A context intention continuation method for in-vehicle navigation voice

Abstract

This application relates to the field of voice navigation technology and discloses a method for contextual intent continuation in vehicle navigation voice. This method collects navigation task stage information corresponding to the current navigation session, historical voice interaction results, the most recent navigation feedback result, and the current route association status; determines the effective context range corresponding to the current round of voice based on the navigation task stage information; performs semantic parsing on the current round of voice input, determines candidate object reference relationships within the candidate object range limited by the most recent navigation feedback result, and generates continuation semantic results by combining the effective context range results and the current route association status; determines the target continuation intent and corresponding intent completion results based on the navigation task stage information, candidate object reference relationships, and continuation semantic results, and then generates a navigation execution request or a voice clarification request; this method can improve the accuracy and consistency of multi-round continuation processing of vehicle navigation voice.

Description

Technical Field

[0001] This application relates to the field of voice navigation technology, and more specifically, to a method for contextual intent continuation in vehicle navigation voice. Background Technology

[0002] Existing in-vehicle navigation voice interaction technologies typically support processes such as destination retrieval, candidate result announcement, route planning, route execution, and service query along the way. In actual use, the subsequent voice responses given by users after the previous round of navigation feedback are often not a complete restatement of the navigation needs, but rather a continuation of the previous interaction content using ordinal expressions, indicative expressions, abbreviated expressions, or constraint adjustment expressions. For example, after the system announces multiple candidate destinations, the user only says "the second one"; after the route has been generated, the user only says "don't use the highway"; during navigation execution, the user says "find a charging station ahead" or "change that one to a closer one." Whether such subsequent voice responses can be processed correctly depends not only on the current round of voice responses themselves, but also on which of the previous interaction contents are still valid, which have become invalid, which still need further confirmation, and whether the current navigation processing is in the candidate object feedback stage, route planning stage, route execution stage, or navigation broadcast interruption recovery stage.

[0003] In existing technologies, one type of solution focuses on returning candidate objects after recognizing the user's voice and waiting for the user to continue selecting; another type focuses on inheriting entity information, slot information, or intent information from the previous round in multi-turn dialogue processing; and yet another type combines contextual information, location status, and device status for scene recognition and skill arbitration. Although the above solutions can support continuous voice interaction to a certain extent, for application scenarios such as in-vehicle navigation with clear task progression stages and route status changes, they usually lack a dedicated context processing mechanism for navigation task stages. It is difficult to perform stable differentiation processing of historical interaction content based on the current stage of the navigation task. In other words, existing technologies usually cannot accurately determine, before the current round of voice processing, which historical interaction content should continue to participate in the current round of semantic interpretation, which historical interaction content should be excluded from the current round of processing, and which historical interaction content cannot be directly inherited and needs to be further clarified in subsequent objects or scopes before participating in processing.

[0004] In this scenario, when the current round of speech may simultaneously correspond to candidate object selection, destination change, route constraint adjustment, service retrieval along the way, or navigation instruction query, existing technologies are prone to carrying over historical content that has become obsolete into the current round of processing, or prematurely deleting historical content that should still be retained. This leads to unclear object reference scope, semantic deviation in continuation interpretation, and confusion in the direction of continuation processing in the current round of speech. For example, the candidate object set may have changed, but the content of the previous round of candidate objects may still be used to interpret "the second one"; the route status may have changed, but the original constraints may still be used to interpret "don't use the highway"; the range of services along the way may have changed, but the original search range may still be used to interpret "find a charging station". These situations prevent the current round of speech from being accurately interpreted in the actual operating state of the current navigation session, thereby affecting the accuracy, consistency, and continuity of subsequent navigation execution.

[0005] Therefore, how to accurately determine the inheritable context range of the current round of speech based on the current navigation task stage and the current route status, and complete the determination of candidate object reference relationships, the generation of continuation semantics, and the diversion processing of continuation intent within this range, so as to generate the correct navigation execution result or speech clarification result, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of existing technologies and achieve the above objectives, this application provides the following technical solution: By identifying the navigation task stage of the current navigation session, determining the effective range of historical interaction content, judging the reference relationship of candidate objects, generating continuation semantics, and processing continuation intent diversion and completion, the interpretation process of the current round of speech is simultaneously constrained by the current navigation task stage, the still valid historical interaction content, and the candidate object range limited by the most recent navigation feedback result. This allows for the inclusion of omitted expressions, referential expressions, ordinal expressions, and constraint adjustment expressions into the same continuation processing chain. This application discloses a method for contextual intent continuation in vehicle navigation voice commands, including: Collect navigation task stage information corresponding to the current navigation session, historical voice interaction results, the most recent navigation feedback result, and the current route association status; Based on the navigation task stage information, the historical voice interaction results, the most recent navigation feedback results, and the historical interaction content corresponding to the current route association status are classified as retained, cleared, and confirmed to determine the inheritable context range corresponding to the current round of voice and generate the context effective range result. Semantic parsing is performed on the current round of voice input. Within the candidate object range defined by the most recent navigation feedback result, the reference relationship of the candidate object is determined, and the continuation semantic result is generated by combining the effective range result of the context and the current route association status. Based on the navigation task stage information, candidate object reference relationships and continuation semantic results, the continuation intent corresponding to the current round of speech is divided, judged and completed to determine the target continuation intent and the corresponding intent completion result; Based on the target continuation intent and the corresponding intent completion result, generate a navigation execution request or a voice clarification request.

[0007] Furthermore, methods for generating context-validated range results include: Based on navigation task stage information, the most recent navigation feedback result, and the current route association status, historical interaction content is classified into retention, clearing, and confirmation categories, resulting in directly inherited content, excluded content, and content pending confirmation. Merge directly inherited content, excluded content, and content to be confirmed, and determine the scope of inheritable context based on the merging results, generating the valid scope of context results.

[0008] By performing retention, clearing, and confirmation classifications on historical interaction content, we can distinguish between historical content that can still participate in the semantic interpretation of the current round, historical content that should be removed from the current round of processing, and historical content that needs to be confirmed in subsequent rounds, so as to provide the current round of speech with a stage-constrained inheritable context scope.

[0009] Furthermore, methods for determining candidate object reference relationships include: The range of candidate objects is determined based on the feedback type of the most recent navigation feedback result; The current round of speech input is parsed to obtain the ordinal segment, explicit object segment, and referential segment; Within the candidate object range, each fragment is compared with the content of each candidate object in the candidate object range in the order of ordinal fragment, explicit object fragment, and referential fragment. The candidate object reference relationship is determined when only one candidate object content corresponds to the smallest level ordinal number.

[0010] By performing object correspondence judgment within the candidate object range defined by the most recent navigation feedback result, the ordinal expression, explicit object expression, and pointer expression in the current round of speech are constrained within the current set of valid candidate objects, so as to suppress extended interpretations that deviate from the current feedback result.

[0011] Furthermore, methods for generating continuation semantic results include: Read the candidate object reference relationships, the context scope results, and the current route association status; extract the directly inherited results, excluded results, and pending confirmation results from the context scope results. The content corresponding to the current round of speech input and the content of the candidate object reference relationship in the directly inherited results will be incorporated into the semantic interpretation scope of the current round, the content corresponding to the excluded results will be removed, and the results to be confirmed will be retained; Based on the semantic interpretation scope of the current round, the results to be confirmed, and the current route association status, a continuation semantic result is generated.

[0012] By incorporating directly inherited results into the semantic interpretation scope of the current round, removing excluded results and retaining results awaiting confirmation, the semantic interpretation of the current round of speech is completed by combining the currently inheritable historical content and the current route association state, so as to form a continuation semantic basis for subsequent intent-based traffic splitting processing.

[0013] Furthermore, methods for defining target continuation intentions include: Based on navigation task stage information, candidate object reference relationships, continuation semantic results, current route association status, and direct inheritance results and pending confirmation results in the context's effective range, determine the candidate continuation intent category; The target continuation intent is determined based on the candidate continuation intent category, the results to be confirmed, and the results corresponding to multiple candidate objects and the results without corresponding objects in the candidate object reference relationship.

[0014] By combining navigation task stage information, candidate object reference relationships, continuation semantic results, and direct inheritance and pending confirmation results in the effective range of context, the continuation processing direction corresponding to the current round of speech is constrained and determined, so as to distinguish different continuation processing directions such as candidate object selection, destination change, route constraint adjustment, service retrieval along the way, and navigation instruction query.

[0015] Furthermore, methods for determining the corresponding intent completion results include: When forming a target continuation intent, the corresponding intent completion result is determined based on the target continuation intent, the continuation semantic result, and the direct inheritance result in the context's valid scope result. When no target continuation intent is formed, the corresponding intent completion result is determined based on the candidate continuation intent category, the unconfirmed results in the context's valid scope, and the corresponding results and no corresponding results of multiple candidate objects in the candidate object reference relationship.

[0016] By adding the necessary content when the target continuation intention has been formed, and retaining the content to be confirmed and the content that has not been uniquely determined when the target continuation intention has not been formed, the clear parts and the parts that still need to be clarified in the current round of speech can be processed separately.

[0017] Furthermore, the methods for generating navigation execution requests include: When the target continuation intent and intent completion result meet the preset execution integrity conditions, the request category is determined based on the target continuation intent; Extract the execution object content and execution scope content from the intent completion results based on the request category; Based on the request type, the content of the object to be executed, and the scope of execution, a navigation execution request is generated.

[0018] By extracting the content of the execution object and the execution scope based on the target continuation intent and the intent completion result, the continuation processing result is transformed into request content that can be directly received by the navigation engine, so that the semantic continuation result can be implemented in the specific navigation execution process.

[0019] Furthermore, methods for generating voice clarification requests include: Based on the pending confirmation results in the context's valid range results, the corresponding results of multiple candidate objects in the candidate object reference relationship, and the results without corresponding results, or the content in the minimum execution content corresponding to the target continuation intent that does not contain the intent completion result, determine the content of the query object; Generate a voice clarification request based on the content of the inquiry.

[0020] The content of the inquiry object is determined by the results to be confirmed, the results corresponding to multiple candidate objects, the content of the minimum execution content corresponding to the target continuation intention that does not contain the intention completion result, so as to supplement the confirmation of the execution object, execution scope or execution content that has not yet been uniquely determined in the current round of speech.

[0021] Furthermore, methods for determining the scope of inheritable contexts based on the merging results include: Based on the excluded content, the directly inherited content and the content to be confirmed are removed; Based on the content to be confirmed, the content that is not distinguished in the current round of voice input will be transferred to the content to be confirmed. The scope of inheritable context is determined based on the directly inherited content after the removal process and the content to be confirmed after the transfer process.

[0022] By performing removal and transfer processing on directly inherited content, excluded content, and content to be confirmed, the directly callable content and the content to be confirmed are further separated, so as to stably limit the inheritable context boundary of the current round of speech.

[0023] Furthermore, the method for comparing each fragment with the content of each candidate object in the candidate object range includes: Based on the matching relationship between each segment and the content of each candidate object, multiple corresponding levels are determined for each candidate object content; the multiple corresponding levels are compared according to a preset sorting order. When the corresponding level of only one candidate object is in a lower position than the corresponding level of other candidate objects in the preset sorting order, the candidate object reference relationship is determined. If the corresponding level of two or more candidate objects is in the same position in the preset sorting order and is lower than the corresponding level of other candidate objects, multiple candidate object results are retained.

[0024] By assigning multiple corresponding levels to the content of each candidate object and comparing them in a preset order, the process of determining the reference relationship of candidate objects has a unified comparison basis, so as to output different subsequent processing results in the two cases of unique correspondence and multiple object parallel correspondence.

[0025] Compared with related technologies, this application has the following advantages: This application collects navigation task stage information corresponding to the current navigation session, historical voice interaction results, the most recent navigation feedback results, and the current route association status, so that the subsequent processing of the current round of voice is based on the same navigation session, the same time base, and the same navigation operation status. This ensures from the source that the data source on which the subsequent context processing is based is consistent and the processing object is clear, providing a stable input basis for subsequent continuation processing.

[0026] Based on navigation task stage information, this application performs retention classification, clearing classification, and confirmation classification on historical interaction content, and generates a context validity range result. This prevents historical interaction content from being indiscriminately included in the current round of semantic interpretation. Instead, it distinguishes historical interaction content into directly inheritable content, excluded content, and pending confirmation content according to the current navigation task stage and current route status. This reduces the interference of invalid historical content on the current round of semantic interpretation and improves the accuracy of context inheritance processing.

