Sentence processing device, sentence processing method, and recording medium

Abstract

A sentence processing device according to the present disclosure comprises: a mask means for masking a proper noun included in a text sentence; a prediction means for predicting, by using a pre-trained model for natural language processing of the text sentence, a word to be pseudonymized into a masked portion; a replacement means for replacing the proper noun with the predicted word; an execution means for executing, by using a specialized-type trained model obtained by specializing the pre-trained model for a specific task, natural language processing of the text sentence in which the proper noun has been replaced; a restoration means for restoring, to the original proper noun, the word that has been replaced according to the processing result; and an output means for causing the restored processing result to be outputted.

Description

Text processing device, text processing method, and recording medium

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

[0002] There is a technique for preprocessing text to be summarized in order to improve the summarization accuracy of a trained model generated by machine learning.

[0003] For example, Patent Document 1 discloses that in order to improve the summarization accuracy of a trained model, a process of replacing predetermined words in a text to be summarized with predetermined phrases, symbols, etc. is carried out.

[0004] Japanese Patent Application Laid-Open No. 2019-016181

[0005] However, rather than generating a model from scratch, a technique has been developed that utilizes a pre-trained model using an MLM (Masked Language Model) and fine-tunes it for a specific task. In this technique, if the content of the words to be replaced is not taken into consideration, as in the invention described in Patent Document 1, it may be impossible to maintain the performance of sentence processing unless the replaced words are re-trained during fine-tuning.

[0006] An example of an objective of the present disclosure is to provide a text processing device or the like that can maintain the accuracy of text processing when fine-tuning without having to learn words to be replaced.

[0007] A text processing device according to one aspect of the present disclosure includes a masking means for masking proper nouns included in a text sentence, a prediction means for predicting words to be pseudonymized in the masked portion using a pre-trained model for natural language processing of the text sentence, a replacement means for replacing the proper nouns with the predicted words, an execution means for performing natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task instead of the pre-trained model, a restoration means for restoring the replaced words in the processing result to the original proper nouns, and an output means for outputting the restored processing result.

[0008] In one aspect of the present disclosure, a text processing method involves a computer masking proper nouns included in a text sentence, predicting words to be pseudonymized in the masked parts using a pre-trained model for natural language processing of the text sentence, replacing the proper nouns with the predicted words, performing natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task instead of the pre-trained model, restoring the replaced words in the processing result to the original proper nouns, and outputting the restored processing result.

[0009] In one aspect of the present disclosure, a recording medium stores a program that causes a computer to perform the following steps: mask proper nouns included in a text sentence; predict words to be kanamanized in the masked parts using a pre-trained model for natural language processing of the text sentence; replace the proper nouns with the predicted words; perform natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task instead of the pre-trained model; restore the replaced words in the processing result to the original proper nouns; and output the restored processing result.

[0010] According to one example of the effect of the present disclosure, when fine-tuning, the accuracy of sentence processing can be maintained without having to learn replacement words.

[0011] Fig. 1 is a block diagram showing the configuration of a text processing device according to the present disclosure. Fig. 2 is a diagram showing a hardware configuration in which the text processing device according to the present disclosure is realized by a computer device and its peripheral devices. Fig. 3 is a flowchart showing the operation of text processing according to the present disclosure. Fig. 4 is a block diagram showing the configuration of the text processing device according to the present disclosure. Fig. 5 is a block diagram showing the configuration of the text processing device according to the present disclosure.

[0012] Hereinafter, with reference to the drawings, embodiments of a text processing device, a text processing method, a program, and a non-transitory recording medium for recording the program according to the present disclosure will be described in detail. The disclosed technology is not limited to these embodiments.

[0013] First Embodiment Fig. 1 is a block diagram showing the configuration of a text processing device 100 according to the present disclosure. As shown in Fig. 1, the text processing device 100 includes a mask unit 101, a prediction unit 102, a replacement unit 103, an execution unit 104, a restoration unit 105, and an output unit 106. The text processing device 100 is a device for performing natural language processing such as summarization on any text. The text is not particularly limited, and may be, for example, a news article. The text may also be a transcript of a meeting or interview.

[0014] 2 is a diagram showing an example of a hardware configuration in which the text processing device 100 according to the present disclosure is realized by a computer device 500 including a processor. As shown in Fig. 2, the text processing device 100 includes a processor 501, memories such as a ROM (Read Only Memory) 502 and a RAM (Random Access Memory) 503, a storage device 505 such as a hard disk for storing a program 504, a communication I / F (Interface) 508 for network connection, and an input / output interface 511 for inputting and outputting data.

