Information processing system

Abstract

To provide an information processing system capable of obtaining a text that a user is looking for.SOLUTION: A text proofreading system obtains a plurality of proofread texts based on the text input by the user, and suggests a proofread text to the user that is closest to an average in a dispersion expression of the proofread texts.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] One aspect of the present invention relates to an information processing system and an information processing method.

[0002] Note that one embodiment of the present invention is not limited to the above technical field. The technical field of one embodiment of the invention disclosed in this specification relates to an object, a method, or a manufacturing method. Alternatively, one embodiment of the present invention relates to a process, a machine, a manufacture, or a composition of matter. Therefore, more specifically, examples of the technical field of one embodiment of the present invention disclosed in this specification include a semiconductor device, a display device, a light-emitting device, a power storage device, a memory device, a driving method thereof, or a manufacturing method thereof. [Background technology]

[0003] In recent years, the development of language models using neural networks has been actively carried out, and large-scale language models (LLMs) in particular have attracted attention. A large-scale language model is a natural language processing model trained using a large amount of data. The large-scale language model can realize a dialogue model that responds to user instructions, for example. Non-Patent Document 1 discloses GPT-4 (Generative Pre-trained Transformer 4) as a large-scale language model, and ChatGPT as a dialogue model. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Summary of ChatGPT / GPT-4 Research and Perspective Towards the Future of Large Language Models, Yiheng Liu et al. (Submitted on 4 Apr 2023, [online], Internet<URL:https: / / arxiv.org / abs / 2304.01852> Summary of the Invention [Problem to be solved by the invention]

[0005] By using an application that uses generative AI such as ChatGPT, it is possible to create text in an interactive format and it can also be used as a text proofreading tool.

[0006] However, conversational sentences and written sentences do not necessarily match. For example, when exchanging sentences with someone, the optimal sentence will differ depending on whether the person you are exchanging with is a superior, or whether they speak the same native language, but since a generative AI only presents one sentence derived from a large amount of dialogue data, the sentence will not be influenced by the other person, and there is a risk that it will not be the sentence the user desires.

[0007] In addition, in the case of complex sentences, there is a risk of misunderstanding. Even in the case of complex sentences, if only generative AI is used, the sentence obtained by the database and program is easily determined, and the sentence that the AI ​​thinks is the best is provided.

[0008] Generative AI can also be used as a tool for creating machine language programs such as Python and C++. The output depends on whether the generative AI correctly understands the user's query and whether it has been sufficiently trained to produce the optimal output.

[0009] In this way, generation AI alone may not produce the sentences that users are looking for.

[0010] In addition, because generative AI requires a large number of calculations based on a large number of parameters, it is difficult to predict what impact changes to the user's input conditions will have.

[0011] If the sentence the user is looking for is not provided, the user must reconsider the input conditions or commands and try again, but may still not obtain the desired result. It is difficult for the user to determine what conditions and what to input so that the sentence will be changed by the generation AI and the sentence that is being output. [Means for solving the problem]

[0012] Therefore, multiple prompts are created from the sentences entered by the user. These multiple prompts are created to be similar to the sentences entered by the user. For example, if the sentence entered by the user is "Please proofread the following sentence. The sentence to be proofread is 'The main components of the said substance are lithium and sulfur'", the prompts are "Please proofread the following sentence. The sentence to be proofread is 'The main components of the said substance are alkali metals and sulfur'", "Please proofread the following sentence. The sentence to be proofread is 'The said substance contains alkali metals and sulfur'", etc.

[0013] The prompt may include an instruction such as "Please proofread the following sentence," or data that explains and supplements the target sentence, such as "The main components of the above substance are lithium and sulfur." ​​Each prompt is input into the generation AI, which outputs multiple sentences, and among the multiple sentences, multiple sentences that are close to the average are prepared, and the user can select from among them to obtain the sentence they are looking for in this information processing system. A sentence proofreading system is one example of an information processing system.

[0014] Furthermore, by averaging multiple sentences output by different generation AIs, the probability of providing the user with the sentence they are looking for is further increased.

[0015] The problem with using AI to create sentences is that it may output sentences that are incorrect or misleading, which is different from the user's intention. Therefore, it is desirable for the user to ultimately check and select the correct sentence.

[0016] The text proofreading system obtains a plurality of proofread texts based on the text input by the user, and suggests to the user a proofread text that is close to the average in the distributed representation of the proofread texts.

