JPS60229184A - Word processor for inputting undifined information - Google Patents

Abstract

PURPOSE:To improve the recognition efficiency of a character or a word inputted as a handwritten character or a voice by displaying the character on the basis of statistical information obtained from a matrix stored in a storage device in addition to a recognized result from a recognition device. CONSTITUTION:If a character K is applied to the recognition device 11, the device 11 executes recognition processing and sends five characters K1-K5 to a control device 13 in the similar order to the inputted character. The device 13 displays the character having the highest probability on the basis of the statistical information obtained from the matrix stored in the storage device 14 by a program in a control memory 12. Namely, the character or word having the highest probability is displayed on the basis of the actually inputted undifined information and the statistical information obtained by a computation matrix obtained by expressing a corresponding table between the input information and the recognized result by the device 11 as a matrix. Consequently, the recognition efficiency of a character or a word inputted as a handwritten character or a voice can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a word processor that receives amorphous information such as handwritten characters or speech as input.

[Prior art]

Conventional 1 For example, a word processor that uses a handwritten character input device has a recognition device that recognizes input handwritten characters in real time, a control memory that stores the entire program, and overall control according to the program in the control memory. It consists of a control device (central processing unit), a display device that displays characters of the processing results, etc. If the letter k is entered,
The recognition device selects the characters closest to the input character k in order of likelihood, 2. 3. 4. 5 (assumed to be the top five here) are sent to the control device.

The control device sends only the most probable character 1 to the display device according to the program in the control memory, stores the remaining characters 2 and 5 in the control memory, and uses them as the next candidate character for correction etc. The characters 2 to 2 are sent to the display device only when there is a request. The display device displays 4 or 2 to 5 to the sent characters. Here, a method has been described in which 5 in 2 to 5 is displayed as the next candidate character only when requested, but there is also a method in which 5 in k7 to 5 is displayed at the same time as □.

Note that the same applies when inputting single syllables using a voice input device.

However, this type of conventional word processor has the following drawbacks.

(1) Since there is no feedback as to whether or not the recognition result is really correct, the recognition rate is limited to a value specific to the recognition device and cannot be improved any further.

(2) Because the documents, number of strokes, or sounds differ for each individual,
The recognition rate varies greatly depending on the user.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide an amorphous information input word processor that solves the above-mentioned conventional drawbacks and improves the recognition rate of characters or words input by hand or by voice.

[Summary of the invention]

The present invention stores, in the form of a matrix (referred to as a confusion matrix), a correspondence table between amorphous information such as letters or words actually input by hand or voice and the recognition results of the input information by a recognition device. A storage device is provided to display the characters or words with the highest probability of correct answer based not only on the recognition results from the recognition device but also on statistical information obtained from the confusion matrix stored in the storage device. This is what I decided to do.

[Embodiments of the invention]

An embodiment of the present invention will be described below using a word processor using a handwritten character input device as an example.

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, in which a recognition device 11.1 recognizes in real time F characters input from a handwritten character input device (not shown). Control memory 12 that stores the entire program. A control device 13 that performs overall control according to a program in a control memory 12 stores a correspondence table between actually input characters and recognition results of the input characters by the recognition device 11 in the form of a matrix (confusion matrix). It is composed of a storage device 14 for storing characters, and a display device 15 for displaying characters of the processing results. FIG. 2 shows the contents of the storage device 14, ie, the confusion matrix. In Fig. 2, n, , indicates the recognition device 11 when the character i is input.
indicates the number of times that the character j was output as a recognition result.

FIG. 3 shows the recognition/display processing flow of this word processor, and FIG. 4 shows the automatic creation processing flow of the confusion matrix niJ. The operation of this embodiment will be explained below with reference to FIGS. 3 and 4.

First, the recognition/display processing will be explained with reference to FIG.

Although the explanation will be limited to the input of katakana characters for i11'-, the process is exactly the same even if kanji characters are input. now.

Assume that the character rkJ is given to the recognition device 11. The recognition device 11 performs recognition processing, and selects the characters rk, .

Five characters 1, 2, 5J (assumed to be the top 5 candidates here) are sent to the control device 13 (step 101).

The control device 13 uses the confusion matrix niJ in the storage device 14 according to the program in the control memory 12,
Probability P that the recognition device 11 outputs the character j as a recognition result when the character i is input.

