Sentence correction method

Abstract

Provided is a method in which a computer: receives a correction request for a first sentence; extracts a keyword and / or a key phrase included in the first sentence; acquires a first sentence key that is the keyword and / or the key phrase in the first sentence; creates a plurality of correction sentence candidates which include the first sentence key; and selects, as the corrected first sentence, a sentence corresponding to the correction request for the first sentence, from the plurality of correction sentence candidates.

Description

How to correct a statement

[0001] The present invention relates to a method for automatically correcting sentences using a computer.

[0002] Japanese Patent Publication No. 7176443 describes a recommendation sentence generation device. The recommendation sentence generation device described in this publication can automatically generate a single sentence. However, this device cannot generate sentences that comply with various requirements. Furthermore, this device can only generate a single sentence.

[0003] Patent No. 7176443

[0004] The present invention aims to provide a method for correcting sentences that can generate sentences that meet various requirements. The present invention also aims to provide a method that can generate multiple candidates for corrected sentences while maintaining keywords, etc.

[0005] The first invention relates to a method for correcting a sentence by a computer. This method includes a first sentence input step (S110), a first sentence correction request receiving step (S120), a corrected first sentence acquisition step (S130), and a corrected first sentence display step (S140). The first sentence input step (S110) is a step in which a first sentence is input to a computer. The first sentence correction request receiving step (S120) is a step in which the computer receives a request to correct the first sentence. Examples of the correction request include one or more of the following: positive, grammatically correct, more accurate, more understandable, and more commonly used. The corrected first sentence acquisition step (S130) is a step in which the computer corrects the first sentence in accordance with the first sentence correction request and acquires the corrected first sentence. This process includes a first in-sentence key extraction process (S131), a corrected sentence creation process (S132), and a corrected first sentence selection process (S133). The first in-sentence key extraction process (S131) ​​is a process of extracting either or both of a keyword and a key phrase contained in the first sentence and acquiring a first in-sentence key which is either or both of a keyword and a key phrase in the first sentence. The corrected sentence creation process (S132) is a process of creating multiple corrected sentence candidates including the first in-sentence key. The corrected first sentence selection process (S133) is a process of selecting a sentence corresponding to the correction request from the multiple corrected sentence candidates as the corrected first sentence. The corrected first sentence display process (S140) is a process of displaying the corrected first sentence by the computer.

[0006] A preferred aspect of the above method further includes a talk input step (S210), a cluster classification step (S220), and a first sentence extraction step (S230). The talk input step (S210) is a step in which a conversation or talk is input to a computer. The cluster classification step (S220) is a step in which the computer classifies the conversation or talk into a plurality of clusters, each of which includes a different topic or point of discussion. The first sentence extraction step (S230) is a step in which the computer extracts a first sentence from one of the plurality of clusters. The first sentence extracted in this step becomes the first sentence input to the computer in the first sentence input step (S110). This aspect is characterized by conversation clustering.

[0007] The next invention relates to a program for causing a computer to execute the above-mentioned sentence correction method, and a non-transitory information recording medium on which the program is recorded.

[0008] The present invention can provide a sentence correction method that can generate sentences that meet various requirements. The present invention can provide a method that can create multiple corrected sentence candidates while maintaining keywords, etc.

[0009] FIG. 1 is a flowchart showing a sentence correction method. FIG. 2 is a block diagram for explaining a sentence correction system. FIG. 3 is a conceptual diagram showing an example of a file being input into the sentence correction system. FIG. 4 is a diagram showing an example of a conversation between two people. FIG. 5 is a conceptual diagram showing an example of a diagram of cluster classification. FIG. 6 is a diagram showing the flow of topics in clustered topic groups.

[0010] The following describes embodiments of the present invention with reference to the drawings. The present invention is not limited to the embodiments described below, but also includes appropriate modifications of the embodiments that are obvious to those skilled in the art.

[0011] FIG. 1 is a flowchart showing a sentence correction method. A computer-based sentence correction method will be described. This method involves a computer correcting a sentence and displaying the corrected sentence on a display unit. As shown in FIG. 1, this method includes a first sentence input step (S110), a first sentence correction request receiving step (S120), a corrected first sentence acquisition step (S130), and a corrected first sentence display step (S140). This method may further include a talk input step (S210), a cluster classification step (S220), and a first sentence extraction step (S230). This method is a method in which a computer executes various steps.

