Methods for operating and / or controlling a dialogue system

Abstract

Method for operating a dialog system (DS), wherein the dialog system (DS) comprises a local dialog unit (LD) for capturing dialog inputs and outputting dialog outputs, and an external dialog unit (ED) connected to the local dialog unit (LD) for analyzing captured dialog inputs and determining dialog outputs based thereon, comprising the following steps performed by the local dialog unit: - Receipt (12) of at least one future probable dialogue sequence calculated by the external dialogue unit (ED), consisting of dialogue inputs and / or dialogue outputs; - Saving (13) the received dialog history; - Retrieving (14) the stored dialog history when the data connection between the local dialog unit and the external dialog unit is interrupted; and capturing dialog inputs and / or outputting dialog outputs (15) based on the retrieved dialog history.

Description

[0001] The present invention relates to a method for operating a dialog system and a method for controlling a dialog system. The invention further relates to a system in which such a method according to the invention is used.

[0002] Dialogue systems enable a user to communicate with an automated system, whereby inputs and outputs can be provided in a variety of ways, such as text, speech, graphics, or gestures.

[0003] For example, speech dialogue systems allow users to conduct partially or fully automated dialogues in largely natural language via a voice interface. Such speech dialogue systems have long been known, for example, in the telecommunications industry. In such applications, all speech processing is performed by a computer system on the service provider's side.

[0004] A similar approach is used, for example, for voice-controlled assistance systems integrated into internet-connected speakers for music playback, and also for applications in consumer electronics, where speech processing is primarily performed by computer systems on the side of the intelligent personal assistant provider. This ensures high-quality recognition and interpretation of user voice input, since the high computing power required for this does not have to be provided in the end device but can be provided by the provider. Only for activating speech processing does a limited amount of speech recognition take place on a user-side device.

[0005] Voice dialogue systems are also increasingly being used in the automotive industry. These systems can be used, for example, to enter destinations and provide navigation instructions using voice in navigation systems, to select media in infotainment systems, or to operate an integrated telephone function. These voice dialogue systems are traditionally embedded locally in the vehicle and are therefore always available. However, such completely local systems also have a number of disadvantages, including poorer speech recognition, difficult software updates, limited opportunities to optimize system behavior, or even to learn from multiple systems simultaneously.

[0006] There are also approaches to implementing purely online voice systems in cars. For example, mobile devices can be linked to the vehicle's infotainment system to operate it by voice. While voice input and output are handled internally, speech processing is performed externally. While purely online-based systems do not have the disadvantages of embedded systems, they do suffer from the well-known problems of connection drops and poor network availability.

[0007] When using speech dialogue systems in motor vehicles, the highest possible reliability is desirable so that the driver is not unnecessarily distracted by malfunctions and can focus their attention on the driving task. For this reason, approaches using so-called hybrid speech dialogue systems are also being pursued, in which speech processing is partially embedded locally in the vehicle and partially performed by an external computer system in the so-called cloud.

[0008] For example, EP 2 909 833 B1 describes a method for speech recognition in a motor vehicle. A user's speech inputs are fed to both an onboard speech recognition system inside the vehicle and an offboard speech recognition system outside the vehicle. The speech text is then determined based on the recognition results of the speech recognition systems. The method does not rely on all speech inputs always being processed by both the onboard speech recognizer and the offboard speech recognizer. Rather, the method provides for complete onboard speech recognition, complete offboard speech recognition, sequential speech recognition, and parallel speech recognition. One of these modes is selected depending on a control parameter.For example, if there is no radio connection from the vehicle to a data network via which the off-board speech recognition system can be reached, recognition can still be carried out based on the on-board speech recognition system.

[0009] Furthermore, DE 10 2016 003 903 A1 discloses a method for speech recognition in a motor vehicle, in which speech recognition results are generated from a speech input made by a user by means of a vehicle-internal speech recognition device and are interpreted syntactically / synemically, wherein an unrecognized part of the speech input is completed in a vehicle-external speech recognition device taking into account the syntactical / synematic interpretation.

