Information processing device

Abstract

To provide high-precision dialogues while suppressing costs.SOLUTION: A processing device: recognizes contents of a user's utterance; generates a prompt text including a list of multiple first language models specialized for understanding intent in multiple predetermined domains and the contents of the utterance; inputs the prompt text into a second language model; identify the first language model with a domain corresponding to the content of the utterance based on the contents of the response obtained from the second language model; and transfers the contents of the utterance to the identified first language model.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to dialogue techniques. [Background technology]

[0002] There is a technology in which a chatbot generates an answer based on an input sentence. In this regard, for example, Patent Document 1 discloses a system that selects a chatbot appropriate for an input question in an environment where multiple chatbots are accessible. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6555838 Summary of the Invention [Problem to be solved by the invention]

[0004] The present disclosure aims to provide highly accurate dialogue while keeping costs down. [Means for solving the problem]

[0005] One aspect of the present disclosure is to The information processing device has a control unit that executes the following operations: recognizing the content of an utterance made by a user; generating a prompt text including a list of a plurality of first language models each specialized for intent understanding in a plurality of specified domains and the content of the utterance; inputting the prompt text into a second language model, and identifying the first language model having a domain corresponding to the content of the utterance based on the content of a response obtained from the second language model; and transferring the content of the utterance to the identified first language model.

[0006] Further, other aspects include a method executed by the above-mentioned device, a program for causing a computer to execute the method, or a computer-readable storage medium non-transitoryly storing the program. Effect of the Invention

[0007] According to the present disclosure, it is possible to provide highly accurate dialogue while keeping costs down. [Brief description of the drawings]

[0008] [Figure 1] 1 is a schematic diagram of a dialogue system according to a first embodiment. [Diagram 2] FIG. 2 is a hardware configuration diagram of the in-vehicle device 10. [Diagram 3] FIG. 2 is a software configuration diagram of the in-vehicle device 10. [Figure 4] 2 is a diagram showing a flow of processing executed by a control unit 11 of an in-vehicle device 10. [Diagram 5] 4 shows an example of a prompt text generated by the dialogue relay unit 112. [Figure 6] 4 is a flowchart of a process executed by a control unit 11 of the in-vehicle device 10. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] In recent years, with the development of machine learning, the number of products equipped with language models is increasing. For example, by incorporating a large language model (LLM), whose accuracy has been improved by machine learning using a large dataset, it is possible to add a natural language dialogue function to the target product.

[0010] An example of a car where natural language dialogue is useful is when it is equipped with an LLM. For example, by installing an LLM in an in-vehicle device, it becomes possible to obtain information without operating a touch panel or the like. Such a function is particularly useful when you cannot take your hands off the wheel, such as when driving.

[0011] On the other hand, trained LLMs have a large capacity and are difficult to install in embedded devices due to cost issues. Therefore, a method has been devised to access LLMs via a network. By accessing LLMs via a network, it becomes possible to use large-scale language models that cannot be installed in local storage.

[0012] In addition, by using a network, it is possible to selectively access an LLM specialized in a certain field. For example, if multiple LLMs are available, such as an LLM specialized in route guidance, an LLM specialized in local information guidance, and an LLM specialized in casual conversation, it becomes possible to select an LLM that matches the user's intention. In this regard, a technology is known that determines the category of a user's utterance based on the content of the user's utterance and automatically selects an LLM that matches the category.

[0013] However, when multiple LLMs are selectively used, another problem arises in that it is difficult to select the LLM that best suits the user's utterance. For example, consider the case where a rule is set such as "if keywords such as 'route' and 'to (name of destination)' are detected from the user's utterance, select the route guidance LLM." However, such a rule-based method cannot handle ambiguous utterances that do not follow the rules, such as "Will I be able to get to XX by XX?" There is also a method to have the user select the utterance category, but this method impairs the user experience. The information processing device according to the present disclosure solves such problems.