[0017] The configuration disclosed in this specification is an information processing system that includes an input / output interface that accepts input sentences from a user and shows answer sentences to the user, a generation unit that generates a prompt based on the input sentence, a sentence generation unit that generates multiple answer sentences, and a comparison unit that obtains distributed representations of sentences for the answer sentences, calculates the average value of the distributed representations, and selects the answer sentences, wherein the input / output interface outputs the input sentences to the generation unit, the generation unit transmits the prompts to the sentence generation unit, and the comparison unit obtains the answer sentences from the sentence generation unit and selectively outputs the answer sentences to the input / output interface.

[0018] It is also possible to construct an information processing system via an external sentence generation unit and a network. Another configuration disclosed in this specification is an information processing system that has an input / output interface that accepts input sentences from a user and shows answer sentences to the user, a generation unit that generates a prompt based on the input sentence, and a comparison unit that obtains distributed representations of sentences for the answer sentences, calculates the average value of the distributed representations, and selects the answer sentence, wherein the generation unit transmits the prompt via the network to the sentence generation unit that generates multiple answer sentences, and the comparison unit obtains the answer sentences from the sentence generation unit via the network and selectively outputs the answer sentences to the input / output interface.

[0019] In the above configuration, the sentence generation unit is a plurality of types of deep learning engines having a language model. By using a plurality of types of deep learning engines, the variation of sentences created increases, and the user can select the most suitable sentence.

[0020] The deep learning engine uses an artificial intelligence-based algorithm (reinforcement learning algorithm or deep learning algorithm) to build a language model for the corresponding user. The programs and huge amounts of data used in the deep learning engine are stored on a dedicated server, and users can access the deep learning engine via the Internet. Effect of the Invention

[0021] The information processing system disclosed in this specification performs processing using a distributed representation technique in natural language processing, and can select a sentence that is close to the average in the distributed representation from multiple sentences. It can also suggest a sentence that is close to the selected average to a user. If the information processing system is a text proofreading system, the proofread text is less likely to be misunderstood by the recipient.

[0022] The text proofreading system disclosed in this specification allows a user to select the most appropriate sentence from among the answer sentences created by the deep learning engine, based on the content of the original text. [Brief description of the drawings]

[0023] [Figure 1] FIG. 13 is a diagram showing a process flow of text proofreading according to one embodiment of the present invention. [Diagram 2] 1 is a diagram showing an example of a network configuration of a text proofreading system according to an embodiment of the present invention; [Diagram 3] 1 is a block diagram of a text proofreading system according to an embodiment of the present invention; [Figure 4] 1 is a diagram showing an example of a network configuration of a text proofreading system according to an embodiment of the present invention; [Diagram 5] 1 is a diagram showing an example of a network configuration of a text proofreading system according to an embodiment of the present invention; [Figure 6] 1 is a diagram showing an example of a network configuration of a text proofreading system according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] Hereinafter, the embodiments of the present invention will be described in detail with reference to the drawings. However, the present invention is not limited to the following description, and it will be easily understood by those skilled in the art that the form and details of the present invention can be modified in various ways. Furthermore, the present invention is not to be interpreted as being limited to the description of the embodiments shown below.

[0025] (Embodiment 1) In this embodiment, an information processing system according to one embodiment of the present invention will be described with reference to Fig. 1, Fig. 2, and Fig. 3. In this embodiment, a text proofreading system which is an example of the information processing system will be described.

[0026] An information processing system according to one aspect of the present invention includes at least an input / output interface, a generator, a sentence generator that outputs a plurality of answer sentences, and a comparator.

[0027] The information processing system is capable of transmitting and receiving data to and from an information processing device capable of performing processing using a language model via a network. In the information processing system according to one embodiment of the present invention, a sentence generation unit performs processing using a language model via a network.

[0028] <Input / output interface> The input / output interface is a device that allows a user to input original text data and shows the user the proofread text data. Therefore, the input / output interface has at least a display unit that displays the proofread text data. The input may be made by the user's voice.

[0029] The input / output interface is an information processing device that can use chat software such as InCircle and has a display unit. For example, it is a mobile information terminal such as a smartphone or a tablet computer, or an information processing device such as a desktop computer or a notebook computer that has a keyboard. These information processing devices have at least a display unit that can display proofread text data.

[0030] A user can input original text into the chat software and send the proofread text data obtained by the process described below to the other party.

