JPH0632091B2 - Japanese sentence reader - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an apparatus for reading Japanese sentences in daily use, and in particular, recognizes characters constituting Japanese sentences character by character, and recognizes the grammatical characters from the recognition result. The present invention relates to a Japanese sentence reading device that selects a certain result as a reading result.

[Conventional technology]

FIG. 2 is a block diagram of a conventional Japanese sentence reading device. In the figure, (1) is a sticker, (2) is scanning means, and (3) is character recognition by cutting out character by character and recognizing candidates. The character recognition means for outputting characters, (9) is, for example, “Error correction processing method of kanji recognition by candidate character complement and morphological analysis” (1985 National Institute of Electronics and Communication Engineers Information Systems Division S4-4) or “ Post-processing of character recognition using linguistic information and recognition information "
As shown in (The Institute of Electronics and Communication Engineers Research Institute material PRL82-76), punctuation marks or character transitions from, for example, hiragana to kanji are detected from recognition candidate characters, and collation with a word dictionary and grammatical connection verification of words are performed. Processing range determining means for determining the range to be performed, (4) is a word dictionary that stores words used in Japanese sentences and their grammatical attributes, and (5) is a range of the range determined by the processing range determining means (9). A word matching means that matches the recognition candidate characters for the character string with the word dictionary (4) and selects words existing in the word dictionary (4). A grammar dictionary storing connection information, (8) is a word connection verification means for selecting a word string that can be connected grammatically by referring to the grammar dictionary (7).

Unlike the languages such as English, Japanese sentences do not have the habit of writing words in word units, and in general, multiple sentences are written in solid form. Therefore, in the Japanese sentence reader, it is not efficient to process the sentence to be read, which is usually composed of multiple sentences, from the beginning to the end in one step because it increases the storage capacity and processing time. Unfortunately, it is necessary to divide the Japanese sentence to be read into appropriate lengths for processing.

The conventional Japanese sentence reading device is configured as described above, and from the recognition candidate characters of the characters constituting the Japanese sentence, the character positions determined as punctuation marks are used as delimiters, and further, for example, the characters determined as hiragana are determined as kanji. Such as turning into a letter
Japanese characters were read using the recognition candidate characters for the characters existing between the boundaries, with the transitions of the character types as the boundaries.

Fig. 3 shows an example of the input Japanese sentence printed or entered on the sticker (1) that is input to the Japanese sentence reading device and each character of the input Japanese sentence is recognized by the character recognizing means (3). This is an example of the recognition candidate characters that can be obtained. In the figure, (10) is the character string that constitutes the input Japanese sentence, “In the big city, there are many humans, but it is interesting.” (11) is the input Japanese sentence (10). It is a recognition candidate character string for each character string of.

FIG. 4 shows the processing range determined by the processing range determining means (9) in the conventional Japanese sentence reading device. In the figure, (12) to (15) are delimiter positions, and (16) to (19) are This is the processing range.

Figures 5 and 6 are examples of word dictionaries (4). In the figure, (20)
~ (59) are word items and grammatical attributes.

Figures 7, 8 and 9 are examples of grammar dictionaries (7), where the ◯ mark indicates that grammatical attributes can be connected, and the X mark indicates that grammatical attributes cannot be connected. Indicates that there is.

FIG. 10 shows recognition candidate characters and word dictionary for a character string within the processing range (16) determined by the delimiter (12) and the delimiter (13).
It is a word selected by collating (4).

FIG. 11 shows the result of selecting, by the word connection verification means (8), a grammatically connectable word for the word selected in the processing range (16) while referring to the grammar dictionary (7). is there.

FIG. 12 shows recognition candidate characters and word dictionary for a character string within the processing range (17) determined by the delimiter (13) and the delimiter (14).
It is a word selected by collating (4).

FIG. 13 shows recognition candidate characters and word dictionary for a character string within the processing range (18) determined by the delimiter (12) and the delimiter (14).
It is a word selected by collating (4). FIG. 14 shows the word connection verification means for the words selected in the processing range (18).
This is the result of selecting the items that can be connected grammatically while referring to the grammar dictionary (7) in (8).

FIG. 15 shows a recognition candidate character and a word dictionary for a character string within a processing range (19) determined by the delimiter (14) and the delimiter (15).
It is a word selected by collating (4).

Next, the operation will be described.