[0027] This application determines the reference relationship of candidate objects within the scope of the candidate objects limited by the most recent navigation feedback result, and generates the continuation semantic result by combining the effective scope result of the context and the current route association state. This ensures that the ordinal expression, referential expression and ellipsis expression in the current round of speech are all interpreted within the scope of the current effective objects, avoiding the object reference being extended and interpreted outside the current navigation feedback result, thereby improving the accuracy of candidate object reference determination and the pertinence of continuation semantic generation.

[0028] Based on navigation task stage information, candidate object reference relationships, and continuation semantic results, this application performs flow determination and completion on the continuation intent corresponding to the current round of speech. It can distinguish the current round of speech into different continuation processing directions such as candidate object selection, destination change, route constraint adjustment, service retrieval along the way, navigation instruction query, or navigation cancellation. When necessary, it combines directly inherited content to complete the execution content, thereby improving the accuracy of continuation processing direction determination and the coherence of multi-round navigation voice interaction processing chain.

[0029] This application generates a navigation execution request based on the target continuation intent and the corresponding intent completion result, or generates a voice clarification request when the execution object, execution scope or execution content is not yet uniquely determined. This ensures that continuation results that are not uniquely determined are not directly sent to the navigation execution stage, but are first further supplemented through voice clarification, thereby improving the reliability of navigation execution results and enhancing the consistency and stability of in-vehicle navigation voice continuation processing.

[0030] In summary, this application addresses the main problem in the process of voice continuation in vehicle navigation: the unclear inheritance boundary of historical content after the current navigation task stage changes, which leads to inaccurate object references and confusion of continuation intentions. It constructs a continuous processing chain from determining the effective scope of the context, determining the candidate object reference relationship, generating continuation semantics, to the diversion of continuation intentions and execution output, which can improve the accuracy, stability and interactive continuity of voice continuation processing in vehicle navigation. Attached Figure Description

[0031] Figure 1 A schematic diagram of the overall process for a contextual intent continuation method for in-vehicle navigation voice provided in this application; Figure 2 The data processing flowchart for generating the valid range of context results provided in this application; Figure 3 A flowchart illustrating the candidate object reference relationships, target continuation intent, and data processing generated for this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] Example 1: Please see Figure 1 As shown, this embodiment provides a method for contextual intent continuation in vehicle navigation voice, including the following steps: The reason for collecting navigation task stage information, historical voice interaction results, the most recent navigation feedback result, and the current route association status corresponding to the current navigation session is that, for the current round of voice input, a basic input result that can be directly called by subsequent steps under the same current navigation session is first formed. This ensures that subsequent steps have a unified data source and a unified time correspondence when performing retention classification, clearing classification, and confirmation classification on historical interaction content. Furthermore, it ensures that subsequent steps, when determining candidate object reference relationships, generating continuation semantic results, and determining target continuation intent, are all based on the actual operating state of the current navigation session. In some implementations, the implementation steps include: Step 101: The processing object is the current navigation session; specifically, the current running identifier, the current voice interaction round identifier, and the current voice input time of the in-vehicle navigation application are read; navigation voice interaction rounds that are consistent with the current running identifier before the current voice input time are merged into the same session range to determine the current navigation session.

[0034] The current navigation session is used to limit the scope of subsequent reading objects; all subsequent navigation task stage information, historical voice interaction results, the most recent navigation feedback results, and the current route association status are generated with the current navigation session as the reading scope.

[0035] Step 102: The object being processed is navigation task stage information; specifically, within the current navigation session, the feedback type of the most recent navigation feedback result before the current voice input time is read, whether a route result already exists, whether a route recalculation status exists, and whether a broadcast interruption status exists; the navigation task stage information is determined according to preset stage determination rules.

[0036] The input objects for the preset stage determination rules are the feedback type of the most recent navigation feedback result, whether a route result already exists, whether a route recalculation status exists, and whether a broadcast interruption status exists. The preset stage determination rules are formed based on the differences in the status presented by each stage in the navigation processing flow. The preset stage determination rules are invoked by executing stage judgments in a preset order. The output of the preset stage determination rules is navigation task stage information.

[0037] In some implementations, the preset sequence includes: when the feedback type of the most recent navigation feedback result is candidate object set feedback and no unique navigation object is currently determined, the navigation task stage information is determined to be the candidate object feedback stage; when the unique navigation object has been determined and there is a route recalculation state, the navigation task stage information is determined to be the route planning stage; when there is a route result and there is no broadcast interruption state, the navigation task stage information is determined to be the route execution stage; when there is a route result and the broadcast is detected to be interrupted, interrupted, or occupied by other voice interactions within a preset time range, the navigation task stage information is determined to be the navigation broadcast interruption recovery stage.

[0038] This step generates navigation task phase information, which can then be used in subsequent steps.

[0039] Step 103: The processing object is the historical voice interaction results; specifically, within the current navigation session, the preset rounds of voice interaction content before the current voice input time are read in chronological order, and speech recognition and semantic correction are performed on each round of voice interaction content to form historical voice interaction results.

[0040] The input object for semantic integration is the speech recognition text of each round of voice interaction; the semantic integration is based on the preset semantic integration rules in the navigation domain; the preset semantic integration rules in the navigation domain are invoked by recognizing each of the location name words, ordinal words, pronouns, route constraint words, service words along the way, description query words, and cancellation words in the speech recognition text; the output of the preset semantic integration rules in the navigation domain is the object content and action content in the corresponding round of voice interaction.

[0041] In some implementations, the content of each round in the historical voice interaction results includes at least the voice recognition text of that round, the location content appearing in that round of voice, the candidate object sequence content appearing in that round of voice, the referential content appearing in that round of voice, the route constraint content appearing in that round of voice, the service content along the way appearing in that round of voice, and the action content appearing in that round of voice.

[0042] This step generates historical voice interaction results, which can be used by subsequent steps to generate contextual valid range results.

[0043] Step 104: The object being processed is the most recent navigation feedback result. Specifically, within the current navigation session, the navigation feedback content most recently returned by the navigation engine and broadcast or displayed by the vehicle terminal before the current voice input occurs is read. The navigation feedback content is then categorized by type and content to form the most recent navigation feedback result.

[0044] Among them, type sorting is used to determine which type of feedback result the most recent navigation feedback result belongs to: candidate object set feedback result, route feedback result, service along the way feedback result, or navigation description feedback result; content sorting is used to arrange the feedback objects in the most recent navigation feedback result according to the output order of the navigation engine.

[0045] When the most recent navigation feedback result is a candidate object set feedback result, the content tidying should retain at least the output order, object name, location constraint content, and category constraint content for each candidate object; when the most recent navigation feedback result is a route feedback result, the content tidying should retain at least whether the route has been generated, the route constraint response content, and the route description content; when the most recent navigation feedback result is a service feedback result along the route, the content tidying should retain at least the output order, object name, location constraint content, and category constraint content for the service objects along the route.

[0046] This step generates the most recent navigation feedback result, which is used by subsequent steps to generate the context's effective scope and determine candidate object reference relationships.

[0047] Step 105: The processing object is the current route association status; specifically, within the current navigation session, the current vehicle position and the current route result are read; when the current route result exists, the current vehicle position is matched with the path point sequence or segment sequence corresponding to the current route result to determine whether the current vehicle position is within the path range corresponding to the current route result, and the route segment corresponding to the current vehicle position is determined; when the current route result does not exist, it is directly determined that the current route does not exist in the current route association status.

[0048] The current route association status is used to indicate whether the current route exists, whether the current vehicle position corresponds to the current route, and the positional relationship of the current vehicle position within the current route.

[0049] This step processes the current route association state, which is then used to generate the context's valid range, determine candidate object reference relationships, and generate navigation execution request calls.

[0050] Step 106: Link the navigation task stage information, historical voice interaction results, the most recent navigation feedback result, and the current route association status to the current voice input time, remove content whose occurrence time is later than the current voice input time, and retain content within the same current navigation session that is prior to the current voice input time; form the basic input result for the next step; the basic input result includes navigation task stage information, historical voice interaction results, the most recent navigation feedback result, and the current route association status.

[0051] Based on navigation task phase information, the historical interaction results, the most recent navigation feedback results, and the historical interaction content corresponding to the current route association status are classified into retention, clearing, and confirmation categories to determine the inheritable context range corresponding to the current round of voice input and generate a valid context range result. The purpose is to distinguish the historical interaction content in the current navigation session into directly inheritable content, content that should be excluded from the current round of processing, and content that needs to be further confirmed in conjunction with the current round of voice input before it can be inherited, so as to avoid indiscriminately incorporating all historical content into the current round of semantic parsing process. Among them, retention, clearing, and confirmation categories constitute one of the core processing actions of this application.

[0052] In some implementations, the steps for generating the context validity range result include: See Figure 2 As shown, in step 201, based on the historical voice interaction results, the most recent navigation feedback results, and the current route association status, extract the historical interaction content related to the current round of voice continuation in the current navigation session, and label the content type and source round for each historical interaction content.

[0053] The content type is used to characterize the content category of the current item's historical interaction content in the current navigation session; in some implementations, the content type includes at least one of destination content, candidate object content, route constraint content, along-the-way service content, and description query content.

[0054] The source round is used to represent the position where the current item's historical interaction content is generated in the current navigation session; the source round is the round number corresponding to each round of voice interaction in the current navigation session according to the order of occurrence; the larger the source round number, the later the current item's historical interaction content is generated in the current navigation session.

[0055] When historical interaction content originates from historical voice interaction results, the round number of the corresponding voice interaction round is marked as the source round; when historical interaction content originates from the most recent navigation feedback result, the round number of the voice interaction round that triggered the most recent navigation feedback result is marked as the source round.

[0056] In some implementations, the destination content includes at least a standardized location name and location-specific content; the candidate object content includes at least an object name, output order, location-specific content, and category-specific content; the route constraint content includes at least a constraint type, constraint direction, and object of application; the services along the route content includes at least a service category and scope of application; and the query description content includes at least the query object content.

[0057] Among them, the constraint type is used to characterize the restriction category corresponding to the route constraint content; in some implementations, the constraint type includes at least one of avoiding highways, prioritizing short distances, avoiding congestion, and avoiding tolls.

[0058] The constraint direction is used to characterize the direction in which the route constraint content affects the constraint type; in some implementations, the constraint direction includes at least one of adding constraints, deleting constraints, and replacing constraints.

[0059] The target content is used to characterize the object on which the route constraint content applies; in some implementations, the target content includes at least one of the current destination content, the current route, or the target route.

[0060] The service category content is used to characterize the types of services corresponding to the services along the route; in some implementations, the service category content includes at least one of refueling, charging, parking and catering.

[0061] The scope of application is used to characterize the search range corresponding to the service content along the route; in some implementations, the scope of application includes at least one of the current route, the target route, the vicinity of the current location, and the vicinity of the target location.

[0062] The query object content is used to characterize and describe the query object corresponding to the query content; in some implementations, the query object content includes at least one of the following: remaining distance, remaining time, next navigation action, and current route status.

[0063] This step generates historical interactive content with content type, source round, and corresponding content, which can be called by steps 202 to 206.

[0064] Step 202: Read the historical interaction content with content type and source round generated in step 201, and process each historical interaction content in descending order of source round number; for the current historical interaction content, read the navigation task stage information, the most recent navigation feedback result and the current route association status in sequence, and execute the stage corresponding judgment, subsequent replacement judgment, route role judgment and unique pointing judgment in a fixed order to determine whether the current historical interaction content enters the retention category, clears the category or confirms the category.

[0065] The specific methods for determining the corresponding stage include: Read the content type of the current item's historical interaction content and the set of content types allowed to participate in the current round of voice processing corresponding to the navigation task stage information; compare the content type of the current item's historical interaction content with the set of content types allowed to participate in the current round of voice processing item by item; if the content type of the current item's historical interaction content belongs to the set of content types allowed to participate in the current round of voice processing, determine that the current item's historical interaction content has passed the stage correspondence judgment; if the content type of the current item's historical interaction content does not belong to the set of content types allowed to participate in the current round of voice processing, determine that the current item's historical interaction content has not passed the stage correspondence judgment, and transfer the current item's historical interaction content to the clearing category.