[0010] Another alternative hybrid architecture for speech dialog systems is described in DE 10 2015 212 650 A1. In this case, the substeps of a speech dialog system are executed completely in parallel, both locally embedded and online on an external computer system. After each substep, a coordination point waits for both sub-results, and then the next substep is started in parallel.

[0011] Furthermore, US 2016 / 0 132 342 A1 shows a computing device that receives a trigger to display commands. A possible set of commands is identified and divided into categories. A category is displayed for user interaction.

[0012] It is an object of the present invention to provide an improved method for operating and / or controlling a dialogue system that can be used in particular in a motor vehicle.

[0013] This object is achieved by a method for operating a dialogue system having the features of claim 1, by a method for controlling a dialogue system according to claim 8. Preferred embodiments of the invention are the subject of the dependent claims.

[0014] In the method according to the invention for operating a dialogue system, the dialogue system comprises a local dialogue unit for capturing dialogue inputs and outputting dialogue outputs, as well as an external dialogue unit connected to the local dialogue unit for data processing for analyzing captured dialogue inputs and determining dialogue outputs based thereon, wherein the following steps are carried out by the local dialogue unit: - Receipt of at least one likely future dialogue progression calculated by the external dialogue unit and consisting of dialogue inputs and / or dialogue outputs; - Saving the received dialogue history; - Retrieving the stored dialogue history in the event of an interruption of the data connection between the local dialogue unit and the external dialogue unit; and - Capturing dialog inputs and / or outputting dialog outputs based on the retrieved dialog history.

[0015] In this way, a dialog can be initiated even if the connection to an online dialog system is interrupted. Nevertheless, the advantages of processing dialog inputs, especially in the form of voice inputs, by an external computer system in the cloud, such as simple update options, are retained. A further advantage over the hybrid architecture disclosed in DE 10 2015 212 650 A1 is that no technically complex and execution-slowing coordination is required. Instead of a local and an external dialog manager, the architecture proposed here requires only one dialog manager.

[0016] According to one embodiment of the invention, the local dialogue unit additionally performs the following steps: - Capturing a user's dialog input; - Transferring the captured dialog input to the external dialog unit.

[0017] This makes it possible to take the previous dialogue into account in the future dialogue course calculated by the external dialogue unit.

[0018] Advantageously, in the method, in an online mode, based on the dialogue input captured by the local dialogue unit and transmitted to the external dialogue unit, a dialogue output generated by the external dialogue unit is received and output, and in an offline mode, the future probable dialogue course received and stored by the external dialogue unit is retrieved in the local dialogue unit.

[0019] In this case, the dialogue history received from the external dialogue unit preferably represents a probable continuation of the previous dialogue for the captured dialogue input, whereby the previous dialogue is continued based on this dialogue history.

[0020] This ensures that ongoing dialogues do not have to be interrupted, but can be continued meaningfully.

[0021] According to a further embodiment of the invention, the dialogue history is received before a dialogue has been started.

[0022] In this way, a dialogue can take place even if it had not yet begun before the connection was interrupted.

[0023] Advantageously, updated versions of the likely future dialogue flow are received from the external dialogue unit at regular intervals and stored in the local dialogue unit.

[0024] Likewise, a likely future dialog progression can advantageously be retrieved from the external dialog unit and stored in the local dialog unit if an interruption of the connection is anticipated.

[0025] In the method according to the invention for controlling a dialogue system, the dialogue system comprises a local dialogue unit for capturing dialogue inputs and outputting dialogue outputs, as well as an external dialogue unit connected to the local dialogue unit for data processing for analyzing captured dialogue inputs and determining dialogue outputs based thereon, wherein the following steps are carried out by the external dialogue unit: - Calculating at least one likely future dialogue course consisting of dialogue inputs and / or dialogue outputs; and - Transferring the calculated dialog history to the local dialog unit.

