JPH0746373B2 - Word recognizer - Google Patents

Description

TECHNICAL FIELD The present invention relates to a word recognition device for recognizing a word from an input character string represented by a sequence of character codes, and particularly in a character recognition process, a voice recognition process, and the like. A word recognition device that efficiently recognizes words when word delimiters in an input character string are unknown and error characters are included in the input character string, or when each character that makes up the input character string is uncertain Is.

[Conventional technology]

In a conventional word recognition device, using a word dictionary in which the arrangement of each character code forming the recognition target word is associated with each of the words, the arrangement of the character code of the input character string and the character code of each word in the word dictionary The candidate words were extracted by sequentially comparing the sequences and evaluating the similarity between the input character string and the words. There is a drawback in performing such a comparison with respect to all words that the processing time becomes huge when the number of words to be registered is large. As a method for solving this, as shown in Japanese Patent Publication No. 57-29745 and Japanese Patent Publication No. 59-188783, a word table in which characters at one or more specific positions are classified as keys is used. There is known a method of executing a search for a word to be compared at high speed. For example, the word table is searched by the remaining characters excluding the characters rejected by the character recognition from the characters at a plurality of specific positions, the word is extracted, and the input character string is input to the extracted word. Are making a comparison.

[Problems to be solved by the invention]

In the conventional word recognition device that has been speeded up in this way, a word to be compared is searched with a small number of characters at a specific position as a key, and therefore when the key character includes many error characters or reject characters. Has a problem that it cannot correctly recognize a word without searching for a correct word. Further, when there are a large number of candidate characters for each character of the input character string, there is a problem that it takes a lot of time to compare the character codes of the searched word and the input character string.

An object of the present invention is to provide a character code and a word for each position of a character string even if the input character string includes many error characters or reject characters, or if there are many candidate characters for each character of the input character string. Using the word table in which the codes correspond to each other, the word code is extracted for each character at each position of the input character string, and the candidate word is selected according to the size of the evaluation value obtained by accumulating a predetermined judgment value for each extracted word code. An object of the present invention is to provide a word recognition device capable of extracting a candidate word with high accuracy and high speed by making a decision.

[Means for solving problems]

The word recognition device according to the present invention classifies the words into the used character types according to the positions of the arrangement of the character strings forming the word to be recognized that is set in advance, and for each position of the arrangement of the input character string, Selection processing means including a word selection table configured to output a word code designating a word including a character of the character code at the position when a character code forming the input character string is input, and a judgment value accumulating process. And a sort processing means, and the selection processing means selects and reads a word code corresponding to the character code of the input character string from the word selection table for each position of the arrangement of the input character string, and inputs the input character string. The judgment value set for each character and the read word code are output to the judgment value accumulation processing means, and output from the selection processing means by the judgment value accumulation processing means. Based on the word code, the judgment value is accumulated for each word to obtain an evaluation value, and the sorting processing means extracts the candidate word according to the magnitude relation of the evaluation value for each word output from the judgment value accumulating processing means. .

[Action]

In this invention, since the evaluation value indicating the similarity between the input character string and each word is obtained by the cumulative calculation of the judgment value for the word selected at each position of the arrangement of the character string, the error character Even if an input character string or a reject character is included in the input character string, word recognition processing is performed with high accuracy and at high speed.

〔Example〕

 Examples of the present invention will be described below.

The word recognition device according to the present invention, when an input character string is input, performs recognition processing for determining which word the input character string corresponds to from among preset words, and outputs a candidate word code as a recognition result. It has a function to do. Here, the maximum number of characters of a word is m, and the number of recognition target words is n.
Is. The configuration and the recognition processing procedure of the word recognition device of the present invention will be described below.

The word recognition device of FIG. 1 includes a selection processing unit 2 including a word selection table 1, a judgment value accumulation processing unit 3 including an evaluation value table 3a, and a sorting processing unit 4 including a candidate word table 4a.
Composed of and.

