Ancient Chinese translation method, system and device based on large model and storage medium

Abstract

The application belongs to the technical field of natural language processing, and particularly relates to an ancient text translation method, system, device and storage medium based on a large model. The method comprises the following steps: performing context coding processing on the basis of an ancient text processing result, fusing character form information and textual criticism information, and combining diachronic semantic modeling to obtain context semantic representation corresponding to an ancient text paragraph; inputting the context semantic representation and an evidence set into a translation and annotation dual-track decoder to generate a plurality of modern Chinese translation tracks and modern Chinese annotation tracks; performing constraint screening on the modern Chinese translation tracks and the modern Chinese annotation tracks according to the evidence set to obtain candidate translation and annotation results meeting evidence coverage requirements; performing text atmosphere rearrangement according to the candidate translation and annotation results, and performing executable processing on traceable identifiers in the annotation tracks to output a translation result containing a modern translation and annotations. The translation obtained by the method has excellent language sense and clear and operable annotations, and meets both reading experience and academic tracking requirements.

Description

Technical Field

[0001] This invention belongs to the field of natural language processing technology, and in particular relates to a method, system, device and storage medium for translating ancient Chinese texts based on a large model. Background Technology

[0002] With the development of artificial intelligence and natural language processing technologies, translation techniques based on Large Language Models (LLMs) have emerged. By relying on neural network structures trained on large-scale corpora, these models possess strong contextual modeling and semantic reasoning capabilities, enabling relatively fluent and natural translations between modern languages. Compared to traditional rule-based or statistical translation methods, LLM translation demonstrates significant advantages in semantic understanding, contextual coherence, and the naturalness of expression.

[0003] In existing research on the translation of classical Chinese, the usual approach is to rely on manually compiled dictionaries, rule systems, and limited bilingual parallel corpora. This involves constructing ancient and modern dictionaries and rule templates to map classical Chinese vocabulary to modern Chinese definitions, and then using syntactic analysis to complete sentence-level combination and reconstruction. Alternatively, one approach attempts to introduce statistical translation or neural network translation frameworks, treating classical Chinese as a low-resource language and modeling and translating it using a small amount of aligned corpora.

[0004] However, the aforementioned methods suffer from the scarcity and lack of systematic annotation of classical Chinese corpora, making it difficult for statistical and neural models to fully learn the unique semantic features of classical Chinese. Classical Chinese language exhibits strong diachronic differences, with the same word potentially having different meanings in different historical periods. Existing translation methods mostly rely on single corpora or dictionary resources, lacking cross-document and cross-domain knowledge integration, which can easily lead to insufficient accuracy in translations. Traditional neural network translation methods focus more on formal syntactic transformations, while lacking the ability to model the tone, style, and cultural background inherent in classical Chinese, resulting in translations that lack overall stylistic consistency and cultural interpretability. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, device, and storage medium for translating ancient Chinese texts based on a large model that can meet the requirements of accuracy, contextual consistency, and knowledge coverage, in order to address the aforementioned technical problems.

[0006] Firstly, this application provides a method for translating classical Chinese texts based on a large model, including:

[0007] The process involves acquiring the classical Chinese text to be translated and performing higher-level semantic standardization on the text to obtain a classical Chinese text processing result with structured information. The classical Chinese text processing result includes character-level information, structure recognition results, and dynasty markers. Higher-level semantic standardization corresponds to sentence segmentation, structure recognition, and dynasty markers.

[0008] Based on the results of the ancient text processing, context encoding is performed to integrate glyphic and exegetical information, and combined with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph;

[0009] Based on the contextual semantic representation, a similarity search is performed in the ancient Chinese knowledge graph to obtain a set of evidence corresponding to the ancient text paragraphs;

[0010] Input the contextual semantic representation and evidence set into the translation and annotation dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks;

[0011] Based on the evidence set, the Modern Chinese translation track and the Modern Chinese annotation track are constrained and screened to obtain candidate translation and annotation results that meet the evidence coverage requirements;

[0012] Based on the candidate translation and annotation results, the text is rearranged in terms of style and the source identifiers in the annotation track are made executable, and the translation result containing the modern translation and annotations is output.

[0013] In one embodiment, context encoding is performed based on the ancient text processing results, fusing glyphic and philological information, and combined with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph, including:

[0014] The stroke sequence and component structure of characters in the ancient text paragraph are extracted based on character-level information, and the stroke sequence and component structure are encoded to obtain character shape vectors;

[0015] Based on character-level information, the corresponding historical commentaries and illustrative texts of the ancient text paragraphs are encoded to obtain semantic vectors;

[0016] By fusing the glyph vector and the semantic vector, the corresponding fused representation is obtained;

[0017] The dynasty time-stamped vector is obtained based on the dynasty marker, and the diachronic semantic modeling is performed on the fused representation based on the dynasty time-stamped vector to obtain the diachronic enhanced representation;

[0018] By jointly modeling the diachronic augmentation representation and the structure recognition results, a contextual semantic representation is obtained.

[0019] In one embodiment, a similarity search is performed in the Classical Chinese knowledge graph based on the contextual semantic representation to obtain a set of evidence corresponding to the Classical Chinese passage, including:

[0020] Based on the knowledge graph of classical Chinese, similarity retrieval is performed on the context semantic representation and the node vector representation to obtain candidate nodes;

[0021] Based on the candidate nodes, obtain the allusion nodes, annotation entries and quotation fragments corresponding to the ancient text paragraphs to form an initial set of evidence;

[0022] The evidence in the initial evidence set is weighted according to similarity score, consistency of dynasty timescale, genre matching degree and entity alignment relationship, and then sorted according to the weighting results to obtain the evidence set corresponding to the ancient text passage.

