Method and device for controlling the rendering of text generated from textual data

Abstract

The present invention relates to a method for controlling the rendering of text generated from input text data from at least one vehicle's embedded system. The method obtains (11) a sequence of input text data from a language model based on a trained neural network; obtains (12) a sequence of predicted text data output from the trained neural network when fed the obtained input text data sequence; obtains (13) prediction intervals for said input text data; generates (14) text from the predicted text data sequence and the prediction intervals; and renders (15) the generated text from the vehicle's embedded system. Figure 2

Description

Title of the invention: Method and device for controlling the rendering of text generated from textual data 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.

[0003] More specifically, the present invention lies at the intersection of the fields of quantifying the uncertainty of models trained to provide predictions and learning in context of these models. Technological background

[0004] 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 passengers. Some of these systems are known as ADAS (Advanced Driver-Assistance System).

[0005] Some AD AS systems can use artificial intelligence to implement generators and other natural language processing according to particular road situations.

[0006] 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).

[0007] 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).

[0008] The fabrication of LLM models is related to the more general problem of neural networks 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.

[0009] However, LLM models can also respond to texts that may not correspond to the textual data presented as input. This is an identified problem with LLM models, which is similar to quantifying the uncertainty in the response of these LLM models.

[0010] There are several ways to quantify the response uncertainty of neural networks, but they generally require training these networks multiple times with different parameter configurations. This is the case, for example, with the so-called model ensemble method, which consists of training several models on the same task and combining their predictions. The variance of the results provided by these model ensembles can then be used to estimate the uncertainty of the results (more details in the article by Abdar, M., Pourpanah, F., Hussain, S., Rezazadegan, D., Liu, L., Ghavamzadeh, M.,... & Nahavandi, S. (2021). A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges. ArXiv preprint:2011.06225).

[0011] Other methods for quantifying response uncertainty use Bayesian neural networks based on parameter distributions, such as the method described in the article by Tishby, Levin, and Solia (“Consistent inference of probabilities in layered networks: predictions and generalizations.” In: International 1989 Joint Conference on Neural Networks, 403-409 vol.2, 1989). In this type of method, the parameters of the LLM model are treated as random variables with associated distributions. Bayesian models naturally provide uncertainty estimates by integrating these distributions.

[0012] Prior art methods can quantify response uncertainty, some even offering theoretical guarantees, but they all require computing resources that are proportional to the number of parameters. Consequently, they become more expensive as the number of parameters in the LLM model increases, rendering them unusable for many applications, particularly those requiring fast and efficient responses, such as the control of ADAS systems embedded in vehicles. Summary of the present invention

[0013] One object of the present invention is to solve at least one of the problems of the technological background described above.

[0014] According to a first aspect, the present invention relates to a method for controlling the rendering of text generated from textual input data from at least one embedded system of a vehicle, said method comprising the following steps for each current textual input data: - obtaining a textual input data sequence from a transformer-based language model, said textual input data sequence being obtained from a textual context data sequence defining a context inside and / or outside the vehicle and the current textual input data, the transformer being a deep neural network trained during a so-called context-based learning phase which uses said textual context data sequence and a corresponding expected output textual data sequence; - obtaining a sequence of predicted textual data at the output of the driven transformer when it is supplied at the input by the sequence of input textual data obtained; - obtaining a prediction interval for the current input text data from the context text data sequence, the corresponding expected output text data sequence, a confidence level and the input text data from said at least one embedded system; - generation of text from the predicted textual data sequence and the prediction intervals obtained for the input textual data from said at least one embedded system, and - rendering of text generated from an on-board vehicle system.

[0015] The method is fast and offers a distribution-free approach to uncertainty quantification, with solid theoretical guarantees and applicable to several scenarios and data modalities (including tabular data), making it a solution that overcomes the limitations of Bayesian inference and model ensemble methods.