[0015] The processor 501 controls the entire computer device 500. The processor 501 may be, for example, a central processing unit (CPU), a digital signal processor (DSP), a graphics processing unit (GPU), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, or a combination thereof.

[0016] The processor 501 runs an operating system to control the entire text processing device 100 according to the present disclosure. The processor 501 also reads programs and data into memory from a recording medium 506 attached to a drive device 507, for example. The processor 501 also functions as the mask unit 101, prediction unit 102, replacement unit 103, execution unit 104, restoration unit 105, and output unit 106 according to the present disclosure, or as part of these units, and executes processing or instructions in the flowchart shown in FIG. 3, which will be described later, based on the program.

[0017] The recording medium 506 is, for example, an optical disk, a flexible disk, a magneto-optical disk, an external hard disk, or a semiconductor memory. A part of the recording medium in the storage device is a non-volatile storage device, and the program is recorded therein. The program may also be downloaded from an external computer (not shown) connected to a communication network.

[0018] The input device 509 is realized by, for example, a mouse, a keyboard, built-in key buttons, etc., and is used for input operations. The input device 509 is not limited to a mouse, a keyboard, or built-in key buttons, and may be, for example, a touch panel. The output device 510 is realized by, for example, a display, and is used to check output.

[0019] As described above, the text processing device 100 shown in Fig. 1 is realized by the computer hardware shown in Fig. 2. However, the means for realizing each unit of the text processing device 100 in Fig. 1 is not limited to the configuration described above. The text processing device 100 may be realized by a single physically coupled device, or may be realized by two or more physically separated devices connected by wire or wirelessly. For example, the input device 509 and the output device 510 may be connected to the computer device 500 via a network. The text processing device 100 shown in Fig. 1 may also be configured using cloud computing or the like.

[0020] The masking unit 101 is a means for masking proper nouns included in a text sentence. The masking unit 101 extracts proper nouns such as names that represent entities, such as personal names, company names, and organization names, place names, and product names, using named entity recognition (NER) from the text sentence to be processed, which is stored in a server device or the like, and masks the extracted proper nouns. The masking unit 101 may mask only specific types of proper nouns. For example, the masking unit 101 may mask only personal names. In this specification, a masked portion of a text sentence may be referred to as [masked].

[0021] The prediction unit 102 is a means for predicting words to be kananamed in masked portions using a pre-trained model for natural language processing of text sentences. The pre-trained model is a model pre-trained by MLM using text sentences in which parts of the text sentences are masked. That is, for training input sentences in which parts of the text sentences are masked, the prediction unit 102 solves so-called fill-in-the-blank problems, in which the prediction unit 102 predicts words to be kananamed by solving fill-in-the-blank problems for the masked text sentences using the pre-trained model. In this case, the prediction unit 102 predicts proper nouns learned by the pre-trained model as words to be kananamed. The training data of the pre-trained model is large-scale, and its aggregation and analysis are time-consuming. This allows learned words to be obtained from the pre-trained model without much effort.

[0022] The replacement unit 103 is a means for replacing proper nouns with predicted words. The replacement unit 103 replaces proper nouns in a text sentence with predicted words, and records information on combinations of the proper nouns and the replaced words in the storage device 505 or the like.

[0023] The execution unit 104 is a means for performing natural language processing of a text sentence in which proper nouns have been replaced, using a specialized trained model that is a pre-trained model specialized for a specific task. The text sentence here refers to the text sentence to be processed. In the present disclosure, the execution unit 104 performs summarization of the text sentence in which proper nouns have been replaced, using, for example, a specialized trained model specialized for a text summarization task. In addition to summarization, the execution unit 104 may also perform processes such as machine translation, question answering, or information search as natural language processing.

[0024] As each trained model, a known machine learning engine or natural language processing algorithm can be used as appropriate. Furthermore, the trained model may be a large-scale language model trained with a large amount of text data, or a trained model obtained by transfer learning of the large-scale language model. For example, GPT-2 (Generative Pre-Training-2), GPT-3, or GPT-4 can be used as the large-scale language model. Furthermore, T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), or ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately) may be used as the large-scale language model.

