tsu, n.'' However, not all of the 18 types of kana form the kanji preposition for the first kana. For example, if the first character is A, Ai... in front of a kanji (see 1: Kiseki) Aku... not in front of a kanji Aki... 4゛ in front of a kanji Aku - in front of a kanji (see -1 - Kiseki) ) Achi... It becomes like kanji before.

Separate the input string by - before 222 or more kanji characters,
By creating a search string with that as the east position,
This eliminates the possibility of misinterpreting something that is originally part of the reading of a kanji as a case particle. Furthermore, the above Kanji-mae is never used in @1doku, but is always used in conjunction with other Kanji-mae. Therefore, as a result of matching the input string of 3 or more characters with the characters before these kanji, if the first character is matched but the second and third characters are not, the first character is matched before the kanji. It can be inferred that there is a high possibility that it is a kana, such as an attached word, rather than a kana. For this reason, there is no need to store all kanji fronts including the sound of a single kanji in the kanji front for kanji front search, as shown in Figure 4 (
It is sufficient to store two or more kanji characters as shown in aL (b'') and (c).

Conventionally, the following preprocessing has been performed using kanji zenfu as shown in FIGS. 4(a), (b), and (c).

For example, the kanji maefu is accessed based on the input kana character string ``Bunjo no Sakusei is very good.'', and the following separation is performed using the kanji mae and kana as the minimum units.

[jin/sho/no/saku/sei/ga/hi/jiyou/ni/you/i/de/a/ru/. ” However, katakana indicates before kanji, and hiragana indicates kana.

After the above delimiting process, attributes are added to the human character string as follows, depending on how the kanji and kana characters are arranged.

(before kanji) + (before kanji) 10~ ・・・・・・TYPE
] (before kanji) + (kana) 10~ ・・・・・・TYPE
2C kana) + (before kanji) 10~ ・・・・・・TYPE
3 (kana) 10 (kana) 10 ~ ・・・・・・TYPE
4 “Bun + Sho 10~” is (before kanji) + (before kanji) 10~
Therefore, the above example sentence becomes TYPEI.

Next, a searched character string is created in the following manner according to the attributes TYPEI to TYPE4 of the input character string.

For TYPEI... (D (before kanji) + (before kanji) ■ (before kanji) For TYPE 2... σ) (before kanji) + (kana) ■ (before kanji) TYPE 3 In the case of...■ (kana) + (before kanji)
■ (kana) For TYPE 4...■ (kana) ■ (kana) 10 (kana) ■ (kana) + (kana) + (kana) In the example sentence above, it is TYPEI, so it is written as follows. Set the search string to .

[Character string to be searched... (1) Let's try it ■
Next, the word dictionary is searched according to the set character string to be searched, the obtained candidate word group is evaluated, and the most suitable m candidate words are selected.

Here, it is assumed that "sentence" is extracted as the optimal candidate word. In this case, the next character string to be analyzed has a very strong r. Since it is j, we apply the kuari again to this with the minimum qt level of the kanji sounds.

[No/saku/sei/ga/hi/jimiu/ni/you/i/de/a/ru. ” Since “no+saku 10~” is (kana) 10 (kanji sound) 10~, the attribute of the input character string is TYPE3. TYP
When a searched character string is created according to the attribute of E3, the searched character string becomes as follows.

In the same way, after the C2 of the set search character string Cr), the remaining input strings are separated using Kanji sounds and kana as the minimum unit, and the arrangement of Kanji sounds and kana is Creates an appropriate search string according to the attributes of the input string.

The reason for creating a search string using the minimum 11th position of two characters is to take into account the fact that most of the J1 words registered in the m-word dictionary have a minimum of 2 or less characters (corresponding to the 11th position). This is to speed up dictionary searches.

However, the above method has the following drawbacks.

This will be explained using the input character string "Kakko de ~" as an example.

In this case, pre-processing using kanji sounds allows [Gara/Kora/
Delimitations are added such as "de/ha~". Since "Gac+Kouju~" is C kanji sound) + f kanji sound) → -~, it belongs to T''i'PE1, and (1) Gatsukou■Katsu is set as the searched character string.

By the way, the ``J'' in the above [■ GA] width is a consonant, but this is originally a phonetic change of the kanji sound ``kaku (gaku)'', and the kanji sound ``ga''. may exist, but there is no independent word corresponding to it. therefore,
Such headings and words corresponding to the headings are not registered in the terminology dictionary.

In the conventional method, as in ``1'', as a result of changes in the reading due to phonetic changes, etc., even though the kanji sound after the change may exist, there is no independent word corresponding to that kanji sound. In this case, the wasteful process of creating a search string containing only the kanji sound and performing a dictionary search using the search string is performed.In order to further improve the processing speed of 41 word extraction, This problem needs to be resolved.

Purpose The purpose of the present invention is to solve the problems of the prior art as described in 1. 1. The object of this invention is to set the searched character strings used when searching W-books to the minimum necessary, and to improve the processing speed of m-word extraction.

