Sentence processing device, sentence processing method, and program

Abstract

A sentence processing device 100 according to the present disclosure comprises a structuring unit 121 that analyzes the structure of a first preregistered sentence and an inputted second sentence, and a detection unit 122 that detects a part of sentence in the first sentence as a candidate for the sentence to be corrected in the second sentence on the basis of the structure obtained by analyzing the first sentence and the second sentence. For example, by detecting the correction candidate on the basis of the structure of the first sentence and the second sentence analyzed using a machine learning model, it is possible to assist in decision making when correcting a sentence.

Description

Text processing device, text processing method, and program

[0001] The present disclosure relates to a text processing device, a text processing method, and a program.

[0002] Speech recognition is commonly used when inputting text into an information processing device. Accordingly, techniques for correcting errors in speech recognition have also been studied. For example, in Patent Literature 1, the topic of a conversation that is speech-recognized is identified, and terms related to that topic are presented as correction candidates.

[0003] Japanese Patent Application Laid-Open No. 2021-117475

[0004] However, the technology described in Patent Document 1 requires grasping the topic of the conversation, and if the topic cannot be grasped from the conversation, it is difficult to correct errors with high accuracy. Furthermore, it is difficult to accurately correct errors such as incorrect input or incorrect conversion, not only for text input by speech recognition but also for text input by any method.

[0005] Therefore, an object of the present disclosure is to solve the above-mentioned problem that it is difficult to correct text with high accuracy.

[0006] A sentence processing device according to one embodiment of the present disclosure is configured to include: a structuring unit that analyzes the structure of a pre-registered first sentence and an input second sentence; and a detection unit that detects a portion of the first sentence as a candidate sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence.

[0007] Furthermore, a text processing method according to one embodiment of the present disclosure is configured to analyze the structure of a pre-registered first text and an input second text, and based on the analyzed structures of the first text and the second text, detect a portion of the first text as a candidate for a text to be corrected in the second text.

[0008] Furthermore, a program according to one embodiment of the present disclosure is configured to cause a computer to execute a process of analyzing the structure of a pre-registered first sentence and an input second sentence, and detecting a portion of the first sentence as a candidate sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence.

[0009] With the above-described configuration, the present disclosure can improve the accuracy of corrections to text.

[0010] FIG. 1 is a block diagram showing the configuration of a first text processing device according to the present disclosure. FIG. 2 is a diagram showing the processing state of the first text processing device according to the present disclosure. FIG. 3 is a diagram showing the processing state of the first text processing device according to the present disclosure. FIG. 4 is a diagram showing the processing state of the first text processing device according to the present disclosure. FIG. 5 is a flowchart showing the processing operation of the first text processing device according to the present disclosure. FIG. 6 is a diagram showing the processing state of the second text processing device according to the present disclosure. FIG. 7 is a diagram showing the processing state of the second text processing device according to the present disclosure. FIG. 8 is a block diagram showing the hardware configuration of a third text processing device according to the present disclosure. FIG. 9 is a block diagram showing the configuration of the third text processing device according to the present disclosure.

[0011] First Embodiment A first embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings may be relevant to any embodiment.

[0012] [Configuration] The text processing device 10 in this embodiment is used to support the correction of errors in input text information. For example, the text processing device 10 in this embodiment is used to present sentences such as words that are candidates for correction for input text information when text information describing a doctor's examination status of a patient is input into an electronic medical record at a hospital. However, the text information for which the text processing device 10 supports the correction of errors is not limited to text information input into an electronic medical record, and may be any text information.

[0013] The text processing device 10 is composed of one or more information processing devices each having a calculation device and a storage device. As described above, the text processing device 10 in this embodiment is used when a doctor inputs text information into an electronic medical record, and is therefore installed, for example, in an examination room in a hospital. However, the text processing device 10 may be installed anywhere, and may be configured as a cloud server on a network.

[0014] As shown in Figure 1, the text processing device 10 includes an acquisition unit 11, a structuring unit 12, a detection unit 13, and an output unit 14. The functions of the acquisition unit 11, structuring unit 12, detection unit 13, and output unit 14 can be realized by the calculation device executing a program for realizing each function stored in a storage device. The text processing device 10 also includes a structured model storage unit 16. The structured model storage unit 16 is configured from a storage device. The text processing device 10 is also connected to an input device 21, a display device 22, and an electronic medical record storage device 30. Each component will be described in detail below.

