Natural language processing device and computer-readable recording medium

Abstract

A natural language processing device according to the present disclosure comprises: an input unit that receives an input related to a processing parameter and text data for learning; an extension unit that executes extension processing of the text data for learning on the basis of the processing parameter by using a learning model that takes text data as input and outputs text data; and an output unit that outputs the result of the extension processing. The extension unit includes: a preprocessing unit that executes preprocessing for the text data for learning on the basis of the processing parameter; a generation unit that executes generation processing on the text data for learning that has been pre-processed using the learning model; and a post-processing unit that executes post-processing on the text data obtained as the result of processing by the generation processing.

Description

Natural language processing device and computer-readable recording medium

[0001] The present disclosure relates to a natural language processing device and a computer-readable recording medium.

[0002] In recent years, attempts at DX using natural language processing techniques such as generative AI (Artificial Intelligence) and large language models (LLM) have been made in various industrial fields such as manufacturing (see, for example, Patent Literature 1). There is also a movement to apply LLMs and the like to business operations at manufacturing sites such as factories.

[0003] Natural language processing technologies using these techniques have higher inference accuracy than conventional machine learning and can flexibly analyze irregular and ambiguous input sentences, including conversational ones. Furthermore, the inference process is complex, making it difficult to explain results based on internal specifications. Generative AI processing, in particular, is highly dependent on random numbers and requires a lot of computation time and memory, often requiring the use of external resources such as cloud environments. Many of the internal specifications of the models, applications, and infrastructure used in external resources are not publicly disclosed, making the processing results difficult to explain. Furthermore, compared to conventional machine learning, a larger amount of data is required for training, and the effort required to adjust models for individual use cases tends to be significant.

[0004] Japanese Patent Application Laid-Open No. 2020-135457

[0005] When applying large-scale language models to industrial operations, there is a lot of industry-specific or site-specific terminology and noise (abbreviations, spelling habits, etc.), and even if publicly known general methods are used, accuracy may be insufficient. Furthermore, there are many situations where high-precision recognition is required for specific frequently used input types for each individual business use case. In the drive control of industrial machinery, operational errors and incorrect settings can result in significant losses. Ensuring explainability along with accuracy is essential. Currently, to address these issues, settings must be manually adjusted for each site, and these adjustments must be done through guesswork and trial and error, leaving room for improvement in terms of performance.

[0006] The natural language processing device disclosed herein solves the above problem by expanding the training data through preprocessing based on user input for a group of setting candidates based on the business requirements to be satisfied when constructing a large-scale language model tailored for that business based on individual business data, thereby improving generalizability when unknown data not included in the training data is input in the inference process of the trained large-scale language model.

[0007] One aspect of the present disclosure is a natural language processing device comprising: an input unit that accepts inputs related to processing parameters and training text data; an extension unit that performs extension processing of the training text data using a learning model that inputs text data and outputs text data based on the processing parameters; and an output unit that outputs the results of the extension processing, wherein the extension unit comprises: a pre-processing unit that performs pre-processing on the training text data based on the processing parameters; a generation unit that performs generation processing on the pre-processed training text data using the learning model based on the processing parameters; and a post-processing unit that performs post-processing on text data obtained as a processing result of the generation processing based on the processing parameters.

[0008] FIG. 1 is a schematic hardware configuration diagram of a natural language processing apparatus according to an embodiment of the present disclosure. FIG. 2 is a block diagram showing schematic functions of a natural language processing apparatus according to a first embodiment. FIG. 3 is a block diagram showing schematic functions of an extension unit. FIG. 4 is a table diagram showing an example of processing parameters of a pre-processing unit. FIG. 5 is a table diagram showing an example of processing parameters of a generation unit. FIG. 6 is a table diagram showing an example of processing parameters of a post-processing unit. FIG. 7 is a block diagram showing schematic functions of a natural language processing apparatus according to a second embodiment.

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following description, components having the same or similar functions will be denoted by the same reference numerals. Duplicate descriptions of those components may be omitted.

[0010] In this application, "based on XX" means "based on at least XX," and includes cases where it is based on other elements in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on XX that has been calculated or processed. "XX" is any element (for example, any information).