[0025] The restoration unit 105 is a means for restoring the words replaced in the processing result to the original proper nouns. The restoration unit 105 restores the original proper nouns based on the combination of the recorded proper nouns and the replaced words.

[0026] The output unit 106 is a means for outputting the processing result after restoration. The output unit 106 displays the processing result of the text sentence on a display device such as a display. The output unit 106 may display the processing result in response to a user's operation such as summarizing the text sentence.

[0027] The operation of the text processing device 100 configured as above will be described with reference to the flowchart of FIG.

[0028] 3 is a flowchart showing an outline of the operation of the text processing device 100 according to the present disclosure. Note that the processing according to this flowchart may be executed based on program control by the processor described above.

[0029] As shown in FIG. 3 , first, the masking unit 101 masks proper nouns included in the text (step S101). Next, the prediction unit 102 predicts words to be kananamed in the masked portion using a pre-trained model for natural language processing of the text (step S102). Next, the replacement unit 103 replaces the proper nouns with the predicted words (step S103). Next, the execution unit 104 performs natural language processing of the text with the replaced proper nouns using a specialized trained model that is specialized for a specific task instead of the pre-trained model (step S104). Next, the restoration unit 105 restores the replaced words in the processing result to the original proper nouns (step S105). Finally, the output unit 106 outputs the restored processing result (step S106). This concludes the text processing operation of the text processing device 100.

[0030] In the text processing device 100, the prediction unit 102 predicts the word to be kana-nized in the masked portion using a pre-trained model for natural language processing of text sentences. The replacement unit 103 then replaces the proper noun with the predicted word, and the execution unit 104 performs natural language processing of the text sentence in which the proper noun has been replaced using a specialized trained model specialized for a specific task instead of the pre-trained model. Because the replaced word is trained when the pre-trained model is generated, there is no need to train the word when generating the specialized trained model. This makes it possible to maintain the accuracy of text processing during fine-tuning without having to train the words to be replaced.

[0031] [Modification 1] Next, each modification of the present disclosure will be described in detail with reference to the drawings. Each modification can be implemented in combination. Below, to the extent that the description of this embodiment is not unclear, description of content that overlaps with the above description will be omitted. As with the computer device shown in FIG. 2, the functions of each component in each embodiment of the present disclosure can be realized not only by hardware but also by a computer device or software based on program control.

[0032] Fig. 4 is a block diagram showing the configuration of a text processing device 110 according to the present disclosure. With reference to Fig. 4, the text processing device 110 will be described, focusing on the parts that differ from the text processing device 100. The text processing device 110 includes a mask unit 111, a reception unit 112, a prediction unit 113, a replacement unit 114, an execution unit 115, a restoration unit 116, and an output unit 117. The configuration of this embodiment, other than the reception unit 112 and the prediction unit 113, is substantially the same as the corresponding components in the first embodiment.

[0033] The receiving unit 112 is a means for receiving clue information that serves as a clue for predicting words from the training data of the pre-trained model. The receiving unit 112 receives input of clue information from a user on, for example, an operation screen of an application program that summarizes text sentences. If the proper noun is a person's name (surname), the receiving unit 112 receives clue information for the replacement word, such as, for example, "[Mask] is a name commonly used in Japan." Furthermore, if the proper noun is an organization name, the receiving unit 112 receives clue information for the replacement word, such as, for example, "[Mask] is the organization name."

[0034] The receiving unit 112 may receive clue information through a dialogue with another language model different from the pre-trained model and the specialized trained model. The receiving unit 112 receives clue information, for example, based on a user's answer to a question about the clue information posed by the language model. In this case, a question from the language model is displayed on the screen of an application program that summarizes text sentences. The user then answers the question by operating the screen. Suppose the proper noun is a person's name (surname), and for example, the question displayed by the language model is, "Is [mask] a commonly used name in Japan?" In this case, if the user answers "Yes," the information about the replacement word is received as "a commonly used name in Japan." On the other hand, if the user answers "No," the language model asks another question about the clue information.

[0035] The reception unit 112 may acquire, as clue information, information indicating characteristics of the word to be predicted. For example, if the proper noun is a person's name, the reception unit 112 may acquire characteristics unique to that person, such as attributes, titles, and personality. For example, if the proper noun is a person's name, the reception unit 112 may acquire information indicating characteristics of the word to be replaced, such as "[Mask] is a very capable manager." For example, if the proper noun is the name of an organization, the reception unit 112 may acquire information indicating characteristics, such as "[Mask] is a very well-known company."