[0026] According to one embodiment of the invention, the following step is additionally performed in the external dialogue unit: - receiving a captured dialog input from the local dialog unit;wherein calculating the dialog history is based on the received dialog input and the dialog history represents a likely continuation of the previous dialog.

[0027] According to a further embodiment of the invention, the dialogue flow is calculated before a dialogue has started.

[0028] Preferably, the dialogue flow is calculated based on one or more of the following parameters: - the current location of the local dialogue unit, - the current local time of the local dialogue unit, - Information on current political, economic, social, cultural, sporting or other events, - information commonly requested by the user, - information requested by other users.

[0029] The dialogue system can be designed in particular as a speech dialogue system in which dialogue inputs are made by speech recognition and interpretation of speech inputs from a user and / or dialogue outputs are made by speech synthesis.

[0030] According to one embodiment of the invention, the calculated dialogue progression includes, in addition to a probable continuation of the previous dialogue for a recognized dialogue input, also a probable continuation of the previous dialogue for a phonetically similar dialogue input.

[0031] Particularly advantageously, in the method according to the invention, the local dialogue unit is implemented in a vehicle and the external dialogue unit is implemented in a server external to the vehicle, which are connected via the Internet.

[0032] Preferably, the calculated dialogue flow is designed in the form of a hierarchical tree structure.

[0033] In the system according to the invention for operating and controlling a dialogue system, the dialogue system comprises a local dialogue unit for capturing dialogue inputs and outputting dialogue outputs, as well as an external dialogue unit connected to the local dialogue unit for data processing, for analyzing captured dialogue inputs and determining dialogue outputs based thereon, wherein the system is configured to carry out a method according to the invention.

[0034] Further features of the present invention will become apparent from the following description and claims in conjunction with the figures. Fig. 1 shows a schematic flow diagram for a method for operating a dialogue system based on a likely future dialogue course; Fig. 2 shows a schematic flow diagram for a method for controlling a dialogue system by calculating a likely future dialogue course; Fig. 3 shows schematically a system for operating and controlling a dialogue system; Fig. Figure 4 shows an example of a hierarchical tree structure of a dialog history for media playback.

[0035] To better understand the principles of the present invention, embodiments of the invention are explained in more detail below with reference to the figures. It is understood that the invention is not limited to these embodiments and that the described features may also be combined or modified without departing from the scope of the invention as defined in the claims.

[0036] Fig. Figure 1 schematically shows a method for operating a dialog system based on a likely future dialog progression. The method is executed by the local dialog unit, which can be embedded in a motor vehicle. The dialog is controlled by a dialog manager located at an external dialog unit, in particular an external computer system in the cloud.

[0037] In a first step 10, a user's dialog input is recorded. Depending on the user interface used, the dialog input can be performed in various ways. For example, inputs can be made using a mechanical control element to select one of several options displayed on a screen, or through gestures, or especially through voice input.

[0038] The dialogue input is then transmitted to an external dialogue unit in a subsequent step 11, possibly after signal processing, recognition, and interpretation steps not shown here. The transmission can be carried out in particular using wireless communication technologies, for example, via a mobile network in accordance with the 4G or 5G mobile communications standards.

[0039] The dialog input is then analyzed by the dialog manager in the external dialog unit. The dialog manager controls the further dialog in an online mode based on the dialog input captured by the local dialog unit and transmitted to the external dialog unit. It also generates appropriate dialog outputs, such as responses to user search queries, which are transmitted to a local dialog unit and output there. In online mode, updates to the prediction of the future dialog flow can be sent to the internal dialog unit along with the system response to each user input. This is advantageously the standard case when an online connection is available.

[0040] According to the invention, a dialogue predictor in the external dialogue unit additionally calculates future probable dialogue courses for a potentially occurring offline mode of the local dialogue unit, which can occur due to an interruption of the connection between the local and external dialogue unit.

