JPS63116283A - Japanese sentence reader - Google Patents

Abstract

PURPOSE:To reduce the number of proposed words and storage capacity and to shorten a processing time by providing the titled device with a proposed word selecting means for selecting a word indicating that the sum of the similarities of recognized proposed characters exceeds a prescribed threshold as a proposed word. CONSTITUTION:A word collating means 6 compares a character string obtained by combining respective recognized proposed characters with a word having the same length in a word dictionary 5 in each character position, and at the time of coincidence, the word, its grammatical property and the similarity of the recognized proposed word are sent to a proposed word selecting means 7. When the sum of respective similarities exceeds the prescribed threshold regulated by the number of characters in the word, the means 7 decides the word as a proposed word and stores the word, the grammatical property of the word and the similarity as an evaluated value in a proposed word storing means 8. In other cases, the word is not decided as a proposed word. Consequently, the number of proposed words to be the objects for the inspection of word connection based upon a word connection inspecting means 10 can be reduced, the storage capacity of the means 8 can be reduced and respective processing times can be shortened.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention is a method of reading Japanese sentences by selecting grammatically probable results from the results of character-by-character recognition of the characters that make up Japanese sentences. It concerns a reading device.

[Conventional technology]

FIG. 6 is a block diagram showing the configuration of a conventional Japanese text reading device. In Fig. 6, 1 is a form, 2 is a scanning means, and 3 is a character recognition means that cuts out and recognizes characters constituting an input Japanese sentence character by character, and outputs recognition candidate characters and their similarity to the input characters. , 4 for input characters. <J.R.
Recognition result storage means for storing candidate characters and their similarity; 5
6 is a word dictionary that stores words used in Japanese sentences and their grammatical attributes; 6 is a word dictionary that matches a character string that is a combination of recognition candidate characters in the recognition result storage means 4 and a word in the word dictionary 5; 8 is a candidate word storage means for storing candidate words selected by the word matching means 6; 9 is a grammar dictionary containing information on whether consecutive word strings are grammatically connectable; 10 is word connection testing means for selecting word strings that can be connected grammatically from the candidate words stored in the candidate word record 1 means 8 using the grammar dictionary 9;
Reference numeral 11 denotes word string storage means for storing connectable word strings, and reference numeral 12 denotes reading result determination means for uniquely determining the reading result for the input Japanese sentence from the word strings in the word string storage means 11.

FIG. 7 is a diagram showing an example of an input Japanese sentence input to the Japanese sentence reading device of FIG. 6 and its recognition candidate characters. 7th
The figure shows an example of an input Japanese sentence printed or written on a form 1 that is input into a Japanese sentence reading device, and an example of recognition candidate characters obtained by recognizing each character of the input Japanese sentence with the character recognition means 3. ing. In Fig. 7, 13 is the input Japanese sentence "read the characters", 14 to 19 are each character composing the input Japanese sentence 13, and 20 to 41 are recognition candidate characters for each character 14 to 19. .

FIG. 8 is a diagram showing the storage contents of the recognition result storage means in the Japanese text reading device of FIG. 6. FIG. 8 shows an example in which recognition candidate characters 20 to 41 obtained by recognizing each character 14 to 19 of an input Japanese sentence 13 by the character recognition means 3 and their similarity are stored in the recognition result storage means 4. ing. In FIG. 8, 5ij (1≦i≦6, 1<j<7) is the degree of similarity of each recognition candidate character 20 to 41. Note that the area marked with ◆ indicates that there is no recognition candidate character, and the degree of similarity thereto is set to O.

9 and 10 are diagrams showing examples of word dictionaries in the Japanese sentence reading device of FIG. 6, respectively. In FIGS. 9 and 10, 42 to 62 are words. Note that here, for convenience, period marks, verb stems, and conjugated endings of verbs are also treated as words.

FIG. 11 is a diagram showing the results of word matching performed by the word matching means in the Japanese sentence reading device shown in FIG. 11th
In the figure, in the word matching means 6, each recognition candidate character 20 to 4
This is an example in which a character string combining 1 and a word in the word dictionary 5 are selected and stored in the candidate word storage means 8. In FIG. 11, 42 to 62 are candidate words.