[0015] The electronic medical record storage device 30 is composed of a storage device and an information processing device that stores electronic medical record data of patients that have been registered up to now. The electronic medical record includes text information describing the status of past examinations and treatments for the patient, entered by the doctor who examined the patient or the medical professional who performed the treatment, in association with the patient's name and identification information. The electronic medical record may be stored in the storage device of any information processing device, or may be stored in the text processing device 10.

[0016] The input device 21 is a device capable of inputting information, such as a keyboard or microphone, and is used by a doctor who has just examined a patient or a medical professional who has performed treatment to input written information describing the status of the patient's examination or treatment. For example, if the input device 21 is a keyboard, the written information is input as text information, and if the input device 22 is a microphone, the written information is input as voice information. Note that the written information input by the doctor or medical professional may be the content of the medical interview (question) for the patient itself, i.e., the voice information itself when speaking to the patient, or may be information representing the content of the doctor's or medical professional's opinion.

[0017] The display device 22 is a device capable of displaying information such as a display, and displays information contained in the electronic medical record. As will be described later, the display device 22 also displays text information indicating the status of new examinations and treatments to be entered into the electronic medical record, as well as information regarding corrections to the newly entered text information.

[0018] The acquisition unit 11 acquires the electronic medical record of a target patient, such as a patient who has been newly diagnosed by a doctor or treated by a medical professional, from the electronic medical record storage device 30. As a result, the acquisition unit 11 acquires electronic medical record information, which is text information (first text) that describes the past condition of the target patient. The acquisition unit 11 is not limited to acquiring the electronic medical record information of the target patient, but may also acquire electronic medical record information of other patients related to the target patient, for example, other patients who correspond to the target patient by having the same medical condition or disease name. For example, if the target patient is a new patient, the acquisition unit 11 may acquire electronic medical record information of other patients.

[0019] Furthermore, when a doctor examines a patient or a medical professional performs treatment on the patient, the acquisition unit 11 accepts and acquires newly input text information (second text) that describes the status of the examination or treatment. In this embodiment, the newly input text information is particularly assumed to be voice information. In this case, the acquisition unit 11 performs voice recognition processing on the acquired voice information and converts it into input information, which is text information. However, the newly input text information may be the text information itself input from a keyboard or the like.

[0020] The structuring unit 12 performs a text structuring process that analyzes the structure of the electronic medical record information of the target patient acquired as described above. Specifically, the text structuring process by the structuring unit 12 involves structurally analyzing the sentence information, which is the electronic medical record information, to extract words and assign a label to each word corresponding to the word. At this time, the label corresponds to a pre-set type that classifies the word according to its content, such as "symptoms," "time," "disease name," and "condition (medical)." A structured model, which is correspondence information between words and labels, is prepared in advance, and the structuring unit 12 performs the text structuring process using this structured model. At this time, the structured model is pre-stored in the structured model storage unit 16.

[0021] FIG. 2 shows an example of text structuring processing of electronic medical record information by the structuring unit 12. First, when the electronic medical record information is text information as shown on the left side of FIG. 2, the structuring unit 12 extracts each word from the text information, such as "recently," "fatigue," "dizziness," "feeling," "heavy," "one month ago," "occasionally," "bloody stool," "rectal cancer," "diagnosis," "hospitalization," etc. Then, the structuring unit 12 assigns labels corresponding to the extracted words using a structured model. For example, as shown on the right side of FIG. 2, the structuring unit 12 assigns the label "symptom" to the words "fatigue," "dizziness," and "bloody stool," the label "time" to the word "one month ago," the label "disease name" to the word "rectal cancer," and the label "condition (medical)" to the word "hospitalization." At this time, there may be words that do not have labels assigned. Note that the electronic medical record information, extracted words, and pre-set labels described above are merely examples and may represent any information. The above-mentioned words are not limited to character strings consisting of words only, but may include phrases in which other characters are added to words.

[0022] Here, the structured model used in the text structuring process by the structuring unit 12 described above may be a machine learning model. In this case, the machine learning model is generated by performing supervised machine learning using training data in which many words extracted from the electronic medical record information described above are each assigned a label as training data. Therefore, the structuring unit 12 can input words extracted from the electronic medical record information into the machine learning model, and obtain labels to be assigned to the words as output from the machine learning model. Note that the machine learning model may perform machine learning on not only the words and labels but also sentence information containing the words, and may be trained to output labels to be assigned to the words taking into account the sentence information containing the words. Note that the structured model used in the text structuring process by the structuring unit 12 may be a rule-based model.

[0023] The structuring unit 12 also performs a text structuring process that analyzes the structure of new input information for the target patient acquired as described above. Specifically, the text structuring process by the structuring unit 12, similar to the text structuring process for the electronic medical record information described above, analyzes the structure of the sentence information that is the input information, extracts words, and assigns a label to each word according to that word. The structuring unit 12 then performs the text structuring process using the structure model described above.