[0041] The dialogue predictor can be integrated into the dialogue manager, but can also be located separately in the external dialogue unit and then interact with the dialogue manager to predict future user inputs and system reactions and proactively send this information to the local dialogue unit.

[0042] The likely future dialog courses can, in particular, be a continuation of an ongoing dialog. One or more of these likely future dialog courses are transmitted from the external dialog unit to the local dialog unit and received by the latter in step 12.

[0043] Likewise, the progress of a dialog that has not yet begun, which is considered likely depending on various parameters to be explained later, can also be received by the local dialog unit. In this case, the previous steps 10 and 11 are optional.

[0044] The likely future dialog progression received from the external dialog unit is then stored in a memory of the local dialog unit in step 13. As long as a data connection exists between the local dialog unit and the external dialog unit, this stored dialog progression does not need to be accessed. Instead, the appropriate subsequent dialog steps are retrieved online from the external dialog unit based on the user's dialog input, ensuring that the dialog continues correctly.

[0045] However, if the data connection between the local dialogue unit and the external dialogue unit is interrupted, for example because the required mobile network is not available locally or is temporarily interrupted due to technical problems, the stored dialogue history is retrieved from the memory in a step 14.

[0046] Subsequently, in step 15, further dialog inputs and / or dialog outputs are performed based on the retrieved dialog history. The dialog history can contain multiple dialog steps. This results in a hierarchical tree structure of dialog steps, since there are often multiple options for operating functions and search queries.

[0047] When the data connection between the local dialog unit and the external dialog unit is restored, the dialog system returns to online mode.

[0048] Furthermore, the dialogue system transmits the history of the dialogue conducted locally in offline mode to the external dialogue unit so that it can improve its prediction (“data mining”).

[0049] The procedure carried out by the external dialogue unit for controlling the dialogue system by means of a calculation of a probable future dialogue course is shown schematically in Fig. 2 shown.

[0050] First, a user's dialog input is received in step 20. In the external dialog unit, the dialog manager generates dialog outputs matching the currently received dialog inputs and transmits them to the local dialog unit for playback in online mode.

[0051] In addition, in step 21, at least one likely future dialogue progression consisting of dialogue inputs and / or dialogue outputs is calculated for a potential interruption of the data connection.

[0052] Predefined dialogue models can be used for this purpose. This is particularly suitable for dialogue flows with only a few possible options. For dialogue flows with a large number of possible options, such as user search queries, the dialogue models can be determined using an artificial neural network, for example, which learns the probabilities of typical dialogue flows or search queries. To train the neural network, it can be fed past dialogue flows from the respective user or from a large number of users. In this way, the neural network is continually optimized, enabling increasingly accurate predictions of the user's future dialogue flow.

[0053] In this way, the likely dialog steps following the last received dialog input can be determined. Instead of continuing a current dialog in this way, a probable dialog progression can also be calculated before the dialog begins. Various external parameters can then be used for this purpose.

[0054] For example, since information about so-called points of interest (POIs) in a user's vicinity is often requested, the current location of the local dialog unit can be evaluated to provide possible queries about the current points of interest. For this purpose, the location information can be determined, for example, by a GPS sensor integrated into the local dialog unit and transmitted to the external dialog unit. This location information can also be determined from the radio cell in which the local dialog unit is currently located.

[0055] Likewise, for example, probable dialogue processes related to the current local time of the local dialogue unit, information on current political, economic, social, cultural, sporting or other events, information usually requested by the user or information requested by other users can also be included.

[0056] Likewise, several of these parameters can be considered simultaneously to enable the best possible dialog prediction. For example, if information is available for a large number of users about which dialog entries or search queries they have made at a specific location at a specific time of day in the past, one or more likely dialog trajectories can be predicted based on this information for a user who is at that location at the same time of day.

[0057] One or more calculated, future probable dialog histories are then transferred from the external dialog unit to the local dialog unit in step 22. The transfer requires an existing connection between the external dialog unit and the local dialog unit and can occur at regular intervals, with older versions of the dialog history being replaced by updated versions.