[0036] The prediction unit 113 predicts a word to be used to kananameize a proper noun based on clue information. For example, the prediction unit 113 predicts a word that satisfies the conditions of the clue information. If there are multiple proper nouns that satisfy the conditions of the clue information, the prediction unit 113 may predict multiple kananame candidates. For example, if the receiving unit 112 receives clue information such as, “[Mask] is a name commonly used in Japan,” the prediction unit 113 may predict a word using multiple kananame candidates and numerical values ​​indicating the confidence levels of each kananame candidate, such as “Sato: confidence level 0.6, Suzuki: confidence level 0.4,” using a pre-trained model. In this case, the replacement unit 114 may replace the proper nouns in descending order of confidence level, and the execution unit 115 may sequentially perform natural language processing.

[0037] In this modification, the receiving unit 112 receives clue information that serves as a clue for predicting words from the training data of the pre-trained model. The predicting unit 113 then predicts words to be converted into kana based on the clue information. This allows proper nouns to be replaced with words that are more suitable for kana conversion, thereby improving processing performance for text sentences.

[0038] [Modification 2] Next, another modification of the present disclosure will be described in detail with reference to the drawings. Below, descriptions of content that overlaps with the above description will be omitted to the extent that the description of this embodiment is not unclear. As with the computer device shown in FIG. 2, the functions of each component in each embodiment of the present disclosure can be realized not only by hardware but also by a computer device or software based on program control.

[0039] Fig. 5 is a block diagram showing the configuration of a text processing device 120 according to the present disclosure. With reference to Fig. 5, the text processing device 120 will be described, focusing on the parts that differ from the text processing device 100. The text processing device 120 includes a mask unit 121, an estimation unit 122, a prediction unit 123, a replacement unit 124, an execution unit 125, a restoration unit 126, and an output unit 127. The configuration of this embodiment, other than the estimation unit 122 and the prediction unit 123, is the same as the corresponding components in the first embodiment.

[0040] The estimation unit 122 is a means for estimating the type of the identified proper noun using natural language processing technology. Specifically, the estimation unit 122 uses NER, dependency analysis, or the like to identify proper nouns from the target text sentence and classify them into categories such as person names, organization names, places, dates, etc. The categories to be classified may be categories with lower concepts than those described above. For example, the estimation unit 122 may classify an organization name into categories with lower concepts such as corporate names, company names, and joint stock company names.

[0041] The prediction unit 123 predicts a word to be used to kananameize a proper noun based on the type. For example, the prediction unit 123 selects a word in the same category as the source proper noun from fill-in candidates in the pre-trained model. If the classified category is names of corporations, the prediction unit 123 predicts a corporation name such as "XX Corporation."

[0042] In this modification, the estimation unit 122 estimates the type of the identified proper noun using natural language processing technology. Then, the prediction unit 123 predicts the word to be used to kanamagari the proper noun based on the type. This allows the proper noun to be replaced with the same type of proper noun as before the replacement, thereby improving processing performance for text sentences.

[0043] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

[0044] For example, although multiple operations are described in a sequence in the form of a flowchart, the sequence does not limit the order in which the multiple operations are performed. Therefore, when implementing each embodiment, the sequence of the multiple operations can be changed within the scope that does not affect the content.

[0045] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0046] (Supplementary Note 1) A text processing device comprising: a masking means for masking proper nouns included in a text sentence; a prediction means for predicting words to be pseudonymized in the masked part using a pre-trained model for natural language processing of the text sentence; a replacement means for replacing the proper nouns with the predicted words; an execution means for performing natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task as the pre-trained model; a restoration means for restoring the replaced words in the processing result to the original proper nouns; and an output means for outputting the restored processing result.

[0047] (Supplementary Note 2) The text processing device according to Supplementary Note 1, wherein the pre-trained model is used to solve a fill-in-the-blank question for the masked text sentence, thereby predicting words to be kana-converted.

[0048] (Supplementary Note 3) The text processing device according to Supplementary Note 1 or Supplementary Note 2, wherein the execution means performs summarization of text sentences in which proper nouns have been replaced using a specialized trained model that is specialized for a text summarization task as the pre-trained model.

[0049] (Supplementary Note 4) A sentence processing device according to any one of Supplementary Notes 1 to 3, further comprising a receiving means for receiving clue information that serves as a clue for predicting the word from the training data of the pre-trained model, wherein the prediction means predicts the word to be converted into kana from the proper noun based on the clue information.