[0031] <Generation part> The generation unit receives text data from the input / output interface.

[0032] The text data includes multiple words. The generation unit generates multiple target sentences based on the text data. Each target sentence is generated by replacing at least one of the multiple words in the text data. For example, a target sentence can be generated by obtaining embedded representations of words included in the text data and replacing the words with words that have similar embedded representations, i.e., words that are similar in meaning. Furthermore, an instruction sentence, for example "Please proofread this target sentence," is added to the target sentence to create a prompt.

[0033] A distributed representation is a representation of the meaning of a word as a low-dimensional real-valued vector. Algorithms such as Word2Vec, FastText, and GloVe can be used to obtain a distributed representation of a word. If the replacement words are different, multiple different target sentences will be generated. The system has the function of sending multiple prompts generated from these multiple target sentences to an AI sentence proofreading tool, which is the sentence generation unit, online, or the prompts can be sent via a network to a device with the same function (sentence generation unit).

[0034] <Sentence generation section> When the sentence generation unit receives multiple prompts online from the generation unit, multiple sentence data are created based on each of the multiple prompts. The sentence generation unit performs processing using a language model (also called AI sentence proofreading). The sentence generation unit is a large computer such as a server computer or a supercomputer. The sentence generation unit can also be called an AI sentence proofreading tool. Multiple different AI sentence proofreading tools can also be used.

[0035] <Comparison section> The comparison unit processes the proofread text received from the online text generation unit (AI proofreading tool) using a distributed representation technique. Distributed representations can use doc2vec, SentenceBERT, Universal Sentence Encoder, BERTScore, etc. Distributed representations can also be compared and the average calculated.

[0036] It is preferable to obtain the likelihood of the answer sentence when calculating the distributed representation of the proofread text.

[0037] FIG. 1 shows an example of a processing flow of text proofreading when a user uses an information processing system according to an embodiment of the present invention.

[0038] First, a user inputs first text data or at least one first character string using the input / output interface (step S11).

[0039] Next, a plurality of different prompts are created based on a portion of the first text data or a portion of at least one of the first character strings (step S12).

[0040] Next, a plurality of different second text data are obtained by sending a plurality of prompts to the language model (step S13).

[0041] Next, a similarity calculation is performed for the acquired plurality of second text data, and a plurality of second text data close to the average value is extracted (step S14).

[0042] Next, a plurality of second text data close to the average value are displayed on the display unit for the user to select (step S15).

[0043] By following the above flow, the user can select the proofread text data that he or she desires to obtain.

[0044] FIG. 2 shows an example of a network configuration diagram for executing the above flow.

[0045] 2 shows information terminal 20a, information terminal 20b, information terminal 20c, and information terminal 20d as examples of input / output interfaces. Information terminal 20a, information terminal 20b, information terminal 20c, and information terminal 20d can transmit and receive data to and from information processing device 10 via network 31. Information processing device 10 can also transmit and receive data to and from information processing device 40 via network 30.

[0046] Network 31 is a computer network smaller than network 30. Typically, network 30 is a global network, and network 31 is a local network. It is preferable to use the Internet, which is the foundation of the World Wide Web (WWW), as network 30. It is preferable to use an intranet or an extranet as network 31. In this embodiment, an example will be mainly described in which the Internet is used as network 30 and an intranet is used as network 31.

[0047] In addition, the network 31 may be a computer network such as a Personal Area Network (PAN), a Local Area Network (LAN), a Campus Area Network (CAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), or a Global Area Network (GAN).

[0048] In addition, when performing wireless communication, communication standards such as the fourth generation mobile communication system (4G), the fifth generation mobile communication system (5G), and the sixth generation mobile communication system (6G), or specifications standardized by IEEE such as Wi-Fi (registered trademark) and Bluetooth (registered trademark), can be used as communication protocols or communication technologies.

[0049] A user of the information processing system can access the information processing device 10 from the information terminals 20a to 20d, etc., via the network 31. Then, the user can receive a service using the information processing method according to one aspect of the present invention.

[0050] A user can use the information processing device 10, for example, by using dedicated software or an application installed on the information terminal 20a. Alternatively, a user can use the information processing device 10, for example, by using the information terminal 20a from the network 31 or a web browser.

[0051] An organization such as a company that provides a service using an information processing method according to an embodiment of the present invention (hereinafter also referred to as a service provider) can provide the service using an information processing device 10. The information processing device 10 is a device that can execute processing using an information processing method according to an embodiment of the present invention. In Fig. 2, a server computer is illustrated as an example.