[0058] Likewise, the transmission can occur instead or additionally when a connection interruption is anticipated. For example, a local dialog unit implemented in a motor vehicle can determine, based on the vehicle's current GPS coordinates and digital maps of the current vehicle surroundings, that the vehicle will soon enter a long tunnel and that a connection interruption will likely occur. In this case, the vehicle can then retrieve a likely future dialog progression from the external dialog unit at a suitable time before entering the tunnel.

[0059] Both predictive dialog continuations for an ongoing dialog and simple dialogs for other topics can be sent to the local dialog unit. This allows for appropriate responses in both cases when network availability is lacking. On the one hand, interruptions to ongoing dialogs are prevented, and on the other hand, more meaningful or useful messages can be issued to the user than, for example, "Service unavailable" or "Unfortunately, I can't do that at the moment."

[0060] Fig. 3 schematically shows a system for operating and controlling a speech dialogue system that has a local dialogue unit LD in a frontend F and an external dialogue unit ED in a backend B.

[0061] The frontend F can be integrated, in particular, into a motor vehicle as part of an infotainment system and, in addition to the local dialogue unit, can include various infotainment components such as a local media database (MDB) and a navigation database (NDB). Additionally, a communication unit (KF) is provided in the frontend, which establishes a wireless connection to a backend in the form of an online server.

[0062] In addition to a local logic unit L, the local dialogue unit has a speech recognition unit ASRF (Automatic Speech Recognition ASR), a speech understanding unit NLUF (Natural Language Understanding NLU) and a speech synthesis unit TTSF (Text-To-Speech Synthesizer TTS).

[0063] The local logic unit L ensures synchronization with the assigned dialog manager, the output of dialog outputs, and the execution of corresponding commands, and switches between online and offline modes depending on whether a data connection to the backend exists. The speech recognition unit ASRF converts a sequence of feature vectors into a sequence of words, the meaning of which is then determined through semantic interpretation in the speech understanding unit NLUF. Since dialog flows often involve querying information to execute an action, such as searching a media database to play a piece of music, it is necessary to determine the intent contained in the user's voice input. After determining the intent, parameters (slots) are extracted from the voice input to more precisely specify the intended query or action.

[0064] Additionally, a separate signal processing unit (not shown) can be provided, which converts the speech signals captured by one or more microphones into feature vectors. However, this signal processing unit can also be integrated into the automatic speech recognition unit. Because the local dialogue unit has a local speech recognition unit and a local speech understanding unit, it is ensured that speech input can continue to be processed even if the connection to the backend is interrupted.

[0065] The speech synthesis unit (TTS) is used to generate dialogue output in the form of artificial speech output from word sequences available locally or received from the backend. Furthermore, a storage unit M is provided, in which, among other things, the dialogue predictions received from the backend are stored.

[0066] The backend features a dialog manager (DM) and a dialog predictor (DP), which are used to manage dialogs in an online mode and predict future dialog progressions. A corresponding communication unit (KB) is provided in the backend for connection to the frontend, corresponding to the communication unit (KF) in the frontend. Furthermore, a speech recognition unit (ASRB), a speech understanding unit (NLUB), and / or a speech synthesis unit (TTSB) can also be present in the backend. Due to the high computing power and storage capacity available on the online server, these can offer more extensive functionality than the corresponding components in the frontend.

[0067] Such a backend system can, for example, be set up or made available by a vehicle manufacturer for the motor vehicles manufactured and distributed for the vehicle manufacturer.

[0068] To illustrate this, the following examples show a local dialogue unit integrated into a motor vehicle. The driver of the vehicle searches for music in their local music library and has already seen a list of answers displayed on the infotainment unit's display.

[0069] Possible dialogue continuations are now e.g. a) “Play line 2” (a generic input), b) “Play the album” (not generic, because it depends on the search result), or c) “no, I meant $InterpretLocalDB” (generic because it contains a variable for the artist in the local music collection).