[0050] (Supplementary Note 5) The text processing device according to Supplementary Note 4, wherein the accepting means acquires, as the clue information, information indicating characteristics of a word to be predicted.

[0051] (Supplementary Note 6) The text processing device according to Supplementary Note 4 or Supplementary Note 5, wherein the accepting means accepts the clue information through a dialogue with a language model.

[0052] (Supplementary Note 7) A text processing device according to any one of Supplementary Notes 1 to 6, further comprising an estimation means for estimating the type of the identified proper noun using natural language processing technology, wherein the prediction means predicts a word to be used to kana-encode the proper noun based on the type.

[0053] (Supplementary Note 8) A text processing method in which a computer masks proper nouns contained in a text sentence, predicts words to be kananamed in the masked part using a pre-trained model for natural language processing of the text sentence, replaces the proper nouns with the predicted words, performs natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task as the pre-trained model, restores the replaced words in the processing result to the original proper nouns, and outputs the restored processing result.

[0054] (Supplementary Note 9) The text processing method according to Supplementary Note 8, wherein the pre-trained model is used to solve a fill-in-the-blank question in the masked text sentence, thereby predicting words to be kana-ized.

[0055] (Supplementary Note 10) The text processing method according to Supplementary Note 8 or Supplementary Note 9, wherein the pre-trained model is a specialized trained model specialized for a text summarization task, and a summary of the text sentence in which proper nouns have been replaced is performed.

[0056] (Supplementary Note 11) A text processing method according to any one of Supplementary Notes 8 to 10, comprising receiving clue information that serves as a clue for predicting the word from training data of the pre-trained model, and predicting a word to convert the proper noun into a kana name based on the clue information.

[0057] (Supplementary Note 12) The text processing method according to Supplementary Note 11, wherein information indicating characteristics of a word to be predicted is acquired as the clue information.

[0058] (Supplementary Note 13) The text processing method according to Supplementary Note 11 or Supplementary Note 12, wherein the clue information is received through a dialogue with a language model.

[0059] (Supplementary Note 14) The text processing method according to any one of Supplementary Note 8 to Supplementary Note 13, which uses natural language processing technology to estimate the type of the identified proper noun, and predicts a word to convert the proper noun into a kana based on the type.

[0060] (Supplementary Note 15) A recording medium storing a program that causes a computer to perform the following steps: masking proper nouns included in a text sentence; predicting words to be kananamed in the masked part using a pre-trained model for natural language processing of the text sentence; replacing the proper nouns with the predicted words; performing natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task as the pre-trained model; restoring the replaced words in the processing result to the original proper nouns; and outputting the restored processing result.

[0061] (Supplementary Note 16) The recording medium according to Supplementary Note 15, wherein the pre-trained model is used to solve a fill-in-the-blank question in the masked text sentence, thereby predicting words to be kana-ized.

[0062] (Supplementary Note 17) The recording medium according to Supplementary Note 15 or Supplementary Note 16, wherein the pre-trained model is a specialized trained model specialized for a sentence summarization task, and a summary of a text sentence in which proper nouns have been replaced is performed.

[0063] (Appendix 18) A recording medium described in any one of Appendices 15 to 17, which receives clue information that serves as a clue for predicting the word from training data of the pre-trained model, and predicts a word to kananameize the proper noun based on the clue information.

[0064] (Supplementary Note 19) The recording medium according to Supplementary Note 18, wherein the clue information is received through a dialogue with a language model.

[0065] (Supplementary Note 20) A recording medium according to any one of Supplementary Note 15 to Supplementary Note 19, which uses natural language processing technology to estimate the type of a specified proper noun, and predicts a word to convert the proper noun into a kana based on the type.

[0066] Some or all of the configurations described in Supplements 2 to 7 that are dependent on Supplement 1 above may also be dependent on Supplement 8 and Supplement 15 in the same dependent relationship as Supplements 2 to 7. Not limited to Supplement 1, Supplement 8, and Supplement 15, some or all of the configurations described as Supplements may be made dependent on various hardware, software, various recording devices for recording software, or systems, within the scope of each of the above-mentioned embodiments.