Configuration To achieve the above object, according to the present invention! 111! Fi
The extraction method is a kana-kanji conversion processing device that has a kanji sound size and a first means that uses the kanji sound size to separate an input character string using kanji sounds and kana as the minimum unit. A second means for determining the arrangement of kanji sounds and kana in a separated input character string, and a third means for determining whether there is an independent word having the same pronunciation as the first kanji sound of the character string to be analyzed. means, and if the first minimum unit of the character string to be analyzed is a kanji sound, and the kanji sound has the same reading as the independent word, the kanji sound and the second minimum jp position. The present invention is characterized by providing a fourth means for setting only a combination of , as a searched character string.

Note that since the input character strings are sequentially analyzed, naturally the input character strings that should be analyzed for the time being change one after another.

In this specification, the input character string to be analyzed for the time being is referred to as the character string to be analyzed.

Hereinafter, the configuration of the present invention will be explained in detail using an example. I will clarify.

FIG. 1 is a block diagram of a kana-kanji conversion processing device to which an m-word extraction method is applied according to an embodiment of the present invention.

In Fig. 1, 1 is a kana character input section for inputting a kana character string corresponding to the Japanese sentence to be created;
3 is an input character string storage part that temporarily stores the input kana character string, and 3 is a minimum of 81 kanji and kana characters for the input character string.
4 is a kanji row table with information on whether there is an independent word with the same pronunciation as the kanji table, and a 5-digit kanji table is used to separate the sections. The attribute judgment part of the input string that determines the attribute of the takana character string based on the kanji stand and kana (7), 3flbi form; 6 is the same as the kanji stand at the beginning of the character string to be analyzed. The attributes of the kanji table are determined based on whether or not an independent word with a reading exists. The attribute determination unit 7 of the first kanji board is the attribute determination unit 5 of the input character string and the attribute determination unit 6 of the first kanji board.
[Search string setting section that sets the minimum necessary search string according to the information to be searched, 9 is a 111-word dictionary, 8
10 is a dictionary search unit that searches for a word dictionary M19 based on the character string to be searched, and t! from the dictionary search unit 8. a candidate word storage unit that stores the selected candidate words; a candidate word evaluation unit that evaluates the 11 candidate words and selects the most appropriate candidate word;
Reference numeral 13 denotes an optimal candidate raw word storage section that stores the optimal candidate eye words selected by the candidate word evaluation section 11, and a display section 13 that displays the optimal candidate middle words as a kana-kanji conversion result.

Based on the information obtained from the kanji line table 4, the attribute determination unit 5 on the kanji stand determines whether the first kanji stand of the character string to be analyzed is the following 2.
Determine which of the two attributes it belongs to.

Kanji table attribute A: There is an independent word that has the same reading as the kanji table.

Examples: Ai (love), aku (evil), etc. Attribute B of the kanji table: There is no independent word that has the same reading as that of the kanji table.

Examples: ga, zu, zun, se, za, etc. FIG. 2 is a diagram showing part of the contents of a kanji row table according to an embodiment of the present invention.

As shown in Figure 2, in addition to the Kanji Zeno 2 column and the coat column, this kanji Zeno has an independence display that indicates whether there is an independent word that has the same reading as the reading on the Kanji stand. A column is provided. In this Kanji Zeno, independence is expressed as 1 bit I ′,
It is indicated by 'O', and when it is 1', it means that it has independence, that is, it is a kanji stand with attribute A, and when it is 0', it does not have independence.

That is, it means that it is a kanji stand of attribute B.

FIG. 3 is a flowchart showing the operation of the tIi word extraction method according to an embodiment of the present invention.

First, there is a 3-character delimiter with a kanji stand, and an input character string storage part 2.
The input kana character string sent from
'l. At this time, the independence information of each kanji stand is also read out from the kanji row table 4 at the same time.

The input character string attribute 4'1 determination unit 5 determines that the attribute of the input character string (character string to be analyzed), which is separated with the kanji base and kana as the minimum middle rank, is determined by the arrangement of the kanji base and kana. F
: Determine whether it is I or TYPE2 (30
2'). In cases other than TVPT':I and TYPE2, the searched character string is set using the conventional method (303).

1"VPEI or TYPE2, the following character string setting process unique to this embodiment is executed. The reason is:

'I'/PEI... (Kanji stand) + (Kanji stand) 10~'r
V l) E 2... If the first minimum unit is a kanji stand, such as (Kanji stand) + (kana) 10~, there is a possibility that the kanji stand may lose its independence due to a change in tone, etc. Because there is.

T'YPEI or T'YPE2 by step 302
If it is determined that it is, it is further determined whether or not it is TYPEl (302, 304). Next, in the case of TYPET, the first kanji character attribute determination unit 6 determines whether the first kanji character character string (first minimum unit) of the character string to be analyzed has attribute B (30/l .