[0070] In order to be understood by the speech understanding unit in the vehicle, the dialogue continuations must also contain the necessary filler information for the parameter slots of the carrier sentences in the speech understanding unit ('album' in (b)) or be filled locally (in (a), (c)). The filler information is then loaded into the local speech understanding unit before recognition. In cases where the user makes an input to play music, possible further continuations can include stopping, skipping, pausing, etc., the music output.

[0071] For cases where a new search is started with "no, I meant...", the predictive dialogue manager could send further dialogue continuations, e.g. d) “No, I meant Michael Headson”

[0072] Such a variant could be based on phonetic similarity to the original query, i.e., it would not be generic and could be based on typical confusions detected in the cloud. In contrast to the local interpreters (c), which are present in every case, the case where "Michael Headson" is not present must also be considered as a continuation of (d).

[0073] For the input examples mentioned, the actual dialog management logic and the possible response sentences must also be sent from the external dialog unit to the local dialog unit. The dialog management logic of a dialog step can then contain specific conditions, for example, that the name "Michael Headson" exists in the local database. The possible response sentences, which replace a more complex local language generation component, contain, as with the input sentences (ad), a mixture of variables for specific names, etc., for example: e) “Here are all the tracks from $InterpretLocalDB in your music collection” (generic continuation of (c). f) “Sorry, I can’t find anything by Michael Headson in your music collection, and I can’t search online right now.” (a continuation of (d) with the proviso that “Michael Headson” is not included in the local music collection).

[0074] The following example further illustrates the transition from online to offline mode. System S is initially in online mode, with user N entering the following voice command: (1) N: “Play something by Michael Jackson.”

[0075] The request to play music can be recognized by both the local dialog unit and the external dialog unit. The artist's name can also be recognized by both the local and external dialog unit, especially if the local media database contains music by the artist. However, it is also possible that the local dialog unit can only recognize the request but not the artist's name. In this case, however, the external dialog unit can recognize the artist's name in online mode.

[0076] After the desired media playback has been recognized by the dialogue manager in the external dialogue unit, the user is asked to select one of several alternatives, either through a voice output or a display on a graphical user interface: (2) S: “Please select one of the following options.”

[0077] In addition, the available alternatives, either from the local media database or from offers from online media content service providers, are offered on the graphical user interface: 1. Michael Jackson - Thriller 2. Michael Jackson - Best of Compilation 3. Michael Jackson - Stream from MusicStreamer.com

[0078] The user can respond to this in various ways, such as (3) N: “Play song” / “Play album / Online” or: (4) N: ‘Line 2’ or: (5) N: “No, I meant Michael Headson.”

[0079] The last input discards all three offered options and re-enters the original query, with the parameter or slot ARTIST_NAME now containing the new artist name (“Michael Headson”).

[0080] Similar dialogs can also be found in other domains, such as navigation, for example, when the user searches for points-of-interest POls that may or may not be present in the embedded database. Due to the large number of possible POls, embedded voice systems generally have only limited options for finding them, in contrast to online systems, which are capable of finding almost all POls.

[0081] To prepare for the interruption of a connection during an ongoing dialog, the dialog predictor in the backend creates a hierarchical data structure of a dialog history in the form of one or more branching "trees," which cover the dialog states for the most likely future dialog scenarios. The dialog continuations are to be understood as instantiations of a general dialog manager, i.e., no generic code is sent to the local dialog unit, but rather concrete forms for the currently conducted dialog, including, for example, proper names. An example of the above dialog is shown in Fig. 4 shown.

[0082] The dialogue predictor can also generate dialogue continuations for misunderstood user intents, for example, if the user didn't want to listen to music, but the speech recognition unit incorrectly interpreted this as such. Likewise, additional dialogue continuations can be generated for misunderstood parameters, such as an incorrectly recognized artist name. In the example above, the dialogue predictor would also search for artists similar to "Michael Jackson" or for user intents that are more frequently confused with "playing music," and generate branches for both.