[0014] An information processing device according to a first aspect of the present disclosure has a control unit that executes the following operations: recognizing the content of an utterance made by a user; generating a prompt text including a list of a plurality of first language models each specialized for intent understanding in a plurality of specified domains and the content of the utterance; inputting the prompt text into a second language model; identifying the first language model having a domain corresponding to the content of the utterance based on the content of a response obtained from the second language model; and transferring the content of the utterance to the identified first language model.

[0015] The multiple first language models are language models specialized for understanding intent in multiple domains. The first language model may be a large language model. The large language model is, for example, a language model trained to be capable of executing a dialogue task in natural language. The multiple domains are, for example, source domains such as "route guidance," "vehicle guidance," "tourist information," and "chat."

[0016] The control unit generates a prompt text including a list of first language models and the content of the utterance made by the user (the spoken sentence). The prompt text may, for example, be a natural language inquiry as to which of the multiple first language models is most suitable for inputting the content of the utterance made by the user. The list of first language models may, for example, include an identifier or name of the first language model and a feature of each language model (for example, a sentence describing the domain in which each language model specializes).

[0017] The generated prompt text is input to a second language model. The second language model is a language model capable of performing natural language dialogue tasks similar to the first language model, but is not domain-specifically trained and is generated at a relatively low cost. The second language model is generated based on the list of first language models and the spoken sentence. The control unit then connects to the first language model that is estimated to be most relevant to the content of the utterance, and transfers the content of the utterance.

[0018] According to this configuration, it is possible to accurately specify the language model to which the utterance should be transferred at a low cost. Also, by executing such processing in the background, it is possible to carry out a dialogue smoothly.

[0019] Specific embodiments of the present disclosure will be described below with reference to the drawings. Unless otherwise specified, the hardware configuration, module configuration, functional configuration, and the like described in each embodiment are not intended to limit the technical scope of the disclosure to only those.

[0020] First Embodiment [System Overview] An overview of a dialogue system according to a first embodiment will be described. The dialogue system according to this embodiment includes an in-vehicle device 10 mounted in a vehicle. The in-vehicle device 10 can access other external devices that provide dialogue services using a predetermined language model via a network (e.g., a mobile communication network).

[0021] The in-vehicle device 10 is mounted on a connected vehicle capable of communicating with any device via wireless communication. The in-vehicle device 10 may be a data communication module (DCM) for connecting a component (e.g., an ECU, an in-vehicle terminal, etc.) of the vehicle to a network. In this embodiment, the in-vehicle device 10 can access the Internet via a predetermined mobile communication network and connect to an external device that provides an interactive service.

[0022] A vehicle (or a component mounted on the vehicle) can provide various services by communicating with external devices via the in-vehicle device 10. Examples of the various services include a navigation service, a remote control service (e.g., remote air conditioning, etc.), an in-vehicle Wi-Fi (registered trademark) service, an emergency call service, and the like.

[0023] Furthermore, the in-vehicle device 10 has a voice input / output function and can communicate with vehicle occupants in natural language.

[0024] FIG. 1(A) is a schematic diagram illustrating an in-vehicle device 10 according to a conventional embodiment. The in-vehicle device 10 is configured to include a control unit for inputting and outputting voice and performing voice recognition, and a local language model. The local language model is a language model that has been trained to enable dialogue in natural language. The local language model may provide predetermined functions, such as route search and vehicle operation (air conditioner, etc.), through dialogue. In the illustrated embodiment, the local language model is stored in a storage device included in the in-vehicle device 10.

[0025] On the other hand, in such a form, it is difficult to provide a large-scale language model. For example, there are a wide variety of conversation categories, such as local information, tourist guides, vehicle information (operation manuals), and casual conversations. In order to cover all of these categories, it is necessary to prepare a language model trained using large-scale training data. Such language models can be several gigabytes to hundreds of gigabytes in size, and storing them in local storage is not realistic from the perspective of cost.