[0066] The specific methods for subsequent replacement judgment include: Once the current historical interaction content passes the stage correspondence judgment, the content type, source round, and target content of the current historical interaction content are read. Within the current navigation session, other historical interaction content with a source round number greater than the current item's source round number and the same content type as the current historical interaction content is searched. For the found other historical interaction content, its target content is compared to see if it matches the target content of the current historical interaction content. If other historical interaction content with a source round number greater than the current item's source round number, the same content type, and the same target content exists, and this other historical interaction content constitutes reverse overwriting, similar replacement, or scope overwriting of the current historical interaction content, it is determined that the current historical interaction content has been replaced, and the current historical interaction content is moved to the clear category. If no other historical interaction content constituting reverse overwriting, similar replacement, or scope overwriting of the current historical interaction content is found, it is determined that the current historical interaction content has not been replaced.

[0067] Among them, reverse overlay is used to indicate that the content type of the subsequent historical interaction content is the same as that of the current historical interaction content, the content of the target is the same, and the direction of action is opposite; similar replacement is used to indicate that the content type of the subsequent historical interaction content is the same as that of the current historical interaction content, the content of the target is the same, and the new content replaces the original content; range overlay is used to indicate that the scope of the subsequent historical interaction content includes the scope of the current historical interaction content.

[0068] The specific methods for determining the role of a route include: If the current historical interaction content has not been replaced, read the content type of the current historical interaction content, and read the current route existence result, the route segment corresponding to the current location, and the current destination content from the current route association status; if the current route existence result indicates that there is no current route, read the target content or scope content from the current historical interaction content; if the target content or scope content requires the current route to exist or the target route to exist, determine that the current historical interaction content no longer has an effect in the current round of speech processing, and transfer the current historical interaction content to the clearing category; if the current route existence result indicates that there is a current route, apply route constraint content. The system compares whether the content of the target object corresponds to the current route or the current destination, and compares whether its constraint type will still affect the recalculation or adjustment of the current route. For service content along the route, it compares whether the content of its scope corresponds to the current route, the target route, the vicinity of the current location, or the vicinity of the target location. When the comparison results show that the current historical interaction content still corresponds to the current route or the target route, it determines that the current historical interaction content continues to play a role. When the comparison results show that the current historical interaction content no longer corresponds to the current route and no longer corresponds to the target route, it determines that the current historical interaction content no longer plays a role, and transfers the current historical interaction content to the clearing category.

[0069] The specific methods for determining the unique pointer include: When the current historical interaction content continues to be effective, the scope of subsequent processing objects corresponding to the current historical interaction content is read; for destination content, it is determined whether only one destination content is retained in the current navigation session; for candidate object content, it is determined whether only one candidate object corresponds to the current historical interaction content within the candidate object scope limited by the most recent navigation feedback result; for route constraint content, it is determined whether the current historical interaction content only applies to one destination content, one current route, or one target route; for along-the-way service content, it is determined whether the current historical interaction content corresponds to only one service category content and only one scope content; when the current historical interaction content corresponds to only one subsequent processing object, it is determined that the current historical interaction content passes the unique pointing judgment and is transferred to the retention category; when the current historical interaction content corresponds to two or more subsequent processing objects, it is determined that the current historical interaction content does not pass the unique pointing judgment and is transferred to the confirmation category.

[0070] In some implementations, the execution order of stage correspondence judgment, subsequent replacement judgment, route function judgment, and unique pointing judgment is fixed as follows: first, stage correspondence judgment is executed, then subsequent replacement judgment is executed, then route function judgment is executed, and finally unique pointing judgment is executed; when the previous judgment is sufficient to determine that the current historical interaction content enters the clearing category, the execution of subsequent judgments is stopped; when none of the previous three judgments cause the current historical interaction content to enter the clearing category, the unique pointing judgment is executed to determine whether the current historical interaction content enters the retention category or the confirmation category.

[0071] This step generates a classification result for each historical interaction, which is then used in steps 203 to 205.

[0072] Step 203: Determine the retention category for historical interaction content. Specifically, read the historical interaction content with content type and source round generated in step 201, and perform retention category determination item by item according to the source round number from largest to smallest. For each historical interaction content, read the navigation task stage information, the most recent navigation feedback result, and the current route association status in sequence, and perform the judgment in the following order: first perform content type matching judgment, then perform same type replacement judgment, then perform route continuation judgment, and finally perform unique object correspondence judgment.

[0073] Among them, the content type matching judgment is used to determine whether the content type of the current item's historical interaction content belongs to the content type that is allowed to continue to participate in the current round of voice processing under the current navigation task stage; the same type replacement judgment is used to determine whether the current item's historical interaction content has been replaced by historical interaction content with a source round number greater than the current item's source round number, the same content type, and the same target; the route continuation judgment is used to determine whether the current item's historical interaction content still has an effect on the current route or the target route under the current route association state; and the unique object correspondence judgment is used to determine whether the current item's historical interaction content corresponds to only one subsequent processing object in the current navigation session.

[0074] In some implementations, the specific methods for content type matching judgment include: when the navigation task stage information is in the candidate object feedback stage, only the candidate object content and the destination content corresponding to the candidate object content are allowed to continue participating in the current round of voice processing; when the navigation task stage information is in the route planning stage, only the current destination content and the currently effective route constraint content are allowed to continue participating in the current round of voice processing; when the navigation task stage information is in the route execution stage, only the current destination content, the currently effective route constraint content, and the service content along the way corresponding to the current route's associated state are allowed to continue participating in the current round of voice processing; when the navigation task stage information is in the navigation broadcast interruption recovery stage, only the current destination content, the currently effective route constraint content, the service content along the way corresponding to the current route's associated state, and the description query content corresponding to the current route description are allowed to continue participating in the current round of voice processing.

[0075] In some implementations, the specific methods for determining the same type of replacement include: For destination content, read the normalized location name in the current project's destination content, and search for other historical interaction content in the current navigation session whose source round number is greater than the current item's source round number and whose content type is destination content; when the normalized location name in the found destination content is different from the normalized location name in the current project's destination content, and the found destination content has been received by the navigation engine as a new navigation object, it is determined that the current project's destination content has been replaced. For route constraint content, read the constraint type, constraint direction, and target content of the current item's route constraint content, and search for other historical interaction content within the current navigation session that has a source round number greater than the current item's source round number and whose content type is route constraint content; when the found route constraint content has the same target content as the current item's route constraint content, the same constraint type, and opposite constraint direction, or when the found route constraint content has already covered the restriction range corresponding to the current item's route constraint content, it is determined that the current item's route constraint content has been replaced; For the service content along the way, read the service category content and scope content of the current item's service content along the way, and search for other historical interaction content in the current navigation session whose source round number is greater than the current item's source round number and whose content type is service content along the way; when the found service content along the way is the same as the service category content of the current item's service content along the way, and the scope content corresponding to the found service content along the way has been covered by the scope content corresponding to the current item's service content along the way, it is determined that the current item's service content along the way has been replaced; For candidate content, read the identifier of the most recent navigation feedback result to which the current candidate content belongs, and compare it with the identifier of the most recent navigation feedback result before the current round of voice input; when the two are inconsistent, determine that the current candidate content will no longer be used as directly inheritable content to continue participating in the current round of voice processing.

[0076] In some implementations, the specific methods for determining the continuation of the route include: When the current route association status indicates that no route exists, read the content type of the current item's historical interaction content; when the content type is route constraint content, read the target content of the current item's route constraint content; when the target content is the current route or the target route, determine that the current item's route constraint content will no longer be effective in the current round of speech processing; when the content type is along-the-way service content, read the scope content of the current item's along-the-way service content; when the scope content is the current route or the target route, determine that the current item's along-the-way service content will no longer be effective in the current round of speech processing.

[0077] When the current route association status indicates that a route exists, read the content type from the current item's historical interaction content; when the content type is route constraint content, read the constraint type, constraint direction, and target content from the current item's route constraint content, and read the current route existence result, the route segment corresponding to the current location, and the current destination content from the current route association status; when the target content is the current route or the current destination content, and the restriction content corresponding to the constraint type still affects the recalculation or adjustment of the current route, determine that the current item's route constraint content continues to take effect; when the target content no longer corresponds to the current route and the current destination content, or the restriction content corresponding to the constraint type is already included in the current route result, and the current round of voice input does not provide new, deleted, or replaced restrictions of the same constraint type again, determine that the current item's route constraint content no longer takes effect.

[0078] When the content type is "service along the route," the service category and scope content of the current item's service along the route are read, and the current route existence result, the route segment corresponding to the current location, and the current destination content are read from the current route association status. When the scope content is the current route, and the current route existence result indicates that a route exists, the current item's service along the route is determined to continue to function. When the scope content is the target route, and the navigation task stage information is in the route planning stage or route execution stage, and the current destination content is still retained in the directly inherited content, the current item's service along the route is determined to continue to function. When the scope content does not correspond to the current route, nor to the target route that still needs to be generated or is still being executed, the current item's service along the route is determined to cease to function.

[0079] In some implementations, the specific methods for determining the unique object correspondence include: For destination content, determine whether there are two or more destination contents that are simultaneously in the pending confirmation content in the current navigation session; if there are no two or more pending confirmation destination contents, determine that the destination content of the current project uniquely corresponds to the subsequent processing object; For candidate object content, determine whether the current item candidate object content corresponds only to one candidate object content within the candidate object range limited by the most recent navigation feedback result; if it corresponds only to one candidate object content, determine that the current item candidate object content uniquely corresponds to the subsequent processing object; For route constraint content, determine whether the current item's route constraint content applies only to a current destination, a current route, or a target route; if it applies only to a single object, determine that the current item's route constraint content uniquely corresponds to the subsequent processing object; For the service content along the route, determine whether the current item's service content along the route corresponds to only one type of service content and only one scope of content; if it corresponds to only one type of service content and only one scope of content, determine that the current item's service content along the route uniquely corresponds to the subsequent processing object.

[0080] When a historical interaction is determined to have been replaced by content type matching, not replaced by the same type, and continues to function through the route, and is also determined by unique object correspondence, the current historical interaction will be classified as retained. After processing in this step, the directly inherited content corresponding to the retained category will be formed, which can be called by steps 206, 304, 403 and 501.

[0081] Step 204: Determine the clearing category for historical interaction content; specifically, read the historical interaction content generated in step 201, which includes content type and source round, and perform clearing category determination item by item according to the source round number from largest to smallest; when the current historical interaction content meets any of the following conditions during the determination process in step 203, the current historical interaction content will be cleared and categorized: In the first scenario, the current item's historical interaction content failed the content type matching judgment. In the second scenario, the current item's historical interaction content is determined in the same type replacement judgment to have been replaced by historical interaction content with a source round number greater than the current item's source round number, the same content type, and the same target. The third scenario is that the current item's historical interaction content does not continue to function through the route judgment; In the fourth scenario, although the historical interaction content of the current item has not been replaced, the candidate object set to which the current item's candidate object content belongs has been replaced by the new, most recent navigation feedback result, and the navigation task stage information is no longer the candidate object feedback stage.

[0082] In some implementations, when the navigation task phase information has moved from the candidate object feedback phase to the route execution phase, the candidate object content in the previous candidate object set is cleared and categorized; when the route constraint content with a larger source round number has replaced the same type of route constraint content with a smaller source round number, the replaced route constraint content is cleared and categorized; when the current route association status indicates that there is no route, and the scope of a certain service content along the route is the current route or the target route, the current service content along the route is cleared and categorized; after this step, the excluded content corresponding to the clearing and categorization is formed and called by steps 206 and 304.

[0083] Step 205: Confirmation and classification determination of historical interaction content; specifically, read the historical interaction content with content type and source round formed in step 201, and perform confirmation and classification determination item by item according to the source round number from largest to smallest; when the current historical interaction content has been determined to have been replaced by the content type matching determination in step 203, has not been replaced by the same type determination, and has been determined to continue to act by the route determination, but has not been determined to have a unique object correspondence determination, the current historical interaction content is confirmed and classified.

[0084] In some implementations, failure to determine unique object correspondence includes at least one of the following situations: In the first scenario, there are more than two candidate objects within the range of the most recent navigation feedback result, and the current item's historical interaction content can only determine that it affects one of the two or more candidate objects, but it is not possible to determine which one it is based solely on the current item's historical interaction content. In the second scenario, there are two or more similar services along the way in the current navigation session, and the current item's historical interaction content only indicates that the same type of service along the way will continue to be processed, but the specific service object or specific scope of the content cannot be determined based solely on the current item's historical interaction content. The third scenario is that there are more than two route constraints that may still be effective in the current navigation session, and it is not yet certain which route constraint should be retained in the current historical interaction content. The fourth scenario is when there are two or more destinations to be confirmed in the current navigation session, and the current item's historical interaction content does not provide standardized location names, location restrictions, or category restrictions that are sufficient to distinguish each destination.

