Patent Application
20240143942
The present disclosure relates to an information processing device and an information processing method.
When a user writes a text such as a script of a movie or an animation or a novel in which characters appear, it is necessary to consider consistency of the characters in the entire text to make the entire text consistent. To achieve consistency of the characters, an expression that is not characteristic of the character is found in a line part of the character or a part indicating a character's speech and behavior in a descriptive part, and is rewritten to an expression that is characteristic of this character.
Such various methods for generating sentences characteristic of a character are conventionally known. Examples of the methods include a method for manually rewriting sentences, a method for automatically converting sentences based on a rule (e.g., Patent Literature 1), a method for converting sentences based on machine learning (e.g., Patent Literature 2), and the like.
The method for manually rewriting the sentences has high accuracy, yet requires high cost in terms of time and cost, and is likely to overlook a part that can be mechanically extracted. On the other hand, the method for automatically converting sentences based on the rule disclosed in Patent Literature 1 and the method for automatically converting sentences based on machine learning disclosed in Patent Literature 2 seemingly requires low cost in terms of time and cost. However, according to these methods for rewriting sentences by automatically converting the sentences, it is necessary to develop rules and machine learning models meeting purposes, and there is a concern that cost increases eventually, and inappropriate sentences such as non-sentences are generated.
An object of the present disclosure is to provide an information processing device and an information processing method that can assist creation of a text in which characters appear at relatively low cost.
For solving the problem described above, an information processing device according to one aspect of the present disclosure has a detection unit that detects an expression based on a feature amount extracted from a text, and character information including information of a character using a learning model learned in advance, the expression being included in the text and indicating character likeness of the character; and a generation unit that generates a different expression that is different from the expression and indicates the character likeness based on the expression detected by the detection unit and the character information, and presents the generated different expression, wherein the detection unit relearns the learning model according to a user's reaction to the different expression presented by the generation unit.
Hereinafter, an embodiment of the present disclosure will be described in detail with reference to the drawings. Note that, in the following embodiment, the same components will be assigned the same reference numerals, and redundant description will be omitted.
Hereinafter, the embodiment of the present disclosure will be described in the following order.
First, the embodiment of the present disclosure will be schematically described. The present disclosure relates to an assist tool that assists a user to write text content such as a script of a movie, an animation, a drama, or a novel in which characters have a conversation. Here, the characters are not limited to persons, and include anthropomorphized animals, plants, and inorganic materials, simulated personalities assumed to be generated by programs, and the like. Hereinafter, these various characters are collectively referred to as characters.
In general, a script is composed of “lines” and “stage directions”, and a novel is composed of “conversational sentences” and “descriptive parts”. In the script, a “line” is a sentence for instructing words to be uttered by a character, is often enclosed in parentheses (“ ”), and is given a character name. A “stage direction” is a sentence for instructing a motion or a behavior of a character. Note that, although the script includes a “slug line” that designates a time and a place, a “slug line” is omitted here.
Furthermore, in a novel, a “conversational sentence” is a sentence indicating a conversation between a character and another character, and a descriptive part is a sentence other than the conversational sentence in the novel. The descriptive part may include a monologue of the character, in other words, a sentence from a character's viewpoint. In a novel, which character makes a conversation indicated by a “conversational sentence” is not clearly indicated in some cases. In many cases, readers of a novel can grasp which character utters a conversation indicated by the “conversational sentence” by following the context.
Note that a context (context) indicates the degree of connection of semantic contents in a flow of a text, and is formed by a logical relationship between a sentence and a sentence or a semantic association between a word and a word in many cases. Even the same word may have a different meaning depending on a context.
When, for example, writing a script and creating text data 20 of the written script, the user 30 activates the assist tool according to the embodiment of the present disclosure in the information processing device 10, and inputs the text data 20 to the information processing device 10. Note that the user may create the text data 20 outside the information processing device 10 or using the information processing device 10.
In step S10, the information processing device 10 reads the input text data 20. In next step S11, the information processing device 10 analyzes the text data read in step S10.
For example, during the analysis processing in step S11, the information processing device 10 extracts a stage direction sentence or a descriptive part, and a line sentence from a text included in the text data 20. For example, the information processing device 10 analyzes, for example, the extracted line sentence, and detects an expression that is included in the line sentence, made by a character who utters a line of the line sentence, and matches character likeness. For example, the information processing device 10 is included in the line sentence. An expression characteristic of this character or an expression not characteristic of the character is detected. In the embodiment, the information processing device 10 detects this expression based on, for example, a learning model learned in advance. The information processing device 10 is not limited to this, and may detect this expression according to a predetermined rule.
The information processing device 10 further generates a different expression from the detected expression. In a case where, for example, the expression is an expression characteristic of the character, the information processing device 10 generates this expression more characteristic of the character as the different expression. On the other hand, in a case where the expression is an expression not characteristic of the character, the information processing device 10 generates this expression characteristic of this character as a different expression.