[0016] The method combines a conformal prediction mechanism with a transformer that replaces a traditional learning model of a deep neural network. This substitution leverages the pre-trained prediction capabilities of a transformer while benefiting from the rigorous uncertainty quantification provided by the conformal prediction mechanism. To achieve this, the trained transformer generates predictions for each sequence of input text data, which are then evaluated to form prediction intervals.

[0017] Contextual learning provides a certain flexibility of use while Conformal prediction, as implemented by the method, provides robust estimates of uncertainty. The combination of context learning and conformal prediction, as described in the method, yields prediction intervals that allow for quantifying the uncertainty of transformer-based models.

[0018] The method produces accurate predictions and also provides reliable measures of response uncertainty, thus improving the interpretability and reliability of prediction models such as LLM models.

[0019] According to a particular and non-limiting embodiment of the present invention, the confidence level is chosen by a vehicle user.

[0020] According to a particular and non-limiting embodiment of the present invention, the transformer is a linear self-attention transformer.

[0021] According to a particular and non-limiting embodiment of the present invention, the context textual data sequence and the corresponding expected output textual data sequence are generated synthetically.

[0022] According to a particular and non-limiting embodiment of the present invention, the textual data of the context textual data sequence and the data of the expected output textual data sequence are linked by linear relationships.

[0023] According to a particular and non-limiting embodiment of the present invention, the textual data of the expected output textual data sequence are noisy with additional noise.

[0024] According to a second aspect, the present invention relates to a rendering control device for text generated from a sequence of 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.

[0025] 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.

[0026] 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.

[0027] Such a computer program can use any programming language, and be in the form of source code, object code, or a intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0028] 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.

[0029] 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, RAM, CD-ROM or a microelectronic circuit-type ROM, or a magnetic recording means or a hard disk drive.

[0030] 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.

[0031] 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

[0032] 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 4, in which:

[0033] [Fig-1] schematically illustrates a transformer learning process in context linear self-attention according to a particular and non-limiting embodiment of the present invention.

[0034] [Fig.2] schematically illustrates a diagram of the steps of a method for controlling the rendering of a text generated from input textual data from at least one embedded system of a vehicle according to a particular and non-limiting embodiment of the present invention.

[0035] [Fig.3] schematically illustrates the process of [Fig.2] according to a particular and non-limiting embodiment of the present invention.

[0036] [Fig.4] schematically illustrates a device configured to control the rendering of text generated from input 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

[0037] A method and a rendering control device for text generated from input textual data from at least one vehicle on-board system will now be described in the following with joint reference to Figures 1 to 4. The same elements are identified with the same reference signs throughout the following description.

[0038] The terms "first," "second" (or "firsts," "seconds"), etc., are used in this document by arbitrary convention to allow for the identification and distinction of different elements (such as operations, means, etc.) implemented in the embodiments described below. Such elements may be distinct or correspond to a single element, depending on the embodiment.

[0039] 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.

[0040] The vehicle includes at least one embedded system which is configured to provide textual data, i.e. either sequences formed from one or more words of natural language.

[0041] According to a particular and non-limiting embodiment of the present invention, the textual data may be representative of a state of at least one component of the vehicle and / or of a road situation in which the vehicle is located.

[0042] According to a particular and non-limiting embodiment of the present invention, one of these embedded systems may include a touchscreen and memory configured for input and recording natural language words. Another example may be a voice recording system comprising a microphone and memory.

[0043] According to a particular and non-limiting embodiment of the present invention, an on-board vehicle system may be an ADAS system. For example, the system may be 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' may be provided by this ADAS system as soon as an 80 km / h speed limit sign is detected. Another ADAS system may provide word sequences as soon as weather conditions change.

[0044] 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.

[0045] According to a particular and non-limiting embodiment of the present invention, an embedded vehicle system may be a display system comprising interface means with a vehicle user that enable the retrieval of textual data and the audio rendering of text. These interface means may, for example, include a touchscreen, a microphone and / or a speaker controlled by a computer, or a voice control system comprising a microphone and / or a speaker. The embedded system may include a memory configured to store natural language words.