[0083] As a result, and especially with complex topics or a large number of topics to be considered and a large number of dialogue steps or hierarchies, the generated tree structure can potentially be very extensive. It is therefore also possible to transmit only the dialog continuation that is statistically most likely (also known as the "happy path") to the local dialog unit, instead of the entire tree structure with all possible dialog states.

[0084] The dialog tree structures and the respective entities, such as media, POIs, navigation addresses, etc., can also be separated. This allows the backend to enrich the frontend's embedded databases with additional probable content, while the tree structures are generic and contain only entity types instead of explicit information.

[0085] Although it may be necessary for the dialog predictor to narrow its search space accordingly, particularly due to limited connection bandwidth and memory capacity of the frontend, since the tree structures essentially contain metadata in plain text, a large number of such tree structures can be stored in the frontend's memory, which can be updated by the backend within seconds.

[0086] The dialog predictor regularly updates the tree structure for a currently running dialog, for example, after each dialog step, or depending on the situation. The dialog predictor can also create the tree structure for dialogs that have not yet been initiated by the user but are likely to occur, or it can use existing tree structures for such dialogs.

[0087] For example, a query about nearby restaurants based on the current position of the vehicle, which can only be reliably solved online, is mentioned here: (6) N: “Find me an Italian restaurant”

[0088] Other search domains could include queries related to current news, for example (7) N: "Who won in the category 'Best Actress'?"

[0089] The likelihood that the latter dialogue will be initiated will increase significantly during a film awards ceremony, as well as shortly before and after it.

[0090] Likewise, possible dialogue trajectories or the corresponding tree structures can be created based on the user's previous dialogue trajectories using personalized prediction models. Finally, possible dialogue trajectories can be determined based on the previous dialogues of a large number of other users, whereby the dialogue models can be continuously updated.

[0091] The generated tree structures are sent from the backend to the frontend in online mode so that they can be used in offline mode.

[0092] If the local dialog unit loses connection to the backend, it immediately switches to offline mode. Normally, the frontend would now be limited to the embedded capabilities, so, for example, a response to the above search query (6) could be: (8) S: “Sorry, the system is currently in offline mode.”

[0093] Or: (9) S: “I’m sorry, I can’t search for POls at the moment”

[0094] However, since, according to the invention, a large number of dialog predictions were sent from the backend to the frontend while the connection between the backend and the frontend was still active, a likely appropriate, cached response can be available for the user.

[0095] For example, for the above query, a list of Italian restaurants can be available in the frontend and the following response can be output: (10) S: “Please select one of the following options”

[0096] For example, a graphical user interface displays three options: 1. Pizza Luigi 2. Pasta & more 3. Roma

[0097] If the names of these POls and their respective addresses were transmitted in advance from the backend, it would then be possible, for example, to use one of the three options as a destination for a navigation system.

[0098] A method according to the invention is particularly advantageously used for voice assistants in a motor vehicle. However, a method according to the invention can also be used in voice assistants of user devices such as mobile phones, tablets, portable computer systems such as wearables or smart home devices, etc. List of reference symbols 10 Capturing a dialog input 11 Transferring the dialog input 12 Receiving a future dialogue history 13 Saving the received dialogue history 14 Retrieving the saved dialogue history 15 Capturing dialog inputs and outputting dialog outputs 20 Receiving a dialog input 21 Calculating at least one future dialogue course 22 Transfer of at least one future dialogue history LD local dialogue unit ED external dialogue unit F Frontend B Backend MDB local media database NDB Navigation Database KF, KB communication unit ASRF, ASRB speech recognition unit NLUF, NLUB Language Comprehension Unit TTSF, TTSB speech synthesis unit L local logic unit M storage unit DM Dialog Manager DP Dialogue Predictor

