A cross-language instant translation collaborative interaction method

Abstract

The application discloses a cross-language instant translation collaborative interaction method, and relates to the technical field of natural language processing, which comprises the following steps: receiving a session creation request and generating a session identifier, registering a participant identifier, a source language identifier, a target language identifier and a role permission set to form session metadata; applying a ruling semantic patch to a corresponding candidate translation fragment according to a preset priority order to form an effective translation, and writing the update record of the effective translation into an event log as an event; when the candidate translation fragment causes the alignment relationship of the target translation fragment to change, performing rollback processing on the applied ruling semantic patch according to the event log and generating a new candidate translation fragment. The application structures the translation modification process for record and orderly application, so that the translation evolution has clear structure boundaries and evolution paths, and the controllability, explainability and overall stability of the translation update process in the cross-language instant translation collaborative interaction are improved.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a cross-language real-time translation collaborative interaction method. Background Technology

[0002] With the development of information technology, the application of natural language processing (NLP) in cross-language communication has been continuously deepened. Related technologies have evolved from early rule-driven translation to automatic translation systems based on statistical and neural network models. Current cross-language real-time translation technologies typically combine speech recognition, text processing, and semantic modeling to achieve real-time understanding and output of multilingual input content. These technologies are widely used in scenarios such as collaborative meetings, online communication, and multi-party communication, with NLP playing a core role in semantic parsing, context modeling, and translation generation.

[0003] However, existing cross-language real-time translation technologies based on natural language processing typically handle translation modifications in a linear or overall overlay manner in multi-party collaborative interaction scenarios. This makes it difficult to manage concurrent modification behavior in a structured way. Especially when fine-tuning the local semantics of the translation is required, it can easily disrupt the internal grammatical consistency and semantic correspondence of the translation, thereby affecting the stability and traceability of the translation evolution process. This problem has become a major technical bottleneck restricting the collaborative translation interaction experience and quality control. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a cross-language real-time translation collaborative interaction method to solve the problems of insufficient semantic consistency of translation and difficulty in controlling the translation evolution process in scenarios with concurrent modifications by multiple participants.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a cross-language real-time translation collaborative interaction method, which includes: receiving a session creation request and generating a session identifier; registering participant identifiers, source language identifiers, target language identifiers, and a set of role permissions to form session metadata; accessing the input stream based on the session metadata and performing turn-taking and semantic fragmentation processing to generate candidate translation fragments corresponding to the source language fragments, and publishing the candidate translation fragments as read-only translations; receiving the modification intentions of the participants based on the read-only translation state, and converting the modification intentions into translation semantic patches pointing to the target translation fragments; constructing a conflict domain for multiple translation semantic patches pointing to the same target translation fragment, and performing adjudication processing on the semantic patches in the conflict domain according to the set of role permissions to generate adjudication semantic patches; applying the adjudication semantic patches to the corresponding candidate translation fragments in a preset priority order to form effective translations, and writing the update record of the effective translations as an event to an event log; when a candidate translation fragment causes a change in the alignment relationship of the target translation fragment, performing rollback processing on the applied adjudication semantic patches according to the event log and generating new candidate translation fragments.

[0007] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the specific steps for forming session metadata are as follows: Parse the session creation request, extract participant identifiers, source language identifiers, target language identifiers, and role and permission information to form a session parameter set; The uniqueness of a session is determined based on the set of session parameters, and a session identifier corresponding to the set of session parameters is generated. The session identifier is associated with the participant identifier, source language identifier, target language identifier, and role permission set in the session parameter set, and then structured to generate session metadata.

[0008] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the specific steps for generating candidate translation fragments corresponding to the source language fragments are as follows: Access input streams from participants in both voice and text formats; The input stream is segmented based on temporal continuity and semantic integrity to form multiple independent talk turns; Semantic boundary recognition processing is performed on each turn of speech to generate a semantically relatively complete source language fragment; Based on the source language identifier and the target language identifier, cross-language translation processing is performed on each source language segment to generate an initial translation; The initial translation is formatted and consistent, and candidate translation fragments corresponding to the source language fragments are output.

[0009] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the modification intention of the participating party includes modification instructions for candidate translation fragments at the levels of vocabulary, grammar, word order and style, as well as position anchor information and semantic modification description attached to the modification instructions.