[0085] When the current item's historical interaction content is classified for confirmation, the current item's historical interaction content is retained as the content to be confirmed for continued use in subsequent steps, but the current item's historical interaction content is not directly incorporated into the object content in the current round of voice processing; after processing in this step, the content to be confirmed corresponding to the confirmation category is formed, which can be called by steps 206, 304, 305, 402, 403, 501 and 504.

[0086] Step 206: Determine the inheritable context scope and the valid context scope results; specifically, read the directly inherited content formed in step 203, the excluded content formed in step 204, and the content to be confirmed formed in step 205; first, perform content merging of the same type on the directly inherited content, excluded content, and content to be confirmed respectively, then determine the inheritable context scope corresponding to the current round of speech based on the merging results, and finally generate the valid context scope results based on the inheritable context scope.

[0087] The specific methods for merging directly inherited content, excluded content, and content awaiting confirmation into similar categories include: Each historical interaction is read from the directly inherited content, excluded content, and pending confirmation content, and the content type, source round, and target content are extracted for each historical interaction. When the content type is a service along the route, the service category and scope content are further extracted. When the content type is a route constraint, the constraint type, constraint direction, and target content are further extracted. When the content type is a candidate object, the object name, output order, location limitation content, and category limitation content are further extracted. When the content type is a destination, the standardized location name and location limitation content are further extracted.

[0088] Historical interaction content with the same content type, source round, and target content will be merged into a single merged content. If multiple historical interaction content before merging are identical in content type, source round, and target content, but differ only in expression form and have the same normalization result, these multiple historical interaction content with the same normalization result will be merged and retained as a single merged content. If multiple historical interaction content before merging are identical in content type and source round, but have different target content, or have the same target content but different constraint type, service category content, object name, or normalization location name, they will not be merged and will be retained as different merged content.

[0089] After the above processing, the directly inherited merged content corresponding to the directly inherited content, the excluded merged content corresponding to the excluded content, and the unconfirmed merged content corresponding to the unconfirmed content are respectively formed.

[0090] The specific methods for determining the inheritable context range corresponding to the current round of speech based on the merging results include: First, read the excluded and merged content, and use the content type, source round, and target content corresponding to each item in the excluded and merged content as the removal conditions; when there is content in the directly inherited merged content or the content to be confirmed that has the same content type, target content, and source round as a certain excluded and merged content, the directly inherited merged content or the content to be confirmed is removed from the candidate set of inheritable context scope and is no longer a component of the current round of speech interpretation scope.

[0091] After removing the excluded and merged content, the content to be confirmed and merged is read, and the content type, source round, and target content corresponding to each item in the content to be confirmed and merged are used as the conditions for retention. When there is a content type and target content that is the same as a certain item in the content to be confirmed and merged in the directly inherited content, and the current round of voice input has not yet provided enough content to eliminate ambiguity, the corresponding directly inherited and merged content is removed from the directly callable part, and the corresponding content is retained in the part to be confirmed and called.

[0092] After removing the excluded merged content and retaining the merged content to be confirmed, the remaining directly inherited merged content is determined as the directly callable part in the inheritable context scope, and the retained merged content to be confirmed is determined as the callable part to be confirmed in the inheritable context scope.

[0093] The directly callable portion represents historical content that can be directly incorporated into the current round of speech interpretation in subsequent steps; the pending confirmation portion represents historical content that can only be called in subsequent steps when the candidate object reference relationship is further clarified, the target continuation intention is further clarified, or a speech clarification request is generated; excluded and merged content does not enter the current round of speech interpretation; after the above processing, the inheritable context range corresponding to the current round of speech is formed.

[0094] The specific methods for generating the valid scope of the context based on the inheritable context scope include: Each item in the directly callable part is written into the direct inheritance result according to its content type, source round, and target content; each item in the pending confirmation part is written into the pending confirmation result according to its content type, source round, and target content; each item in the excluded merged content is written into the exclusion result according to its content type, source round, and target content; and the specific content composition related to the corresponding content type is simultaneously retained in the direct inheritance result, pending confirmation result, and exclusion result.

[0095] In some implementations, when the content type is route constraint content, the constraint type, constraint direction, and target content are simultaneously retained in the direct inheritance result, pending confirmation result, or exclusion result; when the content type is service content along the route, the service category content and scope content are simultaneously retained in the direct inheritance result, pending confirmation result, or exclusion result; when the content type is candidate object content, the object name, output order, location limitation content, and category limitation content are simultaneously retained in the direct inheritance result, pending confirmation result, or exclusion result; when the content type is destination content, the standardized location name and location limitation content are simultaneously retained in the direct inheritance result, pending confirmation result, or exclusion result.

[0096] The results of direct inheritance, pending confirmation, and excluded results are combined to generate results within the context of validity.

[0097] The context valid range result includes at least the directly inherited result, the result to be confirmed, and the excluded result. The directly inherited result is directly called by steps 304, 305, 402, 403, and 501. The result to be confirmed is called by steps 304, 305, 402, 403, 501, and 504. The excluded result is called by step 304 to remove the corresponding historical content from the current round of speech interpretation. After processing in this step, the context valid range result is formed and is called by steps 304, 305, 402, 403, 501, and 504.

[0098] Semantic parsing is performed on the current round of voice input. Candidate object reference relationships are determined within the candidate object range limited by the most recent navigation feedback result. The continuation semantic result is generated by combining the effective range result of the context and the current route association status. The reason is that the current round of voice input is first converted into semantic segments that can participate in object comparison and intent triage. Then, the semantic segments involving object references are restricted to the candidate object range limited by the most recent navigation feedback result to perform object correspondence and form candidate object reference relationships. Then, the direct inheritance result, the pending confirmation result and the exclusion result formed in step 206, as well as the current route association status, are combined to generate a continuation semantic result that can be directly read by subsequent steps, so as to avoid the current round of voice input being interpreted as an extended version of the object range in the current navigation session.

[0099] In some implementations, the steps for generating continuation semantic results include: See Figure 3 As shown, in step 301, speech recognition is performed on the current round of speech input to form the speech recognition text of the current round; according to the preset semantic parsing rules, word segmentation, word class labeling, adjacent word combination recognition and object content normalization processing are performed on the speech recognition text of the current round to form the semantic fragment of the current round of speech input.

[0100] The input object for the preset semantic parsing rules is the speech recognition text of the current round. The preset semantic parsing rules are formed based on a vocabulary of common location names, category names, ordinal words, pronouns, route constraint words, service words along the way, query words, cancellation words, negation words, and location limiting words in in-vehicle navigation speech, as well as the combination relationship of the above vocabulary in navigation statements. The preset semantic parsing rules are invoked by first recognizing location names and category names, then recognizing action words and limiting words, then recognizing ordinal words, pronouns, negation words, and location limiting words, and finally combining adjacent words that act on the same object into the same semantic fragment. The output of the preset semantic parsing rules are explicit object fragments, ordinal fragments, pronoun fragments, route constraint fragments, service words along the way fragments, query words, cancellation fragments, negation fragments, and location limiting fragments.

[0101] When an explicit object fragment is used to represent a location object, location normalization is performed on the location name in the explicit object fragment to form the location content for the current round. The input objects for location normalization are the location name words and location qualifiers in the explicit object fragment. The location normalization is based on the mapping relationship between place names and aliases, the mapping relationship between administrative division names, and the uniformity relationship of location suffixes. The location normalization is invoked by first removing modifiers and redundant modifiers, then replacing aliases with uniform location names, and then recording administrative division names, road names, location names, or doorway names as location qualifiers. The output of location normalization is the normalized location name and location qualifiers.

[0102] After processing in this step, a semantic fragment of the current round of voice input is formed, and the location content of the current round is formed when the explicit object fragment represents the location object; the semantic fragment of the current round of voice input and the location content of the current round are called by steps 302 to 305.

[0103] Step 302: Read the most recent navigation feedback result generated in step 104, and determine the range of candidate objects based on the feedback type of the most recent navigation feedback result.

[0104] When the most recent navigation feedback result is a candidate object set feedback result, all candidate object contents in the candidate object set feedback result are determined as the candidate object range limited by the most recent navigation feedback result; when the most recent navigation feedback result is a service along the way feedback result, all service objects along the way in the service along the way feedback result are determined as the candidate object range limited by the most recent navigation feedback result; when the most recent navigation feedback result is a route feedback result or navigation description feedback result, and no more than two selectable objects are given, the candidate object range limited by the most recent navigation feedback result is determined to be empty.

[0105] Each candidate object in the range defined by the most recent navigation feedback result includes at least the object name, output order, location limitation, and category limitation. The object name comes from the object identifier returned by the navigation engine; the output order comes from the order of the results returned by the navigation engine; the location limitation comes from the administrative division name, road name, distance description, orientation description, or doorway description returned by the navigation engine; and the category limitation comes from the object category description returned by the navigation engine. This step forms the range of candidate objects defined by the most recent navigation feedback result, which is then used in the next step.

[0106] Step 303: Within the candidate object range defined by the most recent navigation feedback result, the candidate object reference relationship is determined in a fixed order based on the semantic fragments formed in step 301. The fixed order includes: first, performing the ordinal fragment correspondence judgment, then performing the explicit object fragment correspondence judgment, and finally performing the reference fragment correspondence judgment. If the previous order has uniquely determined the candidate object reference relationship, the execution of the next order is stopped. If the previous order has failed to uniquely determine the candidate object reference relationship, the execution of the next order continues.

[0107] Before performing the corresponding judgment, the object name, location limitation content, and category limitation content of each candidate object within the candidate object range are first standardized. Specifically, the object name standardization process is used to unify the alternative names, abbreviations, and full names in the candidate object content into the same object name. The location limitation content standardization process is used to unify the administrative division name, road name, door name, distance description, and orientation description into comparable location limitation content. The category limitation content standardization process is used to unify the service category name, location category name, and venue category name into comparable category limitation content.

[0108] When a semantic segment contains an ordinal segment, the ordinal segment represents the sequential position, and the sequential position is compared with the output order of each candidate object within the candidate object range. When only one candidate object's output order matches the sequential position, a candidate object reference relationship based on the ordinal segment is determined between the current round of speech input and that candidate object. When two or more candidate object contents have the same output order as the sequential position, or when no candidate object contents have the same output order as the sequential position, the unique candidate object reference relationship is not output, and multiple candidate object corresponding results or no corresponding results are retained.

[0109] When the ordinal fragment fails to uniquely determine the reference relationship of the candidate object, and there is an explicit object fragment in the semantic fragment, the object name, category name, and position qualifier in the explicit object fragment are read, and normalization processing is performed on the object name, category name, and position qualifier; the normalized object name, category name, and position qualifier are compared item by item with the normalized object name, category qualifier, and position qualifier of each candidate object within the candidate object range.

[0110] For each candidate item, the corresponding level is determined in the following order: if both the object name and the location limitation are the same, it is determined as the first corresponding level; if only the object name is the same, it is determined as the second corresponding level; if both the category limitation and the location limitation are the same, it is determined as the third corresponding level; if only the location limitation is the same, it is determined as the fourth corresponding level; and if only the category limitation is the same, it is determined as the fifth corresponding level.

[0111] In some implementations, the first corresponding level, the second corresponding level, the third corresponding level, the fourth corresponding level, and the fifth corresponding level correspond to level number 1, level number 2, level number 3, level number 4, and level number 5, respectively; the smaller the corresponding level number, the higher the degree of matching between the candidate content and the current round of voice input.

[0112] If only one candidate object content has the smallest corresponding level number among all candidate object content, it is determined that there is a candidate object reference relationship between the current round of voice input and that candidate object content; if two or more candidate object content have the same corresponding level number, and that corresponding level number is the smallest among all candidate object content, multiple candidate object corresponding results are retained.