[0012] A computer has an input unit, an output unit, a control unit, a calculation unit, and a memory unit, and each element is connected by a bus or the like to enable the exchange of information. For example, the memory unit may store a control program or various information. When predetermined information is input from the input unit, the control unit reads the control program stored in the memory unit. The control unit then reads the information stored in the memory unit as appropriate and transmits it to the calculation unit. The control unit also transmits the input information as appropriate to the calculation unit. The calculation unit performs calculation processing using the received various information and stores it in the memory unit. The control unit reads the calculation results stored in the memory unit and outputs them from the output unit. In this way, various processes and steps are performed. Each unit or means executes these various processes. A computer may have a processor, and the processor may realize various functions and steps. A computer may be standalone. A computer may have some of its functions distributed between a server and a terminal. In this case, it is preferable that the server and the terminal can exchange information via a network such as the Internet or an intranet. The computer may include a processor and a memory coupled to the processor. The memory may store instructions that, when executed by the processor, cause the computer to perform various processes or function as various elements. The computer may be provided with various training data to construct a learning model and perform various calculations through machine learning. In this case, the computer may perform various analyses using the learning model created through machine learning and deep learning of AI (artificial intelligence).

[0013] 2 is a block diagram illustrating a sentence correction system. As shown in FIG. 2, the sentence correction system 1 includes a sentence input unit 3, a sentence correction request receiving unit 5, a correction request dictionary 7, a sentence key acquisition unit 9, a sentence key dictionary 11, a corrected sentence creation unit 13, a sentence selection unit 15, and a sentence display unit 17. As shown in FIG. 2, the sentence correction system 1 may further include a talk input unit 21, a voice analysis unit 23, a cluster classification unit 25, and a sentence extraction unit 27. Alternatively, the sentence correction system 1 may include the cluster classification unit 25 and the sentence extraction unit 27 without being based on voice analysis.

[0014] The first sentence input step (S110) is a step of inputting a first sentence to a computer. The sentence may be a single sentence ending with a comma, or a collection of multiple sentences. For example, the first sentence may be input to the sentence correction system 1 by the sentence input unit 3 of the sentence correction system 1. Inputting a first sentence to a computer means enabling the computer to perform arithmetic processing on the first sentence. For example, a user may input a sentence to a computer using a pointing device, thereby inputting the sentence to the sentence correction system 1 implemented by a computer. Alternatively, an audio file or a video file may be input to the sentence correction system 1 by dragging and dropping it. Alternatively, a user may input a search instruction to the sentence correction system 1 using a pointing device, which causes the sentence correction system 1 to obtain sentences related to the search instruction on the Internet, thereby inputting the first sentence to the sentence correction system 1. The first sentence input to the sentence correction system 1 may be stored in a storage unit as appropriate.

[0015] 3 is a conceptual diagram showing an example of inputting a file into the sentence correction system. As shown in FIG. 3, for example, when an audio file or a video file is dragged and dropped into the sentence correction system 1, a number of sentences including a first sentence are input into the sentence correction system 1. Also, when the video file is related to a presentation, audio related to the presentation is input into the sentence correction system 1, resulting in a number of sentences being input into the system 1. In this case, each input sentence may be input into the sentence correction system 1 as a first sentence, a second sentence, a third sentence, etc., or characteristic sentences among the multiple sentences may be input into the sentence correction system 1 as a first sentence, a second sentence, a third sentence, etc.,

[0016] For example, if a user says, "Every day, I practice sales talk at home using my iPhone. I use an app to do dialogue training with AI customers to learn to speak in a more positive way. I want to use more positive expressions," then, through the process described below, "Every day, I practice sales talk at home using my iPhone. I use an app to do dialogue training with AI customers to learn to speak in a more positive way." is input into the sentence correction system 1 as the first sentence, and the first sentence after speech analysis is stored in the memory unit (first sentence input process (S110)).

[0017] The first sentence correction request receiving step (S120) is a step in which the computer receives a first sentence correction request. The first sentence correction request is a request regarding how to change the first sentence. Examples of correction requests include making the first sentence more positive, grammatically correct, more accurate, more understandable, and more commonly used. Other examples of correction requests include summarizing, translating, making the sentence more colloquial, and making the sentence more written. For example, when a correction request is included in a sentence entered with a pointing device or in a conversation, the sentence correction request receiving unit 5 of the sentence correction system 1 may refer to the correction request dictionary 7, analyze the correction request, and obtain a first sentence correction request. The correction request dictionary 7 may store, for example, the above examples, and the sentence correction request receiving unit 5 may obtain a first sentence correction request by examining whether the example is included in the sentence or conversation (whether the terms included in the sentence or conversation match the correction requests included in the correction request dictionary). If a correction request is included in the first sentence, the correction request may be extracted from the first sentence. In this case, the correction request dictionary 7 may not be used. Alternatively, an example of a correction request (e.g., an icon representing a translation into English) may be displayed on the display unit, and the first correction request may be input to the sentence correction system 1 using this example (by clicking the icon). For example, when analyzing a correction request included in an input sentence or conversation, these sentences or conversations may be used as training data to accumulate a learning model and improve the accuracy of machine learning.

