JP2020135523A - Information processor, information processing method and program - Google Patents

Abstract

To enable appropriate translation of a text included in non-structured data.SOLUTION: An information processor 1 includes: an acquisition part configured to acquire non-structured data including a text of a first language; a generation part configured to use a discriminator that discriminates respective elements composing the non-structured data to generate structured data in which the respective elements including the text are structured; and a conversion part configured to convert the structured text into a second language. Accordingly, a text included in non-structured data can appropriately be translated.SELECTED DRAWING: Figure 1

Description

The present invention relates to an information processing device, an information processing method and a program.

In recent years, machine translation technology for translating text in a document into another language has been developed. For example, in Patent Document 1, a method of translating a source language into a target language, in which a text extracted from a document in the source language is translated by a translation engine, and the translated text is converted into the original text format and structural information. A method of editing based on is disclosed.

Special Table 2007-532995

However, if the original document text to be translated is not structured, there is a problem that an inappropriate translated sentence is created.

In one aspect, it is an object of the present invention to provide an information processing device or the like capable of appropriately translating text contained in unstructured data.

The information processing apparatus according to one aspect uses an acquisition unit for acquiring unstructured data including the text of the first language and a discriminator for identifying each element constituting the unstructured data to obtain the text. It is characterized by including a generation unit that generates structured data in which each of the elements including the above is structured, and a conversion unit that converts the structured text into a second language.

On one side, the text contained in unstructured data can be properly translated.

It is a schematic diagram which shows the structural example of the document translation system. It is explanatory drawing which shows an example of the case where the unstructured data is directly translated. It is explanatory drawing which shows an example of the page layout when the translated text is arranged according to the original document. It is a block diagram which shows the configuration example of a server. It is explanatory drawing about a structured table. It is explanatory drawing about the generation processing of structured data. It is explanatory drawing about a document translation process. It is explanatory drawing about layout generation processing. It is explanatory drawing about layout learning processing. It is a flowchart which shows the procedure of layout learning processing. It is a flowchart which shows the procedure of a document translation process. It is a block diagram which shows the configuration example of the server which concerns on Embodiment 2. FIG. It is explanatory drawing about the structured learning process. It is a flowchart which shows an example of the procedure of the structured learning process. It is a functional block diagram which shows the operation of the server of the above-mentioned form.

Hereinafter, the present invention will be described in detail with reference to the drawings showing the embodiments thereof.
(Embodiment 1)
FIG. 1 is a schematic diagram showing a configuration example of a document translation system. In the present embodiment, a document in a predetermined language (first language), which is unstructured data, is structured, and the text in the structured document is converted into another language (second language) to be unstructured. A document translation system that preferably performs machine translation on data will be described. The document translation system includes an information processing device 1 and a terminal 2. Each device is communicated and connected via a network N such as the Internet.

In the following, for convenience of explanation, the unstructured data to be processed will be described as a "document", but this system may process any text contained in the unstructured data as long as it can be translated. The target data is not limited to documents. For example, an illustration image including text may be processed, and unstructured data including text as an element may be processed.

Further, in the following description, for convenience, the language of the original document is referred to as a "source language", and the translated language is referred to as a "target language".

The information processing device 1 is a device that performs various information processing and information transmission / reception, and is, for example, a server device, a personal computer, a multifunctional terminal, or the like. In the present embodiment, the information processing device 1 is assumed to be a server device, and will be read as a server 1 below for the sake of brevity. The server 1 is a document in a source language composed of each element such as text, a photograph, and a chart, and the metadata defining each element is not attached to each element like a PDF (registered trademark) file, for example. The process of structuring unstructured data is performed, and structured data (for example, an XML file) to which metadata is added to each element is generated. Then, after structuring each element including the text, the server 1 converts (translates) the structured text into the target language.

FIG. 2 is an explanatory diagram showing an example of a case where unstructured data is directly translated. When machine translation is performed on unstructured data, it is difficult to perform appropriate translation because no metadata is added to the text in the data. For example, as shown in FIG. 2, a text that should be a single paragraph is broken at an inappropriate place and is mechanically translated as it is, so that an inappropriate translated sentence may be created. ..

Therefore, the server 1 structures the document, which is unstructured data, divides (identifies) each element of a predetermined unit such as a title, paragraph, photograph, chart, caption, etc., which constitutes the document, and then translates the document. As a result, the situation illustrated in FIG. 2 is prevented, and machine translation for unstructured data is preferably executed.