In next step S12, the information processing device 10 displays on the display 11 the expression extracted from the line sentence and a generated different expression from the expression, and presents the expression and the different expression to the user 30.
In next step S13, the information processing device 10 accepts an input of corrections of contents by the user 30 presented in step S12. In a case where, for example, the user 30 has made an input indicating that the user 30 accepts the different expression presented in step S12, the information processing device 10 rewrites and corrects the expression of a corresponding part in a target line sentence to the different expression. Furthermore, in a case where the user 30 has made an input indicating that the user 30 does not accept the different expression presented in step S12, the information processing device 10 rejects the different expression without making any correction.
In next step S14, the information processing device 10 is caused to relearn a learning model used for detecting the expression in step S11 based on a correction result of the user 30 in step S13.
In next step S15, the information processing device 10 determines whether or not to finish correction of the text data 20 read in step S10 in response to the predetermined input of the user 30. In a case where it is determined to finish correction (step S15, “Yes”), the information processing device 10 finishes a series of processing according to this flowchart of
Note that, in a case where it is determined in step S15 not to finish the correction, the information processing device 10 may return the processing to step S11, and perform data analysis again on the corrected text data 20 based on the relearned learning model. Furthermore, the information processing device 10 may execute the relearning processing in step S14 after determining to finish the correction in step S15.
As described above, the information processing device 10 according to the embodiment detects an expression matching the character likeness from the line sentence of the text data 20, generates a different expression from the detected expression, and presents the different expression to the user 30. Furthermore, the information processing device 10 detects the expression based on the learning model, and is caused to relearn the learning data using a selection result of the user 30 in response to presentation of the different expression. Therefore, by applying the information processing device 10 according to the embodiment, it is possible to assist creation of a text in which characters appear at relatively low cost.
Next, an example of a schematic configuration of the information processing device 10 according to the embodiment will be described.
Among these units, the preprocessing unit 110, the detection unit 120, the comparison unit 130, the generation unit 140, and the UI unit 160 are configured by executing an information processing program according to the embodiment on a Central Processing Unit (CPU) included in the information processing device 10. The preprocessing unit 110, the detection unit 120, the comparison unit 130, the generation unit 140, and the UI unit 160 are not limited to this, and may be partially or entirely configured as hardware circuits that operate in cooperation with each other.
The User Interface (UI) unit 160 generates a user interface for the user 30, and controls the overall operation of this information processing device 10. The analysis data storage unit 150 stores information related to the input text data 20. For example, the analysis data storage unit 150 stores in advance information related to characters appearing in a script or a novel of the text data 20.
The preprocessing unit 110 performs processing of dividing the input text data 20 into stage directions or descriptive parts, and line sentences, and converts line sentences divided from the text data 20 into information suitable for processing of the detection unit 120 at a subsequent stage. The detection unit 120 detects an expression included in a line sentence and indicating the character likeness based on the information transferred from the preprocessing unit 110 and the information stored in the analysis data storage unit 150.
The comparison unit 130 refers to the information stored in the analysis data storage unit 150, and compares the character likeness of the specific expression detected by the detection unit 120 between the plurality of characters. The generation unit 140 generates a different expression from the expression detected by the detection unit 120 based on the comparison result of the comparison unit 130, comparison target expressions, and the information stored in the analysis data storage unit 150, and delivers to the UI unit 160 the different expression and the expression that matches the different expression and is the comparison target of the comparison unit 130. The generation unit 140 rewrites the text data 20 according to the different expression according to the instruction from the UI unit 160. The generation unit 140 outputs the rewritten text data 20 as the output data 21.
The storage device 1004 is a non-volatile storage medium such as a hard disk drive or a flash memory. The CPU 1000 controls the entire operation of this information processing device 10 by using the RAM 1002 as a working memory according to programs stored in the ROM 1001 and the storage device 1004.
Note that the above-described analysis data storage unit 150 is formed in, for example, a predetermined storage area in the storage device 1004. The analysis data storage unit 150 is not limited to this, and may be formed in a predetermined storage area of the RAM 1002.
The display control unit 1003 generates a display signal that can be displayed by a display 1020 corresponding to the display 11 in
The input device 1021 corresponds to the input device 12 in
The data I/F 1005 is connected to external equipment by wired or wirelessly or by a connector or the like to transmit and receive data. A Universal Serial Bus (USB), Bluetooth (registered trademark), or the like can be applied as the data I/F 1005. The data I/F 1005 is not limited to this, and may include or be connected to a drive device that can read a disk storage medium such as a Compact Disk (CD) or a Digital Versatile Disk (DVD).
The communication I/F 1006 communicates with a network such as the Internet or a Local Area Network (LAN) by wired or wireless communication.
In the information processing device 10, the CPU 1000 executes the information processing program according to the embodiment to configure the above-described preprocessing unit 110, detection unit 120, comparison unit 130, generation unit 140, and UI unit 160 as, for example, modules on a main storage area in the RAM 1002.