FIG. 12 is a diagram showing an example of a grammar dictionary in the Japanese sentence reading device of FIG. 6. In Figure 12, 63-87 are 2
13 represents connection information indicating whether grammatical attributes of consecutive words can be connected, ○ mark indicates that they can be connected, and X mark indicates that they cannot be connected. FIG. 7 is a diagram for explaining the operation of the word connection test means in the Japanese sentence reading device of FIG. 6. In FIG. 13, "-1" indicates that consecutive words can be connected, and "r----->" indicates that consecutive words cannot be connected. When each word has multiple grammatical attributes, the number of grammatical attributes indicates the same word.

FIG. 14 is a diagram showing the memory contents of the word string storage means in the Japanese sentence reading device of FIG. 6. FIG. 14 shows a word string stored in the word string storage means 11.

FIG. 15 is a diagram showing the reading result of the input Japanese sentence by the reading result determination means in the Japanese sentence reading device of FIG. FIG. 15 shows the reading results for the input Japanese sentences whose reading results have been determined by the reading result determining means 12.

Next, the operation of the conventional Japanese text reading device described above will be explained. The input Japanese text 13 on the form 1 is scanned by the scanning means 2, cut out character by character by the character recognition means 3, and recognized by a method such as pattern matching. Then, a plurality of recognition candidate characters 20 to 41 are outputted in which the degree of similarity is assigned to the order of accuracy as input characters. Each recognition candidate character 20 to 41 output by the character recognition means 3 and its similarity 5ij (1≦
(6.1) (j≦7) is stored in the recognition result storage means 4.

The word matching means 6 compares the character string that is a combination of the recognition candidate characters 20 to 41 in the recognition result storage means 4 and the word in the word dictionary 5 for each character position, and when the comparison results match, The degree of similarity of the recognition candidate character that matches the word and the characters constituting the word, and the grammatical attributes of the word are sent to the candidate word storage means 8. In this example, for example, as shown in ``Character recognition post-processing using language information and recognition information J (child communications society research group material PRL 82-76), combinations of words in the word dictionary 5 and recognition candidate characters are used. When the searched character strings match, all the words are sent to candidate word storage means 8.

In other words, from the first character of the input Japanese sentence 13, the recognition candidate character for the character "Bun" 14 [sentence J20 matches the word "Bun" 59 in the word dictionary 5, so this word "Bun" 59
Then, the similarity degrees S, I and grammatical attributes of the recognition candidate character "sentence" 20 are sent to the candidate word storage means 8.

Furthermore, from the first character of the input Japanese sentence 13, the word "character" 61. Word “character” 60. Word “woman” 52, word “
Husband” 58. Word “Lord” 49. Word “preside” 50. word"
Heaven” 56. Word “large” 53° Word “university” 55. word"
``University'' 54 is selected and sent to the candidate word storage means 8. Similarly, starting from the second character, change the word "ji" 48 to the word "ji" 48.
46. From the third letter, the word "wo" 4
5° Select the word "ma" 43. From the 4th letter, the word "yomi" 57. Word “word” 47. Select the word ``Yu'' 62 and the word ``Sho'' 51. From the 5th letter, the word "mu"
Select 44. From the 6th character onward, the word "." 42 is selected. Then, the candidate words shown in FIG. 11 are stored in the candidate word storage means 8.

In the word connection test means 10, each connection information 6 of the grammar dictionary 9
3 to 84 are used to select connectable word strings and send them to the word string storage means 11. In this example, we test whether or not it is possible to connect consecutive words in order from the beginning of the sentence as shown in Figure 13.For example, the word ``sentence'' 59 selected from the first character. Woman” 52. Word “husband”
58° Word “Lord” 49. Word “heaven” 56. Word "large" 5
3 and the word ``character'' 48 and the word ``gaku'' 46 selected from the second character cannot be connected by each connection information 63 to 66 of the grammar dictionary 9. Also, the word "Character" 61° and the word "Literature" 60. selected from the first character.
Word “preside” 50. Word “Oaza” 55. Word "university" 5
4 and the word “wo” 45 selected from the third letter is,
Connection is possible using each connection information 67 and 69. Similarly, a word connection test is performed by the word connection test means 10 as shown in FIG. 13, and the word string shown in FIG. 14 is stored in the word string storage means 11.