In addition to the above processing, the server 1 generates layout data in which each element including the translated text is rearranged in consideration of the document layout in the target language, that is, the page layout of the translated version document. FIG. 3 is an explanatory diagram showing an example of a page layout when the translated text is arranged according to the original document. As shown in FIG. 3, when each element including the translated text is placed at the position as it is in the original document, adjacent elements (text and charts) overlap each other due to the difference in the number of characters before and after the translation. In some cases, the document page may be created with an inappropriate layout. Therefore, as will be described later, the server 1 determines the layout of each element by using the evaluation model (evaluator) that has learned the layout of the existing document in the target language by machine learning. As a result, the server 1 provides layout data in which the translated document is arranged into an appropriate page layout.

The terminal 2 is a client terminal that communicates with the server 1, and is, for example, an information processing terminal such as a personal computer. For example, the server 1 performs the above processing on the document file uploaded from the terminal 2, performs machine translation to generate layout data in which each element is rearranged, and outputs the layout data to the terminal 2.

FIG. 4 is a block diagram showing a configuration example of the server 1. The server 1 has a control unit 11, a main storage unit 12, a communication unit 13, and an auxiliary storage unit 14.
The control unit 11 has one or more CPUs (Central Processing Units), MPUs (Micro-Processing Units), GPUs (Graphics Processing Units) and other arithmetic processing units, and stores the program P stored in the auxiliary storage unit 14. By reading and executing, various information processing, control processing, etc. are performed. The main storage unit 12 is a temporary storage area for SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), flash memory, etc., and temporarily stores data necessary for the control unit 11 to execute arithmetic processing. Remember. The communication unit 13 is a communication module for performing processing related to communication, and transmits / receives information to / from the outside.

The auxiliary storage unit 14 is a non-volatile storage area such as a hard disk or a large-capacity memory, and stores a program P and other data necessary for the control unit 11 to execute processing. Further, the auxiliary storage unit 14 stores the structured table 141 and the evaluation model 142. The structured table 141 is a table (identifier) that identifies each element constituting the document in the source language and defines rules for assigning metadata. The evaluation model 142 is an evaluator for evaluating the document layout of the target language, and is a trained model generated by machine learning as described above.

The auxiliary storage unit 14 may be an external storage device connected to the server 1. Further, the server 1 may be a multi-computer composed of a plurality of computers, or may be a virtual machine virtually constructed by software.

Further, in the present embodiment, the server 1 is not limited to the above configuration, and may include, for example, an input unit that accepts operation input, a display unit that displays an image, and the like. Further, the server 1 includes a reading unit that reads a portable storage medium P1 such as a CD (Compact Disk) -ROM, a DVD (Digital Versatile Disc) -ROM, and reads and executes a program P from the portable storage medium P1. You can do it. Alternatively, the server 1 may read the program P from the semiconductor memory P2.

FIG. 5 is an explanatory diagram of the structured table 141. FIG. 5 conceptually illustrates the rules for converting a document, which is unstructured data, into structured data.
For example, the server 1 receives a rule setting input for structuring a document in a source language from an administrator terminal (not shown) and stores it in the structuring table 141. Specifically, the server 1 accepts a setting input for setting a rule for identifying each element constituting a document in a source language which is unstructured data and metadata to be given to each element.

The elements of unstructured data are texts, photographs, charts, etc. that compose the document. For example, as shown in FIG. 5 surrounded by a rectangular frame, the document is composed of the title, subtitle, text, photo, figure, etc. It is a basic element to do. The metadata is tag information that defines each element, such as a tag name tagged for each element or an attribute value stored as information in the tag. For example, as shown in FIG. 5, an element corresponding to the title of a document is given a "title", and an element corresponding to a subtitle is given a "subtitle" as metadata (tag name).

The server 1 accepts metadata representing each element constituting the document and setting input of information used as a reference when identifying each element. For example, the server 1 accepts input for setting reference values for text formats, coordinate values, and the like corresponding to each element. For example, the server 1 accepts setting inputs such as a text font and a character size used when describing each element for each element shown in FIG. 5 surrounded by a rectangular frame. Further, the server 1 accepts a setting input regarding the coordinate values (position and range) of the rectangular area corresponding to each element.

The server 1 stores the metadata of each set element in the structured table 141 in association with the information such as the format and the coordinate value corresponding to each element. As a result, the server 1 can identify which metadata should be added to each element according to the characteristics of each element such as the format and layout.