[0011] 1 is a schematic hardware configuration diagram showing the main parts of a natural language processing device according to a first embodiment of the present disclosure. The natural language processing device 1 according to this embodiment can be implemented on a computer such as a personal computer attached to a control device, a personal computer connected to the control device via a wired / wireless network, a cell computer, a fog computer 6, or a cloud server 7. In this embodiment, an example is shown in which the natural language processing device 1 is implemented on a personal computer attached to a control device 3 that controls industrial machinery 4.

[0012] The CPU 11 included in the natural language processing apparatus 1 according to this embodiment is a processor that controls the entire natural language processing apparatus 1. The CPU 11 reads a system program stored in the ROM 12 via the bus 22, and controls the entire natural language processing apparatus 1 in accordance with the system program. The RAM 13 temporarily stores temporary calculation data, display data, various data acquired from the outside, and the like.

[0013] The non-volatile memory 14 is composed of, for example, a battery-backed memory or an SSD (Solid State Drive) (not shown), and retains its stored state even when the power to the natural language processing apparatus 1 is turned off. The non-volatile memory 14 stores programs and data read from an external device 72 via the interface 15, programs and data input via the input device 71, and programs and data acquired from the control device 3 that controls the industrial machine 4 or other devices via the network 5. The programs and data stored in the non-volatile memory 14 may be expanded into the RAM 13 when executed / used. In addition, various system programs such as known analysis programs are written in the ROM 12 in advance.

[0014] The interface 15 is an interface for connecting the CPU 11 of the natural language processing device 1 to an external device 72 such as a USB device. For example, system programs, data, etc. are read from the external device 72. In addition, programs, data, etc. created or edited within the natural language processing device 1 can be stored in external storage means via the external device 72.

[0015] The interface 20 is an interface for connecting the CPU 11 of the natural language processing device 1 to a wired or wireless network 5. The network 5 may be one that communicates using technologies such as serial communication such as RS-485, Ethernet (registered trademark), optical communication, wireless LAN, Wi-Fi (registered trademark), Bluetooth (registered trademark), etc. At least one industrial machine 4 to be controlled, at least one other industrial machine 4, a fog computer 6, a cloud server 7, etc. are connected to the network 5, and data is exchanged between the network 5 and the natural language processing device 1.

[0016] The display device 70 displays various data read into the memory, data obtained as a result of executing programs, etc., output via the interface 17. An input device 71, which is comprised of a keyboard, pointing device, etc., passes instructions, data, etc. based on operations by an operator to the CPU 11 via the interface 18.

[0017] The interactive response device 2 is configured as a device that returns a response sentence in response to a predetermined sentence. The interactive response device 2 includes a large-scale language model that has learned a response sentence for the predetermined sentence. This model can use known models such as BERT (Bidirectional Encoder Representations from Transformers) or a Transformer model. The interactive response device 2 can use the large-scale language model to classify and organize existing text knowledge to support the work of users such as operators at manufacturing sites, a chatbot for referring to manuals for industrial machines 4, NC program creation support for industrial machines 4 using ChatGPT, parameter setting support, operation support, output of some kind of inference result image, and the like. The large-scale language model learns, for each text, the probability that a predetermined sentence will be a response to the given sentence when the given sentence is input. It also further learns the probability that each other text will follow the given text. By repeating this type of learning, the large-scale language model, when a predetermined sentence is input, will output a text string with a high probability as a response to the sentence. Depending on how the document that serves as a response to the predetermined sentence is learned in the learning process, the interactive response device 2 can be used for purposes such as dialogue, question and answer session, sentence summarization, sentence editing, sentence translation, sentence conversion, sentence modification, sentence optimization, sentence interpretation, sentence detection, recognition, prediction, judgment, code generation, image generation, and comprehensive judgment.

[0018] The natural language processing device 1 according to the present disclosure may be connected to an interactive response device 2 via a network 5. The interactive response device 2 includes a model trained based on various information publicly available on the Internet, for example. The information publicly available on the Internet includes information such as manuals and specifications for each control device and industrial machine, operating records, questions and answers, etc. Such information is published on the Internet by manufacturers of each control device and industrial machine. The interactive response device 2 can perform additional training (fine tuning) as needed using various business data from manufacturing sites as training text data. By narrowing down the information used for training, it is possible to prepare an interactive response device 2 that returns responses with a certain degree of improved accuracy tailored to each manufacturing site.