[0018] For example, the sentence correction system 1 may read from the memory unit, "Every day, I practice sales talk at home using my iPhone. I use an app to do dialogue training with AI customers to acquire a positive speaking style. I would like to use more positive expressions," and extract a sentence correction request of "I would like to use more positive expressions."Then, the sentence correction request receiving unit 5 may receive the extracted sentence correction request and store it in the memory unit as appropriate.

[0019] The corrected first sentence acquisition step (S130) is a step in which a computer corrects the first sentence in accordance with a first sentence correction request and acquires the corrected first sentence. This step includes a first sentence key extraction step (S131), a corrected sentence creation step (S132), and a corrected first sentence selection step (S133).

[0020] The first in-sentence key extraction step (S131) ​​is a step of extracting either or both of a keyword and a key phrase included in the first sentence and acquiring a first in-sentence key, which is either or both of the keyword and the key phrase included in the first sentence. For example, the first in-sentence key may be acquired by the in-sentence key acquisition unit 9 of the sentence correction system 1. The sentence correction system 1 may have, for example, an in-sentence key dictionary 11, and the in-sentence key acquisition unit may extract either or both of the keyword and the key phrase included in the first sentence by referring to the in-sentence key dictionary 11. For example, if the first sentence contains a keyword identical to a keyword included in the in-sentence key dictionary, the in-sentence key acquisition unit 9 may extract the keyword as the in-sentence key and acquire the first in-sentence key. Furthermore, the in-sentence key dictionary 11 may store a learning model, and the learning model may be constructed by machine learning using past input sentences and past in-sentence keys as training data. In this way, by using the sentence correction system 1, the accuracy of the sentence correction system 1 can be improved. In this case, the in-sentence key acquisition unit 11 can obtain the first in-sentence key by reading the first sentence from the storage unit and performing machine learning using the read first sentence. The obtained first in-sentence key may be stored in the storage unit as appropriate. The first in-sentence key may also be acquired without using the in-sentence key dictionary 11.

[0021] For example, the in-sentence key acquisition unit 9 reads from the storage unit a first sentence, "Every day, I practice sales talk at home using my iPhone. I use an app to practice conversations with AI customers to acquire a positive way of speaking," and references the keywords and key phrases "iPhone," "sales talk," "app," "acquire a positive way of speaking," and "conversation with AI customers" stored in the in-sentence key dictionary 11, and acquires these as in-sentence keys. The in-sentence key acquisition unit 9 stores the acquired in-sentence keys in the storage unit as appropriate.

[0022] The revised sentence creation step (S132) is a step of creating multiple revised sentence candidates that include the first sentence key. For example, the revised sentence creation unit 13 of the sentence correction system 1 uses the first sentence key acquired by the sentence key acquisition unit 9 to create multiple revised sentence candidates that include the first sentence key in the sentence using a learning model obtained by machine learning. The revised sentence creation unit 13 may create multiple revised sentence candidates that include a term or phrase associated with the first sentence key in the sentence. The revised sentence creation unit 13 appropriately stores the created multiple revised sentence candidates in a storage unit. The revised sentence creation step may be a virtual process that is temporarily performed within a computer. A method of creating sentences in this manner is publicly known, for example, as described in Japanese Patent No. 7176443.