305). The searched character string setting section 7 indicates that the attribute of the kanji stand is B.
``C: A2. In the case, only the combination of two kanji characters should be set as the searched character string305,306), and in the case of attribute A, the two kanji sounds should be combined as before. , and a character string to be searched consisting of the first kanji sound m-doku is created (305°A).

For example, in the case of the input character string "Kakko de ~", the Kanji sound delimiter is [Gara/Kora/de/Ha~], and TY
PEI, as shown in Figure 2, the first kanji sound ``gara'' is 0' in the independence notation column, which means attribute B, and the corresponding heading and kanji corresponding to the heading. Because the sound does not exist in application dictionary 9.

0') is set only to be searched character strings. In this way, by not using the kanji sounds corresponding to m-words that do not exist in the word dictionary 9 as searched character strings, it is possible to omit wasteful dictionary searches and improve the processing speed of dictionary searches.

In the case of the input character string "Kaisatsu de de~", the kanji sound delimiter is "Kai/Sara/ni/te~", and the TYPEI
However, the first kanji sound ``kai'' has attribute A, and the corresponding independent word c kai 2 times, kai.

Since there are two types of search strings, such as 0) Kaisatsu ■ Kai, two types of searched character strings are created as before.

If it is determined in step 304 that it is not T'YPEI, the character string to be analyzed can be identified as T'vPE2, so it is determined whether the first kanji sound is attribute B, and if it is B, is the first kanji sound and the second [minimum 11 of 1]
Only the combination of the digit kana and the digit kana is set as the searched character string (304).

307.30R). If the attribute is A instead of B, as before, in addition to the combination of the first kanji sound and the second minimum unit, kana, a search character string is created using the first kanji sound alone (307. A).

For example, in the case of the input character string "Zashiyori~", the Kanji sound delimiters are "Za/shi/ni/yo/su~", and 1
``VPE2, the first kanji sound ``Zara'' is a phonetic change of the kanji ``Zara'' reading ``Zara''.
Since there is no independent word corresponding to , only ``■'' is set as the character string to be searched.

Finally, the dictionary search unit 8 searches the word dictionary 9 using the obtained searched character string (309).

In this way, if the first kanji sound of the character string to be analyzed in TYPEI and T'YPE2 has attribute B, that is, if there is no independent word corresponding to the kanji sound, then the kanji sound and the second minimum unit are By using only the combined strings as searched character strings, the processing speed of dictionary searches can be improved.

Effects As explained above, according to the word extraction method of the present invention,
When extracting words in the kana-kanji conversion processing device, it is possible to improve the processing speed of five-word extraction by setting the number of searched character strings used at the time of dictionary search to the minimum necessary.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a kana-kanji conversion processing device applying the m-word extraction method according to an embodiment of the present invention, FIG. 3
The figure is a flowchart showing the operation of the word extraction method according to an embodiment of the present invention, and FIG. 4 is a diagram showing the conventional kanji zenhu. 3: Separator by kanji sound, 4: Kanji ex-husband, 5: Attribute determination unit for input character string, 6: Attribute determination unit for first kanji sound 27:
Searched character string setting section, 8: Dictionary search section, 9: l1l
-Word dictionary. Procedural amendment (voluntary) August 8, 1980 Patent application No. 13G1666”i2゜ Title of invention Word extraction method 3 Relationship with the case of the person making the amendment Patent applicant address Naka, Ota-ku, Tokyo Magome 1-3-6
Wow,. , ) (67 → Rico Co., Ltd. - Representative Hama 1) Hiro 5.6? The number of inventions will increase due to amendments.
a) Claim j on page 1 of the specification is amended as follows. [(1) In a kana-kanji conversion processing device having a kanji sound table and a first means for separating an input string using kanji sounds and kana as the minimum units using the kanji sound table, A second step for adding an information column to the kanji sound table indicating whether or not it is a kanji sound having a reading of and a third method that determines whether there is an independent word that has the same pronunciation as the first kanji sound of the character string to be analyzed.
, the minimum unit at the beginning of the character string to be analyzed is a kanji sound, and the kanji sound has the same reading as an independent word.
If it is a kanji sound, the first kanji sound. and No. 2: "A word extraction method characterized by providing a fourth means for setting only a combination of 1 minimum units as a searched character string." (b) Page 5, line 4 of the specification. ``I guess'' of the eyes is ``
I guess that's the case.'' (C) Kanji sound with U in the 7th line from the bottom of page 10 of the specification.”
It is corrected as ``kanji sounds that do not have 1.''

Claims (1)

[Claims] (1) In a kana-kanji conversion processing device having a kanji sound table and a first means for dividing an input string using kanji sounds and kana as the minimum units using the kanji sound table, a second means for adding an information column to the kanji sound table indicating whether or not the kanji sound has a reading, and determining the arrangement of kanji sounds and kana in the input character string separated by the first means; and a third means for determining whether or not there is an independent word having the same pronunciation as the first kanji sound of the character string to be analyzed; A fourth means is provided for setting only the combination of the first kanji sound and the second minimum unit as the searched character string when the kanji sound has the same reading as the independent word. A word extraction method featuring: