JP2023112400A - Input device, method for controlling input device, and information processing device - Google Patents

Abstract

To provide an input device, a method for controlling the input device, and an information processing device that accurately detect wrong key pressing.SOLUTION: An input device 10 includes: a keyboard 110 having a sensor 110a that senses the manner in which a key is pressed; a control unit 120 including a correction detection unit 121 that detects a correction operation that is an operation for correcting input contents with the keyboard 110, an operation information acquisition unit 122 that acquires before-correction operation information including the manner in which the key is pressed and information about the key pressed before the detection of the correction operation, and after-correction operation information that is information about the key pressed after the detection of the correction operation, and a model generation unit 123 that generates a learned model 131 for inferring erroneous input by using the before-correction operation information and the after-correction operation information; a storage unit 130 that stores the learned model; and a display unit 140.SELECTED DRAWING: Figure 2

Description

The present invention relates to an input device, an input device control method, and an information processing device.

Typing speed affects work efficiency in desk work. Mistyping of keyboard keys interrupts input and causes loss of time for correcting characters. In Patent Document 1, for incorrect character string input, a character string before correction and a character string after correction are associated with each other and stored in a memory. A technique for automatically correcting a character string to a corrected character string is disclosed.

JP-A-2003-308316

However, if a character string before correction is subject to correction due to a predetermined number of mistakes, even a meaningful character string that the user intentionally entered may be erroneously and automatically corrected.

An object of the present invention, in one aspect, is to provide a technique for accurately detecting keystroke errors.

In order to achieve the above objects, the present invention employs the following configurations.

A first aspect of the present disclosure includes a keyboard having a sensor that detects how a key is pressed, a correction detection unit that detects a correction operation that is an operation for correcting input content with the keyboard, and a correction operation before detecting the correction operation. an operation information acquisition unit for acquiring pre-correction operation information including information on pressed keys and how to press keys, and post-correction operation information including information on keys pressed after detection of correction operations; using the pre-correction operation information and the post-correction operation information. By learning not only key information but also operation information including how to physically press the keys, the input device can generate a trained model that accurately infers keystroke errors.

The trained model is trained to output correction candidates when inputting operation information including pressed key information and how to press keys, and if the character string corresponding to the key information is incorrectly entered. model. The input device can determine erroneous input and acquire correction candidates by inputting the character strings input by the user and the amount of pressing of the key corresponding to each character to the trained model.

The generation unit may generate a learned model using pre-correction operation information and post-correction operation information by a specific user. The input device can generate a trained model corresponding to the user's peculiar typographical habits by learning information about the incorrect input of a specific user.

The generation unit may generate the learned model using pre-correction operation information and post-correction operation information by a plurality of users. The input device can generate a trained model corresponding to common mistakes by learning various user input error information without distinguishing between users.

The correction operation may be an operation of pressing the Backspace key or the Delete key. The input device can detect that the user has performed an operation to correct the input content by pressing a specific key.

The keyboard may include switches inside the keys, and the sensor may measure the amount or pressure of the switches. The input device can detect how a key is pressed by measuring the amount or pressure of pressing the switch, and can generate a trained model that infers an erroneous input in consideration of the physical way the key is pressed.

Each key of the keyboard may include a plurality of switches, and the operation information acquiring section may acquire the pressing positions of the keys based on the pressing order of the plurality of switches or the pressing amount of the switches. By equipping each key with a plurality of switches, the keyboard can acquire information about which part of the key the user has pressed.

An input device inputs operation information including a keyboard having a sensor for detecting how to press a key and information about the pressed key and how to press the key to a trained model for inferring an erroneous input. A determination unit for determining whether or not a corresponding character string is an input error, an acquisition unit for acquiring correction candidates for the character string determined to be an input error, and a display unit for presenting the correction candidates. may The input device can accurately detect erroneous input and present appropriate correction candidates to the user.

The acquisition unit may acquire the correction candidate using a trained model trained to output the correction candidate when the operation information is input. The input device can acquire appropriate correction candidates by using a trained model learned by machine learning.

Pre-correction operation information including information on keys pressed before detection of a correction operation and how to press the keys, and post-correction operation information that is information on keys pressed after the detection of the correction operation are stored in association with each other. Further comprising a storage unit, the acquisition unit acquires from the storage unit character strings of post-correction operation information associated with pre-correction operation information for the same character string as the character string determined to be an erroneous input, as correction candidates. You may The input device can acquire character strings that have been actually corrected by the user in the past as correction candidates.

The display unit may present correction candidates together with the character string determined to be an incorrect input. By presenting the correction candidate together with the input content, the user can confirm the correction candidate and then determine whether or not to correct the input character string.

The acquisition unit may acquire a plurality of correction candidates, and the display unit may present the plurality of correction candidates so that the user can select them. By presenting a plurality of correction candidates, the user can easily correct the intended input content.

The display unit may delete the character string determined to be an erroneous input and present the correction candidate. By directly correcting a character string determined to be an erroneous input to a correction candidate, the user can reduce the trouble of correction.

The display unit may highlight the corrected portion in at least one of the character string determined to be an erroneous input and the correction candidate. Highlighting is, for example, changing the character color or adding a marker. By highlighting the corrected part, the user can easily grasp the part where the user made a typo.

The keyboard may disable the user's operation for a predetermined period of time after the determining unit determines that the character string has been input incorrectly. The user can avoid continuing input in a state where there is an erroneous input, and can reduce the trouble of deleting the input content before noticing the error.

The key information may include information on at least one of an input character string and key arrangement. The character string is not limited to a Japanese character string, and may be a character string including alphanumeric characters, symbols, and the like. By using a learned model that has learned the information of the input character string and the layout of the keys, the input device can accurately detect user's typing mistakes.

The key pressing method may include information on at least one of the key pressing amount, the key pressing position, and the key input time interval. By using a trained model that has learned how to physically press keys, the input device can detect erroneous inputs in consideration of the habits of how to physically press keys.

The key information may include word-by-word or clause-by-segment character string information extracted by parsing. By using a trained model that has learned meaningful character strings in units of words or clauses, the input device can detect erroneous input in units of properly divided character strings.

The input device may further include a timer that measures time intervals between key inputs. By measuring the time interval between key inputs using a timer and generating a learned model that has learned the time interval between key inputs, the input device can accurately detect erroneous inputs in which the input order is changed.

A second aspect of the present invention includes a keyboard having a sensor that detects how a key is pressed, a correction detection unit that detects a correction operation that is an operation for correcting input content by the keyboard, and a correction operation before detecting the correction operation. an operation information acquisition unit for acquiring pre-correction operation information including information on pressed keys and how to press keys, and post-correction operation information including information on keys pressed after detection of correction operations; using the pre-correction operation information and the post-correction operation information, and the operation information including the pressed key information and the key pressing method are input to the learned model, A determination unit that determines whether or not the character string corresponding to the key information is an erroneous input, an acquisition unit that acquires correction candidates for the character string that is determined to be an erroneous input, and a display unit that presents the correction candidates. and an information processing apparatus.

A third aspect of the present invention is a control method for an input device having a keyboard having a sensor for detecting how a key is pressed, comprising a correction detection step of detecting a correction operation, which is an operation for correcting input content through the keyboard. , pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected. and a generating step of generating a trained model for inferring an erroneous input using the pre-correction manipulation information and the post-correction manipulation information.

Also, in a control method of an input device having a keyboard having a sensor for detecting how to press a key, operation information including information on the pressed key and how to press the key is applied to a trained model for inferring an erroneous input. A judgment step of judging whether or not a character string input and corresponding to key information is an erroneous input; an acquisition step of acquiring a correction candidate for the character string judged to be an erroneous input; and presenting the correction candidate. and a display step for controlling the input device.

The present invention can also be regarded as a program for realizing such a method or a recording medium on which the program is non-temporarily recorded. It should be noted that each of the means and processes described above can be combined with each other as much as possible to constitute the present invention.

According to the present invention, it is possible to accurately detect an erroneous keystroke.

FIG. 1 is a diagram illustrating an application example of an input device. FIG. 2 is a block diagram illustrating the configuration of the input device. FIG. 3 is a diagram illustrating a switch that detects pressing of a key. FIG. 4 is a flowchart illustrating learning processing. FIG. 5 is a flowchart illustrating inference processing. FIG. 6 is a diagram illustrating a plurality of switches for detecting key presses. 7A and 7B are diagrams for explaining acquisition of push amounts of a plurality of switches. FIG. FIG. 8 is a diagram for explaining an erroneous input in which the order of keys to be input is reversed. FIG. 9 is a block diagram illustrating the configuration of an input device according to Modification 3. As shown in FIG.

Hereinafter, embodiments according to one aspect of the present invention will be described based on the drawings.

<Application example>
FIG. 1 is a diagram illustrating an application example of an input device. The input device includes a keyboard or the like, and determines whether or not a key is erroneously typed (erroneous input) based on information about the pressed key and how the key is pressed. The information about the pressed key is, for example, information about the input character string. In the following description, the character string includes numerals and symbols. The way of pressing the key is, for example, the amount of pressing of the key. In character input, if an adjacent key is erroneously pressed, the pressing amount of the erroneously pressed key may be weak. Therefore, by determining an erroneous input in consideration of the amount of depression, the input device can accurately detect an erroneous keystroke.

First, the generation of a trained model (dictionary) used to determine erroneous input will be described. When the input device detects a correction operation such as pressing the Backspace key, the input device acquires information on the character strings input before and after correction and information on the amount of pressing of each key.

FIG. 1 shows an example of mistyping "KIKAI" as "JIKAI". The character string before correction is "JIKAI". If the standard pressing amount is "2" and the pressing amount smaller than the standard is "1", the pressing amount of each key of the character string "JIKAI" before correction is "12222". The corrected character string is "KIKAI", and the pressing amount of each key is "22222". It can be seen that the "J" before the correction was mistakenly pressed for the "K" on the right side, and the pressing amount of the "J" was smaller than the standard.

The input device uses machine learning such as deep learning to generate a learned model for inferring an erroneous input using the character string and the pressing amount of each key before correction and the character string after correction. The trained model may be a model that infers (outputs) whether or not the character string to be judged entered by the user is an incorrect input, and if the character string to be judged is an incorrect input, a correction candidate is displayed. It may be a model that infers strings. The input device generates a trained model using the character string before modification, the pressing amount of each key, and the character string after modification as teacher data.

Next, erroneous input determination using the generated trained model will be described. The input device inputs the information of the input character string and the information of the pressing amount of each key (operation information to be judged) to the trained model, and based on the inference result of the trained model, the input character string is It is determined whether or not it is an erroneous input.

For example, if the character string to be judged is "JIKAI" and the pressing amount of each key is "12222", the trained model infers that the character string to be judged is an erroneous input. On the other hand, if the user intentionally entered the character string "JIKAI", the amount of depression of each key would be "22222". Infer that.

As described above, the input device can detect keystroke errors with higher accuracy by taking into account the physical way of pressing the keys, such as the amount of pressing of the keys. The input device is, for example, a keyboard, and may be a computer (information processing device) having a display device such as a display. The method of pressing the key is not limited to the amount of key pressing, and may be the position of key pressing, the time interval between key inputs, or the like. Also, the amount of key depression is not limited to two steps, and may be three steps or more.

<Embodiment>
The device configuration of the input device 10 will be described with reference to FIG. FIG. 2 is a block diagram illustrating the configuration of the input device 10. As shown in FIG. The input device 10 includes a keyboard 110 , a control section 120 , a storage section 130 and a display section 140 .

The keyboard 110 is an operation member that receives operations from the user, and is, for example, a keyboard. Here, the detection of the key depression amount will be described with reference to FIG. FIG. 3 is a diagram exemplifying the switch 111 that detects pressing of a key. The switch 111 is provided inside the key top and detects pressing of the key. The keyboard 110 is provided with a sensor 110a capable of measuring the pressing amount of the switch 111 (the height at which the switch 111 is pressed) or the pressing pressure on the switch 111, as how to press the key. The amount of key depression may be determined based on the pressure applied to the switch 111 .

The control unit 120 includes a correction detection unit 121, an operation information acquisition unit 122, and a model generation unit 123 for executing learning processing. The control unit 120 also includes an erroneous input determination unit 124 and a candidate acquisition unit 125 for executing inference processing.

In the learning process, the correction detection unit 121 detects a correction operation, which is an operation for correcting the contents of input by the user using the keyboard 110 . The correction operation is an operation of pressing a key for correcting input content, such as a Backspace key or a Delete key. The correction operation may be an operation of selecting the character string before correction with a pointing device such as a mouse and inputting the character string after correction.

The operation information acquisition unit 122 acquires pre-correction operation information including information on the keys pressed before detecting the correction operation and how to press the keys, and post-correction operation information including information on the keys pressed after detecting the correction operation. Get information and

The information of the key is, for example, the information of the input character string, and the character string before correction can be, for example, the character string deleted by the Backspace key or the like. In the example of FIG. 1, after "JIKAI" is entered, the Backspace key is pressed five times to delete "JIKAI" and input for correction is started. JIKAI” is obtained as a character string before correction.

In addition, when only the character to be corrected is corrected, the character string before correction can be a character string including the character to be operated and the subsequently input character. For example, when "J" is deleted and "K" is input after "JIKAI" is input, the operation information acquiring unit 122 retrieves "JIKAI" including "IKAI" input after "J". ” can be obtained as a character string before modification.

For example, when the character string before correction is deleted by continuously pressing the Backspace key, the character string after correction is the same number of characters as the deleted characters among the character strings input after deletion. can be columns. In the example of FIG. 1, when the five characters of "JIKAI" are deleted, the operation information acquisition unit 122 acquires the five characters of "KIKAI" input after the deletion as the corrected character string.

Further, when "J" is deleted and "K" is input after "JIKAI" is input, the operation information acquiring unit 122 determines that the character string "JIKAI" before correction is the same as "JIKAI". "KIKAI", which is a 5-character string, should be obtained as the corrected string.

The pre-correction character string and the post-correction character string may be information on word-by-word or clause-by-clause character strings extracted by syntactic analysis of sentences (natural language) input by the user. Note that the character strings before correction and the character strings after correction are not limited to Japanese character strings, and may include alphanumeric characters, symbols, and the like. Even if the erroneously input character is not at the beginning of a word, the operation information acquisition unit 122 parses the character string before correction and the character string after correction into appropriate units by parsing the input character string. can be obtained at

The method of pressing the key is, for example, the amount of pressing of the key described with reference to FIG. The key depression amount is not limited to two steps, and may be three steps or more. The operation information acquisition unit 122 may associate the acquired pre-correction operation information and post-correction operation information and store them in the storage unit 130 .

A model generation unit 123 (corresponding to a “generation unit”) generates a trained model 131 for inferring an erroneous input using the pre-correction operation information and the post-correction operation information. The model generating unit 123 learns, for example, the uncorrected character string “JIKAI” and the pressing amount “12222”, which are the pre-correction operation information, and the corrected character string “KIKAI”, which is the pre-correction operation information. A finished model 131 is generated.

The model generation unit 123 also determines whether or not to learn the pre-correction operation information and the post-correction operation information as teacher data based on the degree of matching between the pre-correction character string and the post-correction character string. good. For example, when "BOKUWA" is changed to "WATASHIWA", it is considered that the input of "BOKUWA" was not an error, but that the user intentionally changed it to "WATASHIWA". If the degree of matching between a character string before correction and a character string after correction is equal to or less than a predetermined threshold, the model generation unit 123 may exclude these character strings from learning targets.

The degree of matching between a character string before correction and a character string after correction can be defined as, for example, the ratio of the number of characters in the character string before correction that match the character string after correction to the number of characters in the string before correction. can. Specifically, the number of characters in "BOKUWA" is 6, and the number of characters that match "WATASHIWA" is 2, so the degree of matching between "BOKUWA" and "WATASHIWA" is 2/6 (approximately 33%). . Also, since the number of characters of "JIKAI" is 5 and the number of characters that match "KIKAI" is 4, the degree of matching between "JIKAI" and "KIKAI" is 4/5 (80%). When the predetermined threshold is set to 60%, the model generation unit 123 excludes the character strings "BOKUWA" and "WATASHIWA" before and after the correction from learning targets.

In the present embodiment, the model generation unit 123 is assumed to generate the learned model 131 by learning operation information of a specific user who uses the input device 10 . When a plurality of users use the input device 10, the model generating unit 123 identifies the users using, for example, the account information of the users who use the input device 10, and generates a trained model 131 for each user. good too.

In the inference process, the erroneous input determination unit 124 (corresponding to the “determination unit”) inputs operation information including information on the pressed key and how to press the key to the learned model 131 generated by the model generation unit 123. , determines whether or not the character string corresponding to the key information is an erroneous input.

The erroneous input determination unit 124 extracts a character string to be determined as an erroneous input as key information, and acquires the amount of pressing of the key corresponding to each character as how to press the key. The erroneous input determination unit 124 inputs the extracted character string and the acquired key pressing amount to the learned model 131, and determines whether or not the character string to be determined is an erroneous input.

A method for extracting a character string to be judged as an erroneous input will be described. The erroneous input determination unit 124 obtains a character string including a predetermined number of characters (for example, 10 characters) that are last input each time one character is input, and obtains the last two characters, three characters, . A character string of 10 characters is input to the learned model 131 as a character string to be judged along with the amount of pressing. The predetermined number is not limited to 10 characters, and may be determined to be a number of 2 or more characters in consideration of the load of erroneous input determination processing.

In addition, the erroneous input determination unit 124 decomposes the input character string information into word-based or clause-based character strings by syntactic analysis, and uses the character string that was input immediately before as a character string to be subjected to erroneous input determination. may be extracted. The erroneous input determination unit 124 inputs the extracted character string and the pressing amount of the key corresponding to each character to the learned model 131 as operation information. The erroneous input determination unit 124 determines whether or not the extracted character string is an erroneous input based on the inference result of the trained model 131 .

The candidate acquisition unit 125 (corresponding to the “acquisition unit”) acquires correction candidates for the character string determined to be an erroneous input. The candidate acquisition unit 125 acquires correction candidates using a trained model 131 that has been trained to output correction candidates when a character string to be determined and the pressing amount (operation information) of each key are input, for example. . In the example of FIG. 1, the learned model 131 outputs "KIKAI" as a correction candidate for the operation information in which the input character string is "JIKAI" and the pressing amount of each key is "12222".

The candidate acquisition unit 125 is not limited to acquiring correction candidates using the trained model 131, and may acquire correction candidates from the post-correction operation information stored in the storage unit 130 in the learning process. Specifically, the candidate acquisition unit 125 corrects the character string of the post-correction operation information associated with the pre-correction operation information for the same character string as the character string that the erroneous input determination unit 124 has determined to be an erroneous input. It is acquired from the storage unit 130 as a candidate.

Storage unit 130 includes pre-correction operation information including information on keys pressed before detection of a correction operation and how to press keys, and information on keys pressed after detection of a correction operation and how to press keys. It is stored in association with post-correction operation information. The operation information before correction and the operation information after correction are acquired by the operation information acquisition unit 122 . The storage unit 130 also stores the learned model 131 generated by the model generation unit 123 .

A trained model 131 is a trained model for inferring an erroneous input. The learned model 131 uses, for example, user operation information (key information and how to press a key) as input data, and infers whether or not the character string to be determined corresponding to the key information is an erroneous input. Specifically, the trained model 131 outputs the probability (reliability) that the character string to be determined is an erroneous input. In this case, the erroneous input determination unit 124 determines that the input character string is an erroneous input when the probability that the character string to be determined is an erroneous input is a predetermined value (for example, 80%) or more. can be done.

Also, the trained model 131 may be generated so as to infer a correction candidate character string when the character string corresponding to the key information is an incorrect input. In this case, the learned model 131 outputs a character string of a correction candidate and its reliability using operation information by the user as input data. The trained model 131 may output multiple revision candidates and their respective confidences.

For example, when a correction candidate is output and the reliability of the correction candidate is equal to or higher than a predetermined value (eg, 60%), the erroneous input determination unit 124 may determine that the character string to be determined is an erroneous input. can. Further, when no correction candidate is output or when the reliability of the correction candidate is less than a predetermined value, it can be determined that the character string to be determined is not an erroneous input.

The display unit 140 is a display or the like, and displays (presents) the correction candidates acquired by the candidate acquisition unit 125 . The display unit 140 deletes the character string determined to be an erroneous input and automatically corrects it to the acquired correction candidate.

In addition, the display unit 140 may present correction candidates together with the character string determined to be an erroneous input. By presenting correction candidates instead of automatically correcting, the user can determine whether or not to correct the input character string after confirming the correction candidates.

When the candidate acquiring unit 125 acquires a plurality of correction candidates, the display unit 140 may present the acquired plurality of correction candidates so that the user can select them. The user can select a correction candidate closer to the character string to be input from a plurality of correction candidates.

The display unit 140 may highlight the corrected portion in at least one of the character string before correction and the correction candidate. The display unit 140 can highlight the corrected portion by, for example, changing the character color or adding a marker.

Note that each functional unit in FIG. 2 may or may not be separate hardware. Functions of two or more functional units may be implemented by common hardware. Each of the multiple functions of one functional unit may be implemented by separate hardware. Two or more functions of one functional unit may be realized by common hardware. Also, each functional unit may or may not be implemented by hardware. For example, a device may have a processor and a memory in which a control program is stored. The functions of at least some of the functional units of the device may be realized by the processor reading out and executing the control program from the memory.

(learning process)
The learning process will be described with reference to FIG. The learning process shown in FIG. 4 is started and ended by an instruction from the user, for example.

In S101, the keyboard 110 waits for key input from the user. In S102, the correction detection unit 121 determines whether or not a correction operation has been detected by key input from the user. A correction operation is an operation of pressing a Backspace key or the like. If a correction operation is detected, the process proceeds to S103. If no correction operation has been detected, the process returns to S101.

In S103, the operation information acquisition unit 122 acquires operation information before and after correction. The pre-correction operation information is the pre-correction character string and the key depression amount corresponding to each character. The character string before correction is, for example, a character string that is continuously deleted by using the Backspace key or the like, or a character string that includes corrected characters and subsequent characters that have been input.

The corrected operation information is a character string after correction. The post-correction character string is, for example, a character string input after detection of a correction operation, and can be a character string having the same number of characters as the pre-correction character string.

In the example of FIG. 1, the operation information acquisition unit 122 acquires the character string "JIKAI" and the pressing amount "12222" as the pre-correction operation information. Further, the operation information acquisition unit 122 acquires the character string “KIKAI” as the post-correction operation information.

In S104, the model generating unit 123 uses the pre-correction operation information and the post-correction operation information acquired in S103 to generate the learned model 131 for inferring an erroneous input by machine learning. While the user is using the keyboard, the model generation unit 123 collects information on the character strings input before and after the correction operation and information on the pressing amount of each key, and learns the tendency of the user to make mistakes. A trained model 131 can be generated.

Note that the input device 10 may perform the learning process while being set to the learning mode, which is the mode for generating the trained model 131 . On/off of the learning mode can be switched by the user.

(inference processing)
The inference processing will be described with reference to FIG. The inference processing shown in FIG. 5 is started and ended by an instruction from the user, for example.

In S201, the keyboard 110 waits for key input from the user. In S202, the erroneous input determination unit 124 determines whether or not the character string input from the keyboard 110 is an erroneous input using the learned model 131 generated in the learning process.

The erroneous input determination unit 124 acquires the input character string to be determined as the key information, and acquires the key pressing amount as the key pressing method. The erroneous input determination unit 124 can input the acquired character string and the key depression amount of each character to the learned model 131, and determine whether or not the input character string is an erroneous input. If the character string to be determined is an erroneous input, the process proceeds to S203. If the character string to be determined is not an erroneous input, the process returns to S201.

If it is determined in S202 that the character string to be determined is an erroneous input, the keyboard 110 may disable input from the user for a predetermined period of time. By invalidating the input from the keyboard 110, the user can avoid continuous input with an erroneous input, and can reduce the trouble of deleting the input contents before realizing the error. The predetermined time is, for example, the time from when the input is determined to be erroneous to when the user notices the mistake and stops key input.

In S203, the candidate acquisition unit 125 acquires correction candidates based on the character string determined to be an erroneous input in S202. The candidate acquisition unit 125 may acquire correction candidates by inputting a character string determined to be an erroneous input into the learned model 131 . You may acquire the character string of operation information as a correction candidate.

In S204, the display unit 140 presents the correction candidates acquired in S203 on the display. When display unit 140 automatically converts an erroneously input character string into a correction candidate, the user can reduce the trouble of correction. When a correction candidate for an erroneously input character string is presented by the display unit 140, the user can confirm whether the correction is intended or not before converting it to a correction candidate.

Note that the input device 10 may perform inference processing while being set to an inference mode, which is a mode in which erroneous inputs are detected and corrected. The inference mode can be switched on/off by the user.

According to the above-described embodiment, the input device 10 uses the learned model 131 in which the tendency of the user's keystroke mistakes is learned by machine learning, thereby accurately detecting keystroke mistakes. . In addition, the input device 10 can improve typing speed by converting a character string determined to be an erroneous input into a correction candidate or presenting the correction candidate.

<Modification 1>
In the above embodiment, the input device 10 generates the learned model 131 by learning the tendency of a specific user's typographical errors. In Modified Example 1, the input device 10 learns the tendency of multiple users to make typos and generates a learned model 131 . That is, the input device 10 generates a learned model using pre-correction operation information and post-correction operation information input by a plurality of users.

The input device 10 can generate a trained model 131 corresponding to common mistakes by learning information about various user input errors without distinguishing between users. A common mistake is, for example, when trying to press "S", the "A" to the left is pressed with an amount of "1" that is less than the standard pressing amount. By using the trained model 131 corresponding to common mistakes, the user can efficiently correct keystroke mistakes even before learning his/her own mistakes.

<Modification 2>
In the above embodiment, as shown in FIG. 3, the keyboard 110 has one switch 111 inside the key top of each key for detecting key depression. In variant 2, each key has a plurality of switches.

FIG. 6 is a diagram illustrating a plurality of switches for detecting key presses. In the example of FIG. 6, a plurality of switches 111a to 111d are provided inside the key top to detect key depression. By providing a plurality of switches 111a to 111d for each key, the keyboard 110 can obtain information on the pressing position, which is the area of the key surface pressed, as a method of pressing the key.

In Modified Example 2, it is sufficient that the switches 111a to 111d are capable of detecting whether or not they have been pressed, and the input device 10 detects, for example, which of the plurality of switches responds first. The input device 10 can acquire the position of the switch that responds first among the switches 111a to 111d as the key depression position. The model generating unit 123 generates a learned model 131 that has learned the key pressing position as a key pressing method, and the input device 10 can determine an erroneous input in consideration of the key pressing position.

Note that the input device 10 may acquire the pressing order of a plurality of switches included in each key as how to press the keys. The model generating unit 123 generates a learned model 131 by learning the pressing order of the plurality of switches as how to press the keys. will be able to

Further, the input device 10 may acquire the pressing amount of each of a plurality of switches as how to press the key. The model generating unit 123 generates a learned model 131 by learning the amount of depression of a plurality of switches as a method of pressing a key, and the input device 10 determines an erroneous input by considering the amount of depression of a plurality of switches. will be able to

With reference to FIG. 7, acquisition of push amounts of a plurality of switches will be described. FIG. 7 shows an example of mistyping "KIKAI" as "JIKAI". The character string before correction is "JIKAI". If the standard push amount is "2", the push amount less than the standard is "1", and the state of no push is "0", the push amounts of "J" for the switches 111a to 111d are "1". , "1", "0", "0".

In the example of FIG. 7, the switch 111b located on the key side of "K" and the switch 111a located on the key side of "I" to be input next to "K" were depressed less than the standard. indicates On the other hand, the switches 111c and 111d located away from the "K" and "I" keys have not detected their depression.

The model generation unit 123 generates a learned model 131 by learning the amount of depression of each of the plurality of switches as how to press a key, and the input device 10 determines an erroneous input in consideration of the amount of depression of the plurality of switches. be able to

Using the generated trained model 131, the erroneous input determination unit 124 selects "JIKAI" when the pressing amounts of the switches 111a to 111d are "1", "1", "0", and "0". is an erroneous input of "KIKAI". On the other hand, if the user intentionally inputs the character string "JIKAI", the pressing amounts for the switches 111a to 111d will be "2", "2", "2", "2". is assumed. In this case, the erroneous input determination unit 124 can determine that the input of "JIKAI" (pressing amount is "2", "2", "2", "2") is not an erroneous input.

In Modified Example 2, the operation information acquisition unit 122 acquires the key pressing position, the pressing order of the plurality of switches, the pressing amount of the plurality of switches, and the like as the key pressing method. The model generation unit 123 generates a learned model 131 by learning at least one of the information acquired by the operation information acquisition unit 122 as how to press a key. The input device 10 can accurately detect keystroke errors by considering the key depression position, the order of depression of the plurality of switches, the amount of depression of the plurality of switches, and the like.

Also, the number and arrangement of the plurality of switches are not limited to the example in FIG. As many switches as can be placed inside one key may be placed along each side, as in FIG. 7, or placed at each corner or in the center.

<Modification 3>
In the above-described embodiment, the input device 10 detects the amount of pressing of the key as how to press the key. Further, in Modified Example 2, the input device 10 detects the pressing position of the key as the method of pressing the key. In Modified Example 3, the input device 10 further measures the time interval between key inputs and uses it to determine erroneous inputs.

FIG. 8 is a diagram for explaining an erroneous input in which the order of keys to be input is reversed. The time interval between the key input by the right hand and the key input by the left hand is shortened, and the user may input the key to be pressed next with the left hand earlier than the key to be pressed with the right hand. In the example of FIG. 8, the user may mistakenly input "A"→"K" when attempting to input "K"→"A".

If the keys are erroneously input in the order of “A”→“K”, it is assumed that the time intervals between the key inputs of “A” and “K” will be approximately the same timing. Thus, if the time interval between keystrokes is different (eg, shorter) than the user's average time interval between keystrokes, the user may have made a typo. On the other hand, if the key input time interval is substantially equal to the user's average key input time interval, it is considered that the user intentionally input "A"→"K" in that order.

FIG. 9 is a block diagram illustrating the configuration of an input device according to Modification 3. As shown in FIG. The input device 10 includes a timer 150 that measures time intervals between key inputs. Since the configuration other than the timer 150 is the same as the configuration shown in FIG. 2, the contents different from the above embodiment and each modified example will be described.

The operation information acquisition unit 122 acquires the key input time interval measured by the timer 150 as the key press method. The model generation unit 123 generates a trained model 131 that has learned the key input time interval, and the input device 10 can determine an erroneous input in consideration of the key input time interval. Therefore, the input device 10 can accurately detect an erroneous input even when the input order of the keys to be input with the left and right hands is reversed.

<Others>
The above-described embodiment and modifications are merely illustrative examples of the configuration of the present invention. The present invention is not limited to the specific forms described above, and various modifications are possible within the technical scope of the present invention.

Note that in the above embodiment, the learning process and the inference process are described as being executed separately, but they may be executed in parallel. By continuously executing the learning process, the input device 10 can generate the trained model 131 that determines an erroneous input with higher accuracy.

Further, in the above-described embodiment, an example in which the input device 10 generates the trained model 131 has been described, but the trained model 131 may be generated by an external device capable of communicating with the input device 10 . In this case, the external device acquires the pre-correction operation information and the post-correction operation information acquired by the operation information acquiring unit 122 from the input device 10 online. The external device can use the pre-correction operation information and the post-correction operation information acquired from the input device 10 to generate the learned model 131 in the same manner as the model generation unit 123 .

The key information may also include key arrangement information. The key arrangement information is, for example, information as to whether or not the character keys before correction (including keys other than characters) and the character keys after correction are adjacent to each other.

The operation information acquisition unit 122 acquires information on the layout of the keys as information on the keys. The model generation unit 123 generates a trained model 131 that has learned the key arrangement information, and the input device 10 can determine an erroneous input in consideration of the key arrangement information. Since erroneous keystrokes are likely to occur between adjacent keys, the input device 10 can accurately detect erroneous inputs by learning key arrangement information.

<Appendix 1>
a keyboard (110) comprising a sensor (110a) for detecting how a key is pressed;
a correction detection unit (121) for detecting a correction operation, which is an operation for correcting the input content through the keyboard (110);
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition unit (122) that acquires
a generation unit (123) that generates a trained model (131) for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
An input device (10) comprising:

<Appendix 2>
a keyboard (110) comprising a sensor (110a) for detecting how a key is pressed;
Operation information including information on the pressed key and how to press the key is input to a trained model (131) for inferring an erroneous input, and whether or not the character string corresponding to the information on the key is an erroneous input. A determination unit (124) that determines
an acquisition unit (125) for acquiring a correction candidate for the character string determined to be an erroneous input;
a display unit (140) for presenting the correction candidate;
An input device (10) comprising:

<Appendix 3>
a keyboard (110) comprising a sensor (110a) for detecting how a key is pressed;
a correction detection unit (121) for detecting a correction operation, which is an operation for correcting the input content through the keyboard (110);
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition unit (122) that acquires
a generation unit (123) that generates a trained model (131) for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
Operation information including information on the pressed key and how to press the key is input to the learned model (131), and determination is made to determine whether or not the character string corresponding to the information on the key is an erroneous input. a part (124);
an acquisition unit (125) for acquiring a correction candidate for the character string determined to be an erroneous input;
a display unit (140) for presenting the correction candidate;
An information processing device (10) comprising:

<Appendix 4>
A control method for an input device (10) comprising a keyboard (110) comprising a sensor (110a) for detecting how a key is pressed, comprising:
a correction detection step (S102) of detecting a correction operation, which is an operation for correcting the input content through the keyboard (110);
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition step (S103) for acquiring
a generation step (S104) of generating a trained model (131) for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
A method of controlling an input device (10) comprising:

<Appendix 5>
A control method for an input device (10) comprising a keyboard (110) comprising a sensor (110a) for detecting how a key is pressed, comprising:
Operation information including information on the pressed key and how to press the key is input to a trained model (131) for inferring an erroneous input, and whether or not the character string corresponding to the information on the key is an erroneous input. A determination step (S202) for determining
an acquisition step (S203) of acquiring a correction candidate for the character string determined to be an erroneous input;
a display step (S204) of presenting the correction candidate;
A method of controlling an input device (10) comprising:

10: input device, 110: keyboard, 110a: sensor, 111: switch, 111a: switch, 111b: switch, 111c: switch, 111d: switch, 120: control unit, 121: correction detection unit, 122: operation information acquisition unit , 123: model generation unit, 124: erroneous input determination unit, 125: candidate acquisition unit, 130: storage unit, 131: model, 140: display unit, 150: timer

Claims (24)

a keyboard having a sensor that detects how a key is pressed;
a correction detection unit that detects a correction operation, which is an operation for correcting the input content from the keyboard;
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition unit that acquires
a generation unit that generates a trained model for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
An input device comprising When operation information including information on a pressed key and how to press the key is input, the trained model infers a correction candidate when a character string corresponding to the information on the key is an incorrect input. is a model trained on
The input device according to claim 1. The generating unit generates the learned model using the pre-correction operation information and the post-correction operation information by a specific user.
3. The input device according to claim 1 or 2. The generation unit generates the learned model using the pre-correction operation information and the post-correction operation information by a plurality of users.
The input device according to any one of claims 1 to 3. The correction operation is an operation of pressing the Backspace key or the Delete key,
The input device according to any one of claims 1 to 4. The keyboard comprises switches inside the keys,
The sensor measures the amount of depression or pressure of the switch,
The input device according to any one of claims 1 to 5. each key of the keyboard comprises a plurality of the switches;
The operation information acquisition unit acquires the key depression position based on the order in which the plurality of switches are depressed or the amount of depression of the switches.
The input device according to claim 6. a keyboard having a sensor that detects how a key is pressed;
Operation information including information on the pressed key and how to press the key is input to a trained model for inferring an erroneous input, and it is determined whether or not the character string corresponding to the information on the key is an erroneous input. a determination unit;
an acquisition unit that acquires a correction candidate for the character string determined to be an erroneous input;
a display unit that presents the correction candidate;
An input device comprising The acquisition unit acquires the correction candidate using the learned model trained to output the correction candidate when the operation information is input.
The input device according to claim 8. Pre-correction operation information including information on the key pressed before detection of the correction operation and how to press the key, and post-correction operation information that is information on the key pressed after detection of the correction operation. further comprising a storage unit that associates and stores,
The acquisition unit acquires, as the correction candidate, the character string of the post-correction operation information associated with the pre-correction operation information for the same character string as the character string determined to be an erroneous input from the storage unit. do,
The input device according to claim 8. The display unit presents the correction candidate together with the character string determined to be an erroneous input.
The input device according to any one of claims 8 to 10. The acquisition unit acquires a plurality of correction candidates,
The display unit presents a plurality of correction candidates in a user-selectable manner.
An input device according to any one of claims 8 to 11. The display unit deletes the character string determined to be an erroneous input and presents the correction candidate.
The input device according to any one of claims 8 to 10. The display unit highlights a corrected portion in at least one of the character string determined to be an erroneous input and the correction candidate.
Input device according to any one of claims 8 to 13. the keyboard disables the user's operation for a predetermined period of time after the determining unit determines that the character string has been input incorrectly;
Input device according to any one of claims 8 to 14. The key information includes at least one of an input character string and key arrangement,
16. An input device according to any one of claims 1-15. The key pressing method includes information on at least one of key pressing amount, key pressing position, and key input time interval,
17. An input device according to any one of claims 1-16. The key information includes word-based or clause-based character string information extracted by syntactic analysis,
18. An input device according to any one of claims 1-17. It further comprises a timer that measures the time interval between key inputs,
19. An input device according to any one of claims 1-18. a keyboard having a sensor that detects how a key is pressed;
a correction detection unit that detects a correction operation, which is an operation for correcting the input content from the keyboard;
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition unit that acquires
a generation unit that generates a trained model for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
a determination unit that inputs operation information including information on the pressed key and how to press the key into the learned model and determines whether or not the character string corresponding to the information on the key is an input error;
an acquisition unit that acquires a correction candidate for the character string determined to be an erroneous input;
a display unit that presents the correction candidate;
Information processing device. A control method for an input device having a keyboard having a sensor for detecting how a key is pressed, comprising:
a correction detection step of detecting a correction operation, which is an operation for correcting the content entered by the keyboard;
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition step for acquiring
a generation step of generating a trained model for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
A method of controlling an input device including A control method for an input device having a keyboard having a sensor for detecting how a key is pressed, comprising:
Operation information including information on the pressed key and how to press the key is input to a trained model for inferring an erroneous input, and it is determined whether or not the character string corresponding to the information on the key is an erroneous input. a determination step;
an acquisition step of acquiring a correction candidate for the character string determined to be an erroneous input;
a display step of presenting the correction candidate;
A method of controlling an input device including A control program for an input device comprising a keyboard having a sensor for detecting how a key is pressed,
a correction detection step of detecting a correction operation, which is an operation for correcting the content entered by the keyboard;
Pre-correction operation information including information on the key pressed before the correction operation is detected and how to press the key, and post-correction operation information that is information on the key pressed after the correction operation is detected; an operation information acquisition step for acquiring
a generation step of generating a trained model for inferring an erroneous input using the pre-correction operation information and the post-correction operation information;
A control program for an input device including A control program for an input device comprising a keyboard having a sensor for detecting how a key is pressed,
Operation information including information on the pressed key and how to press the key is input to a trained model for inferring an erroneous input, and it is determined whether or not the character string corresponding to the information on the key is an erroneous input. a determination step;
an acquisition step of acquiring a correction candidate for the character string determined to be an erroneous input;
a display step of presenting the correction candidate;
A control program for an input device including

Images