Method for controlling the rendering of text generated from textual data originating from at least one embedded system of a vehicle.

Abstract

The present invention relates to a method and device for controlling the rendering of text generated from textual data, said method comprising the following steps: - receiving (31) textual data from said at least one embedded system; - generating (32) the output text from a trained language model fed as input the received textual data; - obtaining (33) a fabrication score from the output of a trained text classification model fed as input internal parameters of the trained language model, said internal parameters having been obtained during text generation; - rendering (34) the generated text from at least one vehicle embedded interface according to the obtained fabrication score. Figure for the abstract: Figure 3

Description

Title of the invention: Method for controlling the rendering of a text generated from textual data from at least one embedded system of a vehicle. technical field

[0001] The present invention relates to generative artificial intelligence, that is to say, to artificial intelligence systems capable of generating text, images or other media in response to prompts.

[0002] In particular, the present invention relates to generative artificial intelligence used to generate text in various fields such as vehicle driver assistance systems. Technological background

[0003] Modern vehicles incorporate an increasing number of on-board systems that contribute to vehicle control by assisting the driver and / or to the comfort of the driver and vehicle passengers. Some of these systems are known as ADAS (Advanced Driver-Assistance System).

[0004] Some AD AS systems can use artificial intelligence to implement generators and other natural language processing according to particular road situations.

[0005] The emergence of so-called Large Language Models (LLMs) has recently revolutionized the field of natural language processing and generative artificial intelligence because these models are capable of generating texts close to those that a human can generate, answering questions, translating texts into different languages, and even writing poetry. They have been used in various applications ranging from chatbots to writing assistants (Metz, C. (2021). AI language model GPT-3 reproduces human-like biases. MIT Technology Review).

[0006] However, one problem with LLM models is their tendency to hallucinate (fabricate), that is, these LLM models respond with a text (a sequence of words) that is clearly false but is presented as a certain fact. This problem can prove important, particularly in applications where the accuracy and reliability of the generated texts are essential (McCoy, T., Pavlick, E., & Linzen, T. (2021). Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference. arXiv preprint arXiv:2104.08696).

[0007] The fabrication of LLM models is linked to the more general problem of networks of neurons known as out-of-distribution (OOD) generalization. Indeed, like any trained neural network, LLM models perform poorly during the interference phase when the textual data presented to them deviates from the (statistical) probability distributions of the textual data used for their training.

[0008] When an input text data belongs to a distribution of text data used for training a neural network or when this input text data can be interpolated so as to be deducible from data in a distribution of text data used for training this neural network, the neural network of an LLM model provides an output text during the interference phase which is expected but when the text data presented as input is OOD, the neural network of the LLM model extrapolates a text or part of the output text with low accuracy.One of the problems to be solved for the implementation of neural networks used in LLM models is to detect those situations where an input is OOD and to provide an output text that is not extrapolated even partially, or if it is, with sufficient accuracy so that the text is not hallucinatory compared to the input text data. Summary of the present invention

[0009] One object of the present invention is to solve at least one of the problems of the technological background described above.

[0010] Another object of the present invention is to improve the performance of a deep neural network used for the implementation of an LLM model.

[0011] Another object of the present invention is to improve the detection of ODD inputs of a neural network used for the implementation of an LLM model.

[0012] According to a first aspect, the present invention relates to a method for controlling the rendering of text generated from textual data originating from at least one embedded system of a vehicle, said method comprising the following steps: - receiving textual data from said at least one embedded system; - generation of the output text of a trained language model fed with input from the received textual data, the language model having been trained during a first learning phase; - obtaining a fabulation score at the output of a trained text classification model fed as input by internal parameters of the trained language model, said internal parameters having been obtained during the generation of the text, the text classification model having been trained during a second phase of learning; - rendering of the text generated from at least one on-board interface of the vehicle according to the fabulation score obtained.

[0013] The method is advantageous because it allows the classification of text generated by a language model based on a fabrication score calculated from internal parameters used by that language model during text generation. Since the fabrication score is calculated using values ​​of internal parameters of the language model, the text classification is more accurate than a classification based on the detection of false statements from the generated text. The method thus makes it possible to generate texts that can be used in applications implemented by a vehicle such as an ADAS system.