[0113] When an explicit object fragment still fails to uniquely determine the candidate object reference relationship, and a referential fragment exists within the semantic fragment, the position-qualifying fragment, category name, object name, distance description, or location description adjacent to the referential fragment are read, and normalization is performed on the above content. Then, the normalized content is compared item by item with the normalized position-qualifying content, category-qualifying content, and object name of each candidate object within the candidate object range, and a corresponding level is determined for each candidate object content using the same corresponding level determination method as the explicit object fragment correspondence determination. Among these, the first corresponding level, second corresponding level, third corresponding level, etc., are further categorized. The first, fourth, and fifth corresponding levels correspond to level numbers 1, 2, 3, 4, and 5, respectively. The smaller the corresponding level number, the higher the degree of matching between the candidate content and the current round of speech input. When only one candidate content has the smallest corresponding level number among all candidate content, it is determined that there is a candidate reference relationship based on the reference segment between the current round of speech input and that candidate content. When two or more candidate content have the same corresponding level number, and that corresponding level number is the smallest among all candidate content, multiple candidate content corresponding results are retained.

[0114] After step 303 is completed, if the candidate object reference relationship uniquely corresponds to one candidate object content, then the unique candidate object reference relationship is output; if the candidate object reference relationship corresponds to two or more candidate object contents, then the results corresponding to multiple candidate objects are output; if no candidate object content that meets the comparison conditions is found, then no corresponding result is output; after this step, candidate object reference relationships are formed, which can be called by steps 304, 402, 403, 501 and 504.

[0115] Step 304: Read the semantic fragments and location content of the current round of voice input formed in Step 301, the candidate object reference relationships formed in Step 303, the context effective range results formed in Step 206, and the current route association status formed in Step 105; first, read the direct inheritance results from the context effective range results, and perform direct inheritance and inclusion judgment on each item in the direct inheritance results; then, read the exclusion results from the context effective range results, and perform removal processing according to content type, source round, and target content; finally, read the pending confirmation results from the context effective range results, and retain the pending confirmation results as undetermined content to be called in subsequent steps.

[0116] The specific methods for determining direct inheritance include: When a certain item in the direct inheritance result is the destination content, if the current round's location content does not exist, then the destination content is incorporated into the semantic interpretation scope of the current round; if the current round's location content exists, then the normalized location name and location-specific content in the current round's location content are compared with the normalized location name and location-specific content in the destination content; if the comparison result is consistent, the destination content is incorporated into the semantic interpretation scope of the current round; if the comparison result is inconsistent, the destination content is not incorporated into the semantic interpretation scope of the current round.

[0117] When one of the contents in the direct inheritance result is a route constraint, the route constraint segment in the current round of speech input is read. If there is no route constraint segment in the current round of speech input, the route constraint content is incorporated into the semantic interpretation scope of the current round. If there is a route constraint segment in the current round of speech input, the constraint type, constraint direction, and target content of the route constraint segment in the current round of speech input are compared with those of the route constraint content. When the constraint type is the same, the target content is the same, and the constraint direction is different, or the route constraint segment in the current round of speech input has clearly given the replacement constraint content, the route constraint content is not incorporated into the semantic interpretation scope of the current round. In other cases, the route constraint content is incorporated into the semantic interpretation scope of the current round.

[0118] When one of the contents in the direct inheritance result is a service along the route, the service along the route segment in the current round of voice input is read. If there is no service along the route segment in the current round of voice input, the service along the route content is incorporated into the semantic interpretation scope of the current round. If there is a service along the route segment in the current round of voice input, the service category content and scope content in the current round of voice input are compared with the service category content and scope content of the service along the route. When the service category content is the same but the scope content is different, or when the current round of voice input has clearly given a new service category content, the service along the route content is not incorporated into the semantic interpretation scope of the current round. In other cases, the service along the route content is incorporated into the semantic interpretation scope of the current round.

[0119] When a certain item in the direct inheritance result is a candidate object, only the unique candidate object content is retained in the candidate object reference relationship, and the unique candidate object content corresponds to the same candidate object identifier as the candidate object content, and the candidate object content is included in the semantic interpretation scope of the current round.

[0120] The specific methods for performing removal processing based on content type, source round, and target content include: When a certain item in the exclusion result is the same as the historical content in the current round of semantic interpretation, has the same content type and the same target content, and the source round number of the historical content is less than or equal to the source round number of the exclusion result, the corresponding historical content in the current round of semantic interpretation will be removed. When one of the contents in the exclusion results is a candidate object, remove the historical candidate object contents that belong to the same set of replaced candidate objects as that candidate object content. When one of the items in the exclusion result is a route constraint, remove the historical route constraint items that have the same constraint type, the same target content, and a source round number that is less than or equal to the source round number of the exclusion result. When one of the exclusion results is a service along the route, remove the historical service along the route that has the same service category and scope, and whose source round number is less than or equal to the source round number of the exclusion result.

[0121] When multiple candidate object references are retained or no corresponding results are found, the content of any candidate object is not directly merged into the object content. Instead, the corresponding results of multiple candidate objects or no corresponding results are retained as undetermined object content in the results to be confirmed. After processing in this step, the current round semantic merging content is formed for generating continuation semantic results and is called by step 305.

[0122] Step 305: Generate a continuation semantic result based on the current round semantic merging content and the current route association status formed in step 304.

[0123] Among them, the continuation semantic results include at least the object content explicitly given in the current round of voice input, the candidate object content uniquely determined by the candidate object reference relationship, the historical content supplemented by the direct inheritance result, the undetermined object content retained by the pending confirmation result, and the route function content limited by the current route association state.

[0124] When the current round location content exists, it is incorporated into the continuation semantic result; when the current round location content does not exist, and only one candidate object content is retained in the candidate object reference relationship, the only candidate object content is used as the object content for the current round voice input; when the current round location content does not exist, and multiple candidate object corresponding results or no corresponding results are retained in the candidate object reference relationship, multiple candidate object corresponding results or no corresponding results are retained in the continuation semantic result, and the undetermined object content in the pending confirmation result is retained simultaneously.

[0125] When no new location content appears in the current round of voice input, but a route constraint segment appears and the current route association status indicates that a route exists, the current destination content in the directly inherited result will be merged with the route constraint segment to form a route constraint, and the corresponding continuation semantic result will be adjusted. After processing in this step, a continuation semantic result is formed and used by steps 402, 403, 501 and 504.

[0126] Furthermore, to illustrate the process of determining the reference relationships of candidate objects, for example: The most recent navigation feedback result narrowed down the candidate object range to include three candidate objects. The first candidate object is named "Century Avenue Gas Station," with the location specified as "Century Avenue" and the category specified as "gas station." The second candidate object is named "China Petroleum Yanggao Middle Road Station," with the location specified as "Yanggao Middle Road" and the category specified as "gas station." The third candidate object is named "Dongfang Road Gas Station," with the location specified as "Dongfang Road" and the category specified as "gas station." After the current round of voice input is parsed in step 301, the explicit object fragment "China Petroleum Yanggao Middle Road" and the category name "gas station" are formed. In step... In step 303, when comparing the contents of the three candidate objects one by one, the contents of the first candidate object are only consistent in terms of category limitation, so it is determined to be the fifth corresponding level; the contents of the second candidate object are consistent in terms of both object name and position limitation, so it is determined to be the first corresponding level; the contents of the third candidate object are consistent in terms of only category limitation, so it is determined to be the fifth corresponding level; since only the corresponding level number of the contents of the second candidate object is the smallest, it is determined that there is a unique candidate object reference relationship between the current round of voice input and the contents of the second candidate object, and the contents of the second candidate object are used as the candidate object reference relationship result for further invocation in steps 304, 402, 403, 501 and 504.

[0127] Based on navigation task stage information, candidate object reference relationships, and continuation semantic results, the continuation intent corresponding to the current round of speech is determined and completed. The target continuation intent and the corresponding intent completion result are determined. This is because, based on navigation task stage information, candidate object reference relationships, continuation semantic results, and the direct inheritance results and pending confirmation results in the context effective range results, the current round of speech input is assigned to one of the continuation processing directions: candidate object selection, destination change, route constraint adjustment, along-the-way service retrieval, navigation description query, or navigation cancellation. The execution content not directly given by the current round of speech input is supplemented from the direct inheritance results, or the pending confirmation results are retained in the clarification process. This forms the target continuation intent and the corresponding intent completion result for subsequent steps to continue to call.

[0128] In some implementations, the steps for determining the target continuation intent and the corresponding intent completion result include: Step 401: Read the navigation task stage information, candidate object reference relationships, continuation semantic results, current route association status, and direct inheritance results and pending confirmation results in the context valid range results; first, establish a preset continuation intent correspondence relationship, and then, based on the preset continuation intent correspondence relationship, perform preliminary judgment on each of the candidate object selection intent, destination change intent, route constraint adjustment intent, along-the-way service retrieval intent, navigation description query intent, and navigation cancellation intent, and add the continuation intent categories that pass the preliminary judgment to the candidate continuation intent categories.

[0129] Among them, the preset continuation intent correspondence is used to give the correspondence between each continuation intent category and the corresponding judgment element; the judgment element includes at least the navigation task stage element, object content element, route status element, current round semantic element and exclusion element.

[0130] Among them, the navigation task stage element is used to limit the navigation task stage applicable to a certain continuation intent category; the object content element is used to limit the location content, candidate object content, route constraint content, along-the-way service content, or description query content required for the continuation intent category; the route status element is used to limit whether the continuation intent category requires a currently existing route or whether the target route is still to be generated; the current round semantic element is used to limit the semantic content that the continuation intent category should appear in the current round of voice input; and the exclusion element is used to limit conflicting content or pending confirmation results that should not appear simultaneously in the current round of voice input for the continuation intent category.

[0131] In some implementations, the specific methods for performing a preliminary determination on each type of connection continuation intent include: For the candidate object selection intent, read the navigation task stage information, candidate object reference relationship, continuation semantic result and pending confirmation result; when the navigation task stage information is the candidate object feedback stage, the candidate object reference relationship retains the content of a unique candidate object, the continuation semantic result does not retain the content of the current round location, and the pending confirmation result does not retain the content of other candidate objects that belong to the same candidate object set as the unique candidate object content but have different candidate object identifiers, the candidate object selection intent is determined to have passed the preliminary judgment, and the candidate object selection intent is added to the candidate continuation intent category.

[0132] For destination change intent, read the continuation semantic result and the direct inheritance result; when the continuation semantic result retains the current round location content, the direct inheritance result retains the current destination content, and the current round location content is inconsistent with the current destination content after comparison, the destination change intent is determined to pass the preliminary judgment, and the destination change intent is added to the candidate continuation intent category.

[0133] For route constraint adjustment intent, read the current route association status, continuation semantic results, direct inheritance results, and pending confirmation results; when the current route association status indicates that a route exists, route constraint content appears in the continuation semantic results, the current round location content is not retained in the continuation semantic results, the current destination content is retained in the direct inheritance results, and no other route constraint content with the same constraint type, the same object content, and different constraint direction as the current route constraint content is retained in the pending confirmation results, the route constraint adjustment intent is determined to have passed the preliminary judgment, and the route constraint adjustment intent is added to the candidate continuation intent category.

[0134] For the search intent for services along the way, read the current route association status, navigation task stage information, continuation semantic results, and pending confirmation results; when the continuation semantic results contain services along the way, and the current route association status indicates that a route currently exists, or the navigation task stage information indicates that the target route is still to be generated, and the continuation semantic results do not retain the current round location content, and the pending confirmation results do not retain other services along the way with the same service category content but different scope content, the search intent for services along the way is determined to have passed the preliminary judgment, and the search intent for services along the way is added to the candidate continuation intent category.

[0135] For the navigation description query intent, read the continuation semantic results; when the description query content appears in the continuation semantic results and no cancellation content appears in the continuation semantic results, it is determined that the navigation description query intent has passed the preliminary judgment and is added to the candidate continuation intent category.

[0136] For navigation cancellation intents, read the continuation semantic results; when cancellation content appears in the continuation semantic results, determine that the navigation cancellation intent has passed the preliminary judgment, and add the navigation cancellation intent to the candidate continuation intent category.

[0137] If a certain reconnection intent category does not meet the corresponding preliminary judgment condition, the reconnection intent category will not be added to the candidate reconnection intent categories; after processing in this step, candidate reconnection intent categories are formed for use in steps 402 to 404.

[0138] Step 402: Read the candidate continuation intent category, continuation semantic result, current route association status, and the results to be confirmed, the results corresponding to multiple candidate objects, and the results without corresponding objects from the context valid range results; perform a unique determination judgment on the candidate continuation intent category to determine whether a target continuation intent is formed.