[0026] For this reason, a system has been proposed that accesses a language model specialized for dialogue in a specific domain via a network. FIG. 1B shows how the in-vehicle device 10 can access not only the local language model but also the language models on the Internet. 1 is a schematic diagram of a form in which a plurality of language models (hereinafter referred to as remote language models) provided via the Internet can be accessed. For example, a form in which a plurality of remote language models that have been trained to be specialized for a particular domain are stored in an external device connected to the Internet and accessed from the in-vehicle device 10 upon request is conceivable.

[0027] However, this configuration has a problem in that it is difficult to accurately determine the language model to be connected to. For example, the control unit of the in-vehicle device 10 can determine the language model to be connected to (i.e., the language model corresponding to the category of the utterance) according to the content of the utterance made by the vehicle occupant. However, since the utterance made by the vehicle occupant is a natural language, an appropriate language model is not necessarily selected.

[0028] There is a rule-based method for determining the category of an utterance based on the content of the utterance. For example, when words such as "to" and "route" are detected from an utterance, it is determined that the utterance corresponds to the category "route search" and a remote language model specialized for route search is connected. However, with such a method, when an ambiguous utterance such as "I wonder if the road is crowded when I go home now?" is made, it may be determined that a casual conversation has started.

[0029] Therefore, in the in-vehicle device 10 according to the present embodiment, an independent language model is used to determine which language model to connect to among a plurality of large-scale language models. Fig. 1(C) is a schematic diagram of the in-vehicle device 10 according to the present embodiment.

[0030] The gateway language model is a language model into which an utterance made to the in-vehicle device 10 is first input. When the control unit of the in-vehicle device 10 recognizes an utterance, it first inputs the content of the utterance into the gateway language model. The gateway language model is not specialized in a specific field, but is a general-purpose language model that has been trained to the extent that a certain utterance category can be selected from a presented list.

[0031] The control unit inputs the list of language models to the gateway language model together with the content of the acquired utterance, and causes the gateway language model to determine "which language model is most suitable for the utterance." The control unit also transfers the utterance to the language model answered by the gateway language model. This makes it possible to select an appropriate language model from multiple language models and transfer the utterance, thereby achieving both cost reduction and accurate dialogue.

[0032] In the description of this embodiment, the terms "access to a remote language model" and "connection to a remote language model" refer to access to or connection to an external device that provides a dialogue service using the remote language model. In this embodiment, a plurality of dialogue services having remote language models are running in a plurality of external devices, and the in-vehicle device 10 can use a target remote language model by connecting to any dialogue service.

[0033] [Hardware configuration] Next, the hardware configuration of each device constituting the system will be described. Fig. 2 is a diagram showing an example of the hardware configuration of the in-vehicle device 10 according to this embodiment.

[0034] The in-vehicle device 10 can be configured as a computer having a processor (CPU, GPU, etc.), a main memory device (RAM, ROM, etc.), and an auxiliary memory device (EPROM, hard disk drive, removable media, etc.). The operating system (OS), various programs, various tables, etc. are stored, and by executing the programs stored therein, various functions (software modules) that meet specific purposes, as described below, can be realized. However, some or all of the functions may be realized as hardware modules using hardware circuits such as ASICs, FPGAs, etc.

[0035] The in-vehicle device 10 is configured to include, as hardware, a control unit 11, a storage unit 12, a wireless communication module 13, and an input / output unit 14.

[0036] The control unit 11 is a calculation unit that executes a predetermined program to realize various functions of the in-vehicle device 10. The control unit 11 can be realized by a hardware processor such as a CPU. The control unit 11 may also be configured to include a RAM, a ROM (Read Only Memory), a cache memory, and the like.

[0037] The storage unit 12 is a means for storing information, and is configured with storage media such as RAM, a magnetic disk, a flash memory, etc. The storage unit 12 stores programs executed by the control unit 11, data used by the programs, etc.