[0010] As a preferred embodiment of the cross-language real-time translation collaborative interaction method described in this invention, the specific steps for converting the modification intent into a semantic patch pointing to the target translation fragment are as follows: Based on the anchor point information, identify the target translation fragment in the read-only translation state, and read the corresponding grammatical structure representation of the target translation fragment; The semantic modification description is parsed to determine the semantic type of the modification behavior and its position in the target translated text segment; By combining the grammatical structure representation of the target translation fragment, the semantic type and position of action are converted into a sequence of operations that can be executed at the grammatical structure level; The operation sequence is associated with the target translation fragment to form a semantic patch of the translation fragment pointing to the target translation fragment.

[0011] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the specific steps for constructing the conflict domain are as follows: Identify all semantic patches that point to the same target translated fragment, and extract the syntax tree node path operated by each semantic patch; Compare the syntax tree node paths of semantic patches for different translations to determine whether there is overlap in the scope of operations; All semantic patches of translations with overlapping operational scopes are grouped together to form a conflict domain.

[0012] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the adjudication process refers to, based on the set of role permissions, judging the availability and prioritizing of the translation semantic patches in the conflict domain, and selecting the translation semantic patch as the adjudication semantic patch.

[0013] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the preset priority order refers to sorting the adjudication semantic patches according to the predefined role level order in the role permission set, and adjudication semantic patches with the same role level are sorted according to the submission order of the translation semantic patches.

[0014] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the change in the alignment relationship of the target translation fragment refers to the change in the semantic correspondence between the source language fragment and the target translation fragment due to the update of the candidate translation fragment.

[0015] As a preferred embodiment of the cross-language real-time translation collaborative interaction method of the present invention, the specific steps for performing rollback processing on the applied adjudication semantic patch and generating new candidate translation fragments based on the event log are as follows: When a change in the alignment of the target translation fragment is detected, query the event log to retrieve the application record of the adjudication semantic patch associated with the target translation fragment; In the order opposite to the preset priority order, the modifications made to the candidate translation fragments by the semantic patch of the ruling are revoked one by one; New candidate translation fragments are generated based on the withdrawn candidate translation fragments, and the correspondence between the source language fragments and the target translation fragments is updated according to the new candidate translation fragments.

[0016] The beneficial effects of this invention are as follows: by converting the modification intentions of multiple participants into semantic patches of the target translation fragment within a natural language processing framework, and expressing the modification behavior as an operational sequence at the grammatical structure level, the semantic scope of translation modification can be precisely defined, thereby maintaining the stability of the grammatical structure and the consistency of semantic expression of the translation during collaborative interaction; by recording and applying the translation modification process in a structured manner, the evolution of the translation has clear structural boundaries and evolutionary paths, which is conducive to improving the controllability, interpretability and overall stability of the translation update process in cross-language real-time translation collaborative interaction. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for a cross-language real-time translation collaborative interaction method.

[0019] Figure 2 A flowchart for generating session metadata.

[0020] Figure 3 A flowchart for generating semantic patches for the translated text.

[0021] Figure 4 A flowchart for applying the adjudication semantic patch. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a cross-language real-time translation collaborative interaction method, including the following steps: S1: Receives a session creation request and generates a session identifier, registers participant identifiers, source language identifiers, target language identifiers, and role and permission sets, forming session metadata.

[0026] S1.1: Parse the session creation request, extract the participant identifier, source language identifier, target language identifier, and role permission information to form a session parameter set.

[0027] Specifically, it receives a session creation request and performs structured parsing on the request to read the participant identifier field, source language identifier field, target language identifier field, and role permission information field.

[0028] Perform validity checks on the participant identifier field, including at least non-empty participant identifier checks and duplicate participant identifier checks, and retain the validated participant identifiers; perform validity checks on the source language identifier field and the target language identifier field, including at least non-empty source language identifier checks, non-empty target language identifier checks, and checks that the source language identifier and the target language identifier are different, and retain the validated source language identifier and the target language identifier; perform parsing processing on the role permission information field to obtain the role permission set, and perform non-empty checks and duplicate item removal processing on the role permission set.

[0029] The set of role permissions includes translation initiation permissions, translation editing permissions, translation adjudication permissions, version confirmation permissions, and session termination permissions, all associated with the participant's identifier.