[0024] FIG. 3 shows an example of text structuring processing performed by the structuring unit 12 on input information. First, when the input information is text information as shown on the left side of FIG. 3, the structuring unit 12 extracts each word from the text information, such as "1 week," "physical condition," "dizziness," "blood," etc. Then, the structuring unit 12 assigns labels corresponding to the extracted words using a structured model. For example, as shown on the right side of FIG. 3, the structuring unit 12 assigns the label "symptoms" to the word "dizziness" and the label "time" to the word "1 week." Note that the labels "disease name" and "condition (medical)" assigned to the words extracted from the electronic medical record information described above are left as "(blank)" because there are no corresponding words in the input information. Note that the structuring unit 12 can also use the machine learning model described above in the text structuring processing of the input information described above.

[0025] The detection unit 13 detects some sentences, i.e., words, in the electronic medical record information as correction candidates for sentences, i.e., words, to be corrected and entered in the input information, based on the text-structured information obtained by text-structuring the electronic medical record information and the input information, as described above. Specifically, as shown in FIG. 4 , the detection unit 13 compares corresponding words for each label in the text-structured information of the electronic medical record information and the text-structured information of the input information. In particular, the detection unit 13 identifies labels in the electronic medical record information that have corresponding words but do not have corresponding words in the input information. In the example of FIG. 4 , the detection unit identifies the labels "Disease Name" and "Condition (Medical)" that contain the word "(blank)" in the input information. Then, the detection unit detects the words "Rectal Cancer" and "Hospitalization" in the electronic medical record information that correspond to the labels "Disease Name" and "Condition (Medical)" that contain the word "(blank)" as correction candidates for the input information.

[0026] The detection unit 13 may also detect correction candidate words based on the results of comparing the number of words in the electronic medical record information with the number of words in the input information for each label. For example, if the number of words in the input information is less than the number of words in the electronic medical record information for a given label by a threshold or more, the detection unit 13 detects the words in the electronic medical record information for that label as correction candidates. In the example shown in FIG. 5 , for the label "symptoms," the electronic medical record information contains three words, "tiredness," "dizziness," and "bloody stool," while the input information contains one word, "vertigo," meaning that the input information contains two fewer words. In this case, if the threshold value "2" is set, the detection unit 13 detects the words "tiredness" and "bloody stool" in the electronic medical record information corresponding to the label "symptoms" as correction candidates because the number of words in the input information for the label "symptoms" is less than the threshold or more. In this case, the detection unit 13 determines that the word "dizziness" in the electronic medical record information corresponding to the label "symptoms" and the word "vertigo" in the input information are the same word, and detects only words that do not appear in the input information as correction candidates.

[0027] The detection unit 13 is not necessarily limited to detecting words that are correction candidates using the above-described method. For example, the detection unit 13 may detect, for each label, all words in the electronic medical record information that are not in the input information as correction candidates. Furthermore, as described above, the detection unit 13 may detect words in the electronic medical record information as correction candidates using any method depending on the results of comparing words in the input information of the electronic medical record information for each label.

[0028] Furthermore, the detection unit 13 detects the position of a correction candidate in the input information based on the detection result of the correction candidate word. For example, as described above, when a correction candidate word is detected, the position of a word in the input information to which a label corresponding to the correction candidate word is assigned may be detected as the correction candidate position. In the example of FIG. 5, a word corresponding to the label "symptoms" in the electronic medical record information is detected as a correction candidate. In this case, the position of the word "dizziness" corresponding to the label "symptoms" in the input information may be detected as the correction candidate position. As another example, when the detection unit 13 detects a correction candidate word, the detection unit 13 may detect the position of another word in the input information that is phonemically similar to the correction candidate word as the correction candidate position. Note that the detection unit 13 may also detect the position of a correction candidate in the input information based on the detected correction candidate using other methods.

[0029] The output unit 14 outputs the correction candidate words detected as described above to the display device 22 for display. For example, the output unit 14 displays the voice information that is the source of the input information on the display device 22 so that the doctor or medical professional who inputted the information can see it, as described above. At this time, the output unit 14 may also display the input information, which is the voice information input by the doctor or medical professional converted into text through voice recognition, on the display device 22. The output unit 14 may also display the patient's electronic medical record on the display device 22, and display the input information converted into text through voice recognition in a predetermined input field of the electronic medical record so that it can be corrected, and may also display the correction candidate words detected as described above. The output unit 14 may also display the input information and, in addition, may display the correction candidate positions detected as described above on the input information so that they can be identified. Note that the output unit 14 is not necessarily limited to displaying all of the input information, and may display only a portion of the input information, or an extracted summary of the input information. At this time, the output unit 14 may display a portion of the input information around a position where the correction candidate word may correspond, or may display the correction candidate position on the input information so that it can be recognized, and may also display the correction candidate word. This allows, for example, a medical professional to refer to the output correction candidate word and decide whether or not to correct the recognition result, or to correct an error in the recognition result. In this way, it is possible to support the medical professional's decision-making.