The input Japanese sentence (10) on the form (1) is scanned by the scanning means (2),
After performing preprocessing such as binarization with the character recognition means (3),
It is cut out character by character and recognized by a method such as pattern matching. Then, a plurality of recognition candidate characters (11) are output, which are ranked in order of increasing accuracy as input characters. In the processing range determination means (9), first input Japanese sentence (10)
A delimiter (12) is placed at the beginning of, and, for example, where the first position of the recognition candidate character changes from a character that is a punctuation mark to a character that is not, and from the character where the first position of the recognition candidate character is a hiragana character, If the places where the first character of the recognition candidate character changes to a character that is Kanji are set as the transitions of the character types, and a delimiter is set, the delimiter (13) is found first. The word collating means (5) collates the recognition candidate character for the character within the processing range (16) determined by the delimiter (12) and the delimiter (13) with the word dictionary (4) to input the Japanese sentence (10). "Large" as the word starting with the recognition candidate character for the first character of
(49), "Large city" (50), "Sentence" (57), "Student" (58),
"Ta" (47), "Taro" (48), "Tue" (40) From the second character, "City" (51) "City" (52), "Convenience" (53), "Part" ( Five
5), "section" (56), from the third letter, "meeting" (41),
"Go" (43), from the 4th character, "ha" (31), "ba" (32)
From the fifth character, "n" (39), "person" (44), and "on" (54) are selected, respectively. The word connection verification means (8) refers to the grammar dictionary (7) and conducts a connection verification between the words selected by the word matching means (5) to select a word string that enables connection. In this case, connection information (60) to (85) of the grammar dictionary (7)
Thus, the word string shown in FIG. 11 is selected as connectable.

Next, since the delimiter (14) is found, the word matching means (5) determines the processing range (17) determined by the delimiter (13) and the delimiter (14).
Matching the recognition candidate character for the character in 4 and the word dictionary (4), as the word starting with the recognition candidate character for the 6th character of the input Japanese sentence (10), "Ma" (42), "Q" (52) ), 7th
From the 8th character, "ga" (29) from the 8th character, "many" (4
6), from the 9th character to "i" (25), from the 10th character,
"Ga" (29), "ka" (28), from the 11th character, "," (2
Select 1), “to” (33), and “to” (22), respectively. In the word verification means (8), the last words of the processing range (17) are "person" (44) and "on" (54), and the first word of the processing range (17) is "ma" (42). ), "Question" (59) connection verification is performed, but by the connection information (65), (66), (87) to (90) of the grammar dictionary (7),
In both cases, the correct word "human" (45) is not selected because the connection is impossible, and the correct processing result cannot be obtained.

On the other hand, in the processing range determining means (9), for example, when the first sentence of the recognition candidate characters is a punctuation mark, the input character is determined to be a punctuation mark, and a delimiter is placed immediately after that.
First, since the delimiter (14) is found, the word matching means (5) determines the processing range (18) determined by the delimiter (12) and the delimiter (14).
The word candidates shown in FIG. 13 are selected by collating the recognition candidate character for the character in the word dictionary (4). In the word connection verification means (8), the word strings shown in FIG. 14 are selected as connectable by the connection information (60) to (119) of the grammar dictionary (7).

Up to this point, the delimiter (14) has been set correctly, so
The correct read result is included in the connectable word strings shown in FIG.

Next, since the delimiter (15) is found, the word matching means (5) matches the recognition candidate character for the character within the processing range (19) determined by the delimiter (14) and the delimiter (15) with the word dictionary (4). Then, the word shown in FIG. 15 is selected.

The word connection verification means (8) conducts a connection verification of the last word "," (21) of the processing range (18) and the first word "O" (26) of the processing range (19). . Connection information for grammar dictionary (7)
From (120) to (122), "," (21) and "O" (26) can be connected, so "O" (26) and the following "Momo" (35),
Perform a connection test with "mu" (34). However, the grammar dictionary
Which of the grammatical attributes "o" (26) has and the grammatical attributes "mo" (35) and "mu" (34) have according to the connection information (123) to (130) of (7) Since there is nothing that can be connected,
It is not processed correctly. This is because the delimiter (15) is set in the middle of the word "interesting" (27) in the word dictionary (4).

When such an abnormality occurs, it may be processed correctly by, for example, expanding the processing range in which the abnormality has occurred to the next processing range. In that case, further processing time is required. Becomes

[Problems to be Solved by the Invention] In a conventional Japanese sentence reading device, when a recognition candidate character is collated with a word dictionary, a processing range is determined by separating a portion determined to be a punctuation mark or a change in character type. Therefore, if a character in the middle of a word is mistaken for a punctuation mark, or if the character type is wrong, a delimiter is set in the middle of the word and a correct reading result cannot be obtained. . In addition, the conventional Japanese sentence reading device has a problem in that, for example, when a punctuation mark is used as a delimiter, the processing range becomes long, resulting in an increase in storage capacity and processing time required for processing.