[0014] The method significantly improves the reliability of texts generated by language models. By rendering a generated text based on a fabrication score, a user can quickly identify parts or all of a generated text that are potentially unreliable.

[0015] Furthermore, the text classification according to the present invention does not require several responses generated to the same prompt in order to determine a probability of fabrication as found in the prior art.

[0016] The method reduces the costs associated with incorrect or misleading textual information generated by language models. For example, the method can reduce the need for human review and verification of texts generated by language models, or reduce the risk of making decisions based on incorrect textual information.

[0017] According to a particular and non-limiting embodiment, the language model is a large language model comprising a deep neural network whose input is fed with tokens obtained by converting textual data present as input to the language model and adapted to provide as output tokens that correspond to the text generated once these tokens have been converted into textual data.

[0018] According to a particular and non-limiting example of an embodiment, the language model is an autoregressive model based on a transformer-type architecture.

[0019] According to a particular and non-limiting embodiment, the text classification model is a deep neural network.

[0020] According to one variant, the method further comprises a step of comparing the fabulation score generated by the text classification model to a threshold value, and in which the generated text is rendered according to a result of the comparison of the fabulation score generated by the classification model and the threshold value.

[0021] According to a particular and non-limiting embodiment, the threshold value is determined according to a compromise between false negatives and false positives, a false negative corresponding to a situation in which a textual input data of the language model is outside a distribution of the textual data used to train the deep neural network of the language model, while the fabulation score indicates the opposite, and a false positive corresponds to a situation in which a textual input data of the language model is not outside a distribution of textual data used to train the deep neural network of the language model, while the fabulation score indicates the opposite.

[0022] According to a second aspect, the present invention relates to a rendering control device for text generated from textual data from at least one embedded system of a vehicle, the device comprising a memory associated with a processor configured for the implementation of the steps of the process according to the first aspect of the present invention.

[0023] According to a third aspect, the present invention relates to a vehicle, for example of the automobile type, comprising a device as described above according to the second aspect of the present invention.

[0024] According to a fourth aspect, the present invention relates to a computer program which includes instructions adapted for carrying out the steps of the process according to the first aspect of the present invention, in particular when the computer program is executed by at least one processor.

[0025] Such a computer program may use any programming language, and be in the form of source code, object code, or an intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0026] According to a fifth aspect, the present invention relates to a computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the process according to the first aspect of the present invention.

[0027] On the one hand, the recording medium can be any entity or device capable of storing the program. For example, the medium can include a storage means, such as a ROM, a CD-ROM or a microelectronic circuit-type ROM, or a magnetic recording means or a hard disk drive.

[0028] On the other hand, this recording medium can also be a transmissible medium such as an electrical or optical signal, such a signal being able to be transmitted via an electrical or optical cable, by conventional or radio frequency, by self-directing laser beam, or by other means. The computer program according to the present invention can, in particular, be downloaded from an Internet-type network.

[0029] Alternatively, the recording medium may be an integrated circuit in which the computer program is incorporated, the integrated circuit being adapted to execute or to be used in the execution of the process in question. Brief description of the figures

[0030] Other features and advantages of the present invention will become apparent from the description of the particular and non-limiting embodiments of the present invention below, with reference to the attached Figures 1 to 3, in which:

[0031] [Fig-1] schematically illustrates a deep neural network based on an ar transformer type architecture according to a particular and non-limiting embodiment of the present invention.

[0032] [Fig.2] schematically illustrates a device configured to control the rendering of text generated from textual data from at least one embedded system of a vehicle, according to a particular and non-limiting embodiment of the present invention;

[0033] [Fig.3] illustrates a flowchart of the different stages of a process for controlling the rendering of a text generated from textual data from at least one embedded system of a vehicle, according to a particular and non-limiting embodiment of the present invention. Description of examples of achievements

[0034] A method and a rendering control device for text generated from textual data from at least one vehicle's embedded system will now be described in the following with joint reference to Figures 1 to 3. The same elements are identified with the same reference signs throughout the following description.

[0035] A vehicle, according to the present invention, corresponds, for example, to a vehicle with an internal combustion engine, with electric motor(s), or even a hybrid vehicle with an internal combustion engine and one or more electric motors. The vehicle thus corresponds, for example, to a land vehicle, such as a car, a truck, a bus, or a motorcycle.

[0036] The vehicle includes a device 2 of the [Fig.2].

