JPS60225273A - Word retrieving system - Google Patents

Abstract

PURPOSE:To realize more flexible word retrieval and more efficient word retrieval by collating a part preceding a key character of a given character string and a part succeeding this key character with a word dictionary. CONSTITUTION:A character string reading means 31 is started, and a character string to be subjected to word retrieval out of character strings is written on a character string storage means 32. A degree of reliability of character recognition for each character is sent to a key character setting means 33 to determine the key character. A key character retrieving means 34 is started to retrieve words having the key character in key character parts 211. A succeeding expression part collating means 36 is started to collate a partial character string succeeding the key character with rear expression parts 213 of the word dictionary. A preceding expression part collating means 35 is started to collate a partial character string preceding the key character with preceding expression parts 212 of individual words successively, and the word coinciding the input character string is sent back to a retrieval control means 38.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a word search method for searching a word dictionary for a word whose notation matches a substring of a given character string.

Character strings in Japanese sentences are usually written in solid letters, unlike English sentences, which do not have the custom of separating each word. Therefore, when analyzing a Japanese sentence mechanically using an electronic computer, etc., the first step is to search a word dictionary for words that match the partial string of a given string. It is necessary to search for words from the list.

FIG. 1 shows an example of a word dictionary used for word searches as described above. Each record corresponding to a word has a notation part 11 and a word information part 12, and the characters in the notation part 11 are They are stored in the order of code strings.

However, the word information section 12 stores part-of-speech information that is referred to when performing grammatical analysis after a word search, and pronunciation/accent information that is used when outputting audio. This is not information that is considered necessary. Now, conventional word search blades use such a word dictionary and compare and match the character code string corresponding to a given sentence string with the character code string in the notation part of the word dictionary from the beginning of the character code string. By doing this, it detects words with matching notation from the beginning of the given character string.

For example, "rainy season" and "rain" are detected for the character string "when it comes to rainy season." By repeatedly applying this word search method to the string immediately after the detected word, it is possible to divide a given string into word units, but this word search method is Since it detects words that match the notation from the beginning, if the given character string is a string with multiple character candidates, such as the result of character recognition, there may be words that are not included in the word dictionary. This method has the disadvantage that it cannot be applied efficiently when the starting position of the character string to be searched becomes ambiguous due to the presence of too many characters or analysis results.

An object of the present invention is to eliminate such conventional drawbacks and provide a flexible word search method that does not limit matching of a given character string with the notation of a word dictionary from the beginning of the character string. There is a particular thing.

That is, the present invention uses a collocation search method to search a word dictionary for a word whose notation matches a substring of a given character string, and searches for a key character that is a key in the word search for the notation part jJK in the same word dictionary. A front notation part that is a notation part in front of the same key character part and a back notation part which is a notation part after the same key character part are provided, and the key character part in the given character string is provided. key character setting means for setting a key character to be compared with the character string; and searching from the word dictionary by comparing the key character set in the four seen character strings with the key character part. a "-character search means for searching the given 7 character strings; and a forward notation part layering means for matching the forward notation part with a partial character string in front of the key character by 1 in the given 7 character string; The given character string is collated with the notation part in the word dictionary using the backward notation part collation means that matches the partial string after the key character in the string with the back notation part. This is a word search method characterized by searching forward and backward from a character.The present invention will be specifically described below with reference to the drawings.