[0139] The unique determination is determined by the following inputs: candidate continuation intent category, continuation semantic result, pending confirmation result, results corresponding to multiple candidate objects, and no corresponding result. The criteria for determining uniqueness are: whether only one continuation intent category is retained among the candidate continuation intent categories, whether only one execution object content corresponding to that continuation intent category is retained, and whether only one execution scope content corresponding to that continuation intent category is retained. The unique determination is invoked by first determining the number of continuation intent categories retained among the candidate continuation intent categories, then determining whether the execution object content corresponding to that continuation intent category is unique, and finally determining whether the execution scope content corresponding to that continuation intent category is unique. The output of the unique determination is either the target continuation intent or the candidate continuation intent categories that are retained.

[0140] The execution object content is used to represent the processing object that actually acts on the target continuation intention; in some implementations, the execution object content includes at least one of the following: unique candidate object content, current round location content, current destination content, the action object content corresponding to the route constraint content, and the service object content corresponding to the service content along the way.

[0141] The execution scope content is used to indicate the processing range in which the target continuation intention actually takes effect; in some embodiments, the execution scope content includes at least one of the following: current route, target route, current location vicinity, target location vicinity, and candidate object range.

[0142] In some implementations, specific methods for performing a unique determination of candidate continuation intent categories include: When more than two continuation intent categories are retained among the candidate continuation intent categories, it is determined that the continuation processing direction corresponding to the current round of voice input is not unique, and no target continuation intent is formed. Instead, the candidate continuation intent categories are retained.

[0143] When only one continuation intent category is retained among the candidate continuation intent categories, it is further determined whether the content of the execution object corresponding to the unique retained continuation intent category is unique.

[0144] When the only retained continuation intent category is the candidate object selection intent, read the results corresponding to multiple candidate objects and the result without a corresponding object; when multiple candidate object results do not exist, no corresponding result exists, and only one candidate object content is retained in the candidate object reference relationship, determine that the execution object content corresponding to the candidate object selection intent is unique; when multiple candidate object results exist, no corresponding result exists, or only one candidate object content is retained in the candidate object reference relationship, determine that the execution object content corresponding to the candidate object selection intent is not unique.

[0145] When the only retained continuation intent category is a destination change intent, the current round location content in the continuation semantic result and the destination content in the pending confirmation result are read. When only one current round location content is retained in the continuation semantic result, and there are no other destination contents in the pending confirmation result that have the same normalized location name as the current round location content but different location constraints, the execution object content corresponding to the destination change intent is determined to be unique. When there is no current round location content in the continuation semantic result, or there are other destination contents in the pending confirmation result that have the same normalized location name as the current round location content but different location constraints, the execution object content corresponding to the destination change intent is determined to be not unique.

[0146] When the only retained continuation intent category is a route constraint adjustment intent, the route constraint content in the continuation semantic result, the current destination content in the direct inheritance result, and the route constraint content in the pending confirmation result are read. When only one route constraint content is retained in the continuation semantic result, only one current destination content is retained in the direct inheritance result, and there are no other route constraint contents in the pending confirmation result that are the same in constraint type and target content but different in constraint direction as the route constraint content, the execution object content corresponding to the route constraint adjustment intent is determined to be unique. When two or more route constraint contents are retained in the continuation semantic result, or the direct inheritance result does not retain only one current destination content, or there are other route constraint contents in the pending confirmation result that are the same in constraint type and target content but different in constraint direction as the route constraint content, the execution object content corresponding to the route constraint adjustment intent is determined to be not unique.

[0147] When the only retained continuation intent category is the service retrieval intent along the route, the service content along the route in the continuation semantic results and the service content along the route in the pending confirmation results are read; when only one type of service content is retained in the continuation semantic results, and there are no other service contents along the route with the same service category but different scope in the pending confirmation results, the execution object content corresponding to the service retrieval intent along the route is determined to be unique; when two or more types of service content are retained in the continuation semantic results, or there are other service contents along the route with the same service category but different scope in the pending confirmation results, the execution object content corresponding to the service retrieval intent along the route is determined to be not unique.

[0148] When the only retained continuation intent category is the navigation description query intent, the description query content in the continuation semantic result is read; when only one description query content is retained in the continuation semantic result, the execution object content corresponding to the navigation description query intent is determined to be unique; when more than two description query contents are retained in the continuation semantic result, the execution object content corresponding to the navigation description query intent is determined to be not unique.

[0149] When the only retained continuation intent category is navigation cancellation intent, read the current navigation session identifier; when the current navigation session identifier exists, determine that the content of the execution object corresponding to the navigation cancellation intent is unique; when the current navigation session identifier does not exist, determine that the content of the execution object corresponding to the navigation cancellation intent is not unique.

[0150] Once the execution object content corresponding to the uniquely retained continuation intent category is unique, it is further determined whether the execution scope content corresponding to the uniquely retained continuation intent category is unique.

[0151] When the only retained continuation intent category is the candidate object selection intent, the candidate object range corresponding to the candidate object reference relationship is read; when only one candidate object range content corresponding to the unique candidate object content is retained in the candidate object range, the execution range content corresponding to the candidate object selection intent is determined to be unique; when more than two candidate object range contents are still retained in the candidate object range, the execution range content corresponding to the candidate object selection intent is determined to be not unique.

[0152] When the only retained continuation intent category is destination change intent, read the location limitation content in the current round location content; when the location limitation content is empty, or the location limitation content corresponds to only one location range, determine that the execution range content corresponding to the destination change intent is unique; when the location limitation content corresponds to more than two location ranges, determine that the execution range content corresponding to the destination change intent is not unique.

[0153] When the only retained continuation intent category is route constraint adjustment intent, read the current route and target route from the current route association status; when the current route association status indicates that a route currently exists, and the content of the route constraint content applies only to the current route or only to one of the target routes, the execution scope content corresponding to the route constraint adjustment intent is determined to be unique; when the content of the route constraint content applies to both the current route and the target route, the execution scope content corresponding to the route constraint adjustment intent is determined to be not unique.

[0154] When the only retained continuation intent category is the service retrieval intent along the way, the scope content in the service content along the way is read; when the scope content corresponds to only one of the current route, target route, near the current location, or near the target location, the execution scope content corresponding to the service retrieval intent along the way is determined to be unique; when the scope content corresponds to two or more scopes at the same time, the execution scope content corresponding to the service retrieval intent along the way is determined to be not unique.

[0155] When the only retained continuation intent category is navigation description query intent, the query scope corresponding to the description query content is read; when the query scope corresponds to only one query scope in the current route, the execution scope content corresponding to the navigation description query intent is determined to be unique; when the query scope corresponds to more than two query scopes, the execution scope content corresponding to the navigation description query intent is determined to be not unique.

[0156] When the only retained continuation intent category is navigation cancellation intent, since the execution scope of navigation cancellation intent is the navigation process corresponding to the current navigation session, the execution scope content corresponding to navigation cancellation intent is unique when the current navigation session identifier exists.

[0157] When only one continuation intent category is retained among the candidate continuation intent categories, and the execution object content and execution scope content corresponding to this continuation intent category are unique, and the pending confirmation result, the corresponding results of multiple candidate objects, and the absence of a corresponding result no longer affect the execution object content and execution scope content corresponding to this continuation intent category, the uniquely retained continuation intent category is determined as the target continuation intent.

[0158] When more than two continuation intent categories are retained among the candidate continuation intent categories, a target continuation intent is not formed, but the candidate continuation intent categories are retained.

[0159] If only one continuation intent category is retained among the candidate continuation intent categories, but the execution object content corresponding to that continuation intent category is not unique, or the execution scope content is not unique, or the pending confirmation result, the corresponding results of multiple candidate objects, and the absence of a corresponding result still affect the execution object content or execution scope content corresponding to that continuation intent category, then a target continuation intent is not formed, but the candidate continuation intent category is retained. After processing in this step, a target continuation intent is formed when the unique determination condition is met, and a candidate continuation intent category is formed to be retained when the unique determination condition is not met, for use in steps 403 and 404.

[0160] Step 403: Read the target continuation intent formed in step 402 or the candidate continuation intent category that is still retained, the continuation semantic result, and the direct inheritance result and the pending confirmation result from the context effective range result; when a target continuation intent has been formed in step 402, first determine the minimum execution content set corresponding to the target continuation intent according to the preset minimum execution content correspondence, then extract the content that has been clearly given in the current round of voice input from the continuation semantic result, and extract the historical content used to fill in the missing parts in the minimum execution content set from the direct inheritance result to form the corresponding intent completion result; when a target continuation intent has not been formed in step 402, but the candidate continuation intent category is still retained, read the pending confirmation result, the result corresponding to multiple candidate objects, or the result without a corresponding object corresponding to the candidate continuation intent category, and write the candidate continuation intent category, the pending confirmation result, the result corresponding to multiple candidate objects, or the result without a corresponding object together into the corresponding intent completion result.

[0161] The preset minimum execution content correspondence is used to provide the minimum execution content set corresponding to each target continuation intent; the input object of the preset minimum execution content correspondence is the target continuation intent; the preset minimum execution content correspondence is formed based on the execution object, execution scope and execution action requirements corresponding to different target continuation intents; the preset minimum execution content correspondence is invoked by reading the corresponding minimum execution content set according to the target continuation intent type; the output result of the preset minimum execution content correspondence is the minimum execution content set corresponding to the target continuation intent.

[0162] In some implementations, the minimum set of execution content corresponding to the candidate object selection intent includes at least the content of a unique candidate object; the minimum set of execution content corresponding to the destination change intent includes at least the content of the current round location; the minimum set of execution content corresponding to the route constraint adjustment intent includes at least the content of the current destination and the route constraint; the minimum set of execution content corresponding to the along-the-way service retrieval intent includes at least the content of along-the-way services and the scope of application; the minimum set of execution content corresponding to the navigation description query intent includes at least the description query content and the query scope; and the minimum set of execution content corresponding to the navigation cancellation intent includes at least the current navigation session identifier.

[0163] When the target continuation intent selects an intent for a candidate object, the content of the unique candidate object in the continuation semantic result is read; when the content of the unique candidate object has been retained in the continuation semantic result, the content of the unique candidate object is written into the corresponding intent completion result; when the content of the unique candidate object has not been retained in the continuation semantic result, no other candidate object content is added to the corresponding intent completion result, and the missing state is retained for step 501 to judge.

[0164] When the target continuation intent is a destination change intent, the current round location content is read from the continuation semantic result, and the route constraint content is read from the direct inheritance result. If the current round location content is already retained in the continuation semantic result, the current round location content is written into the corresponding intent completion result. If there is route constraint content in the direct inheritance result that has not been replaced by the route constraint segment in the current round voice input in terms of constraint type and target content, the route constraint content is continued to be written into the corresponding intent completion result. If the current round location content does not exist, the old destination content is not added from the direct inheritance result to replace the current round location content.

[0165] When the target continuation intent is a route constraint adjustment intent, read the current destination content from the direct inheritance result and the route constraint content from the continuation semantic result; when only one current destination content is retained in the direct inheritance result, write the current destination content into the corresponding intent completion result; when the route constraint content is retained in the continuation semantic result, write the route constraint content into the corresponding intent completion result; when the route constraint content does not exist, do not automatically replace the new route constraint content not explicitly given in the current round of voice input from the historical route constraint content.

[0166] When the target continuation intent is a service along the route retrieval intent, read the service along the route content and scope content from the continuation semantic result; when the service along the route content and scope content are retained in the continuation semantic result, write the service along the route content and scope content into the corresponding intent completion result; when the service along the route content exists but the scope content is not clear, read the current route content or target route content from the direct inheritance result; when only one of the current route content or target route content can be used as the scope of the service along the route retrieval, write that unique scope content into the corresponding intent completion result.

[0167] When the target continuation intent is a navigation description query intent, read the description query content from the continuation semantic result; if the description query content exists, write the description query content into the corresponding intent completion result; then read the route content from the current route association state and the direct inheritance result; if only one current route content is retained in the current route association state, write the current route content into the corresponding intent completion result; if the current route content does not exist but the target route content exists, write the target route content into the corresponding intent completion result.

[0168] When the target continuation intent is a navigation cancellation intent, read the current navigation session identifier; if the current navigation session identifier exists, write the current navigation session identifier into the corresponding intent completion result.