Since a plurality of word strings exist in the word string storage means 11, the reading result determining means 12 selects as the reading result the word string for which the sum of the similarities of the recognition candidate characters that match the characters constituting each word is maximum. . In this example, the recognition candidate characters that match each character in the word string 61 - word ``wo'' 45 - word ``yomi'' 57 - word ``mu'' 44 - word ``.'' 42 are The sum of similarities becomes maximum, and the reading result shown in FIG. 15 is obtained.

[Problem that the invention seeks to solve]

The conventional Japanese sentence reading device described above is configured as described above, so that when a character string that is a combination of recognition candidate characters matches a word in the word dictionary 5, the matching word is used as a candidate word. Since the word connection test was performed by the word connection test means 10, the storage capacity of the candidate word storage means 8 required to store candidate words increased, and furthermore, the word connection test by the word connection test means 10 increased. There is a problem in that the number of combinations of candidate words increases, resulting in an increase in processing time.

The present invention was made to solve these problems, and an object of the present invention is to provide a Japanese text reading device that can reduce storage capacity and processing time.

[Means for solving problems]

The Japanese sentence reading device according to the present invention provides a method for reading Japanese text when a character string obtained by combining recognition candidate characters obtained by a character recognition means matches a word in a word dictionary. However, candidate word selection means is provided for selecting a word whose value is equal to or greater than a predetermined threshold value as a candidate word.

[Effect]

In the Japanese sentence reading device 1 of the present invention, the candidate word selection means selects as a candidate word a word for which the sum of the similarity degrees of recognition candidate characters is equal to or more than a predetermined threshold value. The number of candidate words to be tested for connectivity can be reduced.

〔Example〕

FIG. 1 is a block diagram showing the configuration of a Japanese text reading device which is an embodiment of the present invention, and the same or equivalent parts as those of the conventional device shown in FIG. 6 are indicated using the same reference numerals.
A detailed explanation thereof will be omitted.

In FIG. 1, 7 is a candidate word selection means for selecting candidate words.

FIG. 2 is a diagram for explaining the operation of the candidate word selection means in the Japanese sentence reading device of FIG. 1. In FIG. 2, ○ indicates C selected as a candidate word by the candidate word selection means 7.
, and × indicates that the candidate word selection means 7 did not select the word as a candidate word.

FIG. 3 is a diagram showing the stored contents of the candidate word storage means in the Japanese sentence reading device of FIG. 1. FIG. 3 shows an example of candidate words selected by the candidate word selection means 7, and these candidate words are stored in the candidate word storage means 8.

FIG. 4 is a diagram for explaining the operation of the word connection test means in the Japanese sentence reading device of FIG. 1. In FIG. 4, "-m-" indicates that consecutive words can be connected, and "-1--→" indicates that consecutive words cannot be connected.

FIG. 5 is a diagram showing the memory contents of the word string storage means in the Japanese sentence reading device of FIG. 1. FIG. 5 shows connectable word strings selected by the word connection testing means 10.