The information processing program can be acquired from an outside (e.g., server device) via a network such as the LAN or the Internet by, for example, communication via the communication I/F 1006, and can be installed on the information processing device 10. The information processing program is not limited to this, and may be provided by being stored in a detachable storage medium such as a Compact Disk (CD), a Digital Versatile Disk (DVD), or a Universal Serial Bus (USB) memory.
Next, the configuration according to the embodiment will be described in more detail.
In
The analysis data storage unit 150 includes a character information storage unit 151, a work setting information storage unit 152, and a plot information storage
The character information storage unit 151 stores character information that is information related to characters appearing in a target work described by the input text data 20. The character information includes, for example, information that indicates a person, an ending of a word, a terminology, a vocabulary range, and the like used in the speech by the character. The character information storage unit 151 can store these pieces of character information as a feature amount.
The item “name” among the respective items of the character information indicates the names of the characters, and the character A is “Tanaka Takashi” and the character B is “Sato Hiroshi”. Note that the names indicated in the item “name” do not need to be specific names, and may be any names that can be used in the target work and can identify the characters.
The item “first person” is a word used by a character to refer to oneself, and the character A uses “I (Boku)” and the character B uses “I (Ore)”. The item “second person” is a word used by a character to refer to an other party of a conversation, and the character A uses “You (Kimi)” and the character B uses “Hey man (Omae)”.
The item “character name” is a word used by a character to refer to another specific character, and the character A calls “Hiroshi” as “Hiroshi” and calls “Jun” as “Senior (Senpai)”. Furthermore, according to the item “character name”, the character B calls “Takashi” as “Takashi” and calls “Jun” as “Senior (Senpai)”. The item “ending of word” is a word frequently used by a character as an ending of a word of a conversation, and the character A uses “I think (Desu)” and the character B uses “I guess (Dana)” and “you know (Dayo)”.
Furthermore, the item “favorite food” among the items of the character information indicates favorite food of a character, and is “apple” in the case of the character A and “melon” in the case of the character B. The item “dislikable food” indicates dislikable food of a character, and is “natto” in the case of the character A and is “okra” in the case of the character B. As described above, the information indicating a character's preference can also be included in the character information as the information that characterizes this character.
The items of the character information stored in the character information storage unit 151 are not limited to the example illustrated in
The work setting information storage unit 152 stores setting information of a target work described by the text data 20.
For example, based on the work setting information stored in the work setting information storage unit 152, it is possible to grasp the role of each character in the target work indicated in the information stored in the above-described character information storage unit 151. Furthermore, although the list of the terms used in the work has been described above as the work setting information, information included in the work setting information is not limited to this example. For example, a background of a story described in this work may be included in the work setting information.
The plot information storage unit 153 stores plot information of the target work described by the text data 20.
In
The above-described character information, work setting information, and plot information are created in advance by an author of the work or the like, and are stored in the character information storage unit 151, the work setting information storage unit 152, and the plot information storage unit 153.
Next, the processing according to the embodiment will be described in more detail.
(4-1. Processing of Preprocessing Unit)
Processing of the preprocessing unit 110 according to the embodiment will be described. The text data 20 of the text that describes the target work is input to the input unit 111. The text described by the text data 20 is assumed to be a script or a novel. The input unit 111 transfers the input text data 20 to the sequence conversion unit 112.
In a case where the text data 20 is the script, the sequence conversion unit 112 divides the text data 20 into stage direction sentences and line sentences, and converts the text data 20 into sequences of stage direction sentences and sequences of line sentences. In the case of the script, speaker information is added to the line sentences, and therefore the sequence conversion unit 112 further divides the line sentences per speaker, and converts the line sentences per speaker.
Furthermore, in a case where the text data 20 is the novel, the sequence conversion unit 112 divides descriptive parts and line sentences into sequences of the descriptive parts and sequences of the line sentences. In this case, speaker information associated with each line sentence is not clearly indicated in novels or the like in many cases. Furthermore, sentences from a speaker's viewpoint are included in a descriptive part in many cases, and the sentence from the speaker's viewpoint included in this descriptive part can be regarded as a line sentence indicating a conversation (speech) of the speaker. In a case where the text data 20 is the novel, the sequence conversion unit 112 analyzes the text data 20 together with the descriptive parts and the line sentences by using clustering and a learned model, and divides the text data 20 into the line sentences.
The sequence conversion unit 112 transfers data converted from the text data 20 to the morphological analysis unit 113. The morphological analysis unit 113 performs morphological analysis on the line sentences in the data transferred from the sequence conversion unit 112, and decomposes the line sentences into morphological sequences. The morphological analysis unit 113 transfers each morphological sequence obtained by decomposing the line sentence to the feature amount extraction unit 114. The feature amount extraction unit 114 extracts a feature amount of an expression related to each morphological sequence, from each morphological sequence transferred from the morphological analysis unit 113. The feature amount is expressed by, for example, a multidimensional vector. The feature amount extraction unit 114 transfers the feature amount extracted from each morphological sequence per line sentence to the detection unit 120.