[0030] [Operation] Next, the operation of the above-mentioned text processing device 10 will be described. First, the text processing device 10 acquires the electronic medical record of a target patient, such as a new diagnosis by a doctor or treatment by a medical professional, from the electronic medical record storage device 30 (step S1 in FIG. 6). The text processing device 10 also accepts input of voice information representing the status of the examination or treatment of the target patient, performs voice recognition processing, and converts it into input information, which is text information (step S2 in FIG. 6).

[0031] Next, the text processing device 10 performs a text structuring process to analyze the structures of the electronic medical record information and the input information. Specifically, the text processing device 10 analyzes the structure of the electronic medical record information and the text information that is the input information to extract words (step S3 in FIG. 6). Then, the text processing device 10 assigns a label corresponding to each word to each word (step S4 in FIG. 6). For example, as shown in FIGS. 2 and 3, the text processing device 10 assigns a label (such as "symptoms," "time," "disease name," or "condition (medical)") to each word included in the electronic medical record information and the input information.

[0032] Next, the text processing device 10 compares the corresponding words for each label between the text structured information of the electronic medical record information and the text structured information of the input information (step S5 in FIG. 6 ). Then, the text processing device 10 detects correction candidates for words in the input information based on the comparison results for each label (step S6 in FIG. 6 ). For example, when a corresponding word exists in the electronic medical record information but not in the input information for the same label, the text processing device 10 detects the word included in the electronic medical record information corresponding to that label as a correction candidate. In the example of FIG. 4 , the words "rectal cancer" and "hospitalization" in the electronic medical record information corresponding to the labels "Disease Name" and "Condition (Medical)" that contain the word "(blank)" in the input information are detected as correction candidates for the input information. Furthermore, in the example of FIG. 5 , the number of words in the label "Symptoms" in the electronic medical record information is three, while the number of words in the input information is extremely small. Therefore, the words in the electronic medical record information for the label "Symptoms" are detected as correction candidates.

[0033] Then, the text processing device 10 displays the correction candidate words detected as described above on the display device 22. At this time, the text processing device 10 may also display input information, which is speech information converted into text, on the display device 22. Furthermore, the text processing device 10 may detect the position of the correction candidate within the input information according to the detection result of the correction candidate word, and display the position of the correction candidate together with the input information.

[0034] As described above, when a doctor or other person inputs a new sentence describing a patient's condition into an electronic medical record, the text processing device 10 detects and provides correction candidates for the newly input sentence based on sentences contained in previously registered electronic medical records. This makes it possible to support the doctor or other person in inputting new sentences into the electronic medical record, and in particular to make highly accurate corrections to the sentences.

[0035] Second Embodiment A second embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings may be relevant to any embodiment.

[0036] [Configuration] The text processing device 10 in this embodiment is configured in the same manner as the text processing device 10 in the first embodiment described above. That is, the text processing device 10 has the configuration shown in Fig. 1. Of these, the text processing device 10 in this embodiment differs from the first embodiment mainly in the functions of the structuring unit 12, the detection unit 13, and the output unit 14. Below, the configuration that differs from the first embodiment will be described in detail.

[0037] The structuring unit 12 performs a text structuring process that analyzes the structure of the acquired electronic medical record information of the target patient. Specifically, the text structuring process by the structuring unit 12 in this embodiment structurally analyzes the text information, which is the electronic medical record information, to extract words and extract the relationship between the words, i.e., between two paired words. At this time, a structured model that outputs the relationship between paired words is prepared in advance, and the structuring unit 12 extracts the relationship between paired words using this structured model. In particular, the structuring unit 12 uses the structured model to output a higher score (degree of relationship) the higher the relationship between paired words. Note that the above-mentioned words are not limited to character strings consisting of words only, and may include phrases in which other characters are added to words.