[0030] The processed participant identifiers, source language identifiers, target language identifiers, and role permission sets are written into the same data structure in a fixed field order to form a session parameter set.

[0031] S1.2: Determine the uniqueness of the session based on the session parameter set, and generate a session identifier corresponding to the session parameter set.

[0032] Specifically, the session parameter set is normalized. The normalization process includes sorting the participant identifiers according to fixed rules and sorting the role and permission sets according to fixed rules to obtain a normalized session parameter set.

[0033] The normalized session parameter set is concatenated into a parameter combination sequence according to a preset fixed field order. The parameter combination sequence is then encoded field by field and concatenated in order to form a normalized string. The normalized string is then subjected to an irreversible digest calculation according to fixed rules to generate a unique and repeatable session identifier.

[0034] The formula for calculating irreversible digests is: ; In the formula, This indicates the generated session identifier. Irreversible digest mapping functions (such as hash functions) are used to map input to fixed-length identifiers. To normalize the total number of parameter fields in the session parameter set, The first one is arranged according to a preset fixed field order. Each session parameter field Indicates the parameter field The deterministic encoding result of the execution, Indicates the relationship with the first The fixed weight coefficients corresponding to each session parameter field are used to distinguish the positional contribution of different fields in the summary calculation.

[0035] It should be noted that, due to Map different types of session parameter fields to a unified dimensionless numerical representation. Dimensionless fixed weighting coefficients The calculation result is a dimensionless scalar, which is obtained through an irreversible abstract mapping function. Even after processing, it remains dimensionless, so all parameters and calculation results in the formula remain consistent in terms of dimensions.

[0036] It should also be noted that the preset fixed field order is arranged according to the field identifier order defined in the normalized session parameter set. The field identifier order is determined by the order in which each parameter field appears in the parameter definition list within the normalized session parameter set and remains unchanged throughout the entire session. The fixed weight coefficient is determined based on the position number of the session parameter field in the preset fixed field order and is taken as a positive integer value corresponding one-to-one with the position number. For example, the value range is... .

[0037] The uniqueness determination is performed based on the session identifier. The uniqueness determination includes at least querying the existing set of session identifiers and determining whether the session identifier already exists. If the session identifier already exists, the existing session identifier is directly used as the session identifier. If the session identifier does not exist, the session identifier is registered in the set of session identifiers.

[0038] S1.3: Establish associations between the session identifier and the participant identifier, source language identifier, target language identifier, and role permission set in the session parameter set, and organize them in a structured manner to generate session metadata.

[0039] Specifically, using the session identifier as the index key, a correspondence is established between the session identifier and the participant identifier, the source language identifier, the target language identifier, and the set of role permissions.

[0040] The establishment of the correspondence includes: writing the session identifier and its corresponding participant identifier, source language identifier, target language identifier, and role permission set as associated fields into the same structured data object, and forming a one-to-one index mapping relationship between the session identifier and each associated field through field binding.

[0041] After completing the index mapping relationship, the structured data object is registered as session metadata.

[0042] S2: Based on the session metadata, access the input stream and perform turn-taking and semantic fragmentation processing to generate candidate translation fragments corresponding to the source language fragments, and publish the candidate translation fragments as read-only translations.

[0043] S2.1: Access the input streams in both voice and text formats from the participants, and perform segmentation on the input streams based on temporal continuity and semantic integrity to form multiple independent talk turns.

[0044] Specifically, based on the participant identifier in the session metadata, the voice input stream and the text input stream are accessed. After the voice input stream is converted into the corresponding text, it is merged with the text input stream in timestamp order to form a unified text input stream.

[0045] A unified text input stream is scanned continuously in timestamp order.

[0046] When the time interval between adjacent texts exceeds the preset interval, the current turn ends and the next turn begins; when the time interval between adjacent texts does not exceed the preset interval, the texts continue to be merged.

[0047] The preset interval is determined during the initialization phase based on the statistical distribution of timestamps of adjacent texts in the unified text input stream. Specifically, it is the time length corresponding to the mean plus standard deviation of the time interval between adjacent texts, with an example value of 2 seconds.

[0048] After completing the initial turn segmentation based on time continuity, semantic integrity segmentation is performed within each turn according to the sentence-ending punctuation and line break position, eliminating cross-sentence merging and obtaining multiple semantically complete turns.