[0038] The wireless communication module 13 is a communication device that performs wireless communication with a predetermined network. In this embodiment, the wireless communication module 13 is configured to be able to communicate with a predetermined mobile communication network. The wireless communication module 13 can be configured to include an eUICC (e.g., a SIM card). The SIM card is configured as a microcomputer equipped with a CPU and a storage device. The SIM card stores information for connecting to a mobile communication network and receiving authentication.

[0039] The input / output unit 14 is a unit that receives input from a user of the device and presents information to the user. The input / output unit 14 is typically configured to include devices for inputting and outputting voice, such as a microphone and a speaker. The input / output unit 14 may also include a device (such as a display) that provides visual information.

[0040] [Software configuration] Next, the software configuration of each device constituting the system will be described below. Fig. 3 is a diagram showing a schematic software configuration of the in-vehicle device 10 according to this embodiment.

[0041] In this embodiment, the control unit 11 of the in-vehicle device 10 is configured to have two software modules: a dialogue receiving unit 111 and a dialogue relay unit 112. Each software module may be realized by the control unit 11 (CPU, etc.) executing a program stored in the storage unit 12. Note that the information processing executed by the software module is synonymous with the information processing executed by the control unit 11 (CPU, etc.).

[0042] The dialogue reception unit 111 acquires utterances made by a vehicle occupant (hereinafter also referred to as a user) via the input / output unit 14, which will be described later. The dialogue reception unit 111 executes a predetermined process on the acquired voice data and executes voice recognition. As a result, the contents of the utterance are converted into text. Furthermore, the dialogue receiving unit 111 transmits text obtained as a result of the speech recognition to the dialogue relay unit 112. The text is transferred to a target language model by the dialogue relay unit 112, which will be described later.

[0043] The dialogue reception unit 111 also outputs a response (hereinafter, a response sentence) from the language model transmitted from the dialogue relay unit 112. The dialogue reception unit 111 outputs The answer sentence is converted into voice and output via the input / output unit 14.

[0044] The dialogue relay unit 112 determines to which of the multiple available language models (local language model and remote language model) the text acquired from the dialogue receiving unit 111 should be transferred. The dialogue relay unit 112 makes an inquiry to the above-mentioned gateway language model, and determines the language model to which the utterance should be transferred based on the response. The dialogue relay unit 112 then transfers the utterance to the determined language model and acquires a response to the utterance. The acquired response is then sent to the dialogue acceptance unit 111.

[0045] The storage unit 12 of the in-vehicle device 10 stores a gateway language model 12A, a local language model 12B, and language model information 12C. The gateway language model 12A is a language model trained to be able to execute a dialogue task in natural language. The gateway language model 12A is a language model generated at a relatively low cost, and has not been trained to perform dialogue specialized for a specific domain. The gateway language model 12A may be any language model for performing general-purpose conversation. For example, a language model that is publicly available as open source may be used as the gateway language model 12A.

[0046] The local language model 12B is a language model trained to be able to execute a dialogue task in natural language, and is a language model for providing information on the functions or driving of a vehicle. The local language model 12B may be capable of, for example, explaining the functions of the vehicle in which the in-vehicle device 10 is mounted, or providing route guidance. The local language model 12B is a lighter language model than a remote language model that can be accessed via a network. By using the local language model 12B, it is possible to carry out a dialogue with good response. For example, it is possible to use the remote language model when a specialized conversation is required, and the local language model 12B for a conversation that requires a high-speed response.

[0047] The language model information 12C is a collection of information on a plurality of language models available to the in-vehicle device 10. In this embodiment, a local language model and a plurality of remote language models are exemplified as the plurality of available language models. The language model information 12C may include, for example, a language model identifier, a name of the language model, features of the language model, information on an access destination (for example, a network address of an external device that provides a remote language model, etc.) The name and features of the language model may be written in a natural language.