[0038] Here, the above-mentioned structured model is a machine learning model generated by supervised machine learning using training data in which two pairs of predetermined words are assigned scores as training data based on, for example, the frequency with which the two words appear in the same electronic medical record information and the distance between the two words in the same electronic medical record information in many past electronic medical record information. In this case, the machine learning model is trained to output a higher score, for example, the more frequently the two pairs of predetermined words appear in the same electronic medical record information and the closer the distance between the two words in the same electronic medical record information. The structured model is pre-stored in the structured model storage unit 16. However, the structuring unit 12 may output the score of the relevance between words using other methods.

[0039] An example of text structuring processing of electronic medical record information by the structuring unit 12 is shown in Fig. 7. First, when the electronic medical record information is text information as shown on the left side of Fig. 7, the structuring unit 12 extracts each word in the text information, for example, "cirrhosis," "hospitalization," "hyperammonemia," "two days ago," "Monilac syrup," "taking," etc. Then, the structuring unit 12 extracts a score representing the association between the extracted pairs of words using the structured model. For example, as shown on the right side of Figure 7, the structuring unit 12 outputs the following: the score of association between the word "cirrhosis" and the word "hyperammonemia" is "0.4", the score of association between the word "cirrhosis" and the word "hospitalized" is "0.08", the score of association between the word "cirrhosis" and the word "two days ago" is "0.0003", the score of association between the word "cirrhosis" and the word "Monilac syrup" is "0.8", the score of association between the word "hyperammonemia" and the word "Monilac syrup" is "0.8", and the score of association between the word "Monilac syrup" and the word "take" is "0.9".

[0040] The structuring unit 12 also performs a text structuring process that analyzes the structure of the new input information for the target patient acquired as described above. Specifically, the text structuring process by the structuring unit 12 extracts words by structurally analyzing the sentence information that is the input information. For example, the structuring unit 12 extracts the words "hyperammonemia," "Eurax," "syrup," "prescription," "administration," and so on, as shown by the dotted oval on the right side of FIG. 7 .

[0041] As described above, the detection unit 13 detects sentences, i.e., words, related to words in the input information as correction candidates based on the relevance score between pairs of words in the electronic medical record information. Specifically, for each word extracted from the input information, the detection unit 13 checks the relevance score between pairs of words in the electronic medical record information containing the word. At this time, if the relevance score between pairs of words in the electronic medical record information containing a word in the input information is equal to or greater than a threshold, the detection unit 13 detects, as correction candidates, words that are not included in the input information, i.e., words that are only included in the electronic medical record information, among the pairs of words that make up the words.

[0042] Here, the process by which the detection unit 13 detects correction candidates will be described using the example shown in FIG. 8 . Assume that the score threshold is set to "0.1." First, it is found that paired words in the electronic medical record information containing the word "hyperammonemia" in the input information include the pair "cirrhosis" and the word "hyperammonemia," and the pair "hyperammonemia" and the word "Monilac syrup," both of which have scores above the threshold. In this case, of these paired words, the words "cirrhosis" and "Monilac syrup," which are different from the word "hyperammonemia" in the input information, are detected as correction candidates. Furthermore, it is found that paired words in the electronic medical record information containing the word "administer" in the input information include the pair "Monilac syrup" and the word "administer," which have scores above the threshold. In this case, of these paired words, the word "Monilac syrup," which is different from the word "administer" in the input information, is detected as a correction candidate. Note that for other words in the input information, no correction candidates are detected because there are no paired words containing those words.

[0043] The detection unit 13 is not necessarily limited to detecting words that are correction candidates using the above-described method. For example, the detection unit 13 may detect, as correction candidates, all of the related pairs of words in the electronic medical record information that include the word in the input information, without requiring that the score be equal to or greater than a threshold.

[0044] The detection unit 13 also detects the position of a correction candidate in the input information based on the detection result of the correction candidate word. For example, when the detection unit 13 detects a correction candidate word as described above, the detection unit 13 may detect the position of a word in the input information that is paired with the correction candidate word as the correction candidate position. In the example of FIG. 8 , the words "cirrhosis" and "Monilac syrup" are detected as correction candidates, but the positions of the words "hyperammonemia" and "administer" that are paired with these in the input information may be detected as correction candidate positions. The detection unit 13 may also detect the position of a word in the input information that is similar to the correction candidate word based on a predetermined criterion, such as a character string or phoneme, as the correction candidate position. In the example of FIG. 8 , the word "Monilac syrup" is detected as a correction candidate, but the positions of the words "Eurax" and "syrup" in the input information that are determined to be similar to the word based on a predetermined criterion may be detected as correction candidate positions. Note that the detection unit 13 may also detect the position of a correction candidate in the input information based on the detected correction candidates using other methods.