In the above, the format and the coordinate values are given as the criteria for identifying each element, but the present embodiment is not limited to this. For example, the server 1 may identify each element based on the distance between the elements (line spacing) or the like. In this way, the server 1 only needs to be able to identify and structure each element according to a predetermined rule, and the content of the rule is not particularly limited.

FIG. 6 is an explanatory diagram relating to the generation process of structured data. The server 1 refers to the structured table 141 set above, and performs a process of structuring a document in the source language which is unstructured data.
For example, the server 1 receives an upload of a document in the source language from the terminal 2 and generates structured data in which the document is structured. The server 1 refers to the structured table 141, identifies each element in the document acquired from the terminal 2, and adds metadata. For example, as shown in FIG. 6, the server 1 refers to the reference value related to the format, the coordinate value, etc. stored in the structured table 141, and the title, subtitle, text, photograph, chart, caption, etc. in the document. Identify each element of. Then, the server 1 assigns the corresponding metadata to each element, and generates structured data in which the elements are associated with each other according to the metadata.

FIG. 7 is an explanatory diagram relating to the document translation process. FIG. 7 illustrates how the text element is converted (translated) into the target language among the elements in the structured document.
Among the elements structured above, the server 1 individually converts the text title, subtitle, body, caption, etc. into the target language for each element. For example, the server 1 individually inputs each element such as a title, a subtitle, and a text to a predetermined translation engine, and acquires the text converted into the target language as an output.

FIG. 7 illustrates a processing image when converting a paragraph in a document identified as the "body" into a target language. According to the header and footer tags assigned to the paragraph as metadata, the server 1 inputs the text located between the tags into the translation engine as a continuous character string without line breaks. Then, the server 1 acquires the text converted into the target language from the translation engine. The server 1 replaces the text stored between the header and footer tags, that is, the paragraph of the source language before translation, with the text after translation and stores it in the structured data.

The server 1 sequentially converts each text constituting the original document into the target language in element units to which metadata is added, and stores the text in the structured data. As a result, the server 1 finally generates structured data of the translated document in which all the text in the document is converted into the target language.

Note that, for example, the server 1 may perform conversion by a different translation algorithm for each text according to the metadata assigned to each text element. For example, the server 1 selects a different translation engine according to the type of metadata assigned to each text, and performs conversion to the target language. As a result, when a document having a significantly different expression depending on the description location of the text, such as a patent specification, is targeted, appropriate translation can be performed in consideration of the difference in expression.

Further, for example, the server 1 may determine whether or not to convert (translate) the corresponding text into the target language according to the metadata. As a result, it is possible to prevent a situation in which untranslated text is converted, for example, a cited document described at the end of a paper. In this way, the server 1 may be capable of performing conversion to the target language according to the metadata.

FIG. 8 is an explanatory diagram relating to the layout generation process. FIG. 8 conceptually illustrates how each element in a structured document is generated by rearranging each element including the translated text to generate layout data of the document. The layout data generation process will be described with reference to FIG.
As described above, the server 1 generates structured data in which the original document is structured, and converts each text in the document into the target language. The server 1 rearranges each element including the converted (translated) text in a predetermined area corresponding to the document page, and generates layout data (image) corresponding to the translated document page.

Specifically, the server 1 randomly determines the arrangement coordinates, sizes, etc. of each structured element, and generates layout data of a plurality of patterns. FIG. 8 illustrates how M ways of layout data are generated. The server 1 generates M different layout information with different layouts by making the coordinates, size, etc. of each element different in each layout pattern. The server 1 generates layout data over one or a plurality of pages in M ways according to the number of pages of the original document and the like.

The server 1 calculates a score that evaluates the certainty of the layout for each of the randomly generated M layout data. Specifically, the server 1 calculates the score of each of the generated M-like layout data by using the evaluation model 142 in which the layout of the existing document in the target language has been learned.

FIG. 9 is an explanatory diagram relating to the layout learning process. FIG. 9 conceptually illustrates how the evaluation model 142 is generated by machine learning. In the present embodiment, the server 1 generates a neural network related to a CNN (Convolution Neural Network) as an evaluation model 142.

In this embodiment, the evaluation model 142 is described as being a CNN, but the evaluation model 142 is another trained model such as another neural network, SVM (Support Vector Machine), Bayesian network, decision tree, etc. You may.

In the present embodiment, the server 1 generates the evaluation model 142 by using the rank learning method. Rank learning is a learning method for learning the order of a data set. As a neural network that performs rank learning by deep learning, for example, DeepLank, SiameseNet, and the like are known. Since rank learning is a known learning method, detailed description thereof will be omitted.