Note that, in the preprocessing unit 110, the feature amount extraction unit 114 can directly extract the feature amount from the data converted by the sequence conversion unit 112.
(4-2. Processing of Detection Unit)
Next, processing of the detection unit 120 according to the embodiment will be described. The feature amount transferred to the detection unit 120 and extracted from each morphological sequence per line sentence is transferred to the character expression detection unit 121. The character expression detection unit 121 detects the character likeness of the expression of the line sentence associated with each feature amount based on each transferred feature amount of the line sentence and the character information stored in the character information storage unit 151.
Here, the character expression detection unit 121 detects the character likeness of the expression using a learning model learned by machine learning. For example, the character expression detection unit 121 uses as labeled data the character information to be stored in the character information storage unit 151 by supervised learning, inputs the feature amount of the expression transferred from the preprocessing unit 110 to the learning model as test data, and obtains a probability of the character likeness of the test data.
As an example, the character information of the character A and the character information of the character B are each used as labeled data, and the feature amount of the expression transferred from the preprocessing unit 110 is input as test data to the learning model. Assuming that the probability P is (0≤P≤1), the learning model outputs, for example, a probability P(A) that an expression has character A likeness as P(A)=0.8, and a probability P(B) that an expression has character B likeness as P(B)=0.2.
The character expression detection unit 121 obtains the character likeness in the line sentence by using one or both of the following two methods indicated as methods (1) and (2).
Method (1): calculates the character likeness per word in a line sentence.
Method (2): calculates the character likeness per sentence in a line sentence.
The method (1) will be described. According to the method (1), the character expression detection unit 121 obtains the character likeness of the word of the specific character for each feature amount of each morpheme transferred from the feature amount extraction unit 114, and indicating the word of each morphological sequence based on the line sentence of the specific character. The character expression detection unit 121 performs threshold determination on each value (e.g., probability) of the obtained character likeness, and detects a word associated with the character likeness whose value is a threshold or more as a word having character likeness of the specific character.
As an example, it is assumed that, in a line sentence [Takashi “I (Boku) don't eat an apple anyway.”] of a character whose name is “Takashi”, values (e.g., probabilities) of “I (Boku)” and an ending of the word “anyway.” are a threshold or more. In this case, the character expression detection unit 121 assumes that these “I (Boku)” and “anyway.” that is the ending of the word are expressions characteristic of the character whose name is “Takashi”.
Note that the character expression detection unit 121 may obtain the character likeness in units finer than words, that is, for example, in units of letters. In this case, for example, it is conceivable that the feature amount extraction unit 114 obtains the feature amount based on connection before and after letters, and the character expression detection unit 121 obtains the character likeness based on the feature amount obtained in these units of letters.
The method (2) will be described. According to the method (2), the connection of the entire line sentence is determined based on each morpheme transferred from the feature amount extraction unit 114, and indicating a word in each morphological sequence based on the line sentence of the specific character. In this case, the character expression detection unit 121 obtains a value (e.g., probability) indicating the character likeness of the line sentence by inputting the entire line sentence as test data to, for example, a learning model obtained by learning the character information as labeled data. When, for example, the obtained value is a threshold or more, the character expression detection unit 121 determines the line sentence as an expression characteristic of the specific character.
More specifically, taking the above-described line sentence of “Takashi” [Takashi “I (Boku) don't eat an apple anyway”] as an example, the character expression detection unit 121 obtains a value indicating the character likeness of this entire line sentence. In a case where the value obtained for “Takashi” is the threshold (e.g., 0.8) or more, the character expression detection unit 121 determines this line sentence “I (Boku) don't eat an apple anyway.” is an expression characteristic of “Takashi”.
The character expression detection unit 121 can further designate whether or not to consider the context for the above-described methods (1) and (2) according to, for example, the user 30 operation. When considering the context, the character expression detection unit 121 obtains the character likeness of the line sentence based on the character of the other party of a line of a target line sentence, a position of the line sentence in an entire text or a chapter including the line sentence, a chronologically preceding line sentence of the line sentence, a stage direction or a descriptive part, and the like. In this case, the character expression detection unit 121 can obtain the character likeness by using a learning model learned based on, for example, a random line sentence, and a text in a predetermined range of a random script or a novel.
In a case where the character likeness is obtained in consideration of the context, the character expression detection unit 121 can present from which element of the context a value (probability) indicating character likeness is calculated. For example, it is conceivable to obtain the character likeness of the expression in the line sentence in consideration of a plurality of dominant elements among a Time, a Place, and an Occasion (TPO), Who, When, Where, What, Why, and How (5W1H), a time zone, and the like in the context. As a specific example, there may be a case where a value indicating the character likeness of a line varies depending on whether a certain line of the character whose name is “Takashi” is a line at night, a line at school, a line at home, or the like. In such a case, the character expression detection unit 121 can present that a ground for obtaining the character likeness is a part indicating “night”, a part indicating “school”, or a part indicating “home” in a stage direction or a descriptive part.