[0045] The output unit 14 outputs the correction candidate words detected as described above to the display device 22 for display. For example, as shown in FIG. 8 , the output unit 14 may display the correction candidate words together with the words of the input information. At this time, the output unit 14 may also display, on the display device 22, input information obtained by converting voice information input by a doctor or medical professional into text through voice recognition. Furthermore, the output unit 14 may display the patient's electronic medical record on the display device 22, and may display the voice-recognized input information in a predetermined input field of the electronic medical record so that it can be corrected, and may also display the correction candidate words detected as described above. Furthermore, the output unit 14 may display the input information and, at the same time, may display the correction candidate positions detected as described above on the input information.

[0046] [Operation] Next, the operation of the above-mentioned text processing device 10 will be described. First, the text processing device 10 acquires the electronic medical record of a target patient, such as a new diagnosis by a doctor or treatment by a medical professional, from the electronic medical record storage device 30 (step S11 in FIG. 9). In addition, the text processing device 10 accepts input of voice information representing the status of the examination and treatment of the target patient, and converts it into text input information by voice recognition processing (step S12 in FIG. 9).

[0047] Next, the text processing device 10 performs a text structuring process to analyze the structures of the electronic medical record information and the input information. Specifically, the text processing device 10 extracts words by structurally analyzing the electronic medical record information and the input information (step S13 in FIG. 9). Then, the text processing device 10 extracts the relationship between words included in the electronic medical record information, i.e., between two pairs of words (step S14 in FIG. 9). For example, the text processing device 10 extracts a score representing the relationship between pairs of words, as shown in FIG. 7, using a machine learning model trained to output a score representing the relationship between pairs of words.

[0048] Next, the text processing device 10 extracts words related to words in the input information from among pairs of related words in the electronic medical record information (step S15 in FIG. 9). That is, the text processing device 10 extracts pairs of words in the electronic medical record information that include words in the input information. Then, if the score of the association between pairs of words in the electronic medical record information that include words in the input information is equal to or greater than a threshold, the text processing device 10 detects, as correction candidates, words that are not included in the input information, i.e., words that are only included in the electronic medical record information, from among the pairs of words that make up the pairs of words (step S16 in FIG. 9). For example, in the example shown in FIG. 8, the words "cirrhosis" and "Monilac syrup" in the electronic medical record information are detected as correction candidates for the word "hyperammonemia" in the input information.

[0049] Then, the text processing device 10 displays the correction candidate words detected as described above on the display device 22. At this time, the text processing device 10 may also display information obtained by converting the audio information, which is the input information, into text on the display device 22. Furthermore, the text processing device 10 may detect the position of the correction candidate within the input information according to the detection result of the correction candidate word, and display the position of the correction candidate together with the input information.

[0050] As described above, when a doctor or other person inputs a new sentence describing a patient's condition into an electronic medical record, the text processing device 10 detects and provides correction candidates for the newly input sentence based on sentences contained in previously registered electronic medical records. This makes it possible to support the doctor or other person in inputting new sentences into the electronic medical record, and in particular to make highly accurate corrections to the sentences.

[0051] Third Embodiment Next, a third embodiment of the present disclosure will be described with reference to the drawings. This embodiment shows an outline of the configuration of the text processing device described in the above-mentioned embodiments. Note that the drawings may be relevant to any of the embodiments.

[0052] First, the hardware configuration of the text processing device 100 will be described with reference to Fig. 10. The text processing device 100 is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; storage device 105 storing programs 104; drive device 106 for reading and writing data from and to a storage medium 110 external to the information processing device; communication interface 107 for connecting to a communication network 111 external to the information processing device; input / output interface 108 for inputting and outputting data; and bus 109 for connecting the various components.

[0053] 10 shows an example of the hardware configuration of an information processing device that is the document processing device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with only a part of the above-described configuration, such as excluding the drive device 106. Furthermore, instead of the above-described CPU, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0054] The document processing device 100 can be equipped with the structuring unit 121 and the detection unit 122 shown in Fig. 11 by having the CPU 101 acquire and execute the program group 104. The program group 104 may be stored in advance in the storage device 105 or the ROM 102, for example, and loaded into the RAM 103 by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, with the drive device 106 reading out the programs and supplying them to the CPU 101. However, the structuring unit 121 and the detection unit 122 may be constructed using dedicated electronic circuits for realizing such means.

[0055] The structuring unit 121 analyzes the structure of a pre-registered first sentence and an input second sentence. Here, the first sentence is, for example, a sentence such as diagnostic information included in a patient's electronic medical record, and the second sentence is a sentence expressing information about the diagnosis and treatment for the patient. Then, as part of the structural analysis of the first sentence and the second sentence, the structuring unit 121 extracts words and assigns labels to them, and extracts relationships between words.