For example, the server 1 collects an existing document in a target language whose attributes are similar to the document in the source language to be translated from the Internet, and uses the collected existing document as teacher data. For example, the server 1 determines the similarity of attributes according to a document field such as a paper, a manual, or a public document. The server 1 may grasp the field of the existing document by means such as character recognition, or may grasp the field of the document collected from the website (for example, the article publication site) from which the existing document is acquired. ..

For example, the server 1 collects, as the above-mentioned teacher data, an existing document which is unstructured data (for example, a PDF file) to which no metadata is added, like a document in the source language to be translated. Next, the server 1 converts the existing document into structured data such as an XML file. For example, the server 1 accepts the setting input for the rules for structuring the target language as well as the case where the document in the source language is structured, and prepares the structured table 141. The server 1 structures by referring to the structured table 141 corresponding to the target language.

The server 1 refers to the structured table 141 for the target language, identifies each element included in the collected existing document from the format, the coordinate value, and the like, and adds metadata to each element. As a result, the server 1 structures texts, photographs, charts, etc. in the existing document in predetermined element units.

The server 1 uses the page image of the collected existing document as the correct answer data of the layout data. In the following description, for convenience, the layout data is referred to as "existing layout data". For example, the server 1 uses the existing layout data as the correct answer data having a score of "1".

Further, the server 1 randomly replaces the arrangement coordinates of each element structured above in each element arranged on the same page, and generates a plurality of layout data in which the arrangement of each element is changed. In the following description, for convenience, the layout data is referred to as "fake layout data". The server 1 uses the fake layout data as incorrect answer data having a score of "0".

The server 1 inputs the existing layout data and the false layout data in which the arrangement of each element is changed from the existing layout data into the neural network, and performs rank learning. Specifically, the server 1 compares the score of the existing layout data with the score of each of the plurality of fake layout data generated above, and the score of the existing layout data is higher than the score of any of the plurality of fake layout data. Learn to be high. As a result, the server 1 generates the evaluation model 142. The server 1 calculates the score of the layout data using the generated evaluation model 142.

In the above, the server 1 is supposed to generate a part of the teacher data (fake layout data) by itself, but the teacher data may be all created manually. In addition, the teacher data may be manually labeled with correct values such as scores and rankings.

Further, the processing entity (server 1) that performs the layout learning process and the generation process does not have to be the same.

The explanation will be continued by returning to FIG. The server 1 inputs each of the M-like layout data generated from the document in the source language into the evaluation model 142, and calculates the score of each layout data. More specifically, the server 1 inputs layout data into the evaluation model 142 for each page for M different layouts, and calculates a total score by adding the scores of all the pages.

The server 1 determines the order of layout data according to M based on the score calculated above. The server 1 outputs the generated layout data to the terminal 2 according to the determined order. For example, the server 1 outputs the top-level layout data to the terminal 2. As a result, the server 1 provides a document file in which the text is converted from the source language to the target language and the layout is optimized.

For example, the server 1 may output not only the top-level layout data but also a predetermined number of high-level layout data. Further, for example, the server 1 may output all the generated layout data and also present (output) the score and the ranking. As described above, the server 1 only needs to be able to output the layout data according to the calculated score (evaluation), and the output mode is not particularly limited.

Further, in the above, a document file is generated as layout data and output to the terminal 2, but the present embodiment is not limited to this, and for example, the coordinates of each text, photograph, chart, etc. related to the top-level layout. , Only data such as size may be output, and the document file may be generated on the terminal 2 (client) side. That is, the server 1 only needs to be able to output information (layout data) regarding the translated document layout, and it is not essential to generate the document file.

From the above, according to the present embodiment, since the document in the source language which is the unstructured data is structured and then converted into the target language, the text contained in the unstructured data can be appropriately translated. .. In addition, the evaluation model 142 can be used to provide layout data in which each element including the translated text is rearranged in an appropriate layout.

FIG. 10 is a flowchart showing the procedure of the layout learning process. Based on FIG. 10, the content of the process of performing machine learning to generate the evaluation model 142 will be described.
The control unit 11 of the server 1 collects the existing document of the target language having the same field as the document of the source language to be translated via the network N (step S11). For example, the control unit 11 collects a page image (existing layout data) of a document which is unstructured data. The control unit 11 refers to the structured table 141 corresponding to the target language, identifies each element such as a text, a photograph, a chart, etc. included in the existing document, and provides structured data to which metadata for defining each element is added. Generate (step S12).