[0049] The various turns obtained from the further segmentation are registered as a turn set in chronological order. Each turn in the turn set contains at least the participant identifier, timestamp range, and turn text content.

[0050] S2.2: Perform semantic boundary recognition processing on each talk turn to generate a semantically relatively complete source language fragment.

[0051] Specifically, the text content of each turn is scanned from front to back, and the positions of sentence-end punctuation and commonly used conjunctions are used as candidate points for semantic boundaries. The text between adjacent semantic boundaries is divided into semantic segments.

[0052] When the semantic segment is too short, it is merged with temporally adjacent semantic segments to obtain a semantically relatively complete source language segment, and the source language segments are registered in chronological order.

[0053] A semantic fragment that is too short means that the number of characters in the semantic fragment is less than the preset minimum length threshold. The minimum length threshold is determined during the initialization phase based on the statistical distribution of the number of characters in the semantic fragment in the turn-by-turn text. The example value is 10 characters. A relatively complete source language fragment means a continuous text fragment that, after merging, has a number of characters that is not less than the minimum length threshold and does not cross the sentence-ending punctuation.

[0054] For example, the turn-taking text "Yes, but further confirmation is needed" can be initially divided into two semantic segments based on semantic boundaries: "Yes" and "but further confirmation is needed". The "Yes" segment has fewer than 10 characters and is considered too short. It is then merged with the adjacent semantic segment to form the source language segment "Yes, but further confirmation is needed".

[0055] S2.3: Based on the source language identifier and the target language identifier, perform cross-language translation processing on each source language segment to generate an initial translation; perform uniform formatting and consistency processing on the initial translation, and output candidate translation segments corresponding to the source language segments.

[0056] Specifically, the source language identifier and target language identifier are read based on the session metadata.

[0057] For each source language segment, sentence-level cross-language translation is performed based on the source language identifier and the target language identifier, generating an initial translation that corresponds one-to-one with the source language segment. Sentence-level cross-language translation refers to using a single source language segment as the smallest translation unit, determining the input language based on the source language identifier and the output language based on the target language identifier, and performing a single translation on the entire source language segment to generate the corresponding translation.

[0058] The initial translation is formatted uniformly to standardize whitespace, punctuation, and line breaks; the formatted initial translation is then processed for consistency to ensure that terminology, names, and numbers are expressed consistently.

[0059] The processed translation is registered as a candidate translation fragment, and a one-to-one correspondence is established with the corresponding source language fragment. The candidate translation fragment is then published as a read-only translation.

[0060] The set of candidate translation fragments is published as a read-only translation fragment, and the read-only translation fragment is associated with the session identifier to receive modification intent and locate the target translation fragment.

[0061] S3: Receive the modification intent from the participants based on the read-only translation state, and convert the modification intent into a semantic patch for the target translation fragment.

[0062] S3.1: The modification intentions of the participating parties include modification instructions for candidate translation fragments at the levels of vocabulary, grammar, word order and style, as well as the positional anchor information and semantic modification descriptions attached to the modification instructions.

[0063] Specifically, based on the location anchor information, the target translation fragment is identified in the read-only translation state, and the corresponding grammatical structure representation of the target translation fragment is read.

[0064] Based on the session identifier, the modification intent submitted by the participant is received. The modification intent includes the modification instruction, location anchor information, and semantic modification description, and the participant identifier is used as the attribution field of the modification intent. Figure 1 And record; among them, semantic modification description refers to the textual description of the expected modification result of the target translation fragment, which is used to characterize the modified semantic content, expression mode or language style.

[0065] The location anchor information is defined as a triple that can be directly located. The triple contains the candidate translation fragment identifier, the start position, and the end position. The start position and end position are represented by the character sequence number within the candidate translation fragment.

[0066] In read-only translation mode, locate the corresponding candidate translation fragment based on the candidate translation fragment identifier, and extract the modified text range according to the start and end positions. The candidate translation fragment containing the modified text range is determined as the target translation fragment.

[0067] The target translation fragment undergoes syntactic parsing, outputting a corresponding grammatical structure representation. This grammatical structure representation is then linked to the target translation fragment and the participant identifiers in the modification intent. The grammatical structure representation is a syntactic tree structure used to characterize the internal grammatical relations of the target translation fragment.