Claims

[1] Method for operating a dialogue system (DS), wherein the dialogue system (DS) comprises a local dialogue unit (LD) for the recording of dialogue inputs and the output of dialogue outputs and an external dialogue unit (ED) connected to the local dialogue unit (LD) for data transmission for the analysis of recorded dialogue inputs and the determination of dialogue outputs based thereon, with the following steps carried out by the local dialogue unit: - reception (12) of at least one future probable dialogue course calculated by the external dialogue unit (ED) and consisting of dialogue inputs and / or dialogue outputs; - Saving (13) the received dialogue history; - Retrieving (14) the stored dialogue history in the event of an interruption of the data connection between the local dialogue unit and the external dialogue unit; and capturing dialogue inputs and / or outputting dialogue outputs (15) based on the retrieved dialogue history. [2] The method according to claim 1, wherein the following steps are additionally carried out: - capturing (10) a user's dialog input; - Transferring (11) the captured dialogue input to the external dialogue unit. [3] Method according to claim 2, wherein in an online mode, based on the dialog input detected by the local dialog unit and transmitted to the external dialog unit, a dialog output generated by the external dialog unit is received and output, and in an offline mode, the future probable dialog history received and stored by the external dialog unit is retrieved in the local dialog unit. [4] The method according to claim 3, wherein the dialogue history received from the external dialogue unit represents a probable continuation of the previous dialogue for the detected dialogue input and the previous dialogue is continued based on this dialogue history. [5] The method of claim 1, wherein the dialogue history is received before a dialogue has been started. [6] Method according to one of the preceding claims, wherein updated versions of the likely future dialogue course are received at regular intervals and stored in the local dialogue unit. [7] Method according to one of claims 1 to 5, wherein a future probable dialogue course is retrieved from the external dialogue unit and stored in the local dialogue unit when an interruption of the connection is anticipated. [8] Method for controlling a dialogue system (DS), wherein the dialogue system (DS) comprises a local dialogue unit (LD) for the recording of dialogue inputs and the output of dialogue outputs and an external dialogue unit (ED) connected to the local dialogue unit (LD) for data transmission for the analysis of recorded dialogue inputs and the determination of dialogue outputs based thereon, with the following steps carried out by the external dialogue unit: - Calculating (21) at least one likely future dialogue course consisting of dialogue inputs and / or dialogue outputs; and - Transferring (22) the calculated dialogue history to the local dialogue unit. [9] Method according to claim 8, wherein the following step is additionally carried out in the external dialogue unit: - receiving (20) a detected dialog input from the local dialog unit; and wherein calculating the dialog history is based on the received dialog input and the dialog history represents a likely continuation of the previous dialog. [10] Method according to claim 8, wherein the dialogue flow is calculated before a dialogue has started. [11] The method of claim 10, wherein the dialogue flow is calculated based on one or more of the following parameters: - the current location of the local dialogue unit, - the current local time of the local dialogue unit, - Information on current political, economic, social, cultural, sporting or other events, - information usually requested by the user, - information requested by other users. [12] Method according to one of the preceding claims, wherein the dialogue system is designed as a speech dialogue system in which dialogue inputs are made by speech recognition and interpretation of speech inputs of a user and / or dialogue outputs are made by speech synthesis. [13] The method of claim 12, wherein the calculated dialogue history includes, in addition to a probable continuation of the previous dialogue for a recognized dialogue input, also a probable continuation of the previous dialogue for a phonetically similar dialogue input. [14] Method according to one of the preceding claims, wherein the local dialogue unit is implemented in a vehicle and the external dialogue unit is implemented in a vehicle-external server which are connected via the Internet. [15] Method according to one of the preceding claims, wherein the calculated dialogue flow is designed in the form of a hierarchical tree structure. [16] System for operating and controlling a dialogue system (DS), which comprises a local dialogue unit (LD) for the recording of dialogue inputs and the output of dialogue outputs and an external dialogue unit (ED) connected to the local dialogue unit (LD) for data technology purposes for the analysis of recorded dialogue inputs and the determination of dialogue outputs based thereon and is set up to carry out a method according to one of the preceding claims.

Images