[0023] In one embodiment, the Classical Chinese knowledge graph is constructed using the following method:

[0024] We acquire ancient text corpora, annotation data, variant character mapping data, phonological data, and data on people, places, and official titles, and perform standardization processing to obtain an index database of knowledge related to ancient texts.

[0025] Nodes and relational edges were extracted from the ancient Chinese knowledge index database. Nodes include characters, words, meanings, structures, allusions, quotations, figures, place names, official titles, dynasties, and annotation entries. Relational edges include diachronic evolution relationships, cognate relationships, classical reference relationships, example relationship relationships, and compositional relationships.

[0026] Generate node vector representations for nodes, and form an ancient Chinese knowledge graph based on nodes, node vector representations of corresponding node attributes, and relation edges.

[0027] In one embodiment, the contextual semantic representation and evidence set are input into a dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks, including:

[0028] Based on contextual semantic representation, modern Chinese translation tracks are generated according to different translation styles;

[0029] Based on the contextual semantics, the sentences are segmented into semantic and syntactic parts to generate annotation tracks, and the source identification corresponding to the evidence set is marked in the annotation tracks to obtain the modern Chinese annotation tracks.

[0030] In one embodiment, the modern Chinese translation track and the modern Chinese annotation track are constrained and screened based on the evidence set to obtain candidate translation and annotation results that meet the evidence coverage requirements, including:

[0031] Generate a constraint lattice based on the evidence set; the constraint lattice contains constraints that must be covered, can be covered, and cannot be covered.

[0032] Based on the constraint grid, the modern Chinese translation track and the modern Chinese annotation track are decoded, and the consistency penalty term is combined to eliminate translations that violate the annotations, resulting in multiple candidate translations. The consistency penalty term includes the penalty weight when the translation and annotation contradict the time and character relationship between the constraint grid.

[0033] Candidate translation annotations are filtered based on evidence coverage to obtain candidate translation annotations that meet the evidence coverage requirements.

[0034] In one embodiment, the textual style is rearranged based on the candidate translation and annotation results, and the source identification in the annotation track is made executable, outputting a translation result containing the modern translation and annotations, including:

[0035] The candidate translation and annotation results are scored for literary style, and the literary style score includes the calculation of parallelism, repetition, rhythm, phonetics and rhetoric preservation.

[0036] The candidate translation and annotation results are rearranged based on the textual style scoring results to obtain the textual style optimized translation and annotation results;

[0037] The source identifier is mapped to the knowledge graph node, and the annotation template is rendered according to the preset user view. The translation result containing the modern translation and annotation is obtained by combining the translation and annotation results.

[0038] Secondly, this application also provides a large-scale model-based ancient Chinese translation system, including:

[0039] The standardization processing module is used to acquire the classical Chinese text paragraphs to be translated and perform higher-level semantic standardization processing on the classical Chinese text paragraphs to obtain classical Chinese text processing results with structured information; the classical Chinese text processing results include character-level information, structure recognition results and dynasty markers; higher-level semantic standardization processing corresponds to sentence segmentation, structure recognition and dynasty markers;

[0040] The semantic extraction module is used to perform context encoding processing based on the ancient text processing results, integrate character shape information and exegesis information, and combine it with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph;

[0041] The evidence module is used to perform similarity retrieval in the ancient Chinese knowledge graph based on the contextual semantic representation, and obtain a set of evidence corresponding to the ancient text paragraph.

[0042] The translation and annotation dual-track module is used to input the contextual semantic representation and evidence set into the translation and annotation dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks.

[0043] The evidence constraint module is used to constrain and filter the modern Chinese translation track and the modern Chinese annotation track based on the evidence set, so as to obtain candidate translation and annotation results that meet the evidence coverage requirements.

[0044] The text rearrangement module is used to rearrange the text based on the candidate translation and annotation results, and to perform executable processing on the source identification in the annotation track, outputting a translation result that includes the modern translation and annotations.

[0045] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above-described ancient text translation methods based on large models.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for translating ancient texts based on a large model.

[0047] The aforementioned methods, systems, devices, and storage media for translating classical Chinese based on large-scale models, through contextual semantic representation and knowledge graph-driven evidence, can reduce ambiguity and mistranslation, ensuring the accurate transmission of classical Chinese meaning in modern Chinese. The close integration of annotation tracks and source identification allows each translation or explanation to be traced back to its original source or related annotations, achieving verifiability for academic-level translation. The dual-track decoder can generate multiple translation tracks based on different styles, while the annotation track provides detailed explanations, satisfying both the needs of general readers for easily readable translations and the needs of academic researchers for detailed interpretation of the original text. Through textual style scoring and rearrangement mechanisms, while ensuring semantic accuracy, the parallelism, rhetoric, rhythm, phonetics, and other rhetorical features of classical Chinese are preserved to the greatest extent, resulting in more natural and elegant translations. From text parsing, semantic modeling, evidence retrieval to annotation generation and optimization, the workload of manual translation and annotation is significantly reduced, improving the efficiency of classical Chinese translation. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating the ancient Chinese translation method based on a large model according to the present invention.

[0050] Figure 2 This is a flowchart illustrating the steps of step S102.

[0051] Figure 3 This is a flowchart illustrating the steps of step S105.