[0037] The vehicle includes at least one embedded system which is configured to provide textual data, i.e. either a word or a sequence of words from natural language.

[0038] According to a particular and non-limiting example of embodiment, the textual data are representative of a state of at least one component of the vehicle and / or of a road situation in which the vehicle is located.

[0039] According to a particular, non-limiting embodiment, one of these embedded systems comprises a touchscreen and memory configured for input and recording of natural language words. Another example might be a voice recording system comprising a microphone and memory.

[0040] According to a particular, non-limiting embodiment, an on-board vehicle system is an ADAS system. For example, the system is a road sign detection system that provides word sequences as soon as it detects a road sign. For example, a word sequence 'speed limit 80 km / h' is provided by this ADAS system as soon as an 80 km / h speed limit sign is detected. Another ADAS system can provide word sequences as soon as weather conditions change.

[0041] These examples of AD AS systems are given only by way of example and the present invention extends to any type of AD AS system which would provide sequences of words according to a state of at least one organ of the vehicle and / or a road situation in which the vehicle is located.

[0042] The vehicle includes at least one human-machine interface configured to render text.

[0043] According to a particular and non-limiting embodiment, an embedded vehicle system is a display system comprising means for interfacing with a vehicle user, such as a touchscreen, a microphone and / or a speaker, or a voice control system comprising a microphone and / or a speaker. The embedded system may include memory configured to store textual data and text.

[0044] A rendering control process for text generated from textual data from at least one vehicle embedded system will now be described according to the present invention.

[0045] In a first operation, text data is received from at least one on-board system of the vehicle. The received text data can be a word or a sequence of words.

[0046] In a second operation, the text is generated as output from a language model fed as input by the received textual data, the language model having been trained during a first learning phase.

[0047] In a third operation, a fabulation score is obtained as output from a trained text classification model and fed as input by internal parameters of the trained language model, said internal parameters having been obtained during the generation of the text, the text classification model having been trained during a second learning phase.

[0048] In a fourth operation, the generated text is rendered from at least one vehicle-mounted interface according to the fabulation score.

[0049] According to a particular and non-limiting embodiment, the language model is an LLM model, that is to say, a language model which includes a deep neural network comprising a significant number of internal parameters. a spike of the order of a billion or more and a converter of textual data into tokens and vice versa.

[0050] LLM models are mathematical functions whose input and output are lists of numbers (tokens). To enable them to accept textual data as input and provide text as output, a conversion is applied to this textual data to obtain tokens (numbers), and a conversion is applied to the tokens to obtain textual data from which the output texts of the LLM model are generated. This conversion is implemented by a bijective function (called a tokenizer) based on a predefined mapping (vocabulary). This bijective function is generally first fitted to the complete training textual dataset and then frozen before the deep neural network is trained.Another characteristic of the bijective function is to compress the textual data presented as input to the LLM model to save computation time during the training and inference of the deep neural network.

[0051] According to a particular, non-limiting embodiment illustrated in [Fig. 1], the LLM model is an autoregressive LLM model based on a transformer-type architecture such as described, for example, in the article by Vaswani et al. (Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A., Kaiser, U., & Polosukhin, I. (2017). Attention is All You Need. In Advances in Neural Information Processing Systems. Curran Associates, Inc.). The transformer architecture is a neural network model that relies on self-attention mechanisms to capture the relationships between tokens in a sequence, enabling parallel processing and efficient context learning.

[0052] Basically, an autoregressive LLM based on a transformer-type architecture comprises an encoder (or 'tokenizer') 10, a decoder (also called a 'tokenizer') 12, and a deep neural network 11. The encoder 10 encodes textual data x, such as words or sentences in natural language, which are present at its input to the LLM model, into a sequence of tokens y, i.e., numbers ('token'), which are present at the input to the deep neural network, based on a predefined mapping (also called a vocabulary) between words and / or sentences in natural language and tokens. The decoder performs the inverse operation by decoding a sequence of tokens z (output of the deep neural network) to obtain textual data x, from which a text is generated.The deep neural network operates on a sequence of tokens y as input (output of the encoder) and generates a new sequence of tokens z as output (input of the decoder).

[0053] The deep neural network comprises a set of layers of ar- neurons tificels.

[0054] For example, each artificial neuron is a perceptron, that is to say a linear classifier comprising generally several inputs and a single output and characterized by an activation function, weights (or synaptic coefficients) and a bias (or threshold).

[0055] For example, a perceptron with n inputs .... x„) and a single output o is defined by n weights ( W], ..., Wn ) and a bias (or threshold) 0: 100561 fl , He o = f{z) = i=ï 1 1 0 otherwise

[0057] The output o then results from the application of the Heaviside function to the postsynaptic potential z given by:

[0058]

[0059] with a non-linear activation function H(x) given for example by:

[0060] w .} ut \ f0 .¥ < 0 VxeR. HLx) = , v 7 11 if x > 1

[0061] The internal parameters of the neural network are then these weights and biases for the set of perceptrons.

[0062] The present invention is not limited to this definition of perceptron or to the use of other basic elements forming a layer of the neural network. It is also not limited to the number of perceptrons (or other basic elements) used per layer or to the number of layers. The internal parameters of the neural network are usually weights and biases, regardless of the basic elements of the neural network layers.

[0063] During the training phase of the deep neural network, the deep neural network provides a set of output tokens when a set of tokens is presented as input. The input and output token sets correspond to words or sequences of words in natural language. The deep neural network is trained when it provides output token sets that correspond to expected words or sequences of words.

[0064] More specifically, during the first training phase, the internal parameters of the language model's deep neural network are optimized using the estimation of a maximum likelihood loss function. The language model calculates the probability distribution over a given vocabulary for each token in a training sequence present as input to the deep neural network, and the objective is to maximize the probability that the next 'true' token in the training sequence is predicted. This is achieved by minimizing a cross-entropy loss function defined between the predicted probabilities and the next real token in the training sequence.

[0065] According to a particular and non-limiting embodiment, during the first learning phase, the internal parameters of the deep neural network can be optimized by backpropagation and gradient descent based, for example, on the method described by Robbins and Monro (Robbins, H. and S. Monro (1951). “A Stochastic Approximation Method.” In The Annals of Mathematical Statistics 22.3, pp. 400-407). The first learning phase continues until the deep neural network reaches a level of performance satisfactory for predicting the next token in a token sequence.

[0066] According to a particular and non-limiting example of an embodiment, the language model can undergo several learning phases: a pre-training phase on large sets of public data such as the 'Common Crawl' database and a refinement phase on sets of data dedicated to a specific application.

[0067] According to a particular and non-limiting embodiment, the text classification model is a deep neural network.

[0068] In a second learning phase, the text classification model is trained using sets of internal parameter values ​​of the language model's deep neural network obtained during the first learning phase. Each set of internal parameter values ​​corresponds to a textual data point used during the first training phase of the language model's deep neural network. Each set of internal parameter values ​​of the language model's deep neural network is associated with a textual input to the language model and is associated with a confabulation score. This confabulation score may indicate that the generated text is hallucinatory compared to the textual data presented as input to the neural network; that is, this confabulation score may indicate that the textual data input to the deep neural network is OOD (out of focus).

[0069] The internal parameters of the deep neural network of the text classification model are optimized using the estimation of a maximum likelihood loss function. The text classification model calculates the probability distribution of fabrication scores over a given set of fabrication scores for each textual data point in the textual dataset presented as input to the deep neural network. The objective is to maximize the probability that the 'true' fabrication score is predicted based on the values ​​of the internal parameters of the deep neural network of the language model; that is, the fabrication score reflects whether or not a textual data point presented as input to the language model belongs to the probability distribution of the textual data used for training the neural network of the language model.

[0070] According to a particular and non-limiting embodiment, the text classification model is trained by minimizing an entropy loss function crossed and using an Adam-type optimizer as described by Kingma et al. (DP Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proceedings of the 3rd International Conference on Learning Representations (ICLR) (2015).

[0071] According to other particular and non-limiting embodiment examples, the text classification model can be based on decision trees, or on vector machines.

[0072] During language model inference, text data is presented as input to the trained language model. The language model then provides text data on the basis of which a text is generated as output from the language model, along with internal parameter values ​​for the deep neural network corresponding to this input text data. The text classification model then evaluates the generated text, providing as output a confabulation score indicating the probability that the generated text is hallucinatory (or not) when the internal parameter values ​​calculated by the language model's deep neural network are presented as input to the trained text classification model.