[0068] S3.2: Analyze the semantic modification description to determine the semantic type corresponding to the modification behavior and its position in the target translation fragment.

[0069] Specifically, the semantic modification description is read, and the modification behavior category, modification content, and scope are parsed according to the pre-defined field order in the semantic modification description. The modification behavior category is limited to one of the following: replacement, insertion, deletion, rearrangement, and style adjustment.

[0070] The modification behavior category is determined as the semantic type, and the scope of application is determined as the position of application. The position of application is represented by the character sequence range within the target translation fragment. The position of application is directly taken from the start and end positions corresponding to the position anchor information. In the grammatical structure representation, the grammatical structure node that overlaps with the position of application is located.

[0071] Among them, a grammatical structure node refers to a structured representation in the grammatical structure representation used to characterize the position of words, phrases or syntactic relations within the target translation fragment.

[0072] S3.3: Combining the grammatical structure representation of the target translation fragment, convert the semantic type and position of action into an operation sequence that can be executed at the grammatical structure level.

[0073] Specifically, based on the syntax structure representation, the syntax structure node corresponding to the position of action is read, and an operation sequence is generated based on the semantic type. The operation sequence consists of atomic operations arranged in sequence.

[0074] When the semantic type is replacement, the operation sequence includes three steps: locating the grammatical structure node, deleting the modified content within the grammatical structure node, and inserting the modified content. When the semantic type is insertion, the operation sequence includes two steps: locating the insertion point and inserting the modified content. When the semantic type is deletion, the operation sequence includes two steps: locating the grammatical structure node and deleting the modified content. When the semantic type is rearrangement, the operation sequence includes two steps: locating the rearrangement scope and adjusting the order of the grammatical structure nodes. When the semantic type is style adjustment, the operation sequence includes two steps: locating the scope of application and replacing the corresponding expression according to the semantic modification description.

[0075] It should be noted that the insertion point is determined based on the character number position corresponding to the position anchor information within the target translation fragment. The reordering range is limited to the character number interval covered by the position anchor information. The location of the grammatical structure nodes is determined based on the nodes in the grammatical structure representation that overlap with the character number interval. Furthermore, the adjustment of the order of the grammatical structure nodes is only performed within the reordering range, thereby ensuring that all types of operations are applied to executable and semantically consistent grammatical structure positions.

[0076] S3.4: Associate the operation sequence with the target translation fragment to form a semantic patch for the target translation fragment.

[0077] Specifically, the operation sequence is associated with the target translation fragment, and the candidate translation fragment identifier, start position, end position, semantic type and participant identifier are written into the same record object to form a translation semantic patch.

[0078] The translated semantic patch is registered as a translated semantic patch pointing to the target translated fragment, and the output of the translated semantic patch is used as the input for subsequent conflict domain construction and adjudication processing.

[0079] A better approach is to convert the modification intentions of the participants into semantic patches of the target translation fragment, and express the modification behavior with atomic operation sequences at the grammatical structure level. This allows the translation modification to be precisely applied to a locatable grammatical position, avoiding the semantic interference problems that are easily generated in existing methods based on full-text replacement or character range modification. This provides a stable and consistent structured foundation for conflict detection, adjudication and rollback processing in multi-participant concurrent collaboration scenarios.

[0080] S4: Construct a conflict domain for multiple semantic patches that point to the same target translated fragment, and perform adjudication processing on the semantic patches in the conflict domain according to the set of role permissions to generate adjudicated semantic patches.

[0081] S4.1: Identify all semantic patches that point to the same target translation fragment and extract the syntax tree node path operated by each semantic patch.

[0082] Specifically, the semantic patch set is filtered according to the candidate translation fragment identifier, and the semantic patches corresponding to the same target translation fragment are grouped according to the start and end positions.

[0083] For each semantic patch in the semantic patch group, read the position and grammatical structure representation recorded in the operation sequence, locate the syntax tree node corresponding to the position in the grammatical structure representation, and record the ordered node identifier sequence from the root node to the syntax tree node as the syntax tree node path.

[0084] S4.2: Compare the syntax tree node paths of different semantic patches in the translation to determine whether there is any overlap in the scope of operation.