[0052] Figure 4 This is a structural diagram of the ancient Chinese translation system based on a large model according to the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, such as Figure 1 As shown, a method for translating classical Chinese texts based on a large model is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0055] S101. Obtain the ancient Chinese text paragraph to be translated, and perform higher-level semantic standardization processing on the ancient Chinese text paragraph to obtain the ancient Chinese text processing result with structured information; the ancient Chinese text processing result includes character-level information, structure recognition result and dynasty mark; higher-level semantic standardization processing corresponds to sentence segmentation, structure recognition and dynasty mark.

[0056] This illustration demonstrates how sentence segmentation algorithms, construction recognition mechanisms, and dynasty marking techniques are used to preprocess original classical Chinese texts through structural processing. Specifically, sentence segmentation is achieved using a joint model based on statistical learning and semantic constraints. This model integrates punctuation marks, common word collocation patterns, and semantic plausibility to automatically determine sentence boundaries in classical Chinese. Construction recognition utilizes a syntactic template library specific to classical Chinese to detect and annotate typical sentence structures in the text, such as subject-object inversion, omitted predicates, and appositive structures. Dynasty marking relies on diachronic dictionaries and corpus statistics to automatically identify the historical context of the text, providing spatiotemporal constraints for subsequent semantic interpretation. After preprocessing, the classical Chinese text is transformed into a processed result with structured information, including character-level information such as part of speech, radical, and glyph features of individual characters; construction recognition results such as syntactic skeletons and semantic relation markers; and dynasty marking, which represents the historical period label corresponding to the text.

[0057] S102. Based on the results of the ancient text processing, context encoding is performed to integrate character shape information and exegesis information, and combined with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph.

[0058] Contextual encoding is performed based on the results of classical Chinese text processing to obtain a deep semantic representation suitable for inputting large models. Specifically, glyphic information and philological information are fused, and diachronic semantic modeling is combined to improve the accuracy of classical Chinese text understanding. For example, the glyphic-philological encoder is based on consistency loss. , For context-aware word representation, For philological information, For glyph information, is the weight coefficient. Among them, the glyph information refers to the visual and structural features of ancient characters in the writing system, such as the stroke configurations of oracle bone inscriptions, seal scripts, or official scripts and their changing rules during the evolution process. The above features are usually modeled through convolutional coding or graph convolutional networks; while the exegetical information refers to the specific meanings and usages of words or phrases in ancient classics throughout the dynasties. It often has polysemy and differences in different eras, so it needs to be represented by embedded dictionary vectors. Diachronic semantic modeling is to capture the semantic evolution trajectory of the same word in different dynasties on the time axis. Exemplarily, a time-aware embedding model is usually adopted to model the semantic distributions of the same word in different periods, and multi-task training is combined to enhance the perception of ancient Chinese grammar phenomena.

[0059] Through the multi-channel fusion mechanism, the above information is jointly encoded into a context semantic representation, thus not only retaining the glyph information of ancient characters but also integrating the knowledge of exegetical interpretations and semantic evolution.

[0060] S103. Perform similarity retrieval in the ancient Chinese knowledge graph according to the context semantic representation to obtain an evidence set corresponding to the ancient text paragraph.

[0061] The context semantic representation will be compared with the external ancient Chinese knowledge graph to obtain a reliable evidence set. The ancient Chinese knowledge graph is a semantic network, whose nodes usually include ancient characters, classics, figures, place names, historical events, exegetical interpretations, etc., and the edges record the semantic or diachronic associations between them. Exemplarily, the corresponding semantic items of the word "benevolence" in The Analects of Confucius and Mencius may be different, and the ancient Chinese knowledge graph can distinguish them by the time-meaning relationship. Specifically, the similarity between the context representation and the nodes of the knowledge graph is calculated through the method of semantic vector retrieval, and vector matching based on cosine similarity or interactive matching based on a deep retrieval network can be used. The finally obtained evidence set usually contains lexical explanations, synonymous substitutions, context examples, and diachronic corresponding relationships related to the ancient text paragraph, thus providing external knowledge constraints for subsequent translation.

[0062] S104. Input the context semantic representation and the evidence set into the translation and annotation dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks.

[0063] The contextual semantic representation and the evidence set are jointly input into the translation-annotation dual-track decoder to simultaneously generate both the translation track and the annotation track for modern Chinese. The translation-annotation dual-track decoder is an improved large-scale model decoding architecture containing two interconnected generation paths: a translation track for generating natural and fluent modern Chinese translations, and an annotation track for generating explanatory annotations. During training, the translation track relies on a large-scale parallel corpus of ancient and modern texts for alignment training, while the annotation track learns from dictionary, classical text annotations, and expert-annotated data to generate explanations, source hints, and diachronic information relevant to the translation. During decoding, the contextual semantic representation primarily provides semantic support for the text to be translated, while the evidence set serves as a retrieval-based external knowledge input to guide the model in maintaining accuracy and consistency when selecting translation terms and generating annotations.

[0064] S105. Based on the evidence set, constrain and filter the modern Chinese translation track and the modern Chinese annotation track to obtain candidate translation and annotation results that meet the evidence coverage requirements.

[0065] Furthermore, the generated translation and annotation tracks are subjected to constraints and screening to ensure the reliability and coverage of the results. Specifically, candidate translation and annotation results are compared with the evidence set. If a translation or annotation fails to cover the core information required by the evidence, it will be eliminated or reweighted. At the same time, different candidate translation and annotation results are assigned weighted scores based on the confidence level of the evidence matching, thereby prioritizing the output of the results with the highest consistency with the knowledge graph. This effectively avoids the problem of large models translating randomly and ensures the accuracy of the translation in terms of semantic interpretation and historical verification.