[0023] For example, the revised sentence creation unit 13 creates the following candidates using the sentence keys "iPhone," "sales talk," "app," "learn positive speaking skills," and "talk to AI customers." Note that in the following example, a sentence is created using the sentence keys as they are. However, the sentence keys may be converted into other words, such as by converting them into synonyms (for example, using a thesaurus dictionary). I practice sales talk every day at home using my iPhone. In order to learn positive speaking skills, I use an iPhone app to conduct dialogue training with AI customers. I practice sales talk at home using my iPhone. In order to learn positive speaking skills, I use an app to conduct dialogue training with AI customers. I practice sales talk at home using an app. In order to learn positive speaking skills, I conduct dialogue training with AI customers. I practice sales talk at home using an iPhone. In order to learn positive speaking skills, I conduct dialogue training with AI customers. I practice sales talk at home using an iPhone. In order to learn positive speaking skills, I conduct dialogue training with AI customers. I practice sales talk using my iPhone at home every day. I use an app to practice conversations with AI customers to learn how to speak more positively. I practice sales talk using an app at home. I practice conversations with AI customers to learn how to speak more positively. I practice sales talk using an app at home. I practice conversations with AI customers to learn how to speak more positively. I practice sales talk using my iPhone at home every day. I use an app to practice conversations with AI customers to learn how to speak more positively. I practice sales talk using my iPhone at home. I practice conversations with AI customers to learn how to speak more positively. I practice sales talk using my iPhone at home while practicing conversations with AI customers. I use an app to learn how to speak more positively.

[0024] This sentence correction method may further include a step of outputting a plurality of candidate corrected sentences. An example of the output is a display on the user's monitor. In this case, the user can see a plurality of candidate corrected sentences. For example, the candidate output unit 14 of the sentence correction system 1 reads out and outputs a plurality of candidate corrected sentences from the storage unit. Then, the monitor of the user's terminal displays a plurality of candidate corrected sentences.

[0025] The first corrected sentence selection step (S133) is a step of selecting a sentence corresponding to the correction request from a plurality of candidate corrected sentences as the first corrected sentence. For example, the sentence selection unit 15 of the sentence correction system 1 selects a sentence corresponding to the correction request from a plurality of candidate corrected sentences. For example, the sentence selection unit 15 reads a plurality of candidate corrected sentences from a storage unit and selects a sentence corresponding to the correction request from the plurality of read candidate corrected sentences. The sentence selection unit 15 may use the correction request to select a sentence corresponding to the correction request from the plurality of read candidate corrected sentences by, for example, machine learning or artificial intelligence. The sentence selection unit 15 may store the selected first corrected sentence in the storage unit.

[0026] For example, the sentence selection unit 15 selects a sentence that corresponds to the correction request from a plurality of candidates for correction sentences, and may select the most positively phrased sentence from among the candidates.

[0027] The corrected first sentence display step (S140) is a step in which the computer displays the corrected first sentence. The sentence display unit 17, for example, reads the corrected first sentence from the storage unit and outputs it to the user's terminal. The user's terminal can then display the corrected first sentence on a monitor. Also, if the sentence correction system 1 is the user's terminal, the sentence display unit 17, for example, reads the corrected first sentence from the storage unit and displays the corrected first sentence on a display device such as a monitor. In this way, the computer can display the corrected first sentence.

[0028] A preferred embodiment of the above method further includes a talk input step (S210), a cluster classification step (S220), and a first sentence extraction step (S230).

[0029] The talk input step (S210) is a step in which a conversation or speech is input to a computer. An example of the talk input unit 21 is a microphone. For example, when a user speaks into the microphone, the speech is input to the sentence correction system 1. The speech input to the microphone is stored in the memory unit as a digital audio signal. The sentence correction system 1 has a speech analysis unit 23, which analyzes the audio signal and appropriately stores the analyzed conversation or speech in the memory unit. For example, it is sufficient to analyze the speech related to the conversation or speech and obtain speech words, which are words included in the conversation or speech. Such speech analysis methods and devices are known. The sentence correction system 1 appropriately stores speech words related to the conversation or speech in the memory unit.

[0030] The cluster classification step (S220) is a step in which a computer classifies a conversation or a story into multiple clusters. Each of the multiple clusters contains a different topic or point of discussion. Cluster classification methods are well known, as described in, for example, Japanese Patent Publication No. 7176443. The cluster classification unit 25 reads speech words related to the conversation or story from the storage unit and classifies the conversation or story into multiple clusters using a learning model. Each cluster may contain one or more sentences. The cluster classification unit 25 may also refer to the in-sentence key dictionary 11, analyze the points of the conversation or story based on keywords or key phrases, and classify the conversation or story into multiple clusters for each keyword or key phrase (or point of discussion). If the cluster classification is correct, the user may be able to input information indicating that the cluster classification is correct to the sentence correction system 1. If the cluster classification is incorrect, the user may be able to input information indicating that the cluster classification is incorrect to the sentence correction system 1. The sentence correction system 1 can improve accuracy by repeating the cluster classification step.