[0019] 2 is a schematic block diagram showing functions of the natural language processing apparatus 1 according to the first embodiment of the present disclosure. Each function of the natural language processing apparatus 1 according to this embodiment is realized by the CPU 11 included in the natural language processing apparatus 1 shown in FIG. 1 executing a system program and controlling the operation of each unit of the natural language processing apparatus 1.

[0020] The natural language processing apparatus 1 of this embodiment includes an input unit 100, an extension unit 120, and an output unit 140. The RAM 13 to the nonvolatile memory 14 of the natural language processing apparatus 1 also include a model storage unit 200, which is an area that stores in advance a trained model related to natural language processing that generates text data from text data.

[0021] The input unit 100 receives input of processing parameters related to natural language processing and training text data from a user. The processing parameters received by the input unit 100 are parameters related to processing performed on the training text data by the extension unit 120. The processing parameters may be, for example, parameters related to preprocessing performed on the training text data. The processing parameters may also be, for example, parameters related to extension processing performed on the training text data that has been preprocessed. The processing parameters may also be, for example, parameters related to postprocessing performed on the extended text data. The training text data may be, for example, text data related to inquiries or commands used in business. Such text data can be extracted from business data.

[0022] The input unit 100 may display an input screen on the display device 70 that accepts input of processing parameters and learning text. In this case, the user can input processing parameters and learning text data by operating the input device 71 while viewing the input screen. The input unit 100 accepts the processing parameters and learning text data from the input. The input unit 100 may accept the processing parameters and learning text data via a voice input device (not shown). Furthermore, the input unit 100 may accept the processing parameters and learning text data from the control device 3, the fog computer 6, the cloud server 7, or the like via the network 5 in accordance with instructions from the user. The input unit 100 outputs the accepted processing parameters and learning text data to the expansion unit 120.

[0023] The expansion unit 120 executes an expansion process on the training text data using a model stored in the model storage unit 200, based on the processing parameters and training text data received by the input unit 100. The expansion process refers to a process of generating at least one piece of derived text data based on the training text data. Through the expansion process, the training text data is expanded into an expanded text data group including at least one piece of derived text data. FIG. 3 is a schematic block diagram showing the functions of the expansion unit 120. The expansion unit 120 includes a preprocessing unit 122, a generation unit 124, and a postprocessing unit 126.

[0024] The preprocessing unit 122 performs preprocessing on the training text data received by the input unit 100 based on the processing parameters received by the input unit 100. The preprocessing performed by the preprocessing unit 122 is a general process of converting text data into a format that allows predetermined natural language processing to be performed on the text data. Examples of preprocessing include segmentation into phrases, removal of unnecessary portions, and character expansion. FIG. 4 is a table diagram showing an example of processing parameters for preprocessing. In the process of segmenting the training text data into predetermined units (sentences, phrases, etc.), processing parameters include, for example, delimiters (punctuation marks, spaces, etc.). In the process of removing unnecessary portions of the training text data, processing parameters include, for example, rules for specifying unnecessary portions (regular expressions, etc.). In addition to these preprocessing steps, the preprocessing unit 122 may also perform other preprocessing steps known in natural language processing. The preprocessing unit 122 outputs the preprocessed training text data to the generation unit 124.

[0025] The generation unit 124 executes a generation process on the training text data preprocessed by the preprocessing unit 122 based on the processing parameters received by the input unit 100. The generation process executed by the generation unit 124 is a general process for generating other text data with the same meaning or similar meaning from text data. For example, this process includes normalizing, denormalizing, correcting, compressing, adding redundancy, simplifying, complicating, smoothing, sharpening, translating into a regional language or another language (e.g., Japanese to English, English to Japanese), summarizing, adding, deleting, or paraphrasing prefixes, suffixes, conjunctions, particles, auxiliary verbs, etc. The generation process is a combination of various natural language processes. The generation unit 124 may further execute a generation process on each piece of derived text data generated from the training text data in one generation process cycle. The generation process includes various processing parameters depending on the process content. FIG. 5 is a table diagram showing an example of processing parameters for the generation process. Examples of processing parameters for the generation process include how many derived text data items to generate in one cycle of the generation process, how many cycles to repeat the generation process, what method to use for tokenization when interpreting the meaning of the training text data, what model task method (translation, summarization, etc.) to use to generate derived text data from the training text data, how much variation in meaning is acceptable, and what scale of model to use.