[0085] Specifically, for each semantic patch of the translation, the scope of operation is determined based on the path of the syntax tree node. The scope of operation is defined as the set of syntax tree nodes consisting of the end syntax tree node of the path and all descendant syntax tree nodes of the end syntax tree node.

[0086] For any two semantic patches of the translation, read the set of syntax tree nodes corresponding to the two operation ranges respectively, and calculate the intersection of the syntax tree node sets; if the intersection of the syntax tree node sets is not empty, it is determined that the operation ranges overlap; if the intersection of the syntax tree node sets is empty, it is determined that the operation ranges do not overlap.

[0087] The determination formula is: ; In the formula, and These represent two semantic patches for the translation. and These represent the sets of syntax tree nodes that represent the scope of operations corresponding to the two semantic patches in the translation. The result of the overlap determination takes the value of a dimensionless scalar, either 0 or 1. The empty set represents the set of nodes in the syntax tree whose operation scopes corresponding to two semantic patches of the translation do not have any common elements, that is, there are no overlapping syntactic structure nodes.

[0088] It should be noted that the subscript... This indicates the first semantic patch selected in the current set of semantic patches participating in overlap determination, according to the determination order. The subscript... This refers to another semantic patch in the same set of semantic patches that is used to determine the overlap relationship with the first semantic patch.

[0089] S4.3: Collect all semantic patches of translations with overlapping operational scopes to form a conflict domain.

[0090] Specifically, within the semantic patch group, an overlap graph is established using the semantic patches as nodes. Any two nodes in the overlap graph satisfy... Establish connections in time.

[0091] Connectivity component extraction is performed on the overlapping graph. The semantic patches of the translated text contained in each connected component are aggregated into a conflict domain, and the conflict domain is associated with the target translated text fragment.

[0092] S4.4: Adjudication processing refers to the process of determining the availability and prioritizing of translation semantic patches within the conflict domain based on the set of role permissions, and selecting the translation semantic patch as the adjudication semantic patch.

[0093] Specifically, a usability assessment is performed on each semantic patch of the translation within the conflict domain. The usability assessment adopts a deterministic verification process, which includes at least the following: the existence of a syntax tree node pointed to by the operation sequence in the syntax structure representation; the existence of a corresponding execution position for each atomic operation in the operation sequence in the syntax structure representation; and the absence of any missing mandatory syntax components after the operation sequence is executed. A semantic patch of the translation that passes the deterministic verification is recorded as a usable semantic patch of the translation.

[0094] Based on the set of role permissions in the session metadata, available translation semantic patches are sorted from high to low role level. Available translation semantic patches with the same role level are sorted from earliest to latest submission order to obtain the priority sorting result.

[0095] Select the top-ranked available translation semantic patch from the priority ranking results, register the selected result as the adjudication semantic patch, establish the association between the adjudication semantic patch and the target translation fragment, and output it.

[0096] For example, if the role levels in the role permission set are, from highest to lowest, reviewer and translator, and there is one semantic patch for the translation submitted by the reviewer and two semantic patches for the translation submitted by the translator, then the priority ranking result for the semantic patches is that the semantic patch submitted by the reviewer has priority, followed by the two semantic patches submitted by the translator in the order of submission time.

[0097] S4.5: Apply the adjudication semantic patch to the corresponding candidate translation fragments in a preset priority order to form the effective translation, and write the update record of the effective translation as an event to the event log.

[0098] It should be noted that the preset priority order refers to the order in which the application processing of the adjudication semantic patch is performed according to the priority sorting result determined based on the role permission set in S4.4.

[0099] Specifically, following the ordered sequence of adjudication semantic patches, the operation sequences corresponding to the adjudication semantic patches are read sequentially, and the operation sequences are applied to the candidate translation fragments corresponding to the target translation fragment.

[0100] After completing the entire sequence of operations, the text content of the candidate translation fragments is updated to obtain the effective translation after the adjudication is applied, and a correspondence is established between the effective translation and the target translation fragment. After generating the effective translation, an effective translation update record is constructed, which includes at least the session identifier, the target translation fragment identifier, the adjudication semantic patch identifier, the effective translation content, and the update time.

[0101] The effective translation update records are written to the event log in the order of generation, so that the event log forms a translation update track ordered by time.