[0066] S106. Based on the candidate translation and annotation results, rearrange the text and make the source identification in the annotation track executable, and output the translation result containing the modern translation and annotation.

[0067] This example illustrates the executable processing of textual rearrangement and source tracing for candidate translations. Textual rearrangement refers to optimizing the style among multiple candidate translations to better align with modern Chinese expression habits while maintaining the original tone and stylistic features of classical Chinese. For instance, when translating a passage from the *Records of the Grand Historian*, it's necessary not only to ensure accurate word meaning but also to maintain the conciseness and solemnity of historical narratives. The executable processing of source tracing involves transforming the source information generated in the annotation track into clickable or callable links, allowing users to directly trace the classical texts or knowledge graph nodes upon which the translation is based, enhancing the interpretability and academic reliability of the translation results.

[0068] The aforementioned large-scale model-based method for translating classical Chinese texts standardizes classical Chinese passages through higher-level semantic standardization, including sentence segmentation, construction recognition, and dynasty marking. This generates structured classical Chinese processing results with character-level information, construction recognition results, and dynasty markings, enabling accurate understanding of the hierarchical structure and linguistic features of the classical texts, thus providing a reliable foundation for subsequent encoding and translation. By fusing glyphic and philological information and combining it with diachronic semantic modeling, a diachronically enhanced contextual semantic representation can be obtained, effectively capturing the polysemy, diachronic evolution, and subtle semantic differences in classical Chinese, thereby improving the accuracy and depth of translation. Similarity retrieval using a classical Chinese knowledge graph matches the contextual semantic representation with knowledge graph nodes, forming an evidence set. This ensures that translation generation not only relies on statistical or deep learning inference but also incorporates documentary and historical evidence, guaranteeing that the translated content conforms to academic and historical logic. A dual-track decoder generates modern Chinese translation and annotation tracks, and a constraint lattice is constructed using the evidence set for filtering, achieving high consistency and traceability between the translation and annotations, avoiding arbitrary interpretations or deviations from evidence. Based on the candidate translation and annotation results, the text style is scored and rearranged, and the source identification in the annotation track is matched with the knowledge graph node and rendered in an executable manner. This results in a translation that is both elegant in tone and clear and operable in annotation, so that the final output satisfies both the reading experience and the academic tracking needs.

[0069] In one embodiment, such as Figure 2 As shown, based on the results of the ancient text processing, context encoding is performed to integrate glyphic and philological information, and combined with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph, including:

[0070] S201. Extract the stroke sequence and component structure of the characters in the ancient text paragraph based on the character-level information, and encode the stroke sequence and component structure to obtain the character shape vector.

[0071] In classical Chinese, the same character often has multiple forms, and the evolution of character forms across different dynasties is quite complex. Relying solely on dictionary-based word mapping often fails to capture the underlying patterns. This paper illustrately describes character form information through two types of structural features: stroke sequence and component structure. The stroke sequence can be extracted from existing character databases, extracting the stroke order of each Chinese character, converting it into a symbol sequence, and inputting it into a sequence encoding model. For example, a Bidirectional Long Short-Term Memory (BiLSTM) network or a convolutional neural network can be used to obtain the stroke-level encoding results. The component structure is obtained by decomposing the character's constituent units, such as radicals, phonetic-semantic combinations, etc., and treating the components as structural diagrams. A graph neural network is used for modeling to obtain a representation reflecting the structural rules of Chinese characters. Concatenating or fusing the stroke encoding and component encoding yields the character form vector.

[0072] Exemplarily, when processing the character "義", extract its stroke sequence, i.e., the order of the left-falling stroke, horizontal stroke, vertical stroke, hook, etc., and identify its component structure as a combination of "羊" and "我". After encoding, a glyph vector that contains both physical features and structural features is obtained.

[0073] S202. Encoding the ancient annotations and example texts corresponding to the words and phrases in the ancient text paragraph according to the word-level information to obtain an item vector.

[0074] The meanings of words in ancient Chinese may vary significantly in different dynasties and contexts. Therefore, relying solely on modern interpretations is difficult to support high-quality translation. Schematically, based on the word-level information, retrieve the pre-constructed database of ancient annotations and example texts. Exemplarily, if the input is the character "仁", extract its interpretations and examples from "Annotations on The Analects of Confucius", "Collected Annotations on Mencius", and the usage examples in ancient classics. Transform the corpus into a vector representation through a Transformer encoder or the representation layer based on a pre-trained large language model, that is, an item vector, which not only contains the item information of the word in ancient texts but also incorporates the context usage rules in historical corpora. Exemplarily, "仁" often corresponds to the core of virtue in the pre-Qin context, while in later Buddhist classics, it may also carry the meaning of compassion. The subtle semantic differences can be characterized by the item vector.

[0075] S203. Fuse the glyph vector and the item vector to obtain the corresponding fused representation.

[0076] Fuse the glyph vector and the item vector to form a more complete symbolic semantic representation. The fusion method can be diverse, such as vector concatenation, gating mechanism weighting, or interaction modeling based on the attention mechanism. The model can not only retain the external morphological features of the text but also combine the philological interpretations in the semantic dimension to achieve cross-modal representation enhancement. Exemplarily, the glyph vector of the character "義" reflects its structure composed of "羊" and "我", while the item vector records its semantics of "justice, fairness". After the two are fused, both the form and semantics can be taken into account.

[0077] S204. Obtain the dynasty time-stamp vector according to the dynasty label, and perform diachronic semantic modeling on the fused representation according to the dynasty time-stamp vector to obtain the diachronically enhanced representation.