[0052] The information processing device 10 generates a prompt based on original text (first text data) input to the input / output interface. The information processing device 10 is a computer with a higher processing capacity than the information terminal 20a. In this embodiment, the information processing device 10 functions as a generating unit. This stage corresponds to step S12 in FIG. 1.

[0053] A plurality of prompts generated by the information processing device 10 are input as data to the information processing device 40 via the network 30. The information processing device 40 can perform information processing such as calculations using the data transmitted from the information processing device 10.

[0054] The information processing device 40 is a large computer such as a server computer or a supercomputer. The information processing device 40 is a computer with a higher processing capacity than the information terminal 20a. The information processing device 40 is also a computer with a higher processing capacity than the information processing device 10. It is preferable that the information processing device 40 has a function as a parallel computer. By using the information processing device 40 as a parallel computer, for example, it is possible to perform large-scale calculations required for AI learning and inference. Note that when both the information processing device 10 and the information processing device 40 have a function as a parallel computer, the information processing device 40 has a higher processing capacity and can perform large-scale calculations compared to the information processing device 10.

[0055] The information processing device 40 can perform processing using a natural language processing model using AI. In addition, examples of the natural language processing model using AI in the information processing device 40 include BART (Bidirectional Auto-Regressive from Transformers) and T5 (Text-to-Text Transfer Transformer). In addition, the information processing device 40 can perform processing using a model using a large-scale language model (such as a sentence generation model or a dialogue model). Examples of the large-scale language model include GPT-3, GPT-3.5, GPT-4, LaMDA (Language Model for Dialogue Applications), PaLM (Pathways Language Model), and PaLM2, and it is preferable to use GPT-4.

[0056] In this embodiment, the information processing device 40 can use an AI text proofreading tool, for example, GPT-4. The service provider does not need to prepare the information processing device 40 by himself, and can use the information processing device 40 via the network 30.

[0057] The information processing device 40 can transmit the result of the information processing to the information processing device 10 via the network 30.

[0058] Inputting data into the information processing device 40 is called prompt transmission.

[0059] [Send Prompt] The prompt is sent online to an AI text proofreading tool in the information processing device 40 as a prompt based on the original text input by the user to the input / output interface. This stage corresponds to step S13 in Fig. 1. The function of the AI ​​text proofreading tool is performed by a text generation unit in the information processing device 40. In this specification, sending a prompt to the information processing device 40 is synonymous with sending a prompt to the text generation unit.

[0060] The target sentence included in the above prompt may be generated by partially replacing it with synonyms. In this case, the generation unit of the information processing device 10 converts the words and phrases input by the user into synonyms to generate multiple target sentences. The advantage of generating multiple target sentences with synonyms is that it allows freedom in the expression intended by the user and provides a breadth of meaning to the multiple proofread texts output by the AI ​​text proofreading tool.

[0061] For example, the synonyms may be words whose similarity to the embedded representations of the text listing the converted words and phrases entered by the user is equal to or greater than a specified value set by the user, obtained by the information processing device 10 for the text listing the converted words and phrases. This similarity may be a cosine similarity or the like.

[0062] Examples of instructions included in the prompt for making a query to the information processing device 40 include "Transform these words or phrases as appropriate, and add words and phrases as appropriate to make a meaningful sentence," "Add words and phrases to these words or phrases as appropriate to make a meaningful sentence," and "Use these words or phrases in the correct order, and add words and phrases as appropriate to make a meaningful sentence." In addition, "Make it in several different sentences" is added to the instructions included in the prompt. This allows multiple proofread texts (answer sentences) to be obtained.

[0063] The above prompt specifies the degree of freedom with which the AI ​​text proofreading tool of the information processing device 40 constructs a sentence. If the user obtains the proofread texts (answer sentences) shown below, and none of them match the user's intention, increasing the degree of freedom specified by the prompt will broaden the range of meaning of the proofread texts (answer sentences) obtained from the AI ​​text proofreading tool, making it easier to obtain the proofread texts (answer sentences) that the user intended.

[0064] A plurality of answer sentences created by an AI sentence proofreading tool of the information processing device 40 are transmitted to the information processing device 10 via the network 30.

[0065] The information processing device 10 receives a plurality of answer sentences from the information processing device 40.