Next, the Japanese sentence reading vc, which is an embodiment of the above-mentioned invention, will be explained.
The operation of the position will be explained below. The character recognition means 3 selects each recognition candidate character 20 for each character 14 to 19 of the input Japanese sentence 13.
~41, and its similarity 5ij (1≦L<6, 1
< j < 7) is determined and stored in the recognition result storage means 4, and then the word matching means 6 sequentially searches from the beginning of the sentence to determine whether the length of the combination of each recognition candidate character 20 to 41 and the length in the word dictionary 5 is the same. words are compared for each character position, and if a match is found as a result of the comparison, a candidate word is selected based on the word, its grammatical attributes, and the degree of similarity between each character that makes up the word and the matching candidate character. Send to means 7. In this example, like the conventional Japanese text reading device, the first part of the input Japanese text 13 is
Recognition candidate character ``sentence J2'' for the character ``sentence'' 14
0 and the word "sentence" 59 in the word dictionary 5, the grammatical attributes thereof are sent to the candidate word selection means 7 together with the similarity Sll. The candidate word selection means 7 selects the word as a candidate word if the similarity Stt of the recognition candidate character "SENTEN" 20 that matches the word "SENTEN" 59 is equal to or higher than a predetermined threshold TH normalized by the number of characters in the word. “Sentence” 59. A similarity of 8+1 is stored in the candidate word storage means 8 as the grammatical attribute of the word "sentence" 59 and the evaluation value of the word "sentence" 59, while if not, it is not selected as a candidate word. Furthermore, for example, the character string "character" which is a combination of the recognition candidate characters "senji J20" and "ji" 28 for the first character "bun" 14 and the second character "ji" 15 is the word Since it matches the word "character" 61 in the dictionary 5, the candidate word selection means 7
send to The candidate word selection means 7 adds the similarity degrees 811 and S2 of each recognition candidate character "sentence J20" and "character" 28 that match the word "character" 61, and calculates the evaluation value of the word "character" 61. do. If this evaluation value is equal to or greater than a predetermined threshold TH, it is stored as a candidate word in the candidate word storage means 8 along with its grammatical attributes and evaluation value.

Then, when the above processing is performed on all the words matched by the word matching means 6 and the candidate words shown in FIG. 2 are selected by the candidate word selection means 7, the contents of the candidate word storage means 8 are as shown in FIG. It will be as shown.

The word connection testing means 10 selects connectable word strings from the candidate word storage means 8 as shown in FIG.

For example, the word "bun" 59 is selected from the first character of the input Japanese sentence 13. The word "woman" 52, the word "husband" 58, and the word "character" 48 selected from the second character. Word “Study” 4
6, neither of them can be connected according to the connection information 63 and 64 of the grammar dictionary 9. Also, the word "character" 61. selected from the first character. The word “literature” 6o and the 3rd
The word "wo" 45 selected from the first character is the connection information 6
7 enables connection. Similarly, the word connection test means 10 performs a word connection test, and the word string shown in FIG. 5 is stored in the word string storage means 11.

Since there are a plurality of word strings in the word string storage means 11, the reading result determining means 12 determines the word string that maximizes the sum of the evaluation values of each word as the reading result. In this example, the word "character" 61 - the word "wo" 45 - which forms the word string
The sum of the evaluation values of the word "yomi" 57 - the word "mu" 44 - the word "." 42 becomes the maximum, and the reading result shown in FIG. 15 is obtained.

Further, when the candidate word selection means 7 uses the sum of the ranks of recognition candidate characters, the following will be described.

In the word matching means 6, when a character string in which recognition candidate characters are combined and a word in the word dictionary S match at each character position, the word and the grammatical attributes of the word and the recognition candidates that match the characters constituting the word are used. The character ranking is sent to candidate word selection means 7. Candidate word selection means 7 calculates the sum of the rankings of recognition candidate characters corresponding to each character of a word and uses it as an evaluation value of the word, and only when the evaluation value is less than a predetermined threshold is selected as a candidate word and stores the candidate word. The information is stored in means 8. Then, the word connection test means 10 tests the connection between consecutive words using the grammar dictionary 9, and selects word strings that can be connected.
It is stored in the word string storage means 11.

The reading result determination means 12 determines the reading result by selecting the word with the minimum sum of the evaluation values of each word constituting the word string.

In the above embodiment, a case was explained in which ordinary Japanese characters are used, but it can also be used to read solid-letter addresses such as prefectures, cities, wards, and counties.

Further, in the above embodiment, the maximum number of recognition candidate characters for an input character is seven, but the apparatus of the present invention is not limited to this.