The character expression detection unit 121 may obtain the character likeness per word or per sentence by further using the work information stored in the work setting information storage unit 152 and the plot information stored in the plot information storage unit 153. The character expression detection unit 121 is not limited to this, and can also convert the character likeness into a numerical value by using various elements (emotions and the like) of a character in addition to the context. For example, even the same line takes different values indicating the character likeness between a case where a character is angry and a case where the character is not angry.
In the detection unit 120, the word level visualization unit 122 visualizes, at a word level, the expression that is detected by the character expression detection unit 121 and is determined to have character likeness in the line sentence.
The UI unit 160 generates the display 123 based on the information transferred from the detection unit 120. The generated display 123 is displayed on the display 1020. For example, based on the display 123 that highlights the expressions Wc1 to Wc3 in this line sentence, the user 30 can grasp based on what the detection unit 120 has detected the character A likeness.
The character expression detection unit 121 transfers, to the comparison unit 130, the expression detected from the line sentence as an expression having character likeness, and a value (e.g., probability) indicating the character likeness of the expression. These items of data transferred to the comparison unit 130 are transferred to the character expression comparison unit 131.
(4-3. Processing of Comparison Unit)
Next, processing of the comparison unit 130 according to the embodiment will be described. In the comparison unit 130, the character expression comparison unit 131 compares the character likeness of a specific line sentence between a plurality of characters appearing in a target script or novel using the transferred expression and the value indicating the character likeness.
In an example, in a case where the character “Takashi” and the character “Jun” are assumed, and a value indicating “Takashi” likeness (probability of Takashi likeness)=0.8 and a value indicating “Jun” likeness=0.2 are obtained for a line sentence [Takashi “I (Boku) don't eat an apple anyway.”] indicating the line of “Takashi”, it is possible to determine that this line sentence sufficiently has the “Takashi” likeness. In another example, in a case where a value indicating the “Takashi” likeness=0.6 and a value indicating “Jun” likeness=0.5 are obtained for a line sentence [Takashi “I (Boku) may not go there.”] indicating a line of “Takashi”, the line sentence is determined as a way of saying similar to the line of “Jun”, and it is difficult to determine that the line sentence sufficiently has the “Takashi” likeness.
In a case where a value indicating the character likeness of each of a plurality of characters is given to a certain line sentence, the character expression comparison unit 131 may determine that the line sentence is an expression that is characteristic of the character and has the largest value.
Furthermore, general expressions such as “Yes.” and expressions commonly used for a plurality of characters appearing in a work can be excluded from comparison targets of the character expression comparison unit 131. That is, it is preferable to distinguish between a general expression and an expression specific to (characteristic of) a character, and specify a correction range of this assist tool (information processing device 10) as an expression characteristic of the character.
In the comparison unit 130, the comparison visualization unit 132 visualizes a comparison result of the character expression comparison unit 131.
In
In
In the example in
The character expression comparison unit 131 transfers, to the generation unit 140, for example, the line sentence 136, each value indicating the character likeness of each of the characters “Takashi”, “Hiroshi”, and “Jun” in the line sentence 136, and information indicating a part serving as a ground of the value indicating the character likeness in the line sentence 136. These items of data transferred to the generation unit 140 are transferred to the character expression conversion/generation unit 141.
(4-4. Processing of Generation Unit)
Next, processing of the generation unit 140 according to the embodiment will be described. In the generation unit 140, the character expression conversion/generation unit 141 generates a different expression from an expression in a target line sentence based on data including the target line sentence transferred from the character expression comparison unit 131 and the character information stored in the character information storage unit 151. The character expression conversion/generation unit 141 presents a rewritten sentence obtained by rewriting an original sentence with the generated different expression.
In an example, a case will be considered where a value indicating character likeness of a character who utters a line in the target line sentence is, for example, 0.2 or 0.3 and is smaller than a predetermined value (e.g., 0.5), and is determined to have no character likeness. In this case, the character expression conversion/generation unit 141 generates and presents a different expression that is different from the expression in the line sentence and has character likeness. In another example, a case will be considered where the value indicating the character likeness of a character who utters this line in the line sentence 136 is, for example, 0.6 or 0.7 and is larger than the predetermined value, yet does not sufficiently indicate character likeness. In this case, the character expression conversion/generation unit 141 generates and presents a different expression that is different from the expression in the line sentence and has more character likeness.
In an example, in the case of the line “I (Watashi) don't eat an apple anyway.” of the character “Takashi”, if the character “Takashi” is a boy (description of a boyhood, etc.), the character expression conversion/generation unit 141 can propose rewriting to the line “I (Boku) don't eat an apple anyway.”. In this case, the character expression conversion/generation unit 141 can generate a different expression for rewriting the expression of the original line based on the character information of the character stored in the character information storage unit 151. The character expression conversion/generation unit 141 is not limited to this, and can generate a different expression corresponding to the expression based on a dictionary of general phrases.