[0046] The vehicle further includes at least one embedded system which is configured to provide context data, i.e. a word or sequence of words from natural language which defines a context relative to textual data provided.

[0047] For example, a user can use a speech recognition system to ask a question about a feature of an ADAS system. The speech recognition system then transcribes this question into a sequence of text data. Another embedded system can then determine the internal and / or external context of the vehicle when the user asked this question. The context can be defined from information relating to ADAS systems that can be activated and configured according to specific parameters. It can also be defined from data provided by other embedded systems such as sensors.

[0048] The present invention uses an LLM type language model which is based on a transformer 201 of [Fig.1].

[0049] Transformer 201 is a multilayer deep neural network.

[0050] The learning of transformer 201 is called In-Context Learning. In-Context Learning, as a form of meta-learning, improves the quality of the responses generated by the LLM model based on transformer 201. Transformer 201 can then learn from the context so that its future responses to user requests are improved without modifying its internal parameters.

[0051] Context learning allows the LLM model to derive accurate predictions from the context, thus improving its ability to generalize to textual input data distinct from the training data. Context learning can follow the usual configuration of mini-batch gradient learning (Herbert, R., Sutton, M. (1951). A Stochastic Approximation Method. The Annals of Mathematical Statistics, Sep., 1951, Vol. 22, No. 3 (Sep., 1951), pp. 400-407) where the internal parameters of a transformer 201 are updated by following the direction of the negative gradient of a loss function 202, calculated for a subset of examples (mini-batches) of the dataset.

[0052] According to a particular and non-limiting embodiment of the present invention, the transformer 201 can be a deep architecture based on a multi-head attention mechanism proposed by Vaswani et al. (Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A., Kaiser, L., & Polosukhin, I. (2017). Attention Is All You Need. arXiv preprint: 1706.03762). Essentially, a sequence of textual input data xA presented as input to the transformer 201 is converted into numerical representations called tokens, and each token is converted into a vector of real values ​​via a lookup in a word integration table. More precisely, each word in a dictionary is represented by a vector of real values, that is, a list of numbers. The vectors all have the same length and encode the meaning of the words from a similarity of use point of view.The more words can be used in the same context (one in place of another in a sentence), the closer their vectors are. This technique is based on the assumption that words appearing in similar contexts have related meanings. The similarity between vectors can be defined as a distance in the vector space where these vectors are defined, or as an angle between these vectors.

[0053] According to a particular and non-limiting example of the present invention, the transformer 201 can be a linear self-attention transformer (LSA) (Angelopoulos, AN, & Bâtes, S. (2021). A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification. arXiv preprint arXiv:2107.07511).

[0054] The linear self-attention mechanism allows simultaneous learning of transformer 201 by updating all tokens corresponding to the input textual data sequence XA through linear transformations.

[0055] The linear self-attention mechanism takes as input the input text data sequence xA, which is formed from a context text data sequence % = (xq, XJ) defining a context of the interior and / or exterior of the vehicle, and an expected output text data sequence y / j corresponding to the context text data sequence. X and a text data xm called query data. The query text data Aw is associated with an expected output text data yM which is not part of the input text data sequence XA (corresponding token equal to 0).

[0056] According to a particular and non-limiting embodiment example, the context textual data sequence X and the corresponding expected output textual data sequence^ can be generated synthetically.

[0057] According to a particular and non-limiting example of embodiment, the textual data of the context textual data sequence X and the data of the expected output textual data sequence Y can be linked by linear relationships.