FIG. 2 is a diagram showing an example of the word f#V Homare used in the word search method of the present invention. Each record corresponding to the word is in the notation section 21 and the word information section 22 such as pronunciation, part of speech, accent, etc. In addition, for the same notation part 21, a key sentence 1 plain part 211 which is a key in word search, a notation part 1μ forward from the same key character part, and a 1 μ forward notation part 212 and a character part from the same key character part. Back notation [backward notation part 213 which is part
It has been established that

The present invention does not impose any restrictions on which characters in the notation of a word are registered in the key character part, but in this example, for example, for <rainfall>, Any character can be registered as a key for searching.Apart from this example, you can also register the last kanji of the word notation in the key character part, or register the prior character of the word notation in the key character part. (The latter corresponds to the conventional word dictionary).Also, in this example, each record corresponding to a word is first stored in the order of the key character part 2110 character code, and the equal The records are arranged in the order of the character code string in the backward notation section 2130, and then in the order of the character code string in the forward notation section 2120.The word information section 22 is arranged in the order of the character code string in the backward notation section 2130, and then in the order of the character code string in the forward notation section 2120.In addition, the word information section 22 is arranged in the order of the character code string in the backward notation section 2130.The word information section 22 is arranged in the order of the character code string in the backward notation section 2130. It is referred to during grammatical analysis and audio output, and is not used within the scope of this word search.

FIG. 3 is a block diagram showing an embodiment of a device implementing the word search method of the present invention. In Fig. 3, 31
is a character string reading means that writes the given character string 320 into the character string memory hand +732 (a), the given character string is converted into a character code string through an input device such as a keyboard, OCR, magnetic tape device, etc. This is the part of the character string that is the target of the word search. In addition, if there is information necessary to set a key character in the given character string, the character string reading means 32 reads that information (for example, when the given character string is input from OCR). , reliability of recognition of each character, etc.) are sent to the key character setting means 33. 32 is a character string storage means for storing the given character string 320, and for example, an IC memory, a magnetic disk device, a magnetic tape, etc. are used. 20 has a key character part 211, a front notation part 212, and a rear notation part 213 in the notation part.
It is a word dictionary storage means for storing the word dictionary 2 as shown in FIG. 2, and has an IC memory and a magnetic disk device like 32.

A magnetic tape device or the like is used. Reference numeral 33 denotes a key character setting means for setting a key character 321 to be compared with the key character part 211 of the word dictionary 2 in a given character string 320, for example, a change in character in a given character string. If a point or a given character string is input from OCR, characters with high reliability of character recognition are set and marketed as key characters. However, in the present invention, no restrictions are placed on the conditions of '-x' for setting key characters. In addition, if a character string in which a key character is set is to be read, the character string reading means 31 and the key character setting means 33 are used.
can be realized as the same component.

34 is a key character search means for searching the word dictionary 2 by comparing the key character 321 with the key character section 211;
5 is a forward notation part collation means for comparing a partial character string 322 in front of a key character in a given character string 320 with the forward notation part 212 of the word dictionary; This is a backward notation part collation means that collates the backward partial character string 323 and the backward notation part 213 of the word dictionary. The key character searching method in the key character searching means 34 and the character string matching method in the forward notation part matching part 35 and the backward notation part matching part 36 are the same as the conventional key character searching method and character string matching method. This is a well-known method. 37 is the given character string 320
This is an information output means for outputting word information regarding the word when a word whose notation matches the partial character string of is detected. The operations of each of the above constituent elements are controlled by the search control means 38, and a series of operations shown in the flowchart in Figure g4 are performed. However, in the flowchart of Figure 4,
The order of TEP4 and TEP5 depends on how the words are arranged in the word dictionary 2. In this embodiment, as mentioned above, words having the same key character part are first arranged in the order of the character code string of the backward notation, then the forward notation. Since the parts are arranged in the order of the character code string, first check the rear notation (5TEP4), then check the front notation (5TEP4).
TEP5) is suitable. On the other hand, if words t that have the same key character part in a word dictionary are arranged first in the order of the character code string of the front notation, then in the order of the character code string of the rear notation, the order of 5TEP4 and 5TEP5 is interchanged. is appropriate.

Here, for example, a case where a word search is performed for the character string "rainfall measurement" inputted from OCR will be explained along with 5TEP of the flowchart of FIG. 4.