[0056] The detection unit 122 detects some sentences in the first sentence as candidates for sentences to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence. For example, the detection unit 122 detects words in the first sentence that have the same label but are not in the second sentence as candidates. Also, for example, the detection unit 122 detects related pairs of words in the first sentence that include words in the second sentence as candidates.

[0057] With the above-described configuration, the present disclosure detects candidates for sentences to be corrected within a pre-registered first sentence in a second sentence based on the structures of the first sentence and the second sentence, thereby enabling highly accurate correction of the sentences.

[0058] In addition, at least one of the functions of the above-mentioned structuring unit 121 and detection unit 122 may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0059] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0060] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0061] <Supplementary Notes> Some or all of the above embodiments can also be described as in the following supplementary notes. Below, an outline of the configurations of a text processing device, a text processing method, and a program according to the present disclosure will be described. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) A text processing device comprising: a structuring unit that analyzes the structure of a pre-registered first sentence and an input second sentence; and a detection unit that detects a part of the first sentence as a candidate for a sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence. (Supplementary Note 2) The text processing device according to Supplementary Note 1, wherein the structuring unit assigns predetermined labels to words in the first sentence and words in the second sentence according to the words, and the detection unit compares the words in the first sentence with the words in the second sentence for each label, and detects the words in the first sentence as the candidates. (Supplementary Note 3) The text processing device according to Supplementary Note 2, wherein the detection unit detects, as the candidate, a word in the first sentence that is not in the second sentence, for a specified label. (Supplementary Note 4) The text processing device according to Supplementary Note 3, wherein the detection unit detects the word in the first sentence as the candidate when, for the specified label, the number of words in the second sentence is less than the number of words in the first sentence by a threshold or more. (Supplementary Note 5) The text processing device according to Supplementary Note 3, wherein the detection unit detects the word in the first sentence as the candidate when, for the specified label, the word in the second sentence is not present. (Supplementary Note 6) The text processing device according to Supplementary Note 1, wherein the structuring unit extracts associations between words in the first sentence, and the detection unit detects, as the candidate, a word in the first sentence that is associated with a word in the second sentence based on the association. (Supplementary Note 7) A text processing device according to Supplementary Note 6, wherein the detection unit detects, as the candidate, a part of the words between the words in the first sentence that include the word in the second sentence, among the words between the words in the first sentence that have the correlation.(Supplementary Note 8) The text processing device according to Supplementary Note 6, wherein the structuring unit extracts degrees of association between the words in the first sentence, and the detection unit detects, as the candidates, words in the first sentence that are related to the words in the second sentence among the words in the first sentence whose degrees of association are equal to or greater than a threshold. (Supplementary Note 9) The text processing device according to Supplementary Note 1, comprising a position detection unit that detects a correction position of the candidate in the second sentence based on a detection result of the candidate. (Supplementary Note 10) The text processing device according to Supplementary Note 9, wherein the position detection unit detects, as the correction position, a position of a sentence in the second sentence that corresponds to the candidate. (Supplementary Note 11) The text processing device according to Supplementary Note 2, wherein the structuring unit obtains labels to be assigned to the words output from the model by inputting the words to a model, and the model is a model generated by machine learning using training data in which the labels are assigned to each of the words as training data. (Supplementary Note 12) The text processing device according to Supplementary Note 1, comprising an output unit that outputs candidates detected by the detection unit to a display device. (Supplementary Note 13) The text processing device according to Supplementary Note 12, wherein the output unit outputs the candidates to support decision-making by a medical professional. (Supplementary Note 14) A text processing method that analyzes the structures of a pre-registered first sentence and an input second sentence, and detects a part of the first sentence as a candidate for a sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence. (Supplementary Note 15) The text processing method according to Supplementary Note 14, wherein a label set in advance according to the word is assigned to a word in the first sentence and a word in the second sentence, and the word in the first sentence is compared with the word in the second sentence for each label, and the word in the first sentence is detected as the candidate.(Supplementary Note 16) The sentence processing method according to Supplementary Note 14, comprising: extracting associations between words in the first sentence; and detecting words in the first sentence that are associated with words in the second sentence as the candidates based on the associations. (Supplementary Note 17) A computer-readable storage medium storing a program for causing a computer to execute a process of: analyzing the structure of a pre-registered first sentence and an input second sentence; and detecting a part of the first sentence as a candidate for a sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence. (Supplementary Note 18) A computer-readable storage medium storing the program according to Supplementary Note 17, comprising: assigning predetermined labels to words in the first sentence and words in the second sentence according to the words, and comparing the words in the first sentence and the words in the second sentence for each label, and detecting the words in the first sentence as the candidates. (Appendix 19) A computer-readable storage medium storing the program described in Appendix 17, which causes a computer to execute the following process: extracting associations between words in the first sentence; and detecting, as the candidates, words in the first sentence that are associated with words in the second sentence based on the associations.