The control unit 11 generates fake layout data in which each element included in the existing document is rearranged (step S13). Specifically, the control unit 11 randomly rearranges each element included in the existing document acquired in step S11 to generate a plurality of fake layout data.

The control unit 11 generates an evaluation model 142 that outputs a score (evaluation) of the layout data when the layout data is input, based on the existing layout data of the document collected in step S11 and the fake layout data generated in step S13. (Step S14). Specifically, the control unit 11 performs rank learning using the existing layout data as correct answer data and the false layout data as incorrect answer data, and generates an evaluation model 142 related to CNN. The control unit 11 ends a series of processes.

FIG. 11 is a flowchart showing the procedure of the document translation process. Based on FIG. 11, the content of the process of translating the text in the document of the source language which is the unstructured data into the target language and rearranging each element including the translated text to generate the layout data will be described.
The control unit 11 of the server 1 acquires a document in the source language, which is unstructured data, from the terminal 2 (step S31). The control unit 11 refers to the structured table 141 corresponding to the source language, identifies each element such as text, a photograph, a chart, etc. that constitutes the acquired document, and adds metadata to each element. Data is generated (step S32).

The control unit 11 converts the text in the structured document into the target language (step S33). In this case, the control unit 11 may refer to the metadata given to each text (element) in the document in step S32, and convert each text into the target language by the translation engine corresponding to the metadata. Good.

The control unit 11 generates layout data in which each element in the document including the converted text is rearranged in a predetermined area corresponding to the original document page (step S34). Specifically, the control unit 11 randomly determines the coordinates for arranging each element and generates a plurality of layout data.

The control unit 11 calculates the score of the layout data generated in step S24 by using the evaluation model 142 in which the layout of the existing document has been learned (step S35). Specifically, the control unit 11 inputs each of the plurality of layout data generated in step S34 into the evaluation model 142, and acquires the score of each layout data as an output. The control unit 11 ranks each layout data according to the calculated score, and outputs the highest-level layout data to the terminal 2 (step S36). The control unit 11 ends a series of processes.

In the above, the evaluation model 142 outputs the score (evaluation value) of the layout data, but a binary parameter indicating the suitability of the layout may be output. That is, the server 1 may be able to acquire the evaluation of the layout data using the evaluation model 142, and the evaluation to be acquired is not limited to the continuous probability value.

From the above, according to the first embodiment, the text included in the unstructured data can be appropriately translated.

Further, according to the first embodiment, the accuracy and convenience of machine translation can be improved by performing translation according to the metadata given at the time of structuring.

Further, according to the first embodiment, by using the evaluation model 142 in which the layout of the existing document has been learned, it is possible to provide the translated document arranged in an appropriate layout.

Further, according to the first embodiment, the layout can be preferably evaluated by ranking a plurality of layouts by using the rank learning method.

Further, according to the first embodiment, by using the evaluation model 142 that has learned the layout of the existing document (sample) in the target language whose attributes are similar to the document (unstructured data) in the source language to be translated. , The layout can be evaluated suitably.

(Embodiment 2)
In the present embodiment, instead of the structured table 141 in which the rules are manually set, the configuration in which the structured model 143 constructed by machine learning is used for structuring will be described. The contents overlapping with the first embodiment are designated by the same reference numerals and the description thereof will be omitted.
FIG. 12 is a block diagram showing a configuration example of the server 1 according to the second embodiment. The auxiliary storage unit 14 of the server 1 according to the present embodiment stores the structured model 143 instead of the structured table 141. The structured model 143 is a trained model constructed by machine learning and is a discriminator for identifying each element in a document in the source language.

FIG. 13 is an explanatory diagram relating to the structured learning process. FIG. 13 conceptually illustrates how the structured model 143 is generated from the teacher data of the document of the source language which is the unstructured data. An outline of the present embodiment will be described with reference to FIG.
As described above, in the present embodiment, the server 1 structures the document by using the structured model 143 constructed by performing machine learning instead of the structured table 141 in which the rules are set manually. For example, the server 1 generates a neural network related to CNN, specifically, a neural network related to semantic segmentation as a structured model 143, and uses it for document structuring processing.

For example, the server 1 learns about each element such as the title and the text constituting the document for the teacher in the source language by using the teacher data associated with the correct answer value of the metadata. For example, as shown in FIG. 13, in the teacher data, a tag name corresponding to the correct answer value of the metadata is associated with the area corresponding to each element (shown by a rectangular frame). The server 1 generates the structured model 143 using the teacher data.