[0078] Furthermore, consider the diachronic evolution problem of semantics. The semantics of ancient texts is highly dependent on historical contexts, and the same vocabulary often has different usages in different dynasties. Schematically, a dynasty time stamp vector can be generated according to the dynasty label in the ancient text processing result, and embedding technology can be used to convert the dynasty information into a vector space representation. Combine the dynasty time stamp vector with the fusion representation, and enhance the temporal sensitivity of semantic representation through diachronic semantic modeling. Diachronic semantic modeling can adopt time-aware language modeling methods, such as introducing a bias in the time dimension in the semantic space, or constructing a time series Transformer, and segmenting the corpora of different periods for training, so that the model can capture the evolution trajectory of semantics. Exemplarily, "doctor" referred to a school official in charge of classics in the Han Dynasty, while it is an academic degree in modern times. After diachronic semantic modeling, the model will be more inclined to generate semantic interpretations consistent with the Han Dynasty context. Schematically, set a learning vector for each dynasty d, and introduce temporal continuity regularization . For the semantic drift of the same word in adjacent dynasties, with known annotations as supervision, perform contrastive learning of approaching the correct semantic direction and moving away from the wrong semantic items, that is, perform contrastive learning through the diachronic contrast loss . Among them, is a specific character or word unit, is the positive sample representation, referring to the semantic representation of the character or word unit in the adjacent or similar historical period d; is the negative sample representation, referring to different other semantic units, or having no semantic correspondence with in different periods.

[0079] S205. Perform joint modeling based on the diachronic enhanced representation and the construction recognition result to obtain the context semantic representation.

[0080] The construction recognition result refers to identifying specific grammatical constructions, such as fixed constructions like "take... as...", "if... then...", etc. by parsing the sentence structure of ancient texts. Constructions often have specific functions and semantics in ancient texts, so they are indispensable in context modeling. By combining the diachronic enhanced representation with the construction recognition result, a multi-channel attention mechanism can be adopted, enabling the large model to not only understand the diachronic semantics of single characters or words but also grasp the meaning of the overall grammatical construction. Exemplarily, in the sentence "subdue people by virtue", the construction "subdue people by..." will prompt the model that this is a causal expression, and together with the diachronic enhanced representation, the final context semantic representation can more accurately reflect the deep meaning of this ancient text paragraph.

[0081] In the above method, the glyph information, exegetical information, and diachronic information are deeply integrated, and combined with the results of construction recognition, finally generating a context semantic representation with rich semantic levels. This representation not only has multiple dimensions of glyphs and semantic items at the single-character level, but also has context constraints of historical evolution and construction rules at the sentence level, thus providing a solid semantic basis for subsequent translation and annotation generation.

[0082] In one embodiment, similarity retrieval is performed in the ancient Chinese knowledge graph according to the context semantic representation to obtain an evidence set corresponding to the ancient Chinese paragraph, including:

[0083] S11. Based on the ancient Chinese knowledge graph, similarity retrieval is performed on the context semantic representation and the node vector representation to obtain candidate nodes.

[0084] Schematically, the context semantic representation is mapped into the same semantic space as the nodes of the knowledge graph. Among them, the nodes of the knowledge graph usually obtain their vectorized expressions through pre-trained graph representation learning methods, such as graph neural networks, TransE, RotatE and other embedding models, so that each allusion, entry or citation fragment can be represented in vector form. The similarity measurement method can use cosine similarity, dot product or distance-based measurement functions. When the similarity between the context semantic representation and some node vectors exceeds a preset threshold, or ranks among the top several in the ranking, the corresponding nodes are selected as candidate nodes. Exemplarily, when the input ancient Chinese is "鸿鹄高飞 (hóng hú gāo fēi)", it may retrieve in the knowledge graph the allusion nodes related to "鸿鹄 (hóng hú)" and the poem citation nodes related to "高飞 (gāo fēi)", thus becoming part of the candidate set.

[0085] S12. According to the candidate nodes, obtain the allusion nodes, annotation entries and citation fragments corresponding to the ancient Chinese paragraph to form an initial evidence set.

[0086] It is necessary to map the knowledge content associated with the candidate nodes into directly usable evidence. Since the nodes in the knowledge graph are often connected to various types of text resources, such as allusion entries, annotations of past dynasties, poem fragments or historical document citations, according to the link relationship and adjacent edges of the candidate nodes, obtain the directly associated materials, and the formed initial evidence set can be either an explanatory entry for single-character exegesis or a cross-sentence citation from ancient classics.

[0087] S13. Weigh each evidence in the initial evidence set according to the similarity score, dynasty time scale consistency, genre matching degree and entity alignment relationship, and sort according to the weighted results to obtain an evidence set corresponding to the ancient Chinese paragraph.

[0088] Weighted screening is carried out through multi-dimensional constraints to ensure the accuracy and relevance of the final evidence set. Specifically, the similarity score is the most basic indicator, which is used to measure the semantic proximity between the context semantic representation and the candidate nodes. The dynasty time stamp consistency constraint is used to ensure that the semantic interpretation conforms to the time background of the target classical Chinese text. For example, the glosses of Han Dynasty texts should not be directly replaced by the annotations of the Song Dynasty. Therefore, it is necessary to match the dynasty time stamp of the evidence with the dynasty label of the input text. The genre matching degree refers to whether the genre attribute of the candidate evidence is consistent with the genre of the text to be translated. For example, for proses, more reference should be made to the annotations of historical proses, while for poems, it is more inclined to cite poem corpora. The entity alignment relationship is used to judge whether the core entities in the evidence correspond to the original text entities. For example, "swan" must be aligned to entities related to birds and cannot be wrongly mapped to irrelevant people or place names.