[0073] According to one variant, the text generated by the trained language model and the fabulation score generated by the text classification model are rendered on at least one on-board interface of the vehicle.

[0074] For example, the generated text can be rendered on an in-vehicle display with a brightness or color that depends on the fantasy score. An icon can be added to the text displayed on a screen when the fantasy score exceeds a threshold value, for example.

[0075] According to a particular, non-limiting embodiment, the fabulation scores generated by the text classification model can be used to guide part of the text generation. A subsequent output token from the language model's deep neural network can be conditioned on the value of the calculated fabulation score.

[0076] This example is advantageous because the texts generated in this way have low fabulation scores without requiring training or tuning of the deep neural network of the language model.

[0077] According to one variant, the process further includes a step of comparing the fabulation score generated by the text classification model to a threshold value, and the generated text is rendered according to a result of the comparison of the fabulation score generated by the classification model and the threshold value.

[0078] According to a particular and non-limiting example of embodiment, the threshold value is determined during a validation phase of the language model by an operator.

[0079] According to a particular and non-limiting example of an embodiment, the threshold value can be determined based on legal provisions concerning false negatives and false Positives: High threshold values ​​produce more false negatives, and low threshold values ​​produce more false positives. A false negative occurs when a text input to the language model falls outside the distribution of text data used to train the language model's deep neural network, even though the confabulation score indicates otherwise. A false positive occurs when a text input to the language model falls within the distribution of text data used to train the language model's deep neural network, even though the confabulation score indicates otherwise.

[0080] According to a particular and non-limiting embodiment, the threshold value is determined according to a compromise between false negatives and false positives.

[0081] [Fig.2] schematically illustrates a device 2 configured to control the rendering of text generated from textual data from at least one embedded system of a vehicle, according to a particular and non-limiting embodiment of the present invention.

[0082] Device 2 corresponds for example to a device embedded in the vehicle, for example a computer.

[0083] Device 2 is, for example, configured to implement at least one step of the process described opposite Figures 1 and 2. Examples of such a device 2 include, but are not limited to, embedded electronic equipment such as a vehicle's on-board computer, an electronic control unit such as an ECU (Electronic Control Unit), a smartphone, a tablet, or a laptop computer. The elements of device 2, individually or in combination, may be integrated into a single integrated circuit, into several integrated circuits, and / or into discrete components. Device 2 may be implemented in the form of electronic circuits or software (or computer) modules, or a combination of electronic circuits and software modules.

[0084] The device 2 comprises one (or more) processor(s) 20 configured to execute instructions for carrying out at least one step of the process and / or for executing instructions from the software embedded in the device 2. The processor 20 may include integrated memory, an input / output interface, and various circuits known to those skilled in the art. The device 2 further comprises at least one memory 21, for example, volatile and / or non-volatile memory, and / or includes a memory storage device that may include volatile and / or non-volatile memory, such as EEPROM, ROM, PROM, RAM, DRAM, SRAM, flash, magnetic disk, or optical disk.

[0085] The computer code of the embedded software(s) including the instructions to be loaded and executed by the processor is for example stored on memory 21.

[0086] According to various specific and non-limiting embodiments, the device 2 is coupled in communication with other similar devices or systems and / or with communication devices, for example a TCU (Telematic Control Unit), for example via a communication bus or through dedicated input / output ports.

[0087] The deep neural network (of the language model and / or the text classification model) can be trained by a remote device, such as a server or a series of servers, communicating with device 2 via a communication network. The internal parameters of the deep neural network (of the language model and / or the text classification model) can then be transmitted by this remote device to device 2, which receives them. Device 2 is then able to implement the rendering control process when text data from an embedded system of the vehicle is presented as input to the trained deep neural network (of the language model and / or the text classification model).

[0088] According to a particular and non-limiting embodiment, the device 2 includes a block 22 of interface elements for communicating with external devices, for example, a remote server or the cloud, other nodes of the ad hoc network. The interface elements of block 22 include one or more of the following interfaces: - radio frequency RF interface, for example of the Wi-Fi® type (according to IEEE 802.11), for example in the 2.4 or 5 GHz frequency bands, or of the Bluetooth® type (according to IEEE 802.15.1), in the 2.4 GHz frequency band, or of the Sigfox type using UBN (Ultra Narrow Band) radio technology, or LoRa in the 868 MHz frequency band, LTE (Long-Term Evolution), LTE-Advanced; - USB interface (from the English "Universal Serial Bus" or "Universal Serial Bus" in French); - HDMI interface (from the English "High Definition Multimedia Interface", or "High Definition Multimedia Interface" in French); - LIN interface (from the English "Local Interconnect Network", or in French "Réseau interconnecté local").