[0031] Figure 4 shows an example of a conversation between two people. Figure 5 is a conceptual diagram illustrating an example of cluster classification. In this example, the conversation between the two people was divided into smaller points of discussion each time the content of the conversation changed, and the points were visualized as clusters. In the graph below, the topics from the conversation between Speaker 0 and Speaker 1 were clustered into eight points of discussion, numbered 0 to 7 on the vertical axis. The horizontal axis represents the number of words. Important keywords and phrases were identified for each point of discussion and classified into topic groups (clusters). Figure 6 shows the flow of topics within the clustered topic groups. It is desirable for the sentence correction system 1 to be able to display a cluster classification diagram like that shown in Figure 5 and a topic group flow diagram like that shown in Figure 6 on a display unit.

[0032] The first sentence extraction step (S230) is a step in which a computer extracts a first sentence from one of multiple clusters. The first sentence extracted in this step becomes the first sentence input to the computer in the first sentence input step (S110). The sentence extraction unit 27 extracts a representative sentence from each cluster. The first sentence may be extracted when the clusters are classified. The first sentence may be, for example, a sentence containing a keyword or key phrase. In this case, a learning model stores the keywords and key phrases, and the sentence extraction unit 27 can extract a first sentence containing those keywords or key phrases from a certain cluster by referring to the learning model. The sentence correction system 1 may also extract a second sentence, a third sentence, etc., which are representative sentences from each of multiple clusters. The second sentence and subsequent sentences may be corrected as required in the same manner as the first sentence.

[0033] The next invention relates to a program for causing a computer to execute the above-described sentence correction method, and a non-transitory information recording medium on which the program is recorded. Examples of recording media include CDs, CD-ROMs, DVDs, USB memory, hard disks, and disks on a server. Such a program is stored in, for example, a memory. The program then causes a processor or computer to execute the sentence correction method, which includes receiving a request to correct a first sentence, extracting either or both of keywords and key phrases contained in the first sentence, obtaining a first sentence key that is either or both of the keywords and key phrases contained in the first sentence, creating multiple corrected sentence candidates including the first sentence key, and selecting from the multiple corrected sentence candidates a sentence that corresponds to the first sentence correction request as the corrected first sentence. The sentence correction method includes various aspects described in this specification.

[0034] The present invention can be used in the information-related industry.

[0035] REFERENCE SIGNS LIST 1 Sentence correction system 3 Sentence input unit 5 Sentence correction request receiving unit 7 Correction request dictionary 9 Sentence key acquisition unit 11 Sentence key dictionary 13 Corrected sentence creation unit 15 Sentence selection unit 17 Sentence display unit 21 Talk input unit 23 Speech analysis unit 25 Cluster classification unit 27 Sentence extraction unit

Claims

1. A method of correcting a sentence by a computer, comprising: a first sentence input step of inputting a first sentence to the computer; a first sentence correction request receiving step of receiving a request to correct the first sentence by the computer; a corrected first sentence acquisition step of correcting the first sentence in accordance with the request to correct the first sentence and acquiring the corrected first sentence by the computer; and a corrected first sentence display step of displaying the corrected first sentence by the computer, wherein the corrected first sentence acquisition step includes a first sentence key extraction step of extracting either or both of a keyword and a key phrase contained in the first sentence and acquiring a first sentence key which is either or both of a keyword and a key phrase in the first sentence; and a corrected sentence creation step of creating a plurality of candidates for corrected sentences including the first sentence key. and a corrected first sentence selection step of selecting a sentence corresponding to the correction request for the first sentence from the plurality of corrected sentence candidates as the corrected first sentence.

2. A sentence correction method as described in claim 1, further comprising: a talk input step in which a conversation or talk is input to the computer; a cluster classification step in which the computer classifies the conversation or talk into a plurality of clusters, each of the plurality of clusters including a different topic or point of discussion; and a first sentence extraction step in which the computer extracts a first sentence from any one of the plurality of clusters, wherein the first sentence input to the computer in the first sentence input step is the first sentence extracted in the first sentence extraction step.

3. A method for correcting a sentence as described in claim 1, wherein the first sentence is a sentence contained in a conversation or story, and the correction request includes a correction request aimed at making the sentence more positive, grammatically correct, more accurate, more understandable, and more commonly used.

4. A program for causing a computer to execute a sentence correction method, the program including: a first sentence input step for inputting a first sentence; a first sentence correction request receiving step for receiving a request to correct the first sentence; a corrected first sentence acquisition step for correcting the first sentence in accordance with the request to correct the first sentence and acquiring the corrected first sentence; and a corrected first sentence display step for displaying the corrected first sentence.

5. A non-transitory information recording medium on which the program according to claim 4 is recorded.