Here, a case will be described in which the character string "digital" is recognized as the input character string "decoral" by character recognition and the result is input as the input character string of the word recognition device of the first embodiment of the present invention. .

The selection processing means 2 performs the following selection processing for each position in the arrangement of the input character strings. First, the word selection table 1 is searched for the character code at each position of the input character string, and the word code corresponding to the character code is read. For example, the word code "digital", "data", and "decoder" are read with the first character code "de," and the word code "decoder" and "record" are read with the second character code "ko." , The determination value determined corresponding to the character position and the read word code are sequentially sent to the determination value accumulation processing means 3. in this case,
The word to be read is an arbitrary number of characters with an unlimited number of characters. In this manner, similar processing is sequentially performed up to the m-th character code or the length of the input character string. Note that Wn (n is 1, 2, ...) Indicates a word code, and indicates that a circled word code, for example, W11, W12, W13 is selected for the first character.

FIG. 2 shows an example of the table configuration for the second position of the character string in the word selection table 1. The search process of the word selection table 1 will be described with reference to FIG. The word selection table 1 can be composed of an address table 1a and a word table 1b as shown in FIG. In the word table 1b, words are classified and arranged by the character code at a predetermined character position (the second position in this example). Also the address table
1a shows the start address of the word table 1b in which a word including a character of the character code at a predetermined position is stored for each character code. For example, as shown in FIG. 1, when "K" is input as the character code at the second position, the addresses A50 and A51 are read from the address table 1a and the addresses A50 to A51-1 on the word table 1b are read. The necessary word code can be read by reading the word code. By such a search process, the selection process can be performed at high speed.

The judgment value accumulation processing means 3 accumulates judgment values on the evaluation value table 3a according to the word code read by the selection processing means 2 to calculate an evaluation value. Here, it is assumed that the determination values are all “1”.

FIG. 3 is a configuration diagram of the evaluation value table 3a. "0" is written in advance from V1 to Vn of the evaluation value table 3a in FIG. Next, the judgment value is accumulated in Vi corresponding to the word code Wi (i is 1, 2, ...) Read by the selection processing means 2. For example, the word code W11 "digital" is read by the processing of the first character, the third character, and the fourth character in the selection processing means 2, so the word code W1 is displayed on the evaluation value table 3a.
The judgment value "1" is accumulated three times for one "digital", and the evaluation value "3" is obtained. In this way, by accumulating the judgment value for each word code, the evaluation value representing the similarity between the input character string and each word is calculated, and the merging process of the word codes selected at each position is performed at the same time. As a result, high-speed processing can be realized.

In the next sort processing, the sort processing means 4 sorts the word codes on the candidate word table 4a described later in order from the one having the largest evaluation value given by the judgment value accumulation processing means 3.

FIG. 4 is a block diagram of the candidate word table 4a. When the rearrangement is completed, the candidate words from the first candidate word having the largest evaluation value to the predetermined rank are output to complete the word recognition processing for one input character string. As a result, "digital" is output as the first candidate word.

In this way, the incorrect character ("ko") in the input string
The correct word "digital" is recognized even if it contains "." In the word recognition device of the present invention, since the selection processing is performed for all positions of the character string, it is possible to correctly recognize the word regardless of the position of an erroneous character or a reject character. Further, by using the certainty of each character of the input character string calculated from the character recognition result or the like as the determination value, more highly accurate word recognition can be performed. For example, it is assumed that the first character, the third character, and the fourth character have a determination value of "1.0" and the second character has a determination value of "0.8". At this time, even if the word "recotal" is used, the evaluation value of "digital" is "3.0" and the evaluation value of "recotal" is "2.8".
And "digital" is the first candidate word.

Next, a second embodiment of the present invention will be described. The word recognition device of the second embodiment has the block structure shown in FIG. 1 as in the first embodiment, but the selection processing means 2
Is characterized in that a plurality of candidate characters are input for the character at each position of the input character string.