[0058] According to a particular, non-limiting embodiment, the expected output text data sequence Y can be noisy with additional Gaussian noise Σ ~ N(0, <7), thus ensuring robustness against noisy context text data conditions: Y = Xw, where w is a matrix defining linear relationships between the context text data sequence X and the expected output text data sequence Y, possibly augmented with additional Gaussian noise. The noise is homoscedastic and maintains a constant standard deviation for all data in the context text data sequence X, thereby ensuring uniform training conditions. This configuration replicates the challenges posed by real-world data, preparing the transformer to efficiently handle variability and noise.

[0059] The linear self-attention mechanism reflects a gradient descent process in which the initial values ​​of the input tokens denoted a matrix transformations of the input textual data sequence xA into tokens) are iteratively refined to reach a predicted textual data sequence ? = ( y„ .... ) X* “ transformer 201 (Aw represents a refinement matrix). The loss function 202 is calculated between the expected output text data yM and the corresponding predicted text data y.

[0060] This method takes advantage of the self-attention mechanism's ability to differentiate the importance of input tokens, by grasping the data dependencies of the input textual data sequence X4-

[0061] The transformer 201 thus trained has the advantage of not having recurrent units and therefore requires less training time than recurrent neural network (RNN) architectures such as, for example, long-term memory neural networks (LSTMs). One of the advantages of context-based learning is that the transformer 201 can then learn from the context so that its future responses to vehicle user requests are improved without modifying its internal parameters.

[0062] Fig. 2 schematically illustrates an example of a diagram of the steps of a method for controlling the rendering of text generated from textual data and context data from at least one embedded system of a vehicle according to a particular and non-limiting embodiment of the present invention.

[0063] The method uses a conformal prediction mechanism.

[0064] Conformal prediction is an automated framework for quantifying uncertainty that produces statistically valid confidence intervals (called prediction intervals) for any underlying point predictor (whether statistical, automated, or deep learning) solely by assuming data interchangeability. In statistical inference, and more specifically in predictive inference, a prediction interval is an estimate of the range within which a future observation will lie, with a certain probability, given what has already been observed.

[0065] Conformal prediction works by calculating conformity scores on previously labeled input text data, and using these conformity scores to create prediction intervals on any new input text data.

[0066] Conformal prediction requires a confidence level (l) that can be chosen by the user of the vehicle for which the algorithm is to produce its predictions. This confidence level limits the frequency of errors that the process is allowed to make. For example, a confidence level of 0.1 means that the process can make a maximum of 10% incorrect predictions. To satisfy this requirement, the output of the trained transformer 201 is an ensemble prediction, instead of a point prediction produced by standard supervised machine learning models. For regression tasks, the results are prediction intervals calculated for a sequence of input text data x1, where a lower confidence level (fewer allowed errors) produces wider prediction intervals that are less specific, and vice versa—more allowed errors produce tighter prediction intervals.

[0067] Conformal prediction offers a distribution-free approach to uncertainty quantification with strong theoretical guarantees, making it a solution that overcomes the limitations of Bayesian inference and so-called model set methods (Angelopoulos, AN, & Bâtes, S. (2021). A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification. arXiv preprint arXiv:2107.07511).

[0068] The method, illustrated in Figure 2, incorporates the conformal prediction principle of the trained linear self-attention transformer 201 to construct prediction intervals Cconf(x) that quantify the prediction uncertainty, offering a robust solution for reliable decision-making. These prediction intervals Cœnj(X) are formulated to contain a 'true' textual datum with a probability predetermined, thus offering statistical assurance regarding the accuracy of the prediction. These prediction intervals Cconf{X) are determined on the basis of a confidence level a. By selecting a confidence level a, it is ensured that the prediction interval C^fix) encompasses a 'true' textual data point at least half the time, thus establishing a measure of confidence a in the predictions output of the trained transformer 201.

[0069] Textual input data vvi are obtained from at least one vehicle on-board system.

[0070] A prediction interval Conf(x) is then calculated for each current input text data x (equal to XH+2, •.. and xM) from steps 11 to 13.