1J Pl, := (+) Calculate in advance. This process is performed only once, only when document creation is started.

Next, the control device I3 controls P, under the condition of J''kr.

Search for Pik, (+ = A, I, Tsu,...
Pi, k,
)Pi7L >Pia )z >Pia lz >P
rIll I, + 131 that satisfies the + condition for i5
Set t4+15J as an input character candidate (step 10
2). Furthermore, among the candidate characters 11 to i5, the control device 13 sends only 11 with the highest probability as an input character to the display device 15 (Step 103), and the remaining candidate characters 1
7 to 15 are stored in the control memory 12 (step 1).
04) When there is a request to display candidate characters from the user of this word processor, the characters 12 to i are read from the control memory 12 and sent to the display device 15. The display device 15 displays the character i or i2 sent from the control device 13.
~i5 is displayed. Next, the control device 13 determines whether it is the last character (step 105), and if it is not the last character.

The process from step 101 is repeated.

Next, according to FIG. 4, the confusion matrix n.

The automatic creation process will be explained. Creating a confusion matrix in the storage device 14 in advance and using its values fixedly has a certain effect, but the effect is dramatically improved by automatically creating a confusion matrix. do.

As mentioned above, when the character rkJ is input, the control device 13 displays 11, which is the most likely input character, on the display device 1.
Send to 5. At this time, the control device 13 checks whether there is a request for the next candidate character (step 201), and if there is a request for the next candidate character, the displayed character iI is Then, the characters 12 to i5 after the next candidate are displayed (step 202), and if a selection instruction is given as to which of the characters i2 to i5 is correct, it is assumed that there has been a correction (step 203). When there is a correction and 1, for example, 12 is selected, the control device 13 determines that +(k, is the closest to the character input by the recognition device 11 in n12 of the confusion matrix in the storage device 14). , add 1 to the value of (
Step 204). On the other hand, if there is no request for the next candidate character and there is no correction to the recognition result, 1 is added to the value of ni in the confusion matrix in the storage device 14 (step 205). After that, the control device 13 checks whether or not there is a possibility that the value of the confluence matrix niJ will overflow (step 206). If there is no possibility, the process ends as is, but if there is a possibility of overflow, After normalizing nij by dividing it by a predetermined constant (step 207), the process ends.

As mentioned above, the confusion matrix is 1 in this word processor.
Although we have explained that there are only two word processors, it is extremely difficult to add an external storage device such as a floppy disk device to one word processor, and configure it so that an individual's confusion matrix can be loaded into the storage device 14 before using the word processor. It's easy.

Furthermore, although a handwritten character input device is used in the embodiment 7, the present invention is also applicable to a voice input word processor that uses a voice input device to input sentences monosyllable by monosyllable.

〔Effect of the invention〕

As explained above, the word processor according to the present invention has
Since the word processor is able to view and display not only the recognition results of the recognition device but also the results of subsequent expansion of the confusion matrix that is automatically updated while the word processor is in use, there are the following advantages.

(1) The more you use this word processor, the better the recognition rate will be.

(2) If the confusion matrix is used by replacing it with individual correspondence, a higher recognition rate can be obtained than when a common confusion matrix is used, and a word processor that takes advantage of the habits of five individuals can be obtained.

[Brief explanation of the drawing]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, and FIG.
FIG. 3 is a flowchart showing the recognition/display processing of the word processor, and FIG. 4 is a flowchart showing the automatic creation processing of the confusion matrix. DESCRIPTION OF SYMBOLS 11... Recognition device, 12... Control memory, 13... Control device, 14... Storage device, 15... Display device. 2, t, ”:a> Figure 2

Claims (2)

[Claims] (1) Input amorphous information such as handwritten characters or audio 1
A recognition device that recognizes characters or words in real time; A storage device that stores data in the form of
The system is equipped with a control memory that stores programs for processing, a control device that performs processing and control according to the program in the control memory, and a display device that displays characters or words as a result of processing by the control device. When fixed information is input, the control device searches the statistical values of the confusion matrix stored in the storage device based on the recognition result of the recognition device, and selects the character or character with the highest probability of correct answer. An amorphous information input word processor characterized by displaying words on the display device. (2) The amorphous information input word processor according to claim 1, wherein the confusion matrix stored in the storage device can be rewritten by each user using an external storage device.