[0048] [Dialogue flow] Next, there will be described an outline of the processing executed by the control unit 11. Fig. 4 is a diagram showing the flow of processing executed by the control unit 11 of the in-vehicle device 10 that has received an utterance from the user.

[0049] The dialogue acceptance unit 111 acquires an utterance from a user via the input / output unit 14. For example, the input / output unit 14 converts the utterance acquired via a microphone or the like into voice data, which is acquired by the dialogue acceptance unit 111. The dialogue acceptance unit 111 executes a predetermined voice recognition process on the acquired voice data, and converts the voice data into text. The dialogue acceptance unit 111 transmits the text obtained as a result of the voice recognition to the dialogue relay unit 112, while retaining the response to the utterance. The text is hereinafter referred to as a "spoken sentence."

[0050] The dialogue relay unit 112, which has received the spoken sentence, makes an inquiry to the gateway language model 12A to determine to which language model the spoken sentence should be transferred. First, the dialogue relay unit 112 executes a query to the gateway language model 12A. The prompt text includes a sentence for inquiring of the gateway language model 12A about an appropriate language model as a destination to which the spoken sentence should be transferred.

[0051] The prompt text includes a sentence to be spoken and a list of available language models. Figure 5(A) shows an example of the prompt text. Here, it is assumed that the following language models are available: (1) Vehicle LLM A language model that can grasp the situation inside the car and configure in-car devices. It has learned the car owner's manual and can also provide information about the vehicle's functions. (2) LLM Chat A language model specialized for non-task-oriented chatter. (3) Regional Information LLM A language model that can provide local area information, including tourist information. (4) Local LLM A language model that deals with topics that do not fall into (1) to (3) above (local language model 12B).

[0052] The prompt text includes the names of these language models and the characteristics of each language model described in natural language. The names and characteristics of each language model may be obtained from the language model information 12C. Furthermore, the prompt text includes a sentence inquiring which language model is appropriate as an input destination for the spoken sentence. For example, the prompt text includes a request to output the name of an appropriate language model, as shown in FIG. 5(A).

[0053] The dialogue relay unit 112 inputs the generated prompt text to the gateway language model 12A and obtains a response. In the example of Fig. 5(A), the gateway language model 12A interprets the utterance as a category of "question for vehicle" and outputs a response of "vehicle LLM". Note that, although the illustrated example has the name of the language model as the response, the target language model may be specified by an identifier or the like. In this example, the name and characteristics of each language model are included in the prompt text, but this information may be held in advance in the gateway language model 12A.

[0054] Secondly, the dialogue relay unit 112 transfers the utterance sentence to the language model answered by the gateway language model 12A. If the target language model is the local language model 12B, the dialogue relay unit 112 transfers the utterance sentence within its own device. If the target language model is a remote language model, that is, a language model accessible via a network, the dialogue relay unit 112 transfers the utterance sentence via the network to an external device that provides the target language model.

[0055] The language model to which the utterance sentence is input outputs a response in response to the utterance sentence. Figure 5(B) is an example of a response to the utterance sentence. In this example, the vehicle LLM that received a question for the vehicle outputs a response.

[0056] When the dialogue relay unit 112 receives a response from the language model to which the spoken sentence has been transferred, the dialogue relay unit 112 transmits the text included in the response (hereinafter, an answer sentence) to the dialogue receiving unit 111. When the dialogue receiving unit 111 receives the answer sentence, it generates voice data based on the answer sentence and outputs it via the input / output unit 14. The dialogue receiving unit 111 may generate only voice data for reading out the answer sentence, or may generate information accompanying the voice data. An example of the information accompanying the voice data is a user interface screen. The user interface screen may include answers written in text. The dialogue system according to this embodiment seamlessly performs the series of processes shown in Fig. 4. This provides an effect as if the user who made the utterance was having a dialogue with the in-vehicle device 10.