[0071] In step 11, a sequence of input textual data X1 = (xh x2, ..., xn, x) is obtained from a context textual data sequence ( vvri used as input text data for transformer 201 "Ak * * - H / A during contextual learning of transformer 201 and a current input text data x which is equal to an input text data vv Y

[0072] The context textual data sequence (xx^ x^) defines a context inside and / or outside the vehicle.

[0073] In step 12, a sequence of predicted textual data O is obtained at the output of the driven transformer 201 Y = h r when the input textual data sequence x7cst presented at the input of the driven transformer 201.

[0074] In a step 13, a prediction interval Cconj(x) is calculated for the current input text data x from the context text data sequence x^, from an output text data sequence expected yl _ corresponding to the textual data sequence of context xx^ and which was used during the context learning of transformer 201, from a confidence level af(0.1) and from the textual input data (rrr Y 1' • • • '

[0075] According to an example embodiment of step 13, a prediction interval (x) is calculated for the current input text data x during substeps 131 to 133.

[0076] In substep 131, for each current expected output text data y of the expected output text data sequence Y1, a text data predicted p ( xi ) i — 1 to H is obtained for each context text data xî and a current predicted text data ^(x) is obtained for the current input text data x.

[0077] For example, a linear function £ is obtained by linear regression of pairs of values ​​(x, y) = 1 to n and the pair of values ​​(x, y) formed by the input text data x and the current expected output text data y. The predicted text data i = 1 to n and the current predicted text data ^(x) are then equal to values ​​of the linear function f. Each pair of values ​​(Xf, y) is formed from a context text data xi from the context text data sequence (vvv 1) and a corresponding text data y from the expected output text data sequence Y1-

[0078] . In step 132, a conformity score Ryj (i = 1 to n) is calculated for each pair of values ​​(Xj, y^ by: [Math.l]

[0079] and for the pair of values ​​(x,y) by:

[0080] [Math.2] Ryy^i = \y-Ry(x) |

[0081] In step 133, a probability of belonging to a prediction interval is calculated for the current expected output data y from the conformity scores calculated Ryj (i=l to n) by:

[0082] [Math.3] æ ( y ) = ( 1 + ( 1 {Ryj Ry»+[ )) / («+ 1)

[0083] In step 134, the prediction interval Conf(x) is calculated by:

[0084] [Math.4] CemfCx) = [y = y.for (i = n+1M) :(n+1)^(y) < FO -o^n4-1) 1}

[0085] This prediction interval calculation is a robust nonparametric method for creating prediction intervals independently of strict data distribution assumptions. It evaluates training data by integrating new textual input data with context data and calculating conformity scores to assess inclusion in prediction intervals. This method guarantees coverage, based on the data exchange assumption.

[0086] In a step 14, a text is generated from the predicted textual data sequence Çr and the prediction intervals Cconf(x) obtained for the input textual data ( rrr Y An+l' An+'2' • • • ' '

[0087] The method indicates to which prediction interval each input text data item belongs and with what probability. Each prediction interval is associated with one of the expected output text data items ( ) with a certain probability. Thus, it is possible to generate -^n+2' ' ' ' ' a text from these prediction intervals and generate a text from the set of expected output textual data associated with these prediction intervals.

[0088] In a step 15, the generated text is rendered from an on-board vehicle system such as an audiovisual text rendering system or display system interface means.

[0089] Figure 4 schematically illustrates a device 4 configured to control the rendering of text generated from input text data originating from at least one embedded system of a vehicle, according to a particular and non-limiting embodiment of the present invention. Device 4 corresponds, for example, to a device embedded in a vehicle, such as a computer.

[0090] Device 4 is, for example, configured to carry out the steps of the processes described opposite Figures 1 to 3. Examples of such a device 4 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 4, individually or in combination, can be integrated into a single integrated circuit, into several integrated circuits, and / or into discrete components. Device 4 can be implemented in the form of electronic circuits or software (or computer) modules, or a combination of electronic circuits and software modules.