(STEPI: 100 in Figure 4) The character string reading means 31 is activated, and the character string "rainfall measurement" that is the target of the word search among the character strings input from the OCR is written into the character string storage means 32. . Further, as information for setting key characters, the reliability of character recognition for each character of "rainfall measurement" is sent to the key character setting means 33. For example, the following 5TEP will be explained assuming that the reliability of each character is given as follows.

Rainfall 0.95 Measurement 0.96 Rain 0.98 Constant 0.94 Amount 0.96 1,', dsTEp2: 101 in Fig. 4) The key character setting means 13 is activated, and the key character, #21 is set. Here, it is assumed that "rain", which has the highest reliability of character recognition, is set as the key character.

(STEP 3: 102 in Figure 4) Key character verification means 3
4 is activated, and words having the key character "rain" in the key character section 211 are examined. In the word dictionary 2 in Figure 2, the words kuuame>, <shower>, etc.

<Rainfall amount>, <Rainfall amount>, . . . are searched, and the start and end positions in the word dictionary of the word group having “Aya” in the key character section 211 are returned to the search control means 38. . In this case, if the word "rain" in the gold key character section 211 does not exist in the word dictionary 2, a search failure is returned to the search control means 38 and the process ends.

(STEP 4: 103 in Fig. 4) The rear notation section collation means 36 is activated, and the key character "Rain J (321)" in "Rainfall amount measurement" is replaced with the character string "Amount measurement" (323).
is compared with the backward notation section 213 of the word dictionary.

This verification is performed on the words between the beginning and end positions of the word group that has "rain" in the key character part obtained in 5TEP3, and the words that have "quantity measurement" in the rear notation part 213, "quantity measurement" in the rear notation part 213, Words with 74jJ in the rear notation part 213, words with "quantity" in the latter part 213, and words whose rear part is not voiced (words with a space in the rear notation part 213) are searched, and the word dictionary 2 their positions in are returned to the search control means 38. In other words, “
``Rain'' is the last letter of the word ``Rokuame'', ``Shower''.

The positions of words such as ``light rain'' and ``rain amount'' in the word dictionary 2 are returned to the search control means 38 .

(STEP 5: 104 in FIG. 4) The front notation part collation means 35 is activated, and the front notation part 212 and the key character 1 in "rainfall measurement" - rain J (321) υ are sequentially activated for the round can word obtained in 8TEP4. The character string in front is “Fuku J”
(322) is performed, and the position of the matched word ) is returned to the search control means 38. In other words, the position of ``rainfall amount'' in the word dictionary 2 is returned to the search control means 38. Also, at this time.

Among the words obtained in 5TEP4, the positions in the word dictionary 2 are also returned to the search control means 38 for words that do not have a front notation (words in which the front notation 212 is a space) <rain> and <rainfall>. Therefore, the words "rain", "rainfall", and "rainfall" were detected at 8TEP5. In this case, if no word is detected,
A detection failure is returned to the search control means 38, and the process ends.

'(STEP 6: 105 in Figure 4) Information output means 37
was activated and the word kurain was detected in 8TEP5〉
, <rainfall>, and word information 22 such as pronunciation, part of speech, and accent regarding <rainfall> are output.

In general, the words obtained through word testing are further determined to be correct through a mechanical connection test based on part-of-speech information regarding the same word, selection by the user, or the like.

Character strings input from OCR may have multiple character recognition candidates, and if the correct Japanese word is not obtained in word search, the character that seems to be misrecognized is considered as another candidate for character recognition. You need to replace it and repeat the word search again. At this time, according to the word verification method of the present invention, character string matching is not limited to the beginning of the character string, but can be performed from any position, so characters with high reliability of character recognition are fixed and It is possible to perform a word search while replacing warm characters, and it is possible to efficiently perform a word search from ambiguous character strings.

Although the above embodiment relates to a word search targeting character strings input from OCR, the word search method of the present invention is applicable not only to character strings input from OCR but also to character strings input from a keyboard, magnetic tape, etc. It is rumored that efficient word searches can be realized for sentences and °11# columns. for example,
Sentence mixed with kanji and kana: ``Zaatto: A shower began to fall.