FIG. 5 shows an example of an input character string having a plurality of candidate characters obtained from the character recognition result and the like. In the word recognition device according to the second embodiment, the selection processing means 2 performs the selection process for each candidate character at each position of the input character string. As the determination value, a value weighted by the candidate rank of each candidate character or the likelihood of each character of the input character string calculated from the character recognition result or the like is used.

FIG. 6 shows an example of the judgment value of each candidate character given in this way. For example, in the selection processing means 2, in the processing of the first character, the word codes "mental" and "mechanism" are read out for the character code "me", and a judgment value of 1.0 is given. Furthermore, the word code "digital", "data", and "decoder" are read for the character code "de", and a judgment value of 0.8 is given. These word codes and judgment values are sequentially sent to the judgment value accumulation processing means 3. In this way, the selection process is sequentially performed up to the second candidate character “ri” of the fourth character. The judgment value accumulating processing means 3 accumulates judgment values to obtain the evaluation values shown in FIG. As a result, the word code “digital” with an evaluation value of 3.3 is obtained as the first-ranked candidate word. Thus, 1 in the 1st place candidate character
Even if only the correct answer character exists, the correct word is recognized by including the correct answer character in the second-ranked candidate characters.

In order to perform such processing, conventionally, for all combinations of character strings of first-ranked candidates and second-ranked candidates, words that match these character strings are searched and evaluated for each searched word. It was done by calculating the value. In the word recognition device according to the second embodiment of the present invention, a high-speed selection process using a word selection table 1 in which a character code and a word code of a word including the character code at predetermined positions are associated with each other, and a merge process. By the judgment value accumulating process that simultaneously calculates the evaluation value and the evaluation value, even if the character at each position of the input character string is a plurality of candidate characters, it can be performed at high speed.

Next, a third embodiment of the present invention will be described. The word recognition device according to the third embodiment has the block structure shown in FIG. 1 as in the first and second embodiments. However, in the selection processing means 2, the character at each position of the input character string and the characters before and after it. It is characterized in that one character at each position is inputted as a candidate character at each position.

FIG. 8 shows each position when "digital" is recognized as "digital, digital" due to an error in the character cutting process in the character recognition process and the result is input as an input character string of the word recognition device of the present invention. The example of the candidate character of and the example of a judgment value are shown. For example, for the third position, the character "," at that position is a candidate character with a judgment value of "1.0", and the characters "Si" and "Ta" at the preceding and following positions are candidate characters with a judgment value of 0.8, respectively. . When an input character string consisting of such candidate characters is input to the word recognition device of the present invention, an evaluation value as shown in FIG. 9 is obtained for each word. As a result, "digital" with an evaluation value of 2.6 becomes the first candidate word.

FIG. 10 shows a case where “digital” is recognized as “deshioraru” due to an error in the character cutting process in the character recognition process, and the result is input as an input character string of the word recognition device of the present invention. The example of the candidate character of each position and the example of the determination value are shown. In this case, in the word recognition device of the present invention,
An evaluation value as shown in FIG. 11 is obtained for each word.
As a result, "digital" with the evaluation value "2.6" becomes the first candidate word.

As described above, the candidate character at each position is the character at each position of the input character string and the character at each position one position before and after each position, so that two certain characters in the character string of the correct word become one character. It is possible to correctly recognize a word by the word recognition device according to the third embodiment of the present invention even if the word is converted into a character or a certain character is missing.

FIG. 12 is a block diagram showing a specific configuration of the word recognition device shown in FIG. In the figure, 5 is a processor, 6 is a program memory, 7 is a general-purpose memory, 8 is a 2-port memory, 9 is an input bus, 10 is a connection bus, 11 is an output bus, 12 is an address bus, 13 is a data bus, and 14 is a A program address line and 15 are program read lines.