[0067] 100, 110, 120 Text processing device 101, 111, 121 Mask unit 102, 113, 123 Prediction unit 103, 114, 124 Replacement unit 104, 115, 125 Execution unit 105, 116, 126 Restoration unit 106, 117, 127 Output unit 112 Reception unit 122 Estimation unit 500 Computer device 501 Processor 502 ROM 503 RAM 504 Program 505 Storage device 506 Recording medium 507 Drive device 508 Communication interface 511 Input / output interface 512 Bus

Claims

1. A text processing device comprising: a masking means for masking proper nouns included in a text sentence; a prediction means for predicting words to be pseudonymized in the masked part using a pre-trained model for natural language processing of the text sentence; a replacement means for replacing the proper nouns with the predicted words; an execution means for performing natural language processing of the text sentence in which the proper nouns have been replaced using the pre-trained model as a specialized trained model specialized for a specific task; a restoration means for restoring the replaced words in the processing result to the original proper nouns; and an output means for outputting the restored processing result.

2. The text processing device according to claim 1, wherein the prediction means predicts words to be kana-converted by solving fill-in-the-blank questions for the masked text sentences using the pre-trained model.

3. A text processing device as described in claim 1 or claim 2, wherein the execution means performs summarization of text sentences in which proper nouns have been replaced using a specialized trained model that is specialized for the task of text summarization instead of the pre-trained model.

4. A text processing device as described in any one of claims 1 to 3, further comprising a receiving means for receiving clue information that serves as a clue for predicting the word from the training data of the pre-trained model, wherein the prediction means predicts the word to be converted into kana from the proper noun based on the clue information.

5. The text processing device according to claim 4, wherein said receiving means acquires, as said clue information, information indicating characteristics of the word to be predicted.

6. The text processing device according to claim 4 or 5, wherein said receiving means receives said clue information through a dialogue with a language model.

7. A text processing device according to any one of claims 1 to 6, further comprising an estimation means for estimating the type of the identified proper noun using natural language processing technology, wherein the prediction means predicts a word to be used to convert the proper noun into a kana based on the type.

8. A text processing method in which a computer masks proper nouns contained in a text sentence, predicts words to be pseudonymized in the masked part using a pre-trained model for natural language processing of the text sentence, replaces the proper nouns with the predicted words, performs natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task as the pre-trained model, restores the replaced words in the processing result to the original proper nouns, and outputs the restored processing result.

9. The text processing method according to claim 8, wherein the pre-trained model is used to solve a fill-in-the-blank question for the masked text sentence, thereby predicting words to be kana-ized.

10. A text processing method as described in claim 8 or claim 9, wherein the pre-trained model is used as a specialized trained model specialized for a text summarization task to perform summarization of text sentences in which proper nouns have been replaced.

11. A text processing method according to any one of claims 8 to 10, comprising receiving clue information that serves as a clue for predicting the word from the training data of the pre-trained model, and predicting a word to convert the proper noun into a kana based on the clue information.

12. The text processing method according to claim 11, wherein information indicating characteristics of a word to be predicted is acquired as the clue information.

13. A text processing method according to claim 11 or 12, wherein the clue information is received through a dialogue with a language model.

14. A text processing method according to any one of claims 8 to 13, which uses natural language processing technology to estimate the type of the identified proper noun, and predicts the word to be used to convert the proper noun into a kana based on the type.

15. A recording medium storing a program that causes a computer to perform the following steps: masking proper nouns contained in a text sentence; predicting words to be kananamed in the masked part using a pre-trained model for natural language processing of the text sentence; replacing the proper nouns with the predicted words; performing natural language processing of the text sentence in which the proper nouns have been replaced using a specialized trained model specialized for a specific task as the pre-trained model; restoring the replaced words in the processing results to the original proper nouns; and outputting the restored processing results.

16. The recording medium according to claim 15, wherein the pre-trained model is used to solve a fill-in-the-blank question in the masked text sentence, thereby predicting words to be kana-ized.

17. A recording medium according to claim 15 or 16, wherein the pre-trained model is used as a specialized trained model specialized for a text summarization task to perform summarization of text sentences in which proper nouns have been replaced.

18. A recording medium according to any one of claims 15 to 17, which receives clue information that serves as a clue for predicting the word from the training data of the pre-trained model, and predicts the word to which the proper noun will be converted into a kana name based on the clue information.

19. The recording medium of claim 18, wherein the clue information is received through interaction with a language model.

20. A recording medium according to any one of claims 15 to 19, which uses natural language processing technology to estimate the type of the identified proper noun, and predicts the word to convert the proper noun into kana based on the type.

Images