[0057] [flowchart] Next, the details of the process executed by the in-vehicle device 10 will be described. Fig. 6 is a flowchart of the process executed by the in-vehicle device 10. The process shown in Fig. 6 is started when a user (a vehicle occupant) makes an utterance. The start of an utterance may be detected by, for example, a predetermined keyword.

[0058] First, in step S11, the dialogue receiving unit 111 recognizes the content of the utterance. The dialogue receiving unit 111 acquires voice data output from the input / output unit 14 and converts the utterance into text by executing a voice recognition process. The converted text (utterance) is transmitted to the dialogue relay unit 112.

[0059] In step S12, the dialogue relay unit 112 generates a prompt text. As described above, the prompt text includes the spoken sentence and a list of available language models. The list of language models includes the name of the language model and a description in natural language of the features of each language model. The name of the language model and the features of each language model can be acquired from the language model information 12C. In addition, the prompt text includes a sentence inquiring which language model is appropriate as a language model for inputting the spoken sentence, as exemplified in FIG. 5(A). The generated prompt text is input by the dialogue relay unit 112 to the gateway language model 12A.

[0060] Next, in step S13, the dialogue relay unit 112 acquires a response from the gateway language model 12A. As shown in Fig. 5(B), the gateway language model 12A indicates the language model to which the utterance sentence should be transferred. The dialogue relay unit 112 determines to which language model the utterance sentence should be transferred based on the response.

[0061] In step S14, the dialogue relay unit 112 transfers the spoken sentence to the language model determined in step S13. If the target language model is a local language model, the dialogue relay unit 112 transfers the spoken sentence within the device. If the target language model is a remote language model, the dialogue relay unit 112 accesses an external device that provides the target language model via the wireless communication module 13, and transfers the spoken sentence to the language model.

[0062] In step S15, the dialogue relay unit 112 acquires a response from the target language model, and transmits an answer sentence included in the response to the dialogue receiving unit 111. The dialogue receiving unit 111 generates voice data based on the answer sentence, and outputs it via the input / output unit 14. The dialogue receiving unit 111 may convert the answer sentence into voice data by voice synthesis technology. The voice data is output via the input / output unit 14 (speaker, etc.) and provided to the user.

[0063] In step S16, the dialogue receiving unit 111 determines whether the dialogue between the user and the language model has ended. For example, the dialogue receiving unit 111 may determine that the dialogue has ended when no utterance has been made for a predetermined time or longer since the answer was output by voice in step S15, or when an utterance for terminating the dialogue has been made (for example, when a predetermined keyword has been detected). If it is determined in this step that the dialogue has ended, the process ends. If it is determined that the dialogue has not ended, the process returns to step S11.

[0064] In addition, when a second or subsequent utterance is made, the dialogue relay unit 112 may change the contents of the process shown in FIG. 6 depending on whether the utterance is a continuation of the previous utterance. For example, in the course of a dialogue, multiple utterances may be made. In this case, if the successive utterances belong to the same category (domain), the dialogue relay unit 112 does not need to make an inquiry to the gateway language model 12A every time. Therefore, the dialogue relay unit 112 may omit the inquiry to the gateway language model 12A (the process of steps S12 to S13) as long as the domain of the dialogue has not changed. In a series of dialogues, whether the domain of the dialogue has changed may be detected by a known method. For example, it may be determined that the domain of the dialogue has changed when a predetermined keyword is detected from the utterance or when a user performs a predetermined operation.

[0065] As described above, the in-vehicle device 10 according to the present embodiment uses a gateway language model to determine to which language model an utterance made by a vehicle occupant should be transferred. This makes it possible to use multiple language models that can handle highly specialized topics without comprehensively defining matching between the content of the utterance and the language model. In particular, by using a language model that has been trained at a relatively low cost as the gateway language model, the cost of the in-vehicle device 10 can be reduced.