The server 1 inputs the page image of the teacher's document into the structured model 143, and acquires the identification result of identifying each element included in the page image as an output. For example, the server 1 acquires the coordinate values of the image area corresponding to each element and the metadata to be given to the elements included in the area as output. The server 1 compares the coordinate values and metadata of the output image area with the correct answer values, and optimizes parameters such as weights used in the calculation in the structured model 143 so that they approximate each other. As a result, the server 1 generates the structured model 143.

When a document in the source language is acquired from the terminal 2 and translated, the server 1 performs structuring using the structuring model 143 generated above. Specifically, the server 1 inputs the page image of the acquired document in the source language into the structured model 143, and acquires the identification result that identifies each element. The server 1 extracts each element in the document according to the identification result and adds metadata. As a result, the server 1 generates structured data in which the document in the source language is structured. After that, the server 1 performs text conversion to the target language and generates layout data as in the first embodiment.

FIG. 14 is a flowchart showing an example of the procedure of the structured learning process. Based on FIG. 14, the contents of the process of generating the structured model 143 by machine learning will be described.
The control unit 11 of the server 1 associates the correct answer value of the metadata of each element constituting the document with the teacher document which is the teacher data for generating the structured model 143 and is the unstructured data. The obtained teacher data is acquired (step S201). The control unit 11 generates a structured model 143 using the acquired teacher data (step S202). Specifically, the control unit 11 inputs the page image of the teacher's document into the structured model 143, and identifies the image area corresponding to each element and the metadata to be given to the element included in the area. Get the result as output. The control unit 11 compares the acquired identification result with the correct answer value, optimizes various parameters such as weights so that the two are approximate, and generates the structured model 143. The control unit 11 ends a series of processes.

From the above, according to the second embodiment, the document in the source language can be structured by using the structured model 143 constructed by machine learning.

(Embodiment 3)
FIG. 15 is a functional block diagram showing the operation of the server 1 in the above-described form. When the control unit 11 executes the program P, the server 1 operates as follows.
The acquisition unit 151 acquires unstructured data including the text of the first language. The generation unit 152 generates structured data in which each element including the text is structured by using a discriminator that identifies each element constituting the unstructured data. The conversion unit 153 converts the structured text into a second language.

The third embodiment is as described above, and the other parts are the same as those of the first and second embodiments. Therefore, the corresponding parts are designated by the same reference numerals and detailed description thereof will be omitted.

The embodiments disclosed this time should be considered to be exemplary in all respects and not restrictive. The scope of the present invention is indicated by the scope of claims, not the above-mentioned meaning, and is intended to include all modifications within the meaning and scope equivalent to the scope of claims.

1 Server (information processing device)
11 Control unit 12 Main storage unit 13 Communication unit 14 Auxiliary storage unit P program 141 Structured table 142 Evaluation model 143 Structured model 2 Terminal

Claims (7)

An acquisition unit that acquires unstructured data including text in the first language,
A generator that generates structured data in which each element including the text is structured by using a classifier that identifies each element that constitutes the unstructured data, and a generator.
An information processing device including a conversion unit that converts the structured text into a second language. The generation unit generates the structured data to which the metadata defining the element is added to each element.
The information processing device according to claim 1, wherein the conversion unit refers to the metadata attached to the text and converts the data into the second language. A layout generator that generates layout data in which each of the structured elements is arranged in a predetermined area,
An evaluation unit that acquires an evaluation of the layout data generated by the layout generation unit using an evaluation device that has learned a plurality of samples of the layout data, and an evaluation unit.
The information processing apparatus according to claim 1 or 2, further comprising an output unit that outputs the layout data according to the evaluation result. The layout generation unit generates a plurality of the layout data in which the arrangement of each element is different.
The evaluation unit acquires the ranking of the plurality of layout data and obtains the ranking.
The information processing device according to claim 3, wherein the output unit outputs the layout data according to the order. The information processing apparatus according to claim 3, wherein the evaluation unit acquires an evaluation of the layout data by using an evaluation device that has learned the sample having similar attributes to the unstructured data. Get unstructured data, including text in the first language,
Using a classifier that identifies each element that constitutes the unstructured data, structured data in which each element including the text is structured is generated.
An information processing method characterized in that a computer executes a process of converting the structured text into a second language. Get unstructured data, including text in the first language,
Using a classifier that identifies each element that constitutes the unstructured data, structured data in which each element including the text is structured is generated.
A program characterized in that a computer executes a process of converting the structured text into a second language.

Images