[0026] The generation unit 124 may generate derived text data from the training text data using a known natural language processing technique. Examples of known natural language processing techniques include a combination of classical natural language processing techniques such as Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF), and a technique using machine learning models related to generative AI, such as a Transformer model and a BERT model. The model storage unit 200 pre-stores trained models used to perform such natural language processing. The generation unit 124 selects an appropriate trained model based on the received processing parameters and configures the selected trained model based on the processing parameters. The set trained model is then used to generate one or more derived text data from the training text data. The generation unit 124 outputs a group of augmented text data including the training text data and one or more derived text data to the post-processing unit 126.

[0027] The post-processing unit 126 performs post-processing on the expanded text data group including the training text data and at least one derived text data generated by the generation unit 124, based on the processing parameters received by the input unit 100. The post-processing performed by the post-processing unit 126 is a general process of extracting data to be used for training from one or more text data. Examples of post-processing include rejecting text data that does not satisfy a condition and aggregating the text data. FIG. 6 is a table illustrating an example of processing parameters for post-processing. In the process of rejecting derived text data, processing parameters include, for example, a limit value for the character string distance (e.g., Hamming distance or Levenshtein distance) of the derived text data to be rejected from the training text data, a limit value for the character string length of the derived text data, and a limit value for the number of scores at the time of generation. In addition, in the process of aggregating the expanded text data group, processing parameters include, for example, a limit value for the number of aggregated frequencies (excluding low-frequency items). In addition to the above post-processing, the post-processing unit 126 may also perform well-known post-processing in natural language processing and statistical processing. The post-processing unit 126 outputs the post-processed expanded text data group to the output unit 140 .

[0028] The output unit 140 outputs the expanded text data group expanded by the expansion unit 120 based on the processing parameters and the learning text data. The output unit 140 may display the expanded text data group on the display device 70. Alternatively, the output unit 140 may record the expanded text data group in the non-volatile memory 14 or the external device 72. Furthermore, the expanded text data group may be transmitted to another computer such as the control device 3 via the network 5. The output unit 140 may output auxiliary information related to the expanded text data group in addition to the generated expanded text data group. Examples of the auxiliary information include score values ​​and frequency values ​​at the time of generation for each piece of text data included in the expanded text data group.

[0029] FIG. 7 is a table diagram illustrating an example of an expanded text data group generated from training text data by the expansion unit 120. In the example of FIG. 7, the leftmost column of the table shows the training text data and the derived text data generated therefrom. The columns "Input" to "Post-Processing" each show the number of text data pieces obtained as a result of the reception process by the input unit 100, the pre-processing by the pre-processing unit 122, the first cycle of the generation process by the generation unit 124, the second cycle of the generation process by the generation unit 124, and the post-processing by the post-processing unit 126. For example, suppose the input unit 100 receives text data "Please tell me the cause of the alarm." as training text data. Furthermore, suppose the processing parameters are set to five times per generation process, two cycles of the generation process, and three or more times as the limit value for the frequency count of text data tallied in post-processing.

[0030] In the example of FIG. 7 , when the input unit 100 receives the training text data, only the training text data "Please tell me the cause of the alarm." is managed (the value in the row "Please tell me the cause of the alarm." in the "Input" column is 1). Next, after preprocessing by the preprocessing unit 122, the generation unit 124 performs a first cycle of generation processing to generate other text data with the same meaning or text data with a similar meaning. As a result, five derived text data, "Please tell me the cause of the alarm.", "I don't know the cause of the alarm," "I don't know the alarm," "What alarm is it?", and "Do something about the alarm." are generated from the training text data "Please tell me the cause of the alarm." (The value in the row of each text data in the "Generation Process 1" column is 1). Then, a second cycle of generation processing is performed on these training text data and derived text data, resulting in the generation of an expanded text data set including 36 text data (the sum of the values ​​in the "Generation Process 2" column is 36). Finally, the post-processing unit 126 performs aggregation processing, leaving only text data with a frequency count of three or more, and excluding other text data from the aggregated text data group (the values ​​in the rows of "Do something about the alarm" and "I want to stop the alarm" in the "Post-processing" column are 0). In this way, by using the natural language processing device 1 according to this embodiment, it is possible to generate an expanded text data group that includes derived text data with the same or similar meaning as the training text data. By setting processing parameters, it is also possible to adjust the extent to which variations in spelling or different wordings are tolerated, the extent to which text data that is too different in meaning from the training text data is accepted, and so on. The expanded text data group generated in this way can be used to build and adjust large-scale language models.

