JPS60124771A - Word extracting system - Google Patents

Abstract

PURPOSE:To attain high speed retrieval of a word including a Kanji (Chinese character) by operating in parallel cut-out of a candidate word by a Kanji sound and accessing of a special word table storing and disabling to extract by the Kanji sound processing. CONSTITUTION:A Kana (Japanese syllabary) is stored in an input character string temporary storage section 2 and fed to a Kanji sound processing section 3. The Kanji sound processing 3 has a classification module and a register, classifies the result of processing according the condition and gives the result to a retrieved character string forming section 4. The retrieved character string forming section 4 sets six characters started from a set analysis start position to the buffer, performs matching processing with a Kanji sound table storage section 5, applies sectioning at an element level and performs Kanji sound processing. Words disable for extraction by the Kanji sound processing are processed for extraction without omission by accessing the special word registered in the special word table in advance in parallel, the candidate is evaluated, a suitable candidate is selected and a character string is formed at th retrieved character string forming section 4.

Description

[Detailed Description of the Invention] Technical Field The present invention relates to a word extraction method in a Japanese word processor, etc., and in particular, introduces a pre-kanji concept into the word extraction process and sets the search string to the minimum necessary when searching in a dictionary. This invention relates to a word extraction method that can reduce erroneous analysis and improve analysis speed by not extracting unnecessary candidates.

Prior Art In conventional kana-kanji conversion processing devices, the algorithm for extracting words from input kana character strings is generally 1.
It was something like this:

(1) Setting analysis star 1 to position for character strings Except for special cases, generally the first character (first character) of a character string is used as the starting position for analysis.

First, if a word is successfully extracted with that position as the first character, the first character of the character string after word extraction is set as the start position of a new analysis.

(Example) Dango's ~ ↑: Initial analysis start position Here, if the extraction of "Kaiseki (analysis)" is successful, the result will be as follows.

By Kaiseki Dango no ~ 18th analysis start position (2) Creation of searched character string for dictionary search Assuming that the reading length in the dictionary is a maximum of 6 characters, in the case of the example sentence above, the following A search string like this is set.

a) Cut out the first word 02) Cut out the first word 02) Cut out the word by cutting out the word after successful "analysis" (3) Set search character string and Dictionary II's 51851-
Matching judgment with 'sentence'f two columns; ttl of candidates according to
+ rB” In the case of the three major examples, do the following;! , 1゜■
Cannot extract candidates with "Kaisekiyo" ■ Cannot extract candidates with "Kaiseki" (■ Cannot extract "Kaiseki", "Analysis", and "Kaiseki" with "Kaiseki") ■ Cannot extract candidates with "Gaise" ■ Extracting ``kai'', ``kai'', ``kai'', ``kai'', etc. with ``kai'' 1 ■ Extracting ``ka'', ``ko'', ``mosquito'', ``section'', etc. with ``汀か'' (4) 3)
Various evaluations are performed on the candidate group extracted in , and the candidate considered to be the most appropriate is determined.

However, the word extraction method as described in -1 has a problem in that depending on the input character string, an extremely large number of candidate groups may be extracted, causing erroneous analysis and a reduction in dictionary search speed.

Purpose The present invention has been made in view of the above circumstances, and its first purpose is to solve the above-mentioned problems in conventional word extraction methods, reduce erroneous analysis, and improve analysis speed. The objective is to provide a word extraction method for

Configuration Hereinafter, the configuration of the present invention will be explained in detail based on examples.

FIG. 1 is a functional block diagram showing an overview of a kana-kanji conversion processing device according to an embodiment of the present invention. As is clear from the figure, the feature of this embodiment is that two word extraction processes are executed in parallel.

That is, the process is as follows.

(1) The access process to the general independent word dictionary that exists as an external dictionary (external file) is performed as described below.
(Bean paste) is used.

(2) Newly perform access processing to an internal dictionary provided separately from the external dictionary.

(:3) Equal evaluation is performed on each of the candidate groups extracted by the above processes (L) and (2), and the most appropriate candidate is selected.

In other words, in addition to effectively using kanji sounds to extract word candidates from a general independent word dictionary with a huge number of registered words, a separate internal By registering in a dictionary and searching without performing kanji sound processing, it becomes possible to extract all necessary candidates.

This example will be explained in more detail below.

FIG. 2 is a diagram showing a processing block that performs processing using the above-mentioned kanji sounds. In the figure, ■ is the keyboard input section, 2
3 indicates an input character string temporary storage section, 3 a kanji sound processing section, 4 a check search string creation section, and 5 a kanji pre-table storage section.

The kanji sound processing unit 3 has a classification module for classifying processing results according to conditions described later, and a register for holding the classification results. Below, this register is r
It's called i'YPEJ.

Here, I will explain the three major kanji sound tables (.

As shown in FIG. 3, this table registers kanji characters with readings of 2 to 3 characters, and IL is used as a unit of word extraction. It should be noted that FIG. 3 merely shows an example of the quintic character sound table, and it goes without saying that the present invention should not be limited to this.

The operation of this embodiment will be described below, using the following example sentence of the 9th order: (According to the result of the M-fold, the household is ~).

The searched character string creation unit 4 stores six characters starting from the analysis star 1 position set in the same way as in the conventional case, into a buffer that has been used in advance (see FIG. 4). This buffer may be one that can one-dimensionally store 6 characters (7y1), and hereinafter this buffer will be referred to as "WINI)OWJ".

Next, a matching process is performed between the character string in the above WINDOW and each element of the kanji front table shown in FIG.
A method that delimits the string in INDOW at the element level before the kanji and expresses the result concretely.
Set it in a buffer prepared in advance. Here, W.I.
A buffer called NDOW2 is prepared that has the representation of an array of size 6 in the - dimension.