FIG. 12 illustrates the configuration and function of each means in the word recognition device.

The selection processing means 2 comprises a word selection table 1, a processor 5 for executing selection processing described later, and a program memory 6 storing a control program for the selection processing. The processor 5 is configured for high-speed control, and is independently connected to the program memory 6 by a program address line 14 and a program read line 15, and to the general-purpose memory 7 by an address bus 12 and a data bus 13. The processor 5 is also connected to the word selection table 1 by an address bus 12 and a data bus 13, which are common. As shown in FIG. 2, the word selection table 1 stores the word code of the word including the character code at a predetermined position corresponding to each position of the character string and each character code. With the above configuration, the judgment value accumulation processing means 3 selects and reads a part of the word code from the word selection table 1 for each position of the input character string according to the character code of each position of the input character string, and the reading process. It implements the function of outputting the issued word code and the judgment value set for each character of the input character string.

The judgment value accumulation processing means 3 includes a processor 5 for executing judgment value accumulation processing, which will be described later, a program memory 6 storing a control program for the judgment value accumulation processing, and a two-port storing evaluation value table 3a shown in FIG. It comprises a memory 8. The two-port memory 8 and the processor 5 include two sets of addresses 12a and 12b of the port a and the port b and the data bus 1
It is connected by 3a and 13b. As a result, in the 2-port memory 8, the read operation from the port a and the write operation to the port b can be executed in parallel with different address designations.
The processor 5 of the judgment value accumulation processing means 3 and the processor 5 of the selection processing means 2 are connected by a connection bus 10. With the above configuration, the judgment value accumulation processing means 3 accumulates judgment values for each word according to the word code read from the word selection table 1 for each position of the input character string by the selection processing means 2 to obtain an evaluation value. It realizes a cumulative processing function. The reading of the word code in the selection process and the accumulation of the judgment value in the judgment value accumulation process can be executed by the pipeline process.

The sort processing means 4 includes a processor 5 for executing sort processing described later, a program memory 6 storing a control program for the sort processing, and a candidate word table shown in FIG.
It comprises a general-purpose memory 7 storing 4a. The processor 5 of the sort processing means 4 and the processor 5 of the judgment value accumulation processing means 3 are connected by a connection bus 10.
The contents of the candidate word table 4a in the general-purpose memory 7 can be output to the outside via the processor 5 and the output bus 11. With the above configuration, the sort processing means 4 is configured to make the judgment value accumulation processing means 3
It realizes the function of extracting candidate words based on the magnitude relation of the evaluation value for each word output from.

In the embodiment of the present invention described above, the device configuration of FIG. 12 is illustrated, but the selection processing means 2, the judgment value accumulation processing means 3, the processor 5 constituting the sorting processing means 4 and the various memories 6 to 8 are A commercially available LSI or the like can be used. Regarding the function sharing among the respective means, the connection relation, and the block configuration of each means, in addition to the configuration of FIG. 12, the word selection table 1 of the selection processing means 2 and the program memory 6 are stored on the common general-purpose memory 7. Various modifications such that the processor 5 of the selection processing means 2 and the processor 5 of the judgment value accumulation processing means 3 are mounted in common, or the 2-port memory 8 of the judgment value accumulation processing means 3 is replaced with a general-purpose memory 7. Different configurations are possible.

Further, in the word recognition device of the present invention, even if all of the input character strings are not prepared, the selection process and the judgment value accumulation process can be sequentially performed from the characters output by the character recognition process,
The character recognition processing and the selection processing and judgment value accumulation processing in the word recognition device of the present invention can be pipelined.