[0031] The natural language processing apparatus 1 according to the present embodiment, having the above configuration, is expected to improve generalizability when unknown data not included in the training text data is input during the inference process of the trained large-scale language model by expanding the training text data based on user input for a set of candidate settings based on business requirements to be satisfied, such as the tolerance for spelling variations for each unique term, when building a large-scale language model tailored to that business based on individual business data from a manufacturing site, for example, a factory. Furthermore, the system outputs processing results and auxiliary information for representative examples of business data (training data) based on the set processing parameters, allowing the user to confirm them. This allows the system to adjust the processing parameters to suit the purpose and obtain the desired expanded text data set tailored to the purpose. In addition to conventional rule-based methods (such as Unicode normalization and stemming), the system also uses a text conversion method using generative AI, enabling business operations to be more in line with the user's intentions and improving the usability of the large-scale language model.

[0032] Second Embodiment A natural language processing apparatus according to a second embodiment of the present disclosure will be described below. The natural language processing apparatus 1 according to this embodiment has the same hardware configuration as the natural language processing apparatus 1 according to the first embodiment.

[0033] 8 is a schematic block diagram showing functions of the natural language processing apparatus 1 according to the second embodiment of the present disclosure. Each function of the natural language processing apparatus 1 according to this embodiment is realized by the CPU 11 included in the natural language processing apparatus 1 shown in FIG. 1 executing a system program and controlling the operation of each unit of the natural language processing apparatus 1.

[0034] The natural language processing apparatus 1 according to this embodiment further includes a learning unit 150 in addition to an input unit 100, an extension unit 120, and an output unit 140. The RAM 13 to the nonvolatile memory 14 of the natural language processing apparatus 1 are provided with a model storage unit 200, which is an area that stores in advance a trained model related to natural language processing that generates text data from text data.

[0035] The input unit 100 and the output unit 140 included in the natural language processing apparatus 1 according to this embodiment have the same functions as those according to the first embodiment.

[0036] The extension unit 120 according to this embodiment has the same functions as the extension unit 120 according to the first embodiment. The extension unit 120 outputs the generated extended text data group to the output unit 140. The extension unit 120 also outputs the generated extended text data group to the learning unit 150.

[0037] The learning unit 150 performs a learning process related to the large-scale language model based on the expanded text data group generated by the expansion unit 120. The learning unit 150 creates, for example, additional training data for the interactive response device 2 based on the expanded text data group. Then, it performs a process related to additional training of the interactive response device 2 using the created additional training data. The additional training data may be created according to the specifications of the interactive response device 2 to be trained. For example, the additional training data is data in JSON format that includes pairs of prompts and answers. The learning unit 150 may perform additional training of a trained model stored in, for example, the model storage unit 200. Alternatively, it may perform additional training of a trained model stored in another computer.

[0038] The natural language processing apparatus 1 according to this embodiment can perform a training process for a large-scale language model based on a set of extended text data. This is expected to improve generalizability when unknown data not included in the training text data is input in the inference process of the trained large-scale language model.

[0039] Other Embodiments In the above-described embodiment, the natural language processing apparatus 1 is described as including a model storage unit 200 that stores trained models. However, the trained models used by the natural language processing apparatus 1 may be managed on a computer other than the natural language processing apparatus 1, such as a fog computer 6 or a cloud server 7. In such a configuration, the natural language processing apparatus 1 references the model storage unit 200 on the other computer via the network 5. With such a configuration, it becomes possible to collectively manage trained models used in the augmentation process.

[0040] Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the individual embodiments described above. Various additions, substitutions, modifications, partial deletions, etc. are possible in these embodiments without departing from the gist of the invention or the idea and intent of the present disclosure derived from the content described in the claims and their equivalents. For example, in the above-described embodiments, the order of each operation and the order of each process are shown as examples and are not limited to these. The same applies when numerical values ​​or mathematical expressions are used in the description of the above-described embodiments.