The present invention has been made in order to solve the problems related to the present invention, and an object of the present invention is to provide a Japanese sentence reader capable of eliminating a decrease in reading accuracy due to an incorrect setting of a delimiter, and reducing a storage capacity and a processing time.

[Means for solving problems]

The Japanese sentence reading device according to the present invention collates the recognition candidate character output from the character recognition means with the word dictionary, detects that there is no word including the recognition candidate character for two consecutive characters, and detects the word between them. It is provided with a word string boundary detection means that assumes that there is a boundary between strings.

[Action]

The word string boundary detection means according to the present invention shortens the character string in the processing range without setting a delimiter of the word connection verification processing range in the middle of the word stored in the word dictionary.

Example of Invention

An embodiment of the present invention will be illustrated and described below. FIG. 1 is a block diagram showing an embodiment of the present invention. The same reference numerals as those in the conventional example shown in FIG. 2 indicate the same components, and the description thereof will be omitted. In the figure, (6) is a word string boundary detection means for detecting the boundaries of word strings.

FIG. 16, FIG. 17, and FIG. 18 are word string boundary detection means.
Recognition candidate characters and word dictionary within the word string boundary detected in (6)
This is the result of matching with (4), and in the figure, (131) to (142)
Is a word string boundary.

FIGS. 19, 20, and 21 are FIGS. 16 and 17, respectively.
Corresponding to the word string boundaries in FIGS.
In (8), this is the result of performing the word connection test with reference to the grammar dictionary (7).

FIG. 22 shows the processing results from the beginning to the end of the input Japanese sentence (10).

FIG. 23 shows the reading result when the sum of the ranks of the recognition candidate characters is used to uniquely determine.

Next, the operation will be described.

After the recognition candidate character (11) for the input Japanese sentence (10) shown in FIG. 3 is obtained by the character recognition means (3), first, the word matching means
In (5), the word dictionary (4) is matched. Words (49), (50), (57), (58), (47), (4 shown in FIG. 16 as words starting with the recognition candidate character for the first character of the input Japanese sentence (10)
8) and (40) are selected. Before selecting a word starting with a recognition candidate character for the second character, a word string boundary detection means
In (6), it is checked whether the first character and the second character form a word string boundary. That is, it is checked whether or not the word including the recognition candidate character for the first character and the second character is selected, and if no such word is selected, it is set as the word string boundary, and it must be so. For example, do not use word string boundaries. In the case of this example, "Big city" (5
Since there is the word 0), it is assumed that there is no word string boundary between the first character and the second character of the input Japanese sentence (10). Then, the word collating means (5) collates with the word dictionary (4) to select a word starting with a recognition candidate character for the second character of the input Japanese sentence (10). Then, the word (5
1), (52), (53), (55), (56) are selected. Similarly, before selecting the word starting with the recognition candidate character for the third character of the input Japanese sentence (10), the word string boundary detecting means is selected.
In (6), it is checked whether there is a word string boundary between the second character and the third character of the input Japanese sentence (10).

In the case of this example, the word "major city" (50) exists across the second and third characters, so it does not become a word string boundary. Therefore, the word matching means (5) (5) refers to the word dictionary (4) and selects words (41) and (43) starting with the recognition candidate character for the third character of the input Japanese sentence (10). To do.
Further, this time, before selecting a word starting with the recognition candidate character for the fourth character of the input Japanese sentence (10), the word string boundary detection means (6) uses the third and fourth characters of the input Japanese sentence. Check if there is a word string boundary between the eyes. In the case of this example, since there is no word that extends across the third and fourth characters, the third and fourth characters of the input Japanese sentence (10)
It is assumed that there is a word string boundary (132) between the letters. When the word string boundary detection means (6) finds the word string boundary (132), the word connection verification means (8) detects the word string boundary (131) that has been set in advance at the beginning of the input Japanese sentence (10). Word string boundary (13
The connection test is performed on the words selected between 2). The connection information (60) to (62) and (64) to (76) in the grammar dictionary (7) provides the word connection test result shown in FIG.

Next, the words (31), (3) shown in FIG. 17 are shown as words starting with the recognition candidate character for the fourth character of the input Japanese sentence (10).
2) is selected.