〔Effect of the invention〕

As explained above, in the Japanese sentence reading device, the sum of the degrees of similarity between the words selected by the word matching means and the recognition candidate characters matched with the words in the word dictionary is obtained as an evaluation value for the word. Since we have provided a candidate word selection means that selects only words whose evaluation value is equal to or higher than a predetermined dark value as candidate words, the storage capacity and processing time are reduced compared to conventional devices of this type. It has excellent effects.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a Japanese text reading device that is an embodiment of the present invention, and FIG. 2 is a diagram for explaining the operation of the candidate word selection means in the Japanese text reading device of FIG. 3 is a diagram showing the memory contents of the candidate word storage means in the Japanese sentence reading device of FIG. 1, and FIG. 4 is a diagram showing the operation of the word connection testing means in the Japanese sentence reading device of FIG. 1. 5 is a diagram showing the storage contents of the word string storage means for the Japanese sentence reading device of FIG. 1, FIG. 6 is a block diagram showing the configuration of the conventional Japanese sentence reading device, and FIG. FIG. 6 is a diagram showing an example of the input Japanese sentence input to the Japanese text reading device of FIG. 6 and its recognition candidate characters; FIG. , FIG. 9 and FIG. 10 are respectively No. 6
A diagram showing an example of a word dictionary in the Japanese sentence reading device shown in FIG.
FIG. 11 is a diagram showing the results of word matching performed by the word matching means in the Japanese sentence reading device shown in FIG. 6, and FIG.
FIG. 6 is a diagram showing an example of the grammar dictionary in the Japanese sentence reading device, FIG. 13 is a diagram for explaining the operation of the word connection test means in the Japanese sentence reading device of FIG. 6, and FIG. 6 is a diagram showing the storage contents of the word string storage means in the Japanese sentence reading device, and FIG. 15 is a diagram showing the reading result for the input Japanese sentence by the reading result determining means in the Japanese sentence reading device of FIG. 6. . In the figure, 1... form, 2... scanning means, 3...
・Character recognition means, 4... Recognition result storage means, 5...
Word dictionary, 6... Word matching means, 7... Candidate word selection means, 8... Candidate word storage means, 9... Grammar dictionary, 10... Word connection testing means, 11... Words Column storage means, 12... Reading result determining means, 13... Input Japanese sentence, 14-19... Characters, 20-41... Recognition candidate characters, 42-62... Words (candidate words) , 63~
87...Connection information. In each figure, the same reference numerals indicate the same or equivalent parts.

Claims (3)

[Claims] (1) a character recognition means that recognizes an input Japanese sentence character by character, determines a plurality of recognition candidate characters and their similarities for each character, and outputs recognition candidate characters and similarities in order of increasing similarity; a recognition result storage means for storing recognition candidate characters and degrees of similarity in association with each character position of an input Japanese sentence; a word dictionary storing words used in Japanese sentences and grammatical attributes for each word; comparing a grammar dictionary storing grammatical connection information between grammatical attributes, a character string of an arbitrary length that is a combination of the recognition candidate characters, and a word of the same length in the word dictionary for each character position; A word matching means for selecting a matched word, and a sum of degrees of similarity between the word selected by the word matching means and the matching recognition candidate character, as an evaluation value for the word, and the evaluation value is set as a predetermined threshold. candidate word selection means for selecting only the words satisfying the above as candidate words; and the candidate word, the grammatical attributes of the candidate word, and the evaluation value obtained by the candidate word selection means, corresponding to the character position of the input Japanese sentence. a candidate word storage means for storing the candidate words in combination with each other; and the sentence dictionary is used to test whether the grammatical attributes of each word are grammatically connectable to each other from the string of candidate words for the input Japanese sentence, and words that can be connected are determined. word connection testing means for selecting columns; word string storage means for storing the connectable word strings and the evaluation value obtained by the candidate word selection means for each word; 1. A Japanese sentence reading device, comprising reading result determining means for selecting a word string with the largest sum of evaluation values as a reading result for an input Japanese sentence. (2) The candidate word selection means calculates the sum of ranks of recognition candidate characters that match the word selected by the word matching means as an evaluation value for the word, and words for which the evaluation value is less than a predetermined threshold. 2. The Japanese sentence reading device according to claim 1, further comprising means for selecting only candidate words. (3) The reading result determining means is a means for selecting a word string having the smallest sum of evaluation values from the connectable word strings as the reading result for the input Japanese sentence. Japanese text reading device according to item 2 of the scope.