[0089] In the weighting process, the weights of each index can be automatically learned through large model training or set according to empirical rules. For example, in poem texts, the weight of genre matching degree is increased, and in gloss text, the weight of dynasty consistency is increased. The weighted scores will comprehensively evaluate the initial evidence set, and the optimal evidence set will be selected through a sorting mechanism. Exemplarily, for the sentence "The swan soars high", the final evidence set may preferentially retain the relevant poem interpretations in "Chu Ci · Jiu Ge" and the allusion entries in "Zhuang Zi · Xiao Yao You", and filter out materials that do not conform to other genres or late literature.

[0090] Through the above method, through the multi-dimensional screening mechanism, the finally obtained evidence set can not only ensure a high semantic correlation with the original text, but also has strong consistency at the time, genre and entity levels, thus providing reliable knowledge support for the generation of translation and annotation.

[0091] In one of the embodiments, the ancient Chinese knowledge graph is constructed by the following method:

[0092] S21. Obtain ancient book corpora, annotation data, variant character mapping data, phonetic data, and character, place name and official position data, and perform standardization processing to obtain a knowledge index library related to classical Chinese.

[0093] Schematically, five types of information sources are utilized, namely ancient book corpora, annotation data, variant character mapping data, phonetic data, and person, place, and official position data. Ancient book corpora refer to text materials from historical classics, historical records, literary collections, and exegetical dictionaries. They often contain mixed use of traditional Chinese characters and variant characters, as well as differences in language habits of different dynasties. Therefore, text cleaning and unified coding are required. Annotation data includes the explanations and commentaries of classical ancient texts by scholars of past dynasties, usually sourced from electronic ancient book collation projects or large-scale digital humanities databases. Variant character mapping data mainly addresses the problem of glyph divergence caused by different writings of the same morpheme, such as the correspondence between "爲" and "为". The mapping is completed by constructing a variant character comparison table and combining it with the "List of Chinese Character Standards". Phonetic data covers the rhyme books and research results of historical phonology of past dynasties. It not only records the changing trajectory of phonetic sounds but also reveals the context of semantic evolution. Person, place, and official position data mainly come from historical geography databases and biographical materials of historical figures. After entity recognition and disambiguation processing, a standardized historical entity library is formed. In the standardization process, all data needs to be unified into the same character set, such as Unicode, and a unified time marking system, such as indexing by dynasty or specific years, and then word segmentation, deduplication, and splitting of multiple semantic items are carried out to form a clean and structured knowledge index library.

[0094] S22. Nodes and relationship edges are extracted according to the ancient text-related knowledge index library; the nodes include characters, words, semantic items, constructions, allusions, citations, persons, places, official positions, dynasties, and annotation entries; the relationship edges include diachronic evolution relationships, homologous variant relationships, allusion relationships, usage example relationships, and rhetorical relationship.

[0095] Furthermore, nodes and relationship edges are extracted based on the knowledge index library. The design of the nodes covers key elements in the semantic system of ancient texts, including characters, words, semantic items, constructions, allusions, citations, persons, places, official positions, dynasties, and annotation entries. It not only ensures a refined representation at the language level but also connects language with culture and historical context. The setting of the relationship edges reflects the multi-dimensional connections of ancient text knowledge. Among them, the diachronic evolution relationship is used to describe the change of the meaning of characters or words over time; the homologous variant relationship is used to represent the consistency of different glyphs or word forms in terms of etymology; the allusion relationship is used to mark the connection between an allusion and its source text; the usage example relationship is used to connect specific citations with the semantic items they explain; the rhetorical relationship reveals the logical connection within the text in terms of rhetoric or text structure.

[0096] S23. A node vector representation is generated for the nodes, and an ancient Chinese knowledge graph is formed based on the nodes, the node vector representations corresponding to the node attributes, and the relationship edges.

[0097] Indicatively, node vector representations are typically generated using graph neural networks or context-based language modeling methods. These methods incorporate the node's textual description, attribute information, and its neighbor structure within the relational network into the model. The resulting node vectors not only reflect the semantic attributes of a single entity but also capture its positional characteristics within the knowledge network. For example, the vector for the node "Confucius" is influenced not only by its biographical description but also by its relationships with nodes such as "Spring and Autumn Period," "State of Lu," and "The Analects," thus possessing a richer semantic representation. After obtaining the node vectors, all nodes and relationships are uniformly encoded and organized into a graph database, forming a knowledge graph of classical Chinese. This supports efficient retrieval and reasoning and provides evidence-based support for subsequent semantic matching and translation inference.

[0098] In one embodiment, the contextual semantic representation and evidence set are input into a dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks, including:

[0099] S31. Based on contextual semantic representation, generate modern Chinese translation tracks according to different translation styles.

[0100] Based on contextual semantic representation, translation tracks with different translation styles are generated. Illustratively, translation styles typically fall into three categories: "faithfulness," "expressiveness," and "elegance." "Faithfulness" emphasizes accurately conveying the original meaning; "expressiveness" emphasizes fluency and fluency; and "elegance" pursues beautiful language and elevated style. To achieve this diverse translation track generation, the training of the translation-annotation dual-track decoder often incorporates style tags. By labeling the corresponding style type in the training corpus, the decoder can automatically favor the appropriate expression style during translation based on the combination of contextual semantic representation and style tags.

[0101] S32. Based on the contextual semantics, the sentences are segmented into semantic and syntactic parts to generate annotation tracks, and the source identification corresponding to the evidence set is marked in the annotation tracks to obtain the modern Chinese annotation tracks.