[0089] Data, for example internal parameters of the trained deep neural network (of the language model and / or the text classification model), are, for example, loaded into the device 2 via the interface of block 22 using a Wi-Fi® network such as, according to IEEE 802.11, an ITS G5 network based on IEEE 802.1 IP, or a mobile network such as a 4G (or 5G) network based on the LTE (Long Term Evolution) standard defined by the 3GPP consortium, including an LTE-V2X network.

[0090] According to another particular and non-limiting embodiment, the device 2 includes a communication interface 23 which allows communication to be established with other devices (such as other computers in the embedded system) via a communication channel 24. The communication interface 23 corresponds, for example, to a transmitter configured to transmit and receive information and / or data via the communication channel 24. The communication interface 23 corresponds, for example, to a wired network of the CAN (Controller Area Network) type, CAN FD (Controller Area Network Flexible Data-Rate), FlexRay (standardized by ISO 17458) or Ethernet (standardized by ISO / IEC 802-3).

[0091] Internal parameters of the trained deep neural network (of the language model and / or the text classification model) can be received via the communication interface 23.

[0092] According to a particular and non-limiting embodiment, the device 2 can provide output signals to one or more external devices, such as a display screen 25, touch or not, one or more speakers 26 and / or other peripherals 27 (projection system, microphone, etc.) via output interfaces 28, 29 and 30 respectively. According to a variant, one or more of the external devices is integrated into the device 2.

[0093] Of course, the present invention is not limited to the embodiments described above but extends to a method for controlling the rendering of text generated from textual data originating from at least one embedded system of a vehicle, which would include secondary steps without departing from the scope of the present invention. The same would apply to a device configured for implementing such a method.

[0094] The present invention also relates to a vehicle, for example an automobile or more generally an autonomous land-powered vehicle, comprising the device 2 of [Fig.2].

Claims

Claims

1. Method for controlling the rendering of a text generated from textual data from at least one on-board system of a vehicle, said method comprising the following steps: - receiving (31) textual data from said at least one on-board system; - generating (32) the text as output from a trained language model and supplied as input by the received textual data, the language model having been trained during a first learning phase; - obtaining (33) a confabulation score as output from a trained text classification model and supplied as input by internal parameters of the trained language model, said internal parameters having been obtained during the generation of the text, the text classification model having been trained during a second learning phase; - rendering (34) the text generated from at least one on-board interface of the vehicle as a function of the confabulation score obtained.

2. The method of claim 1, wherein the language model is a large language model comprising a deep neural network whose input is fed with tokens obtained by converting textual data present as input to the language model and adapted to provide as output tokens which correspond to the text generated once these tokens have been converted into textual data.

3. The method of claim 2, wherein the language model is an autoregressive model based on a transformer-type architecture.

4. Method according to one of the preceding claims, wherein the text classification model is a deep neural network.

5. The method of one of the preceding claims, which further comprises a step (35) of comparing the confabulation score generated by the text classification model with a threshold value, and wherein the generated text is rendered based on a result of the comparison of the confabulation score generated by the classification model and the threshold value.

6. The method of claim 5, wherein the threshold value is determined according to a compromise between false negatives and false positives, a false negative corresponding to a situation in which textual data input to the language model is outside a dis- tribution of the text data used to train the deep neural network of the language model while the confabulation score indicates the opposite, and a false positive corresponding to a situation in which a text data input to the language model is not outside a distribution of the text data used to train the deep neural network of the language model while the confabulation score indicates the opposite.

7. Computer program comprising instructions for implementing the method according to any one of the preceding claims, when these instructions are executed by a processor.

8. A computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the method according to one of claims 1 to 6.

9. Device (2) for controlling the rendering of a text generated from textual data from at least one on-board system of a vehicle, said device (2) comprising a memory (21) associated with at least one processor (20) configured for implementing the steps of the method according to any one of claims 1 to 6.

10. Vehicle (10) comprising the device (2) according to claim 9.

Images