[0102] A better approach is to determine the overlap of the operational scope of semantic patches in the translation at the syntax tree level and construct a conflict domain, and then combine this with a set of role permissions to perform deterministic adjudication. This enables concurrent modifications to the same target translation fragment by multiple parties to automatically identify conflicts and take effect in an orderly manner. Compared with the existing processing method that covers the translation in chronological order, this approach effectively maintains the consistency of the translation structure and enhances the controllability and traceability of the translation evolution process.

[0103] S5: When a candidate translation fragment causes a change in the alignment of the target translation fragment, rollback is performed on the applied adjudication semantic patch based on the event log and a new candidate translation fragment is generated.

[0104] S5.1: When a change in the alignment of the target translation fragment is detected, query the event log to obtain the application record of the adjudication semantic patch associated with the target translation fragment.

[0105] The change in alignment of the target translation fragment refers to the change in the semantic correspondence between the source language fragment and the target translation fragment due to the update of the candidate translation fragment.

[0106] Specifically, after the candidate translation fragments are updated, the semantic correspondence between the current candidate translation fragments and the source language fragments is recalculated based on the established semantic correspondence between the source language fragments and the target translation fragments.

[0107] When the recalculated semantic correspondence is inconsistent with the semantic correspondence recorded before the update, it is determined that the alignment relationship of the target translation fragment has changed.

[0108] After determining that the alignment relationship of the target translation fragment has changed, the event log is queried based on the target translation fragment identifier to obtain the application records of the adjudication semantic patch written in a preset priority order, and a sequence of adjudication semantic patches to be rolled back is formed according to the record order.

[0109] S5.2: Revoke the modifications made to candidate translation fragments by the adjudication semantic patch one by one in the order opposite to the preset priority order.

[0110] Specifically, based on the sequence of semantic patches for the ruling to be rolled back, the ruling semantic patches are read one by one in the order opposite to the preset priority order.

[0111] For each ruling semantic patch, read the corresponding operation sequence and perform the inverse operation of the operation sequence to undo the modifications made by the ruling semantic patch to the candidate translation fragments one by one.

[0112] The inverse operation of the operation sequence refers to performing the operations that are semantically opposite to the corresponding atomic operations in the reverse order of the execution order of the atomic operations contained in the adjudication semantic patch, so as to restore the candidate translation fragment to the text state before the execution of the corresponding atomic operation.

[0113] After completing all the reverse operations, we obtain the candidate translation fragments after revoking all the semantic patch modifications of the rulings.

[0114] S5.3: Regenerate new candidate translation fragments based on the withdrawn candidate translation fragments, and update the correspondence between the source language fragments and the target translation fragments according to the new candidate translation fragments.

[0115] Specifically, the candidate translation fragment generation process is re-executed based on the withdrawn candidate translation fragments.

[0116] The generation process includes performing structured processing, semantic segmentation, and semantic consistency verification on candidate translation fragments to generate new candidate translation fragments.

[0117] Based on the new candidate translation fragments, the semantic correspondence between the source language fragments and the target translation fragments is recalculated, and the updated semantic correspondence is written into the session association record.

[0118] In summary, this invention transforms the modification intentions of multiple participants into semantic patches for the target translation fragment within a natural language processing framework, and expresses the modification behavior as an operational sequence at the grammatical structure level. This allows for precise definition of the semantic scope of translation modification, thereby maintaining the stability of the translation's grammatical structure and the consistency of its semantic expression during collaborative interaction. By structurally recording and orderly applying the translation modification process, the translation evolution possesses clear structural boundaries and evolutionary paths, which is beneficial for improving the controllability, interpretability, and overall stability of the translation update process in cross-language real-time translation collaborative interaction.

[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cross-language real-time translation collaborative interaction method, characterized in that: include, Receive a session creation request and generate a session identifier, register the participant identifier, source language identifier, target language identifier, and role and permission set to form session metadata; Based on the session metadata, the input stream is accessed and the turn-taking and semantic fragmentation processes are performed to generate candidate translation fragments corresponding to the source language fragments. The candidate translation fragments are then published as read-only translation fragments. Based on the read-only translation status, the modification intentions of the participants are received and converted into semantic patches for the target translation fragment. For multiple semantic patches of translated text pointing to the same target translated text fragment, a conflict domain is constructed, and the semantic patches within the conflict domain are adjudicated according to the set of role permissions to generate adjudicated semantic patches; The adjudication semantic patch is applied to the corresponding candidate translation fragments in a preset priority order to form the effective translation, and the update record of the effective translation is written to the event log as an event; When a candidate translation fragment causes a change in the alignment of the target translation fragment, the applied adjudication semantic patch is rolled back based on the event log and a new candidate translation fragment is generated.