[0062] REFERENCE SIGNS LIST 10 Text processing device 11 Acquisition unit 12 Structuring unit 13 Detection unit 14 Output unit 16 Structured model storage unit 21 Input device 22 Display device 30 Electronic medical record storage device 100 Text processing device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Structuring unit 122 Detection unit

Claims

1. A text processing device comprising: a structuring unit that analyzes the structure of a pre-registered first sentence and an input second sentence; and a detection unit that detects a portion of the first sentence as a candidate for a sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence.

2. A text processing device as described in claim 1, wherein the structuring unit assigns predetermined labels to words in the first sentence and words in the second sentence according to the words, and the detection unit compares the words in the first sentence with the words in the second sentence for each label, and detects the words in the first sentence as candidates.

3. A text processing device according to claim 2, wherein the detection unit detects, for a given label, a word in the first sentence that is not in the second sentence as the candidate.

4. A text processing device as described in claim 3, wherein the detection unit detects the words in the first sentence as candidates when, for a given label, the number of words in the second sentence is less than the number of words in the first sentence by a threshold or more.

5. A text processing device according to claim 3, wherein the detection unit detects the word in the first sentence as the candidate when the word in the second sentence does not exist for a given label.

6. A text processing device as described in claim 1, wherein the structuring unit extracts associations between words in the first sentence, and the detection unit detects words in the first sentence that are associated with words in the second sentence as candidates based on the associations.

7. A text processing device as described in claim 6, wherein the detection unit detects as the candidates some of the words between words in the first sentence that include the word in the second sentence, among the words between words in the first sentence that have the correlation.

8. A text processing device as described in claim 6, wherein the structuring unit extracts the degree of association between the words in the first sentence, and the detection unit detects as candidates the words in the first sentence that are related to the words in the second sentence among the words in the first sentence whose degree of association is equal to or greater than a threshold value.

9. A text processing device according to claim 1, further comprising a position detection unit that detects a correction position of the candidate in the second sentence based on a result of the detection of the candidate.

10. A text processing device according to claim 9, wherein the position detection unit detects a position of a sentence in the second sentence that corresponds to the candidate as the correction position.

11. A text processing device as described in claim 2, wherein the structuring unit obtains a label to be assigned to the word output from the model by inputting the word into the model, and the model is a model generated by machine learning using training data in which the label is assigned to each of the words as training data.

12. A text processing device according to claim 1, further comprising an output section for outputting the candidates detected by said detection section to a display device.

13. A text processing device according to claim 12, wherein the output unit outputs the candidates to support decision-making by a medical professional.

14. A text processing method comprising: analyzing the structures of a pre-registered first sentence and an input second sentence; and detecting a portion of the first sentence as a candidate for a sentence to be corrected within the second sentence based on the analyzed structures of the first sentence and the second sentence.

15. A text processing method as claimed in claim 14, further comprising the steps of: assigning predetermined labels to words in the first sentence and words in the second sentence according to the words; comparing the words in the first sentence with the words in the second sentence for each label; and detecting the words in the first sentence as candidates.

16. A text processing method as claimed in claim 14, comprising: extracting associations between words in the first sentence; and detecting, as the candidates, words in the first sentence that are associated with words in the second sentence based on the associations.

17. A computer-readable storage medium storing a program for causing a computer to execute a process of analyzing the structure of a pre-registered first sentence and an input second sentence, and detecting a part of the first sentence as a candidate for a sentence to be corrected in the second sentence based on the analyzed structures of the first sentence and the second sentence.

18. A computer-readable storage medium having stored therein the program described in claim 17, the computer-readable storage medium having stored therein a program for causing a computer to execute a process of: assigning predetermined labels to words in the first sentence and words in the second sentence according to the words; comparing the words in the first sentence and the words in the second sentence for each of the labels; and detecting the words in the first sentence as candidates.

19. A computer-readable storage medium having stored thereon the program described in claim 17, the computer-readable storage medium having stored thereon a program for causing a computer to execute a process of extracting associations between words in the first sentence, and detecting, as candidates, words in the first sentence that are associated with words in the second sentence based on the associations.

Images