[0169] When step 402 does not form a target continuation intent, the corresponding intent completion result simultaneously retains the candidate continuation intent category, the pending confirmation result, the result corresponding to multiple candidate objects, or the result with no corresponding object. Among them, the pending confirmation result is used to indicate historical content for which the execution object or execution scope has not yet been determined, the result corresponding to multiple candidate objects is used to indicate that the same semantic segment can correspond to more than two candidate object contents, and the result with no corresponding object is used to indicate that the object reference content in the current round of voice input fails to correspond to any candidate object content in the candidate object range limited by the most recent navigation feedback result. After processing in this step, the corresponding intent completion result is formed and is called by steps 404, 501, and 504.

[0170] In some implementations, to illustrate the process of forming the corresponding intention completion result, for example: The direct inheritance result corresponding to the current navigation session retains one item for the current destination and one item for the current route. The current destination is "Hangzhou East Railway Station". The current voice input is "Avoid toll fees". After processing in step 305, the continuation semantic result retains one item for route constraints. The constraint type of this route constraint is "Avoid toll fees", and the constraint direction is "Add Restriction". No new location content appears in the current voice input. At this time, in step 401, the candidate continuation intent category only includes route constraint adjustment intent. In step 402, since a route exists and the continuation semantic result only retains one item for route constraints, the direct inheritance result... Only one item of the current destination content is retained, thus forming the target continuation intent as the route constraint adjustment intent; further, in step 403, the minimum execution content set corresponding to the route constraint adjustment intent is read, which includes two items: the current destination content and the route constraint content; since the current round of voice input has clearly given the route constraint content, but has not given the current destination content again, the current destination content corresponding to "Hangzhou East Station" is extracted from the direct inheritance result, and together with the toll-avoidance route constraint content parsed in the current round of voice input, a corresponding intent completion result containing two items is obtained.

[0171] Step 404: Read the target continuation intent or the candidate continuation intent category formed in step 402, and the corresponding intent completion result formed in step 403; when a target continuation intent has been formed in step 402, associate the target continuation intent with the corresponding intent completion result formed in step 403 and output it; when a target continuation intent has not been formed in step 402, but the candidate continuation intent category is retained, associate the candidate continuation intent category with the corresponding intent completion result formed in step 403 and output it.

[0172] Specifically, when a target continuation intent has been formed in step 402, the output object of step 404 is the target continuation intent and the corresponding intent completion result; when a target continuation intent has not been formed in step 402, the output object of step 404 is the intent completion result corresponding to the candidate continuation intent category; after processing in this step, a target continuation intent and the corresponding intent completion result are formed for step 501 to continue calling, or an intent completion result corresponding to the candidate continuation intent category is formed for step 501 and step 504 to continue calling.

[0173] Based on the target continuation intent and the corresponding intent completion result, a navigation execution request or a voice clarification request is generated. The reason is that the formed target continuation intent and the corresponding intent completion result are transformed into execution content that the navigation engine can directly receive. Alternatively, when the current round of voice input is insufficient to support a unique execution object or a unique execution range, supplementary inquiry content for the user is generated based on the result to be confirmed, the corresponding results of multiple candidate objects, or the absence of a corresponding result. This ensures that the continuation processing result of the current round of voice input falls into the specific navigation processing process or the specific voice clarification process.

[0174] In some implementations, the steps of generating a navigation execution request or generating a voice clarification request include: Step 501: Read the target continuation intent and corresponding intent completion result formed in step 404, or the intent completion result corresponding to the candidate continuation intent category; first determine whether the target continuation intent has been formed in the output object of step 404; when the target continuation intent has been formed in the output object of step 404, determine whether to generate a navigation execution request based on the preset execution integrity condition; when the target continuation intent has not been formed in the output object of step 404, but the candidate continuation intent category is still retained, directly determine to generate a voice clarification request, and do not generate a navigation execution request.

[0175] The input objects for the preset execution integrity condition are the target continuation intent and the corresponding intent completion result. The preset execution integrity condition is formed based on whether the minimum set of execution content corresponding to the target continuation intent has been completely written into the corresponding intent completion result, and whether the corresponding intent completion result still retains the pending confirmation result, the result corresponding to multiple candidate objects, or no corresponding result. The preset execution integrity condition is invoked by first reading the minimum set of execution content corresponding to the current target continuation intent, then checking item by item in the corresponding intent completion result to see if it already contains all the content in the minimum set of execution content, and finally determining whether the corresponding intent completion result still retains the pending confirmation result that affects the uniqueness of the execution object or execution scope, the result corresponding to multiple candidate objects, or no corresponding result. The output result of the preset execution integrity condition is to generate a navigation execution request or a voice clarification request.

[0176] In some implementations, when the target continuation intent is to select an intent for a candidate object, a navigation execution request is determined to be generated only if the corresponding intent completion result has a unique candidate object content and there are no corresponding results for multiple candidate objects or no corresponding results.

[0177] When the target continuation intent is a destination change intent, a navigation execution request is generated only if the corresponding intent completion result contains the current round location content and there is no pending confirmation result that has the same standardized location name as the current round location content but whose location limitation content cannot be uniquely determined.

[0178] When the target continuation intent is a route constraint adjustment intent, a navigation execution request is generated only if the corresponding intent completion result contains the current destination content and route constraint content, and no route constraint content to be confirmed is retained.

[0179] When the target continuation intent is the intent to retrieve services along the way, a navigation execution request will be generated only if the corresponding intent completion results contain the content of services along the way and the scope of application, and if multiple search object corresponding results are not retained or there are no corresponding results.

[0180] When the target continuation intent is a navigation description query intent, a navigation execution request is generated only if the corresponding intent completion result contains descriptions of the query content and query scope.

[0181] When the target continuation intent is a navigation cancellation intent, a navigation execution request is generated only if the current navigation session identifier is present in the corresponding intent completion result.

[0182] When the target continuation intent and the corresponding intent completion results still retain results to be confirmed, results corresponding to multiple candidate objects, or no corresponding results, and the above content directly affects the execution object or execution scope of the target continuation intent, a voice clarification request is generated.

[0183] When step 404 outputs the intent completion result corresponding to the candidate continuation intent category, since the current round of voice input has not yet completed the determination of the unique continuation processing direction, a voice clarification request is directly generated. After processing in this step, a judgment result of the navigation execution request generation condition is formed and is called by steps 502 and 504.

[0184] Step 502: When the judgment result formed in step 501 is to generate a navigation execution request, read the target continuation intent and the corresponding intent completion result; first, determine the request category corresponding to the target continuation intent based on the preset request generation correspondence; then, extract the execution object content, execution scope content and execution action content from the corresponding intent completion result based on the request category; and combine the request category, execution object content, execution scope content and execution action content to form a navigation execution request.

[0185] The preset request generation correspondence is used to provide the correspondence between each target continuation intent and the navigation execution request category; the input object of the preset request generation correspondence is the target continuation intent; the preset request generation correspondence is formed based on the different navigation processing actions corresponding to different target continuation intents; the preset request generation correspondence is invoked by reading the corresponding request category according to the target continuation intent type; the output result of the preset request generation correspondence is the request category.

[0186] In some implementations, the request category corresponding to the candidate object selection intent is a candidate object selection request, the request category corresponding to the destination change intent is a destination change request, the request category corresponding to the route constraint adjustment intent is a route constraint adjustment request, the request category corresponding to the route service retrieval intent is a route service retrieval request, the request category corresponding to the navigation description query intent is a navigation description query request, and the request category corresponding to the navigation cancellation intent is a navigation cancellation request.

[0187] When the target continuation intent is a candidate object selection intent, the unique candidate object content in the corresponding intent completion result is read; the object name, location limitation content, category limitation content and output order in the unique candidate object content are extracted as the execution object content; the candidate object range corresponding to the unique candidate object content is taken as the execution range content; selecting the candidate object content and continuing to perform navigation processing is determined as the execution action content; and then the request category, execution object content, execution range content and execution action content are combined to form the candidate object selection request.

[0188] When the target continuation intent is a destination change intent, the current round location content and the route constraint content that continues to be effective are read from the corresponding intent completion result; the current round location content is used as the execution object content, the location range corresponding to the current round location content is used as the execution range content, the current destination content is replaced with the current round location content, and the target route is regenerated based on the route constraint content that continues to be effective and determined as the execution action content; then the request type, execution object content, execution range content and execution action content are combined to form a destination change request.

[0189] When the target continuation intent is a route constraint adjustment intent, the current destination content and the route constraint content that needs to be added, deleted, or replaced are read from the corresponding intent completion result; the current destination content is used as the execution object content, and the current route or target route is used as the execution scope content; based on the route constraint content that needs to be added, deleted, or replaced, the currently effective route constraint content is added, deleted, or replaced to generate the adjusted route constraint content; under the condition that the current destination content remains unchanged, the route is regenerated or adjusted based on the adjusted route constraint content to determine the execution action content; then the request category, execution object content, execution scope content, adjusted route constraint content, and execution action content are combined to form a route constraint adjustment request.

[0190] The specific methods for performing addition, deletion, or replacement processing on currently effective route constraints include: when adding route constraints, merging the new route constraints into the currently effective route constraints; when deleting route constraints, removing the route constraints that correspond to the deleted object in terms of constraint type and affected object content from the currently effective route constraints; and when replacing route constraints, removing the replaced route constraints from the currently effective route constraints and merging them into the replaced route constraints.

[0191] When the target continuation intent is a service retrieval intent along the route, the service content along the route and the scope content in the corresponding intent completion result are read; the service content along the route is used as the execution object content, and the scope content is used as the execution scope content. The service objects corresponding to the service content along the route within the route corresponding to the scope content are retrieved and determined as the execution action content; then the request category, execution object content, execution scope content and execution action content are combined to form a service retrieval request along the route.

[0192] When the target continuation intent is a navigation description query intent, the description query content and the content corresponding to the current route in the corresponding intent completion result are read; the description query content is used as the execution object content, the content corresponding to the current route is used as the execution scope content, and the navigation description corresponding to the output description query content for the content corresponding to the current route is determined as the execution action content; then the request category, execution object content, execution scope content and execution action content are combined to form a navigation description query request.

[0193] When the target continuation intent is a navigation cancellation intent, the current navigation session identifier and cancellation action content are read from the corresponding intent completion result; the current navigation session identifier is used as the execution object content, the navigation process corresponding to the current navigation session is used as the execution scope content, and the cancellation action content is used as the execution action content; then the request type, execution object content, execution scope content, and execution action content are combined to form a navigation cancellation request; after this step, a navigation execution request is formed and is called by step 503.

[0194] Step 503: Send the navigation execution request generated in step 502 to the navigation engine; receive the processing result returned by the navigation engine for the navigation execution request, and include the processing result in the feedback record in the current navigation session as the new round of navigation feedback content in the current navigation session.

[0195] The new round of navigation feedback content serves as one of the input sources when reading the most recent navigation feedback result in subsequent rounds; after processing in this step, the navigation feedback result corresponding to the navigation execution request is formed and can be called in subsequent rounds.

[0196] Step 504: When the judgment result formed in step 501 is to generate a voice clarification request, read the intent completion result corresponding to the candidate continuation intent category output in step 404, or the target continuation intent and its corresponding intent completion result; then read the retained results to be confirmed, results corresponding to multiple candidate objects, results without corresponding results, and the minimum execution content that is missing; and form a voice clarification request according to the preset clarification generation rules.

[0197] The preset clarification generation rule is used to generate supplementary inquiry content based on the content that is not yet uniquely determined in the current round of voice input. The input object of the preset clarification generation rule is the intent completion result corresponding to the candidate continuation intent category, or the pending confirmation result, the result corresponding to multiple candidate objects, the result without a corresponding object, and the minimum execution content in the target continuation intent and the corresponding intent completion result. The preset clarification generation rule is formed based on the following: first, it is determined whether the current round of voice input is not uniquely determined as the continuation processing direction, the execution object content, or the execution scope content, and then a corresponding inquiry is formed for the content that is not uniquely determined. The preset clarification generation rule is invoked by executing the clarification source judgment, the inquiry object judgment, and the inquiry content generation in a preset order. The output result of the preset clarification generation rule is a voice clarification request.

[0198] In some implementations, the preset sequence includes: First, perform a source clarification check; then, perform a query target check; finally, generate the query content.