[0091] The device 4 comprises one (or more) processor(s) 40 configured to execute instructions for carrying out the steps of the process and / or for executing instructions from the software embedded in the device 4. The processor 40 may include integrated memory, an input / output interface, and various circuits known to those skilled in the art. The device 4 further comprises at least one memory 41, corresponding, for example, to volatile and / or non-volatile memory, and / or includes a memory storage device which may include memory volatile and / or non-volatile, such as EEPROM, ROM, PROM, RAM, DRAM, SRAM, flash, magnetic or optical disk.

[0092] 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 41.

[0093] According to various particular and non-limiting embodiments, the device 4 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.

[0094] According to a particular and non-limiting embodiment, the device 4 includes a block 42 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 42 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").

[0095] According to another particular and non-limiting embodiment, the device 4 includes a communication interface 43 which enables communication with other devices (such as other computers in the embedded system) via a communication channel 44. The communication interface 43 corresponds, for example, to a transmitter configured to transmit and receive information and / or data via the communication channel 44. The communication interface 43 corresponds, for example, to a wired CAN (Controller Area Network) or CAN FD (Controller Area Network Flexible Data-Rate) type network. flexible data rate”), FlexRay (standardized by ISO 17458) or Ethernet (standardized by ISO / IEC 802-3).

[0096] According to a particular and non-limiting embodiment, the device 4 can provide output signals to one or more external devices, such as a display screen 45, touch or not, one or more speakers 46 and / or other peripherals 47 via output interfaces 48, 49, 50 respectively. According to a variant, one or more of the external devices is integrated into the device 4.

[0097] 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 text input data from at least one embedded system of a vehicle, which would include secondary steps without falling outside the scope of the present invention. The same would apply to a device configured for implementing such a method.

[0098] The present invention also relates to a vehicle, for example an automobile or more generally an autonomous land-powered vehicle, comprising the device 4 of [Fig.4].

Claims

Demands

1. A method for controlling the rendering of text generated from input text data from at least one vehicle on-board system, said method comprising the following steps for each current input text data: - obtaining (11) a sequence of input text data from a transformer-based language model, said input text data sequence being obtained from a context text data sequence defining a context inside and / or outside the vehicle and the current input text data, the transformer being a deep neural network trained in a so-called context-based learning phase which uses said context text data sequence and a corresponding expected output text data sequence;- obtaining (12) a predicted textual data sequence at the output of the driven transformer when it is fed into the input by the obtained input textual data sequence; - obtaining (13) a prediction interval for the current input textual data from the context textual data sequence, the corresponding expected output textual data sequence, a confidence level and the input textual data from said at least one embedded system; - generating (14) the text from the predicted textual data sequence and the prediction intervals obtained for the input textual data from said at least one embedded system, and - rendering (15) the generated text from an embedded system of the vehicle.

2. A method according to claim 1, wherein the confidence level is chosen by a vehicle user.

3. Method according to claim 1, wherein the transformer is a linear self-attention transformer.

4. A method according to any one of the preceding claims, wherein the context text data sequence and the corresponding expected output text data sequence are generated synthetically.

5. A method according to claim 4, wherein the textual data of the context textual data sequence and the data of the expected output textual data sequence are linked by linear relationships.

6. A method according to claim 5, wherein the text data of the expected output text data sequence are noisy with additional noise.

7. Device (4) for controlling the rendering of text generated from a sequence of textual data from at least one vehicle-mounted system, said device (4) comprising a memory (41) associated with at least one processor (40) configured for carrying out the steps of the method according to any one of claims 1 to 6.

8. Computer program comprising instructions for carrying out the method according to any one of claims 1 to 6, when such instructions are executed by a processor.

9. Computer-readable recording medium on which is recorded a computer program comprising instructions for carrying out the steps of the process according to any one of claims 1 to 6.

10. Vehicle comprising a device according to claim 7.

Images