FIG. 5 shows an example of the divisions applied to the character strings in WINDOW and the contents of WINDOW2. W.I.
The arrow attached to NDOW indicates a division at the level of the three major kanji sounds, and indicates that the kanji that is the content of WI NDOW2 has a kanji front corresponding to the number of characters in the kanji front table. be.

Here, W I N D OW 2 (1) = 2 (r chi)
L,, -tdcorresponding)WrNl)OW2(21=2(r
Seki J 21 above)W I NDOW2(3)=l
(corresponding to "r's") WINI) OW2 (4') = +
(corresponding to rke) W T NL) OW2 (5'1=
O W I N +, )OW2(6)=0.

The results of the above processing are classified according to the following conditions.

(])WINDOW2(1)≧2, KakWIN+,)O
If W2(2)≧2, set rlJ to TYPIΣ.

(2) WINDOW2(1)≧2. AND WINDOW2
(2) If = 1, set TYPE to "2" (3) WI N1) OW2 (+) = 1 and WINI
) If OW2 (2) ≧ 2, set "3" to T' ¥ P E.

(4) In cases other than the above categories (1) to (3) T Y P
Set E ni r 71 J.

Nth place Superintendent's Award: 1 piece of tuna expert plate, 1 piece of paper, 1 piece of sodehen, and 1 piece of tuna expert plate.Create it using the following method.

T Yl) Either "1" to "3" is different from E to Se-No1
~Sa, h, cut out the number of characters corresponding to WINDOW2(1)+WINOW(2) and WINOW2(1) from the input string, and then write in two ways. Create a search string for .

In the case of the above example sentence, T Y P E is "1", so two characters are set as the character strings to be searched: (1) Kai Seki (2) Kai. Using this searched character string, a dictionary search is performed in the same manner as before.

Note that when TYPE is "4", the searched character string is created in exactly the same manner as in the past. .

Specific examples of each classification are given in FIGS. 6(A) to 6(D) to supplement the explanation. (A> is classification 1 (TYPE=1”l, (B
) correspond to classification n (TYPE=2), . . . , respectively. The input sentence and searched character string in each case are as follows: (A): At university (at university), at middle school (at university) (at university) (B): at Riyakugi (in abbreviated form) (,0 Riya Nagi Q) In the case of Riyaku (C): Kanonabaai (if possible) (,1)
In the case of Kanou (Second) (D): In this way (same as left) (]) In this way 2) In this way (Bon this way (4) This way (b) This (Φ different).

"In each of the three major embodiments, WINDOW and WIN
Although DOW2 is a buffer having a size of 6 characters, this is not necessarily limited to 6 characters. Furthermore, instead of the buffer for the WINDOW lJl, a baranoa for setting an input character string, a pointer for the number of sheets indicating the position in the buffer, and a register for reading the value of the pointer may be provided.

On the other hand, words that cannot be extracted by the kanji sound processing are processed as follows.

for example, I'll ask you again.

Considering the input sentence, if this is processed by the kanji sound processing described above, the following error will occur.

That is, as shown in FIG. 7, 'rY P E = 2
Based on this, the search string is (
The character string ``J) ouuka sip) ou'' is extracted, but the word that would be used to obtain the recognition ``visiting you'' is not extracted, resulting in an analysis result such as ``Europeanization''.

This is a negative aspect of using kanji sound processing, but since such cases are very few, in this embodiment, these are registered in the special word table in advance, and the kanji By configuring the special word table to be accessed in parallel with sound processing, it is possible to extract virtually all necessary words without omission.

The comparison and selection of the 941 words extracted by the kanji sound processing and the words extracted by the special word processing is performed, for example, by a method in which the operator selects the words using a key while looking at the display. It is not limited to.

Effects As described above, according to the present invention, when performing word extraction, it is possible to cut out candidate words using kanji fronts and access the special word table that stores m-words that cannot be extracted by kanji sound processing. Since these are operated in parallel, this has the remarkable effect that words containing kanji can be extracted faster and without erroneous extraction.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram showing an overview of the present invention. Figure 2 is a block diagram of the kanji sound processing unit, Figure 3 is a diagram showing part of the contents of the kanji pre-table, Figure 4 @ is a diagram showing an example of the contents of the input character string buffer, and Figure 5 is a diagram showing human input characters. FIG. 6 is a diagram showing a specific processing example, and FIG. 7 is a diagram showing an example in which the kanji cannot be extracted by sound processing. 1: Keyboard input section, 2: Input character string temporary storage section, 3
: Kanji sound processing section, 4: Searched character string creation section, 5: Kanji pre-table storage section. Patent applicant Rico Co., Ltd. - 2..., ζ 1 representative patent attorney Masatoshi Isomura''. Figure 1 (manual character string) Figure 2 Figure 4 Figure 5-Figure 6 Figure 7

Claims (1)

[Claims] (1) A word dictionary storage means for storing a plurality of words in correspondence with character strings representing their pronunciations, a means for temporarily storing input kana character strings, and a word dictionary storage means for storing a plurality of words in correspondence with character strings representing their pronunciations; and means for temporarily storing input kana character strings; In the kana-kanji conversion processing device, the kana-kanji conversion processing device has table storage means in which kanji fronts having two or more pronunciations are registered. means for dividing the input kana character string using the kanji front; means for morphologically classifying the processing results by the dividing means; 1, a special word table storage means storing special words unsuitable for kanji sound processing, and a second search character string accessing the storage means to create a search character string. A column creating means is provided, and the first.
Compare and select the output of the second search string creation means +
し＊＋6