Furthermore, the character expression conversion/generation unit 141 can propose a different expression from the original expression by setting various items. For example, the character expression conversion/generation unit 141 can generate and propose the different expression in consideration of a context.
In an example, a case will be considered where a context is presented which includes the line “I (Watashi) don't eat an apple anyway.” and in which the character “Takashi” speaks to a friend of another character. In this case, the character expression conversion/generation unit 141 generates and proposes, for the line, a line “I (Ore) don't eat an apple anyway.” of a different expression having familiarity to the friend.
In another example where a context is considered, in a case where a script or the like describes in a stage direction sentence that “She got angry”, the character expression conversion/generation unit 141 can generate and propose a different expression “got mad” for the expression “got angry” indicating an emotion in this description. Furthermore, the character expression conversion/generation unit 141 can generate and propose a different expression in the line sentence related to the description according to the description “She got angry” in the stage direction sentence.
In the generation unit 140, the output unit 142 visualizes the different expression generated by the character expression conversion/generation unit 141.
The example in
In this case, as illustrated in
Note that, in practice, the display 123 in
(4-5. Feedback According to User's Instruction)
Next, feedback according to a user 30 instruction will be described. For example, the user 30 can instruct whether or not to apply in the display 143 in
For example, the UI unit 160 updates the display on the display 1020 according to a user's operation to apply the different expression proposed by the generation unit 140 (a specific example will be described later). Furthermore, the UI unit 160 instructs the detection unit 120 to relearn the learning model using the applied different expression according to the user's operation. That is, the user 30 operation on the different expression visualized by the output unit 142 is fed back to the detection unit 120.
Here, the text data 200 includes a descriptive part (or a stage direction sentence) and a line sentence. Furthermore, in the text data 200, the line is uttered by the character “Takashi”, and the character “Takashi” uses “You (Kimi)” as the second person according to the character information illustrated in
The system analyzes the input text data 200, divides the text data 200 into the stage direction sentence and the line sentence, analyzes the line sentence, and extracts a feature amount in units of words, sentences, and the like. The system obtains a value indicating the character likeness of the line sentence based on the extracted feature amount, and generates a different expression from the expression in the line sentence according to the obtained value. The system gives presentation that encourages the user 30 to make correction to the different expression (step S101).
In the example in
In a case where the user 30 rejects the correction in response to the presentation that encourages correction to this different expression (step S102), the system uses the expression Wc10 in the original expression as is as the expression Wr10 of a correction result. That is, the text data 200 is not corrected. In this way, an instruction by the user 30 is fed back (FB) to the system. According to this feedback, the system obtains knowledge KN that “When Takashi gets angry, Takashi may use “You bastard (Kisama)” as the second person”. The system causes the detection unit 120 to relearn the learning model used by the detection unit 120 to detect the expression based on the acquired knowledge KN.
For example, the user 30 inputs, to the system, the text data 200 including the descriptive part and the line sentence similar to
The system analyzes the text data 200 input as described above, obtains a value indicating the character likeness of the line sentence included in the text data 200 based on an analysis result, and generates a different expression from the expression in the line sentence according to the obtained value. The system gives presentation that encourages the user 30 to make correction to the different expression (step S101).
In the example in
In
When the user 30 selects step S102a, the system outputs output data 203a having the same contents as that of the text data 200 without correcting the text data 200. In this way, an instruction by the user 30 is fed back to the system. According to this feedback, the system acquires knowledge KNa that “Takashi uses “Hey man (Omae)” as the second person for Hiroshi (who is a close friend)”. The system causes the detection unit 120 to relearn the learning model used by the detection unit 120 to detect the expression based on the acquired knowledge KNa.
On the other hand, when selecting step S102b, the user 30 corrects the descriptive part or the line sentence according to the system's proposal, and outputs output data 203b obtained by correcting the text data 200. For example, the user 30 corrects the expression Wc20 (“Hey man (Omae)”) for which the correction has been proposed to an expression Wr20 (“Teacher (Sensei)”). Furthermore, the character “Teacher (Sensei)” is set as the senior person for the character “Takashi”, and therefore the user 30 rewrites the other part of the line sentence into a polite language. Furthermore, the user 30 rewrites the expression of the second person in the descriptive part from “Hiroshi” to “Teacher (Sensei)”. In this way, an instruction by the user 30 is fed back to the system. According to this feedback, the system acquires knowledge KNb that “Takashi uses “Teacher (Sensei)” as the second person for the teacher (who is the senior person)”. The system causes the detection unit 120 to relearn the learning model used by the detection unit 120 to detect the expression based on the acquired knowledge KNb.