[0102] The generation of annotation tracks relies on a combination of contextual semantic representation and evidence sets. Specifically, the decoder segments the classical Chinese passages into sentences and further divides them into semantic and syntactic units. The segmentation process can be rule-based, such as utilizing common punctuation marks or function word structures in classical Chinese, or based on statistical models or deep learning methods, such as using syntactic dependency analysis or sequence labeling models to identify sentence boundaries and word functions. After obtaining the segmented units, the decoder generates modern Chinese annotation tracks for each unit, explaining its semantic and grammatical functions.

[0103] Furthermore, the annotation track incorporates source identifiers provided by the evidence set. The evidence set may contain allusions, historical commentaries, or quotations corresponding to the word or phrase. Embedding source information as identifiers within the annotation track ensures that each interpretation has a clear origin.

[0104] As an illustration, the training process of the translation-annotation dual-track decoder, in addition to the regular ancient-to-modern translation corpus, also incorporates manually annotated historical commentaries and knowledge graph evidence as training signals. This allows the model to not only learn to generate translations but also simultaneously generate annotation tracks and evidence labels. Specifically, the translation-annotation dual-track output can be presented as two columns: the left side displays the parallel modern Chinese translation, and the right side displays sentence-by-sentence and word-by-word annotations and source tracing. As an illustration, the joint loss function for training the translation-annotation dual-track decoder is... ,in, The joint loss function is the overall optimization objective, which comprehensively considers the training requirements of the translation track, the annotation track, and their alignment. For the cross-entropy loss of modern translation tracks, where, This factor represents the distributional differences between the model-generated translation track and the reference translation, ensuring that the model can output semantically accurate and fluent modern Chinese translations. To calculate the cross-entropy loss for the annotation track, multiply by a weighting factor. , The distribution difference between the annotation tracks generated for the model and the reference annotations. It is important to control the training of annotation tracks and avoid over-biasing towards translation while neglecting annotations. The alignment loss is multiplied by the weighting coefficient. This ensures consistency between the translation track and the annotation track in terms of semantics and evidence set, for example, by requiring that the terms and philological interpretations in the annotation track remain relevant to the translation track. Control the weight of alignment constraints in the overall optimization.

[0105] In one embodiment, such as Figure 3 As shown, based on the evidence set, constraints are applied to the Modern Chinese translation track and the Modern Chinese annotation track to obtain candidate translation and annotation results that meet the evidence coverage requirements, including:

[0106] S301. Generate a constraint lattice based on the evidence set; the constraint lattice contains constraints that must be covered, can be covered, and cannot be covered.

[0107] A constraint lattice is a multi-dimensional constraint structure in which each piece of evidence is mapped to a type of coverage condition. The core objective of a constraint lattice is to provide a bounded boundary for the translation and annotation process, ensuring that the translation and annotation are not only the result of semantic inference but also an evidence-driven deduction. For example, when the piece of evidence contains the entry "quotation from the *Records of the Grand Historian*, specifically the biography of Xiang Yu," a "must cover" constraint is generated, meaning that the allusion or corresponding semantic scenario must be explicitly or implicitly mentioned in the candidate translation and annotation results. Conversely, if the evidence contains explicit negative content, such as a commentary proving that a certain word's meaning is irrelevant to the passage, a "prohibit coverage" constraint is generated to avoid erroneous source attribution in the translation and annotation track. There exists a type of evidence that does not affect overall correctness but helps enrich the interpretation, such as examples from similar eras or analogies in parallel texts. This is labeled "optional coverage," and its presence or absence in the candidate results does not directly affect the validity of the results but does affect the priority ranking.

[0108] S302. Based on the constraint grid, decode the modern Chinese translation track and the modern Chinese annotation track, and combine the consistency penalty term to eliminate translations that violate the rules, and obtain multiple candidate translation results; the consistency penalty term includes the penalty weight when the translation and the constraint grid contradict each other in terms of time and character relationship.

[0109] Furthermore, when generating each word or phrase, the decoder simultaneously checks whether the output meets the coverage requirements of the constraint lattice. If a translation selection omits evidence that "must be covered," its candidate translation track will be marked with a negative score; if the output contains elements that "must not be covered," a rejection mechanism is directly triggered. Optionally, a consistency penalty term is introduced to resolve potential logical conflicts between translation annotations and evidence. For example, if a translation track attributes an event to "Emperor Wu of Han," while the timestamps and relationships between figures in the evidence set indicate that the event should be attributed to "Qin Shi Huang," it is judged as a "contradiction between time and figure relationship." In this case, the translation annotation result will be given a higher penalty weight, causing the result to drop significantly in the candidate set. The magnitude of the penalty weight is usually determined according to the severity of the contradiction. If the contradiction involves key constraints such as time and dynasty, the penalty value is higher; if it is only a minor inconsistency in genre or tone, the penalty value is lower. In this way, the decoder not only generates translation annotations in terms of linguistic form, but also ensures that they are consistent in terms of historical facts, knowledge logic, and documentary evidence. For example, ,in, Indicates the set of candidate translation annotation results In the process, select the annotation result that maximizes the objective function value. . This is the candidate set generated based on the translation-annotation dual-track decoder and after preliminary constraint screening. Given an input classical Chinese passage At that time, the translation and annotation results The conditional probability logarithm of the translation ensures the semantic matching between the translation and annotation results and the input classical Chinese text, and can be understood as the confidence score of the language model itself. For coverage constraints, To adjust the weights. Evidence set refers to the exegesis, examples, or historical sources corresponding to the classical Chinese passages in the knowledge graph. For translation annotation results The extent to which the evidence set is covered is usually measured by how many evidence points are involved in the translator's notes, and the completeness or relevance of the evidence used. This ensures that the final translator's notes are not only semantically reasonable, but also have sufficient "sources" or "support". This is a weighting factor used to balance the importance of language model confidence and evidence coverage.