[0066] (Modification) The above-described embodiment is merely an example, and the present disclosure can be modified and implemented as appropriate without departing from the spirit and scope of the present disclosure. For example, the processes and means described in this disclosure can be freely combined and implemented as long as no technical contradiction occurs.

[0067] In the embodiment, the remote language model is arranged in an external device connected to the Internet, but the remote language model may be arranged in another device. For example, the remote language model may be arranged in an edge server accessible from a vehicle. The remote language model may be distributed across multiple edge servers or clouds. In addition, although an in-vehicle device has been exemplified in the embodiment, the information processing device according to the present disclosure can also be embodied as a device or equipment that is not mounted on a vehicle.

[0068] In addition, the remote language model may be arranged in a plurality of edge servers. In this case, the in-vehicle device 10 may acquire the location information of the vehicle and identify the edge server in which the target language model is arranged based on the location information. For example, the remote language model providing the regional information may be arranged in a plurality of edge servers arranged in each region. In this case, for example, it is assumed that the gateway language model 12A responds with "regional information LLM". In this case, the in-vehicle device 10 may identify the regional information LLM arranged in the edge server that is geographically closest based on the location information of the vehicle. The regional information handled by the remote language model may be different for each edge server. For example, the edge server arranged in the A area may be arranged with a language model providing the regional information of the A area, and the edge server arranged in the B area may be arranged with a language model providing the regional information of the B area. The in-vehicle device 10 identifies the edge server (or the regional information LLM provided by the edge server) corresponding to the area in which the vehicle is located, and accesses the regional information LLM provided by the edge server to obtain the regional information of the area.

[0069] In addition, the same language model may be distributed and arranged in a plurality of devices, such as a local machine, an edge server, and a cloud server. When a language model is available in multiple devices, it may be determined which device to access based on certain requirements. For example, in an environment where a fast response is required, it may be preferable to connect to an edge server rather than a cloud server. Also, when the local machine has insufficient resources, it may be preferable to use a language model provided in an external device.

[0070] Furthermore, a process described as being performed by one device may be shared and executed by multiple devices. Alternatively, a process described as being performed by different devices may be executed by one device. In a computer system, the hardware configuration (server configuration) by which each function is realized can be flexibly changed.

[0071] The present disclosure can also be realized by supplying a computer program implementing the functions described in the above embodiments to a computer, and having one or more processors of the computer read and execute the program. Such a computer program may be provided to the computer by a non-transitory computer-readable storage medium connectable to the system bus of the computer, or may be provided to the computer via a network. Non-transitory computer-readable storage media include, for example, any type of disk, such as a magnetic disk (floppy disk, hard disk drive (HDD), etc.), an optical disk (CD-ROM, DVD disk, Blu-ray disk, etc.), a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic card, a flash memory, an optical card, and any type of medium suitable for storing electronic instructions. [Explanation of symbols]

[0072] 10...In-vehicle equipment 11. Control section 12...Storage section 13 Wireless communication module 14...Input / output section

Claims

1. Recognizing the content of an utterance made by a user; generating a prompt text including a list of a plurality of first language models each specialized for intent understanding in a plurality of predetermined domains and the content of the utterance; inputting the prompt text into a second language model, and identifying the first language model having a domain corresponding to the content of the utterance based on the content of a response obtained from the second language model; transferring the content of the utterance to the identified first language model; An information processing device having a control unit that executes the above.

2. the list included in the prompt text includes a sentence describing in natural language a domain in which each of the plurality of first language models specializes; The information processing device according to claim 1 .

3. the prompt text includes a sentence inquiring in a natural language about which of the first language models has a domain corresponding to the content of the utterance; The information processing device according to claim 2 .

4. the first language model and the second language model are language models trained to perform a natural language dialogue task; The information processing device according to claim 1 .

5. the control unit outputs a response obtained from the first language model to which the content of the utterance has been transferred as a response to the utterance. The information processing device according to claim 1 .

Images