Next, a fourth embodiment of the word recognition device of the present invention will be described. FIG. 13 is a block diagram showing the fourth embodiment of the word recognition device of the present invention. In the figure, 16 is an input character string extracting means. The input character string extraction means 16 is composed of a general-purpose processor or a general-purpose memory 7. It can also be shared with the processor 5 and the program memory 6 used in the selection processing means 2. The fourth embodiment utilizes the fact that a word can be recognized even if the number of characters of the word in the input character string is unknown, and the input to be processed next is calculated from the number of characters of the candidate word output from the sort processing means 4. The input character string extraction means 16 is added to the first embodiment so that the leading character position of the character string is determined and the character string from the leading character position is output to the selection processing means 2 as a new input character string. It is characterized by that. For example, the word "digital",
A case where the input character string “digital image processing” is input when “image processing” is registered in the word selection table 1 will be described. Initially, "digital image processing" is the input character string to the selection processing means 2. As a result, the word "digital" is output from the sort processing means 4 as a first-ranked candidate and sent to the input character string extraction means 16. In the input character string extraction means 16, since the word "digital" has four characters, the fifth character is set as the first character position of the input character string to be processed next. Then, "image processing" is sent to the selection processing means 2 as a new input character string. As a result, the word "process" is output from the sort processing means 4 as the first-ranked candidate.
In this way, in the fourth embodiment of the word recognition device of the present invention, the input character string in which a plurality of words of unknown character number are arranged by repeating the processing through the input character string extracting means 16 Can be recognized correctly.

The evaluation means will be described.

[Step 1] For each character position in the input character string, a judgment value is given to the word read from the word list.

For example, the following example is given as a method of giving a determination value.

I. The point “1” is fixedly attached to the extracted word,
This is used as the judgment value (FIG. 1 and its description).

B. The judgment value assigned to each weighted character candidate is used to judge the word extracted for that character candidate, so that it is effective when a plurality of character recognition results are extracted for a certain character position. Value (see Figures 5 and 6 and its description).

[Step 2] The judgment values for each word extracted for each character position are totaled and used as the evaluation value.

In the present invention, for example, the word having the second character "ko" includes the word "record" of four characters and the word "octopus" of two characters.

On the other hand, in the prior art as disclosed in Japanese Patent Laid-Open No. 58-39377, for example, if the input character is determined to be 4 characters, "record" is read as the word to be evaluated.
The letter "octopus" is not the only target.

Further, in the above-mentioned conventional technique, when a character string having a number of characters (n) different from the number of characters (m) to be originally input is input, the word dictionary uses a dictionary composed of n characters,
The word consisting of m characters cannot be extracted, and there is a limit to the matching.

There is one character at each character position (r =
In the case of 1), in the present invention, since a plurality of words are extracted for each character position, the word candidate is not specified. On the other hand, in the conventional technique, one word candidate (or 0, There is no corresponding word in the word list).

That is, even if there is a typographical error in the input character string, the present invention is superior in terms of accuracy from the viewpoint that the word candidates can be extracted.

Application to Compound Words In the present invention, since the character string length (number of characters) n is not limited, short unit words can be registered as a word dictionary and used for recognition after compounding.

For example, "digital", "calculator", "analog",
When the "signal" is registered, the following processing can be performed.

a. "Decotal calculator" is entered.

b. “Digital” is recognized. The total number of words to be processed = 3 c. Since this character string length is 4, recognition is performed with 5 or more characters as a new input word.

d. As a result, "computer" is recognized. Total number of processing target words = 3 The calculation amount will be described.

In the present invention, a word candidate is obtained for each character, the judgment values are totaled, and this is used as the evaluation value.

If the number of words in the j-th candidate category at each character position i is mij (for example, m ₁₁ = the number of words having “de” in the first character), If mij = m (the same number, the number of words that include a certain character in a specific position), the number of points added = n × r × m (for the total number of words, assuming that a buffer for accumulation is prepared, Therefore, the present invention can be applied to high-speed processing.