'', if the word ``kuzaat'' is not registered in the word dictionary, conventionally, after the word search from the beginning of the sentence fails, the word search is repeated while shifting the starting position of the word search one character at a time. However, according to the word search method of the present invention, a word search using the koto key, ``rain'', which is a change point in character types, will search for ``rain'', ``shower'', which is unsearched and searched. By repeating the word search using the characters immediately before, before, and after the registered word as key characters, it is possible to perform analysis without reducing efficiency even if unregistered words appear.

Also, the word dictionary used in the word search method of the present invention is not limited to the configuration shown in FIG. 2, but can also be configured as shown in FIG. 5 or 6. is a word dictionary with key character parts separated as tables. 50 is a key character part table, which is composed of a key character part 501 and a pointer 502 to the beginning position of a word group having the same key character part.

The word dictionary 5 has a front notation section 51, a rear notation section 52, and a word information section 53. If such a word dictionary structure is used, the position of a group of words having the same key character part in the word dictionary can be easily known from the pointer 502, so that the search speed is greatly improved.

FIG. 6 is a word dictionary in which the key character part and the rear part have a tree structure. 60 is a key character part table, and a key character part 601 and a pointer 602 to the table for the first character of the backward notation part have the same key character, and a pointer 603 to a word without a notation part is exposed. Become. Reference numeral 61 is a rear notation section, and a rear notation section first character table 611. It has a backward notation part second character table 612, etc., and each table is made up of a notation part, a pointer to the next table, and a pointer to a word.

As explained above, according to the word search method of the present invention, the matching of a given character string with the word dictionary is not limited to starting from the beginning of the character string. Can be performed forward and backward;
Compared to the conventional word search method that searches from the beginning of a character string, more flexible word searches and more effective word searches can be achieved.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of a word dictionary used in a conventional word search method, and FIGS. 2, 5, and 6 are diagrams showing examples of a word dictionary used in the word search method of the present invention. , FIG. 3 is a block diagram showing an embodiment of an apparatus implementing the word search method of the present invention, and FIG. 4 is a flowchart for explaining the embodiment of FIG. 3. In the figure, 1.2,5.6・・・・・・・・・・Word dictionary 1
1.21......Notation section 12.22, 53.63...Word information section 211
．． 501.601...Key character section 212.51.62
......Front notation part 213.52.61...
・・・・・・Rear notation part 31・・・・・・・・・・・・
・・・・・・・・・・・・Character string reading means 3
2・・・・・・・・・・・・・・・・・・・・・・・・
...Character string storage means 33...
......Key character setting means 34...
・・・・・・・・・・・・・・・・・・・・・ Key character search means 35 ・・・・・・・・・・・・・・・・・・
......Front notation section collation means 36...
・・・・・・・・・・・・・・・・・・Backward notation part collation means 37・・・・・・・・・・・・・・・・・・・・・
......Information output means 38...
・・・・・・・・・・・・・・・Search control means. Patent Applicant 1 Rejection Table III Field Bath Festival Figure 1 Figure 3 ill? Zll ZI5 22 Figure 4

Claims (1)

[Claims] A word search method is used to search a word dictionary for a word whose notation matches a substring of a given character string, and the key character part that is the key in word search and the same key character part are used for the notation part in the same word dictionary. A forward notation part, which is a notation part in front of the character string, and a back notation part, which is a notation part after the same key character part, are provided, and the character string is the object of matching with the key character part in the given character string. a key character setting means for setting a key character; a key character search means for searching the word dictionary by comparing the key character set in the given character string with the key character part; forward notation part collation means for matching a partial character string preceding the key character in a given character string with the forward notation part; A word search characterized in that the given character string is compared with the notation part in the word dictionary forward and backward from the key character using a backward notation part collation means for comparing the given character string with the notation part in the word dictionary. method.