[0066] [Receive] The information processing device 10 includes a comparison unit that receives multiple proofread texts (answer sentences) from an AI text proofreading tool online. Then, each proofread text is processed using a distributed representation technique.

[0067] Next, the comparison unit calculates the similarity of the multiple proofread texts and selects the proofread text that is closest to the average. Alternatively, clustering is performed using the k-means method with the number of clusters set to 1, and the proofread text that is closest to the cluster center is selected. This is because a sentence that is close to the average value in the distributed representation is the sentence that is most easily associated with among the sentences generated from the user's original text, and is therefore considered to be in line with the user's intention. In other words, when assuming that an AI text proofreading tool is equivalent to a proofreading process performed by a human, the most people proofread the sentence to be equivalent to the average value in the distributed representation, and it corresponds to the sentence decided by majority vote.

[0068] In the case of distributed representations, this average value is assumed to be distance 0. Assume that the reader will misunderstand by a maximum of E. When a reader is given a proofread text with distributed representation 0, the misunderstanding is E, whereas when a proofread text with distributed representation x is given, the misunderstanding is x+E. In other words, by choosing a proofread text that corresponds to distributed representation 0, the difference between the other person's understanding and the user's understanding can be minimized.

[0069] When the comparison unit can obtain the likelihood of the sentence, it is preferable to select a proofread text on the input / output interface that is close to the average value and has a small likelihood. In this case, the E can be made small, which means that the range of misunderstanding can be narrowed.

[0070] Then, the input / output interface selectively obtains the proofread text from the comparison unit and suggests it to the user. This step corresponds to step S14 in FIG.

[0071] It is acceptable to extract multiple proofread texts. Because only one proofread text is extracted and the one suggested proofread text does not necessarily match the user's intention, if multiple suggestions are made, the user can select the proofread text that matches his / her intention.

[0072] It is preferable that these multiple expressions are separated from each other in the distributed representation, because the meanings of each expression are different and the user can easily make a judgment.

[0073] The multiple proofread text suggestions may be repeated if a single proofread text suggestion fails to select a sentence that matches the user's intention.

[0074] In this way, the user can select the intended sentence from the proofread text displayed on the display unit of the input / output interface. This stage corresponds to step S15 in FIG. 1. For example, the proofread text selected by the user can be sent to a recipient (another user or multiple users) via chat software such as InCircle as the input / output interface. In this embodiment, the input / output interface shows the proofread text to the user when it has obtained it. For example, the input / output interface is provided with a display unit, but this is just one example, and the display unit may be replaced with an output device such as an audio output unit.

[0075] Also, an example of a block diagram of an information processing system is shown in Fig. 3. Also, Fig. 3 shows a block diagram corresponding to Fig. 2, and the same reference numerals are used for the same parts. In Fig. 3, there are an input / output interface 101 to which an input sentence from a user is input, an information processing device 10 that receives the input sentence data, and an information processing device 40 that receives a plurality of prompts from the information processing device 10, generates sentence data (m is a natural number of 2 or more) for each prompt in a sentence generation unit 104, and transmits the generated sentence data to the information processing device 10. The input / output interface 101 has a display unit 105.

[0076] The information processing device 10 includes a generating unit 102 that generates a plurality of prompts, and a comparing unit 103 that extracts answer sentence data close to the average using a distributed representation technique.

[0077] The information processing device 10 creates n prompts in the generation unit 102 and transmits the n prompts (n is a natural number equal to or greater than 2) to the information processing device 40. The information processing device 10 also extracts answer sentence data close to the average using a distributed representation technique in the comparison unit 103, and transmits the data to the input / output interface 101.

[0078] The information processing device 40 transmits m pieces of sentence data (m is a natural number of 2 or more, but m may be the same as or different from n) for each prompt to the information processing device 10 in the sentence generating unit 104.

[0079] Although FIG. 3 shows an example in which two pieces of answer sentence data are extracted, the comparison unit 103 may extract three or more pieces of answer sentence data.

[0080] In this embodiment, an example of use as a text proofreading system is shown, but the present invention is not limited to this.

[0081] In this embodiment, an example is shown in which the hardware constituting the sentence generation unit is a large-scale computer, but this is not particularly limited, and the generation unit 102 may also be processed by the same large-scale computer. The comparison unit may also be processed by the same large-scale computer.

[0082] This embodiment mode can be freely combined with other embodiment modes.