[0041] The following are supplementary notes related to embodiments of the present disclosure. (Supplementary Note 1) A natural language processing apparatus (1) according to one aspect of the present disclosure includes an input unit (100) that accepts inputs related to processing parameters and training text data, an extension unit (120) that performs an extension process on the training text data using a learning model that receives text data as input and outputs text data based on the processing parameters, and an output unit (140) that outputs a result of the extension process, wherein the extension unit (120) includes a preprocessing unit (122) that performs preprocessing on the training text data based on the processing parameters, a generation unit (124) that performs generation processing on the preprocessed training text data using the learning model based on the processing parameters, and a postprocessing unit (126) that performs postprocessing on text data obtained as a processing result of the generation process based on the processing parameters.

[0042] (Supplementary Note 2) The post-processing unit (126) included in a natural language processing device (1) according to another aspect of the present disclosure performs a rejection process on text data obtained as a processing result of the generation process. (Supplementary Note 3) The generation process by the generation unit (124) included in a natural language processing device (1) according to another aspect of the present disclosure includes at least one of a translation process, a summarization process, a simplification process, and a correction process on text data. (Supplementary Note 4) The translation process of the natural language processing device (1) according to another aspect of the present disclosure performs a bidirectional translation process between Japanese and English. (Supplementary Note 5) The natural language processing device (1) according to another aspect of the present disclosure further includes a learning unit (150) that performs a learning process on a large-scale language model using a group of augmented text data obtained as a processing result of the augmentation process.

[0043] (Supplementary Note 6) A computer-readable recording medium according to one aspect of the present disclosure records a program that causes a computer to operate as an input unit (100) that accepts input related to processing parameters and training text data, an extension unit (120) that performs extension processing of the training text data using a learning model that receives text data as input and outputs text data based on the processing parameters, and an output unit (140) that outputs a result of the extension processing, and the extension unit (120) includes a pre-processing unit (122) that performs pre-processing of the training text data based on the processing parameters, a generation unit (124) that performs generation processing of the pre-processed training text data using the learning model based on the processing parameters, and a post-processing unit (126) that performs post-processing of text data obtained as a processing result of the generation processing based on the processing parameters.

[0044] REFERENCE SIGNS LIST 1 Natural language processing device 2 Interactive response device 3 Control device 4 Industrial machine 5 Network 6 Fog computer 7 Cloud server 11 CPU 12 ROM 13 RAM 14 Non-volatile memory 15, 17, 18, 20 Interface 22 Bus 70 Display device 71 Input device 72 External device 100 Input unit 120 Expansion unit 122 Pre-processing unit 124 Generation unit 126 Post-processing unit 140 Output unit 150 Learning unit 200 Model storage unit

Claims

1. A natural language processing device comprising: an input unit that accepts inputs related to processing parameters and training text data; an extension unit that performs extension processing of the training text data using a learning model that takes text data as input and outputs text data based on the processing parameters; and an output unit that outputs the results of the extension processing, wherein the extension unit comprises: a pre-processing unit that performs pre-processing of the training text data based on the processing parameters; a generation unit that performs generation processing of the pre-processed training text data using the learning model based on the processing parameters; and a post-processing unit that performs post-processing of text data obtained as a processing result of the generation processing based on the processing parameters.

2. The natural language processing device according to claim 1, wherein the post-processing executes a rejection process on the text data obtained as a processing result of the generation process.

3. The natural language processing device according to claim 1, wherein the generation process by said generation unit includes at least one of translation process, summarization process, simplification process, and correction process for text data.

4. The natural language processing device according to claim 3, wherein the translation process executes a bidirectional translation process between Japanese and English.

5. The natural language processing device according to any one of claims 1 to 4, further comprising a learning unit that executes a learning process for a large-scale language model using a group of expanded text data obtained as a result of the expansion process.

6. A computer-readable recording medium having recorded thereon a program that causes a computer to operate as: an input unit that accepts inputs related to processing parameters and training text data; an extension unit that performs extension processing of the training text data using a learning model that takes text data as input and outputs text data based on the processing parameters; and an output unit that outputs the results of the extension processing, wherein the extension unit comprises: a pre-processing unit that performs pre-processing of the training text data based on the processing parameters; a generation unit that performs generation processing of the pre-processed training text data using the learning model based on the processing parameters; and a post-processing unit that performs post-processing of text data obtained as a processing result of the generation processing based on the processing parameters.

Images