〔The invention's effect〕

As described above, the word recognition device of the present invention classifies the words into the used character types according to the positions of the arrangement of the character strings forming the word to be recognized set in advance, and Selection processing means including a word selection table configured to output a word code designating a word including a character of the character code at a position when a character code forming an input character string is input for each position; Since it is composed of the value accumulating processing means and the sorting processing means, a word selection table configured by classifying words in advance at each position of the character string by the character code of each position is used, and at each position of the input character string. By performing the cumulative calculation of judgment values for some of the selected word codes, an evaluation value that indicates the similarity between the input character string and the character strings of all words can be obtained. Character is also contained in the input string, it is possible to perform word recognition processing with high accuracy and high speed.

In addition, in the present invention, since a plurality of words are extracted for each character position, word candidates are not specified, and even if there is a typographical error in the input character string, the rate of extracting word candidates is high. .

Further, this word recognition device treats the code string as an input character string if the recognition result is represented by a sequence of codes, so that not only the character string of the character recognition result but also various character recognition devices and image recognition devices can be used. The present invention can be applied to a device, a voice recognition device, etc., and these devices can be realized with high accuracy and high speed.

[Brief description of drawings]

FIG. 1 is a diagram showing a device configuration of a first embodiment of the present invention,
2 to 4 are block diagrams of the table used in the apparatus of FIG. 1, and FIGS. 5 to 7 explain the processing when the character at each position of the input character string is a plurality of candidate characters. Figure, 8th
FIG. 11 to FIG. 11 are views for explaining the processing when the candidate character at each position of the input character string is the character at each position of the input character string and the character at each position before and after that, and FIG. FIG. 13 is a block diagram of the apparatus shown in FIG. 13. FIG. 13 is a diagram showing the apparatus configuration of the second embodiment of the present invention. In the figure, 1 is a word selection table, 2 is selection processing means, 3 is judgment value accumulation processing means, 4 is sorting processing means, 5
Is a processor, 6 is a program memory, 7 is a general-purpose memory, 8 is a 2-port memory, 9 is an input bus, 10 is a connection bus, 11 is an output bus, 12 is an address bus, 13 is a data bus, 14 is a program address line, Reference numeral 15 is a program read line, and 16 is an input character extracting means.

Claims (6)

[Claims] 1. A recognition target for which an input character string is preset by inputting an input character string represented by a sequence of character codes and outputting a predetermined number of word candidates corresponding to the input character string. When a character position in a character string and a character code at the character position are specified, the character of the character code is set to the character position. A word selection table that outputs a word code that is associated with a word that the user has, and a selection processing unit that gives a judgment value to each word code output from the word selection table for each character position of the input character string and outputs the word code. , A judgment value accumulation processing means for totaling and outputting the judgment values for each word code, and a predetermined number of word codes selected and outputted according to the evaluation value of the output of the judgment value accumulation processing means Word recognition device characterized by having a sorting unit. 2. The word recognition device according to claim 1, wherein the determination value given to the word code is the number of output of the word code for the input character string. 3. The character at each position of the arrangement of the input character string is a plurality of candidate characters, and the selection processing means selects each of the candidate characters from the word selection table for each position of the arrangement of the input character string. The word code corresponding to the character code is selected and read out, and the judgment value set for each candidate character of the input character string and the read word code are output to the judgment value accumulating processing means, and judgment is made. The evaluation value is obtained by accumulating the determination value for each word based on the word code output from the selection processing means by the value accumulation processing means to obtain an evaluation value. The word recognition device according to item (2). 4. The determination value is a scale indicating the certainty of each character obtained from the recognition result of each character of the input character string, and the determination value is a scale (1), (2) ) Or the word recognition device according to item (3). 5. The plurality of candidate characters at each position are the character at each position and each character at one or more positions before and after each position. ) The word recognition device described in the section. 6. The selection processing means determines the starting character position of the input character string to be processed next from the number of characters of the candidate words output from the sorting means, and the character string from the starting character position is newly input as the input character string. The word recognition device according to claim (1) or (2), further comprising an input character string extracting means for outputting as.