[0199] The specific methods for clarifying the determination of the source include: When step 404 outputs the intent completion result corresponding to the candidate continuation intent category, read the candidate continuation intent category; when more than two continuation intent categories are retained in the candidate continuation intent categories, determine that the continuation processing direction is not uniquely determined in the current round of voice input, and determine the continuation processing direction as the first query object; When step 404 outputs the target continuation intent and the corresponding intent completion result, and multiple candidate object corresponding results are retained or there is no corresponding result, it is determined that the current round of voice input is not uniquely determined as the execution object content, and the execution object content is determined as the second query object; when step 404 outputs the target continuation intent and the corresponding intent completion result, and a result to be confirmed is retained, and the result to be confirmed corresponds to the route range, scope of action, or location range, it is determined that the current round of voice input is not uniquely determined as the execution range content, and the execution range content is determined as the third query object; when step 404 outputs the target continuation intent and the corresponding intent completion result, and no result to be confirmed, multiple candidate object corresponding results, or no corresponding result are retained, but the missing minimum execution content still exists, it is determined that the current round of voice input is missing as the minimum execution content, and the missing minimum execution content is determined as the fourth query object.

[0200] The specific methods for determining the object of inquiry include: When the first query target is the continuation processing direction, read the continuation intent categories retained in the candidate continuation intent categories; when a candidate continuation intent category retains both the candidate object selection intent and the destination change intent, the query target will be determined by whether to select the previously selected candidate object or change to a new destination; when a candidate continuation intent category retains both the route constraint adjustment intent and the along-the-way service retrieval intent, the query target will be determined by whether to adjust the route conditions or query along-the-way services; when a candidate continuation intent category retains both the navigation description query intent and the navigation cancellation intent, the query target will be determined by whether to query the navigation description or end the navigation.

[0201] When the second query object is the content of the execution object, read the corresponding results of multiple candidate objects or there is no corresponding result; when more than two candidate object contents are retained in the corresponding results of multiple candidate objects, use the object name, position limitation content, category limitation content or output order in each candidate object content as the query object content; when there is no corresponding result, the object name, position limitation content, category limitation content or order position will be re-given as the query object content.

[0202] When the third query object is the execution scope content, read the scope content or location scope content retained in the confirmation result; when the confirmation result retains both the current route and the target route scope content, determine whether the current route or the target route is the query object content; when the confirmation result retains both the current location vicinity and the target location vicinity content, determine whether the current location vicinity or the target location vicinity is the query object content.

[0203] When the fourth query object is a missing minimum execution content, read the content in the minimum execution content set corresponding to the target continuation intent that has not been written with the corresponding intent completion result; when the target continuation intent is a route constraint adjustment intent and the route constraint content is missing, determine which route constraint needs to be added, deleted, or replaced as the query object content; when the target continuation intent is a service retrieval intent along the way and the service content along the way is missing, determine which type of service along the way needs to be queried as the query object content; when the target continuation intent is a navigation description query intent and the description query content is missing, determine whether the remaining distance, remaining time, or next navigation action needs to be queried as the query object content.

[0204] The specific methods for generating query content include: Once the content of the inquiry object has been determined, a voice clarification request is generated based on the object type corresponding to the content of the inquiry object; when the content of the inquiry object corresponds to the direction of continuation processing, inquiry content is generated to distinguish different continuation intent categories; when the content of the inquiry object corresponds to the content of the execution object, inquiry content is generated to distinguish different candidate object content or supplement object name content; when the content of the inquiry object corresponds to the content of the execution scope, inquiry content is generated to distinguish different scope of action content or location range content; when the content of the inquiry object corresponds to the missing minimum execution content, inquiry content is generated to supplement the minimum execution content.

[0205] In some implementations, the specific form in which the query content is generated includes: When the candidate continuation intent category is not uniquely determined, the voice clarification request guides the user to choose among multiple continuation processing directions; when multiple candidate objects have corresponding results, the voice clarification request guides the user to supplement the object name, location limitation content, category limitation content, or sequence position; when no corresponding result exists, the voice clarification request guides the user to resubmit object content that corresponds to the candidate object range defined by the most recent navigation feedback result; when more than two route constraint contents are retained in the pending confirmation result, the voice clarification request guides the user to supplement the route constraint contents that need to be retained, deleted, or replaced; when more than two along-the-way service contents are retained in the pending confirmation result, the voice clarification request guides the user to supplement the service category content or scope content; when the missing minimum execution content exists, the voice clarification request guides the user to supplement the minimum execution content corresponding to the target continuation intent; after this step, a voice clarification request is formed for use in the next step.

[0206] As another example, to illustrate the process of generating a speech clarification request, for example: The most recent navigation feedback result limited the candidate object range to include three candidate objects: the Jing'an Temple store corresponding to output order 1, the Nanjing West Road store corresponding to output order 2, and the Yan'an Middle Road store corresponding to output order 3. The current voice input was "go to that". After parsing in step 301, only referential fragments were formed in the current voice input, without forming other semantic fragments that could uniquely define the object name, location, or sequence position. After processing in step 303, no unique candidate object reference relationship was formed in the candidate object reference relationship; instead, the results corresponding to the three candidate objects were retained. After processing in steps 401 and 402, although the candidate continuation meaning... The image category can be limited to the candidate object selection intent, but since the execution object content is still not unique, the preset execution integrity condition for generating the navigation execution request is not met in step 501. At this time, in step 504, the non-unique determination of the execution object content is determined as the source of clarification, and the output order and position limitation content corresponding to the three candidate object contents are determined as the inquiry object content, further forming the voice clarification request "Is it the first Jing'an Temple store, the second Nanjing West Road store, or the third Yan'an Middle Road store?" When the user's next voice input is "the second one", the candidate object content corresponding to the output order 2 is uniquely determined and continues to participate in the subsequent steps of processing.

[0207] Step 505: The voice clarification request generated in step 504 is delivered to the vehicle voice output component for broadcast; after the broadcast is completed, the next round of voice input from the user is received; the next round of voice input is incorporated into the current navigation session, and the content of the question object, the result to be confirmed, the results corresponding to multiple candidate objects, the result without a corresponding object, and the minimum execution content corresponding to the voice clarification request in step 504 are included as part of the historical voice interaction content corresponding to the next round of voice input in the subsequent processing of steps 103, 201 and 206.

[0208] The specific methods for including the content corresponding to the voice clarification request in the historical voice interaction content of the next round of voice input include: Write the content of the inquiry object that has been identified in step 504 into the historical voice interaction content corresponding to the next round of voice input, so as to identify the type of object targeted by the current round of clarification. The pending confirmation results retained in step 504 are written into the historical voice interaction content corresponding to the next round of voice input, so as to continue to determine whether the execution object content or execution range content is unique after the next round of voice input arrives; The results of multiple candidate objects retained in step 504, or the results of no corresponding objects, are written into the historical voice interaction content corresponding to the next round of voice input, so as to continue to determine the candidate object reference relationship after the next round of voice input arrives; The missing minimum execution content retained in step 504 is written into the historical voice interaction content corresponding to the next round of voice input. This is used to continue to determine whether the minimum execution content corresponding to the target continuation intention is complete after the next round of voice input arrives. After processing in this step, the connection result of subsequent rounds of input is formed, so that the clarification process caused by the result to be confirmed, the result corresponding to multiple candidate objects, the result without corresponding result or the missing minimum execution content can continue to be completed in the next round of voice input.

[0209] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims.

Claims

1. A method for contextual intent continuation in vehicle navigation voice, characterized in that, include: Collect navigation task stage information corresponding to the current navigation session, historical voice interaction results, the most recent navigation feedback result, and the current route association status; Based on the navigation task stage information, the historical voice interaction results, the most recent navigation feedback results, and the historical interaction content corresponding to the current route association status are classified as retained, cleared, and confirmed to determine the inheritable context range corresponding to the current round of voice and generate the context effective range result. Semantic parsing is performed on the current round of voice input. Within the candidate object range defined by the most recent navigation feedback result, the reference relationship of the candidate object is determined, and the continuation semantic result is generated by combining the effective range result of the context and the current route association status. Based on the navigation task stage information, candidate object reference relationships and continuation semantic results, the continuation intent corresponding to the current round of speech is divided, judged and completed to determine the target continuation intent and the corresponding intent completion result; Based on the target continuation intent and the corresponding intent completion result, generate a navigation execution request or a voice clarification request.

2. The method for contextual intent continuation of in-vehicle navigation voice according to claim 1, characterized in that, Methods for generating context-validated range results include: Based on navigation task stage information, the most recent navigation feedback result, and the current route association status, historical interaction content is classified into retention, clearing, and confirmation categories, resulting in directly inherited content, excluded content, and content pending confirmation. Merge directly inherited content, excluded content, and content to be confirmed, and determine the scope of inheritable context based on the merging results, generating the valid scope of context results.

3. The method for contextual intent continuation of in-vehicle navigation voice according to claim 1, characterized in that, Methods for determining candidate object reference relationships include: The range of candidate objects is determined based on the feedback type of the most recent navigation feedback result; The current round of speech input is parsed to obtain the ordinal segment, explicit object segment, and referential segment; Within the candidate object range, each fragment is compared with the content of each candidate object in the candidate object range in the order of ordinal fragment, explicit object fragment, and referential fragment. The candidate object reference relationship is determined when only one candidate object content corresponds to the smallest level ordinal number.

4. The method for contextual intent continuation of in-vehicle navigation voice according to claim 3, characterized in that, Methods for generating continuation semantic results include: Read the candidate object reference relationships, the context scope results, and the current route association status; extract the directly inherited results, excluded results, and pending confirmation results from the context scope results. The content corresponding to the current round of speech input and the content of the candidate object reference relationship in the directly inherited results will be incorporated into the semantic interpretation scope of the current round, the content corresponding to the excluded results will be removed, and the results to be confirmed will be retained; Based on the semantic interpretation scope of the current round, the results to be confirmed, and the current route association status, a continuation semantic result is generated.

5. The method for contextual intent continuation of in-vehicle navigation voice according to claim 4, characterized in that, Methods for determining the target's continuation intent include: Based on navigation task stage information, candidate object reference relationships, continuation semantic results, current route association status, and direct inheritance results and pending confirmation results in the context's effective range, determine the candidate continuation intent category; The target continuation intent is determined based on the candidate continuation intent category, the results to be confirmed, and the results corresponding to multiple candidate objects and the results without corresponding objects in the candidate object reference relationship.

6. The method for contextual intent continuation of in-vehicle navigation voice according to claim 5, characterized in that, Methods for determining the corresponding intent completion result include: When forming a target continuation intent, the corresponding intent completion result is determined based on the target continuation intent, the continuation semantic result, and the direct inheritance result in the context's valid scope result. When no target continuation intent is formed, the corresponding intent completion result is determined based on the candidate continuation intent category, the unconfirmed results in the context's valid scope, and the corresponding results and no corresponding results of multiple candidate objects in the candidate object reference relationship.

7. A method for contextual intent continuation of in-vehicle navigation voice according to any one of claims 1-6, characterized in that, Methods for generating navigation execution requests include: When the target continuation intent and intent completion result meet the preset execution integrity conditions, the request category is determined based on the target continuation intent; Extract the execution object content and execution scope content from the intent completion results based on the request category; Based on the request type, the content of the object to be executed, and the scope of execution, a navigation execution request is generated.

8. The method for contextual intent continuation of in-vehicle navigation voice according to claim 7, characterized in that, Methods for generating voice clarification requests include: Based on the pending confirmation results in the context's valid range results, the corresponding results of multiple candidate objects in the candidate object reference relationship, and the results without corresponding results, or the content in the minimum execution content corresponding to the target continuation intent that does not contain the intent completion result, determine the content of the query object; Generate a voice clarification request based on the content of the inquiry.

9. A method for contextual intent continuation in vehicle navigation voice according to claim 2, characterized in that, Methods for determining the scope of inheritable contexts based on the merge results include: Based on the excluded content, the directly inherited content and the content to be confirmed are removed; Based on the content to be confirmed, the content that is not distinguished in the current round of voice input will be transferred to the content to be confirmed. The scope of inheritable context is determined based on the directly inherited content after the removal process and the content to be confirmed after the transfer process.

10. The method for contextual intent continuation of in-vehicle navigation voice according to claim 3, characterized in that, Methods for comparing each fragment with the content of each candidate object in the candidate object range include: Based on the matching relationship between each segment and the content of each candidate object, multiple corresponding levels are determined for each candidate object content; the multiple corresponding levels are compared according to a preset sorting order. When the corresponding level of only one candidate object is in a lower position than the corresponding level of other candidate objects in the preset sorting order, the candidate object reference relationship is determined. If the corresponding level of two or more candidate objects is in the same position in the preset sorting order and is lower than the corresponding level of other candidate objects, multiple candidate object results are retained.

Images