2. The cross-language real-time translation collaborative interaction method as described in claim 1, characterized in that: The specific steps for generating session metadata are as follows: Parse the session creation request, extract participant identifiers, source language identifiers, target language identifiers, and role and permission information to form a session parameter set; The uniqueness of a session is determined based on the set of session parameters, and a session identifier corresponding to the set of session parameters is generated. The session identifier is associated with the participant identifier, source language identifier, target language identifier, and role permission set in the session parameter set, and then structured to generate session metadata.

3. The cross-language real-time translation collaborative interaction method as described in claim 2, characterized in that: The specific steps for generating candidate translation fragments corresponding to the source language fragments are as follows: Access input streams from participants in both voice and text formats; The input stream is segmented based on temporal continuity and semantic integrity to form multiple independent talk turns; Semantic boundary recognition processing is performed on each turn of speech to generate a semantically relatively complete source language fragment; Based on the source language identifier and the target language identifier, cross-language translation processing is performed on each source language segment to generate an initial translation; The initial translation is formatted and consistent, and candidate translation fragments corresponding to the source language fragments are output.

4. The cross-language real-time translation collaborative interaction method as described in claim 3, characterized in that: The modification intentions of the participating parties include modification instructions for candidate translation fragments at the levels of vocabulary, grammar, word order and style, as well as the positional anchor information and semantic modification descriptions attached to the modification instructions.

5. The cross-language real-time translation collaborative interaction method as described in claim 4, characterized in that: The specific steps for converting the modification intent into a semantic patch pointing to the target translated fragment are as follows. Based on the location anchor information, identify the target translation fragment in the read-only translation state, and read the corresponding grammatical structure representation of the target translation fragment; The semantic modification description is parsed to determine the semantic type of the modification behavior and its position in the target translated text segment; By combining the grammatical structure representation of the target translation fragment, the semantic type and position of action are converted into a sequence of operations that can be executed at the grammatical structure level; The operation sequence is associated with the target translation fragment to form a semantic patch of the translation fragment pointing to the target translation fragment.

6. The cross-language real-time translation collaborative interaction method as described in claim 5, characterized in that: The specific steps for constructing the conflict domain are as follows: Identify all semantic patches that point to the same target translated fragment, and extract the syntax tree node path operated by each semantic patch; Compare the syntax tree node paths of semantic patches for different translations to determine whether there is overlap in the scope of operations; All semantic patches of translations with overlapping operational scopes are grouped together to form a conflict domain.

7. The cross-language real-time translation collaborative interaction method as described in claim 6, characterized in that: The adjudication process refers to determining the availability and prioritizing of translation semantic patches within the conflict domain based on the set of role permissions, and selecting the translation semantic patch as the adjudication semantic patch.

8. The cross-language real-time translation collaborative interaction method as described in claim 7, characterized in that: The preset priority order refers to sorting the adjudication semantic patches according to the predefined role level order in the role permission set, and adjudication semantic patches with the same role level are sorted according to the submission order of the translation semantic patches.

9. The cross-language real-time translation collaborative interaction method as described in claim 8, characterized in that: The change in alignment of the target translation fragment refers to the change in the semantic correspondence between the source language fragment and the target translation fragment due to the update of the candidate translation fragment.

10. The cross-language real-time translation collaborative interaction method as described in claim 9, characterized in that: The specific steps for rolling back the applied adjudication semantic patch based on the event log and generating new candidate translation fragments are as follows: When a change in the alignment of the target translation fragment is detected, query the event log to retrieve the application record of the adjudication semantic patch associated with the target translation fragment; In the order opposite to the preset priority order, the modifications made to the candidate translation fragments by the semantic patch of the ruling are revoked one by one; New candidate translation fragments are generated based on the withdrawn candidate translation fragments, and the correspondence between the source language fragments and the target translation fragments is updated according to the new candidate translation fragments.

Images