For example, similar to
The system analyzes the text data 200 input as described above, obtains a value indicating the character likeness of the line sentence included in the text data 200 based on an analysis result, and generates a different expression from the expression in the line sentence according to the obtained value. The system gives presentation that encourages the user 30 to make correction to the different expression (step S101).
In the example in
When the user 30 selects step S102a, the system outputs output data 203c having the same contents as that of the text data 200 without correcting the text data 200. In this way, an instruction by the user 30 is fed back to the system. According to this feedback, the system obtains knowledge KNc that “When Takashi gets angry, Takashi may use “You bastard (Kisama)” as the second person”. The system causes the detection unit 120 to relearn the learning model used by the detection unit 120 to detect the expression based on the acquired knowledge KNc.
On the other hand, when selecting step S102b, the user 30 corrects the descriptive part or the line sentence according to the system's proposal, and outputs output data 203d obtained by correcting the text data 200. For example, the user 30 corrects the expression Wc21 (“Hey man (Omae)”) for which the correction has been proposed to the expression Wr21 (“You (Kimi)”) according to the proposal. Furthermore, the user 30 corrects the expression “got angry” indicating an emotion of anger in the descriptive part to an expression 205 (“as usual”) indicating an emotion at a normal time (not angry). In this way, an instruction by the user 30 is fed back to the system. According to this feedback, the system acquires knowledge KNd indicating that “a certainty factor that Takashi uses “You (Kimi)” as the second person at the normal times is informed”. The system causes the detection unit 120 to relearn the learning model used by the detection unit 120 to detect the expression based on the acquired knowledge KNd.
Thus, the system according to the embodiment obtains, based on the learning model, a value whose line sentence is included in the input text data and indicates the character likeness of the character who utters the line in the line sentence. The system generates a different expression from the expression in the line sentence based on the value indicating the character likeness, and presents the different expression to the user. The system feeds back an instruction to a user's instruction for the different expression, and relearns the learning model. Consequently, it is possible to assist creation of a text in which characters appear at relatively low cost.
(4-6. Specific Example of Correction of Expression)
Next, a specific example of correction of an expression according to the embodiment will be described.
(Regarding Proposal for Correction Matching Context)
As described above, in the embodiment, the character expression detection unit 121 can present a part in a context that serves as a ground of correction proposed by the system.
In the example in
In section (a) of
Section (b) in
Furthermore, a ground order 241 is displayed at a lower part of the designation part 240 on the left side of the screen 210. When a plurality of items are designated at the designation part 240, the ground order 241 indicates the degree of contribution indicating contribution of each designated item per item as a ground for specifying a part for which correction is proposed. In the example in
Furthermore, for example, the UI unit 160 can change contents of the proposed correction according to this degree of contribution. In the example in
Consequently, the user 30 can know a ground of correction proposal, and easily decide whether or not to accept the correction proposal.
(Regarding Application of Correction)
An operation in a case where correction proposed by the system is applied according to the embodiment will be described with reference to
In the left side of
An example of an operation in a case where the user 30 accepts proposed correction to the expression Ws30 in response to this display of the screen 240 will be described. As illustrated at the center part of
In this state, the user 30 performs a predetermined operation for accepting the correction to the proposed expression Ws30. The predetermined operation is not particularly limited, and is clicking of a right button of the mouse, pushing of a predetermined key of the keyboard, or the like. The UI unit 160 displays an execution button 231 for executing correction according to the predetermined operation of the user 30. When accepting the proposed correction, the user 30 performs an operation of pushing this execution button 231 (e.g., moving the cursor 230 onto the execution button 231 and clicking a left button of the mouse).
When detecting this operation by user 30 on the execution button 231, the UI unit 160 rewrites the corresponding part of the text data 220, and displays a text of the rewritten text data 220. The right side of
As described above, in the embodiment, processing of reflecting correction proposed by the system in the text data can be executed in several steps.
Note that there is also a case where correction to an expression different from the expression Ws30 of proposed correction and the original expression Wc30 is performed. In this case, for example, it is conceivable to correct the corresponding part in the correction proposal item 223 as desired, and then perform the above-described operation.
(4-7. Proposal of Correction According to Transition of Emotion in Story)
Next, a proposal for correction matching transition of an emotion in a story according to the embodiment will be described. In the embodiment, the “emotion” that transitions according to a progress of the story can be reflected in an expression Ws for which correction is proposed.
For example, the detection unit 120 analyzes a stage direction sentence or a descriptive part, and a line sentence in the input text data 20 to detect an expression indicating an emotion. Expressions indicating emotions include expressions related to anger, expressions related to laughter, expressions related to impressions, and the like. The detection unit 120 detects for the text data 20 an expression indicating an emotion based on a word, a phrase, and moreover a context. The detection unit 120 sets a value (referred to as an emotion value) indicating the degree of activation of an emotion to the expression indicating the emotion detected from the text data 20. The detection unit 120 may detect the expression indicating the emotion and set the emotion value based on a specific keyword indicating the emotion, or using a learning model obtained by learning the expression indicating the emotion.