[0083] (Embodiment 2) In this embodiment, an example of a network configuration diagram that is partially different from that in Fig. 2 will be described below with reference to Fig. 4. Note that the same parts as in the first embodiment will be described in a simplified manner.

[0084] FIG. 4 is a network configuration diagram in which one information processing device 10 can transmit or receive data to or from a plurality of information processing devices 40a, 40b via a network 30. In FIG.

[0085] The AI ​​text proofreading tools of information processing device 40 shown in embodiment 1 do not have to be the same. For example, as shown in Fig. 4, a program may be used that requests proofreading from a plurality of different AI text proofreading tools, such as an AI text proofreading tool of information processing device 40a managed by server A and an AI text proofreading tool of information processing device 40b managed by server B.

[0086] Using multiple different AI proofreading tools creates more multiple proofread versions of a text, increasing the likelihood that the proofread version matches the user's intent.

[0087] 4, if the process performed by the generating unit 102 or the comparing unit 103 of the information processing device 10 can be executed by the information terminal 20a which is a desktop computer, the information terminal 20 which is a notebook computer, the information terminal 20 which is a smartphone, and the information terminal 20d which is a tablet computer, the information processing device 10 can be omitted. In that case, the network 30 can be replaced with the network 31.

[0088] In such a configuration, even if the information processing device 40a has insufficient knowledge of the instruction sentence or the target sentence included in the prompt, if the information processing device 40b or other devices have sufficient knowledge, the user can obtain a favorable answer. For example, when the user's instruction sentence is "Please tell me the C++ code that displays the current time" or "Please tell me how to speed up the processing of the following C++ code", only the information processing device 40b has learned the C++ language.

[0089] This embodiment mode can be freely combined with other embodiment modes.

[0090] (Embodiment 3) In this embodiment, an example of a network configuration diagram that is partially different from that in Fig. 2 will be described below with reference to Fig. 4. Note that the same parts as in the first embodiment will be described in a simplified manner.

[0091] A user can access the information processing device 10 through the network 30 from, for example, the information terminals 20a to 20d.

[0092] The information processing device 10 is a computer with higher processing power than the information terminal 20a, and is therefore convenient for obtaining distributed representations that require computational resources; however, when this is done on information terminals 20a to 20d, etc., there is no need to access the information processing device 10.

[0093] With this configuration, a user can receive a service using the information processing method according to one aspect of the present invention.

[0094] (Embodiment 4) In this embodiment, an example of a network configuration diagram that is partially different from that in Fig. 4 will be described below with reference to Fig. 5. Note that the same parts as in the second embodiment will be described in a simplified manner.

[0095] A user can access the information processing device 40a and the information processing device 40b through the network 30 from, for example, the information terminals 20a to 20d.

[0096] With this configuration, a user can receive a service using the information processing method according to one aspect of the present invention. [Explanation of symbols]

[0097] 10. Information processing device 20a Information terminal 20b Information terminal 20c Information terminal 20d Information terminal 30 Network 31 Network 40 Information processing device 40a Information processing device 40b Information processing equipment 101 Input / Output Interface 102 Generation part 103 Comparison unit 104 Text generation unit 105 Display unit

Claims

1. an input / output interface for receiving an input sentence from a user and displaying an answer sentence to the user; a generator for generating a prompt based on the input sentence; a sentence generation unit that generates a plurality of the answer sentences; a comparison unit that obtains a distributed representation of a sentence for the answer sentence, calculates an average value of the distributed representation, and selects the answer sentence; having The input / output interface outputs the input sentence to the generation unit; The generation unit transmits the prompt to the sentence generation unit; the comparison unit acquires the answer sentence from the sentence generation unit, and selectively outputs the answer sentence to the input / output interface; Information processing system.

2. an input / output interface for receiving an input sentence from a user and displaying an answer sentence to the user; a generator for generating a prompt based on the input sentence; a comparison unit that obtains a distributed representation of a sentence for the answer sentence, calculates an average value of the distributed representation, and selects the answer sentence; having the generation unit transmits the prompt via a network to a sentence generation unit that generates a plurality of answer sentences; the comparison unit acquires the answer sentence from the sentence generation unit via the network, and selectively outputs the answer sentence to the input / output interface; Information processing system.

3. In claim 2, The generation unit transmits the prompt to a plurality of the sentence generation units via the network.

4. 3. The information processing system according to claim 1 or 2, wherein the sentence generation unit is a plurality of types of deep learning engines having language models.

Images