Similarly, the word string boundary detection means (6) is used to detect the word string boundary (13) shown in FIG. 17 between the fourth character and the fifth character.
3) was found, and when the word connection verification means (8) performed the word connection verification, the words (31) shown in FIG. 20 were obtained from the connection information (63), (77), (78) of the grammar dictionary (7). ) Is obtained as the word connection test result.

Thereafter, in order, the word string boundaries (133) to (13) shown in FIG.
7), the word string boundaries (138) to (142) shown in FIG. 18 were found, and the words (45), (29), (46), (25), (29), and (2) were found, respectively.
8), (21), (27), (25), and (20) are obtained as connection verification results.

When the input Japanese sentence (10) is processed to the end, the final result shown in FIG. 22 is obtained. In the case of this example, a plurality of word strings are obtained as candidates for the first to third characters of the input Japanese sentence (10) and the tenth character of the input Japanese sentence (10). In such a case, the device may be configured so that a human can select the correct Japanese sentence, or, for example, when there are multiple word strings between word string boundaries, each word string is configured. If the sum of the ranks of the recognition candidate characters is selected to be the smallest, the sum of the ranks of "Big City" (50) and "Large" (49) "City" (52) is 1
+ 1 + 1 = 3 is selected as the minimum, and similarly, "ga"
(29) is selected. In any case, the correct result shown in FIG. 23 is obtained as the reading result for the input Japanese sentence (10).

In addition, in the above-mentioned embodiment, a normal Japanese sentence is explained, but it is possible to read a solid address which is not written separately for each prefecture / city / ward, or a name which is not written separately for each name. It can also be used.

〔The invention's effect〕

As described above, according to the present invention, since the word string boundary detection means is provided so that the processing range is not set in the middle of the words registered in the word dictionary, the processing range can be set accurately and shortly. The effect is to obtain a Japanese sentence reader that requires less storage capacity and processing time for processing.

[Brief description of drawings]

FIG. 1 is a block diagram according to an embodiment of the present invention, FIG. 2 is a block diagram when a conventional method is used, FIG. 3 is a diagram showing an example of an input Japanese sentence and its recognition candidate character, FIG. 4 is a diagram showing the division and processing range in the conventional method, FIGS. 5 and 6 are diagrams showing examples of word dictionaries, and FIGS. 7, 8 and 9 are examples of grammar dictionaries. Figure, Figure 10, Figure 12, Figure 1
3 and 15 are diagrams showing an example in which word matching is performed on the processing range of the conventional device, and FIGS. 11 and 14 are word connection verifications performed in the processing range of the conventional device. FIG. 16, FIG. 17, FIG. 18, and FIG. 18 show examples of word matching performed by the apparatus according to the present invention.
Fig. 21 is a diagram showing an example of performing a word connection test with the device according to the present invention, and Figs. 22 and 23 are diagrams showing the results of reading Japanese sentences according to the present invention. Fig. 22 (3) Is a character recognition means, (4) is a word dictionary, (5)
Is a word matching means, (6) is a word string boundary detection means, (7) is a grammar dictionary, and (8) is a word connection verification means. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] 1. A Japanese sentence reader for recognizing and reading characters constituting a Japanese sentence, and a character recognizing means for selecting, for each input character, a plurality of recognition candidate characters ranked in order of high accuracy as input characters. A word dictionary that stores words used in Japanese sentences and grammatical attributes for each word,
A word existing in the word dictionary is searched from a grammar dictionary that stores grammatical connection information between the grammatical attributes of the words and a character string of any length that combines recognition candidate characters for the character strings that make up the Japanese sentence. The word matching means to be selected and one character from the Japanese sentence one by one from the front, and two consecutive characters from the next character.
Word string boundary detection means for defining a boundary between two consecutive characters as a boundary of a word string when a word of two or more characters including a character string obtained by combining respective recognition candidate characters for a character does not exist in the word dictionary. And a word connection verification means for selecting a grammatically connectable word string from consecutive words using the grammatical connection information in the grammar dictionary, and the word string detected by the word string boundary detection means. It is possible to select a plurality of candidate word strings for the character string delimited by boundaries using the word matching means, and use the word connection testing means for grammatical connection from the selected candidate word strings. A candidate word string is selected, and further, from the selected candidate word strings before and after the word string boundary, a word string that can connect to a Japanese sentence across the word string boundary is selected by using the word connection verification means. ， Selected single Text reader day, characterized by a read result columns for the input date body.