As an example, expressions indicating emotions are classified into about five levels from inactive expressions to activated expressions. It is conceivable to use three to five levels of expressions for a scene with a high emotion value, one to three levels of expressions for a scene with a low emotion value, and two to four levels of expressions for a scene with an intermediate emotion value as the expressions of proposed correction.
Consequently, by properly using the expression of proposed correction according to an emotion per scene, it is possible to propose more appropriate correction, and it is possible to more efficiently execute, on a correction result, relearning of a learning model according to feedback.
(4-8. Example of UI Screen)
Next, an example of a UI screen that is applicable to the embodiment will be described.
When, for example, the assist tool according to the embodiment is activated in the information processing device 10 and the text data 20 that is input to the information processing device 10 and in which, for example, a certain work (a script or a novel) is described is read by the assist tool, the UI unit 160 displays the UI screen 400 on, for example, the display 1020.
In
The area 402 displays a legend 430 for display in the area 404. Furthermore, the area 402 is further provided with a button 431 for editing information stored in the character information storage unit 151, the work setting information storage unit 152, and the plot information storage unit 153.
The area 404 is provided with tabs 440a, 440b, and 440c, and the UI unit 160 performs, on a display area 441, display corresponding to a designated tab among the tabs 440a, 440b, and 440c. In this example, in a case where the tab 440a is designated, the UI unit 160 causes the display area 441 to display the screen presented by the output unit 142 in the generation unit 140. In a case where the tab 440b is designated, the UI unit 160 causes the display area 441 to display the screen presented by the comparison visualization unit 132 in the comparison unit 130. Furthermore, the tab 440c causes the display area 441 to display the screen presented by the word level visualization unit 122 in the detection unit 120. In the example in
Next, a modification of the embodiment will be described. In the above-described embodiment, the assist tool according to the present disclosure is mounted and executed in the local information processing device 10. By contrast with this, according to the modification of the embodiment, the assist tool according to the present disclosure is mounted and executed on a server connected to a network.
The network 301 is, for example, the Internet. The network 301 is not limited to this, and may be a network closed in a predetermined environment such as a Local Area Network (LAN).
The server 302 employs a configuration similar to those of general computers, and includes functions of the preprocessing unit 110, the detection unit 120, the comparison unit 130, the generation unit 140, and the UI unit 160 in the information processing device 10 according to the embodiment illustrated in
Note that the server 320 is illustrated as a single computer in the example in
The terminal device 310 is, for example, a general information processing device such as a Personal Computer (PC), and includes a browser application 311 (displayed as the browser 311 in
According to such a configuration, the user 30 inputs the text data 20 to the terminal device 310. In the terminal device 310, the browser 311 transfers the input text data 20 to the server 320 via the network 301. The server 320 analyzes the text data 20 as described above, and generates a proposal for correction of an expression or the like. The UI unit 160 generates display control information for displaying the UI screen 400 described with reference to
Note that, according to this configuration, the text data 20 may be stored in the server 320 in advance. Furthermore, the output data 21 may be also stored in the server 320. In this case, the server 320 can store a plurality of items of the text data 20 and the output data 21 of respectively different works. In the server 320, for example, the detection unit 120 learns the learning model based on the plurality of items of text data 20, so that it is possible to propose more accurate correction.
On the other hand, when storing the items of the text data 20 and the output data 21 of the plurality of users 30, the server 320 needs to strictly manage the text data 20 and the output data 21 of each user 30 per each user 30 to avoid problems such as plagiarism.
Here, the user 30 generally indicates a user related to writing of a certain work, and is not limited to an individual. For example, the user 30 may be a plurality of users who write the same work together, or may be a plurality of users who write a plurality of works included in the same series. In these cases, by using the assist tool according to the present disclosure, it is easy to commonalize an expression per character in each work.
Next, another application example of the embodiment will be described. Although the above description has described that the assist tool according to the embodiment is applied to a script or a novel, the application range of the assist tool is not limited to the script or the novel.
The assist tool according to the embodiment may be applied to a game operated by a program. In this case, appearances, motions, and the like of characters can be captured as input and output to the assist tool. For, for example, a line “I (Ore) am enjoying so much.” of the character “Takashi”, for example, a line “I (Boku) am enjoying very much.” is proposed as a different expression. On the other hand, a meaning itself of the expression “enjoy” included in these lines does not change, so that it is possible to generate and select the appearance or the motion of the character matching the proposed different expression.
Furthermore, in recent years, there are agents that have virtual personality with character properties, official characters of specific brands, and the like. The assist tool according to the embodiment can be applied to posting to a Social Networking Service (SNS), generation of a message, and the like performed by these agents or official characters.
Note that the effects described in the description are merely examples and are not limited, and other effects may be provided.
Note that the present technique can also have the following configurations.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2021-054158 | Mar 2021 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2022/005255 | 2/10/2022 | WO |