[0110] S303. Based on the evidence coverage rate, the candidate translation and annotation results are screened to obtain candidate translation and annotation results that meet the evidence coverage requirements.

[0111] Furthermore, the candidate annotations, after being constrained and penalized, undergo coverage screening. Evidence coverage is an indicator that measures the extent to which a candidate annotation matches the evidence set, and it can be calculated by the ratio of the number of evidence items involved in the candidate annotation to the total number of evidence items. For example, if a candidate annotation covers 6 out of 8 pieces of evidence, the coverage rate is 75%. Optionally, a coverage threshold can be set, and only candidate annotations that reach or exceed this threshold will be retained. For candidate annotations with the same coverage rate, additional indicators such as language fluency, stylistic diversity, and annotation clarity will be used for further ranking.

[0112] The above method, through evidence-driven constraint lattice generation, multi-dimensional penalty mechanism, and layer-by-layer control of coverage screening, ensures that the final output candidate translation and annotation results are not only linguistically natural and fluent, but also reliable in terms of knowledge logic and evidence consistency. This effectively solves the problems of arbitrariness, de-evidence, and contradiction that are prone to occur in traditional ancient Chinese translation and annotation.

[0113] In one embodiment, the textual style is rearranged based on the candidate translation and annotation results, and the source identification in the annotation track is made executable, outputting a translation result containing the modern translation and annotations, including:

[0114] S41. Score the literary style of the candidate translation and annotation results to obtain the literary style score; the literary style score includes the calculation of parallelism, repetition, rhythm, phonetics and rhetoric preservation.

[0115] The literary style scoring is a comprehensive language evaluation indicator used to measure the quality of a translation in terms of form and artistry. Specifically, the literary style scoring includes multiple dimensions of evaluation to obtain semantic fidelity. Preserving the true essence of literature and readability semantic fidelity Used to measure translation The degree of semantic consistency between the original classical Chinese input and the target text. This can be calculated using the similarity of the context-encoded vectors. (Textual style preservation) Used to measure translation In modern Chinese, naturalness and expressive quality are typically assessed using a large-scale model's perplexity or fluency score. Readability Used to measure translation The evaluation assesses the grammatical correctness of the translation. Specifically, the retention of parallelism and antithesis reflects the consistency of sentence structure and inter-sentence coherence; the retention of rhythm and phonology measures the harmony of words, syllables, and rhyme, which is particularly important for translating classical Chinese poetry or rhymed texts; and the retention of rhetoric focuses on whether rhetorical devices such as metaphor, personification, contrast, and anaphora are preserved in the translation. To achieve this evaluation, rule-based methods can be used, such as statistically analyzing sentence length, detecting sentence repetition, and analyzing syllable patterns, while also incorporating a large-scale model to identify rhetorical features, thereby obtaining a quantitative score for each candidate translation in each dimension. The overall style and tone score forms a vectorized index that reflects both the overall style and tone level and can distinguish subtle differences between candidate translations.

[0116] S42. Based on the textual tone scoring results, the candidate translation and annotation results are rearranged to obtain the textual tone optimized translation and annotation results.

[0117] The goal of the rearrangement process is to prioritize outputting translator's notes with higher style scores. ,in, The goal is to generate candidate translation annotations that ensure the final translation is not only semantically accurate but also more natural and elegant in terms of language and rhetoric. Specific implementation methods may include adjusting word order, optimizing inter-sentence connectors, and modifying rhetorical structure and rhythm. Through this process, multiple candidate annotations can be ranked, and the annotation with the highest overall score can be selected as the final output.

[0118] S43. Match the source identifier with the knowledge graph node, render the annotation template according to the preset user view, and combine the translation and annotation results to obtain the translation result containing the modern translation and annotation.

[0119] The source identifiers in the annotation tracks are made executable to ensure a clear and operable correspondence between annotation information and knowledge graph nodes. Specifically, the source identifiers of each annotation track are mapped to nodes in the knowledge graph and presented in conjunction with the user's preset views. User views can include an academic research view displaying detailed sources and commentaries from various dynasties, a teaching view focusing on easy-to-read annotations and examples, or a general reader view emphasizing word meaning explanations and translation comprehension. Different views adjust the hierarchy, level of detail, and presentation of annotations. Based on the preset views, the annotation content is presented to the final translation page or document using template rendering technology, such as displaying floating annotations, footnotes, or side notes next to the text, while retaining interactive functionality with knowledge graph nodes, allowing users to click on the source identifiers to access the original text or related entries. Through this process, the final output translation result includes both the modern Chinese translation track and the visualized and interactive annotation track, achieving a high-quality integration of translation and annotation.

[0120] In summary, through the optimization of textual style scoring and rearrangement, as well as the executable processing of source identification, this technical solution achieves high-quality integration of the translation track and the annotation track, enabling the translation results to achieve a balance in accuracy, readability, and knowledge traceability, thereby significantly improving the overall practical value of the classical Chinese translation system.

[0121] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0122] Based on the same inventive concept, this application also provides a large-model-based ancient Chinese translation system for implementing the aforementioned large-model-based ancient Chinese translation method. The solution provided by this system is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more large-model-based ancient Chinese translation system embodiments provided below can be found in the limitations of the large-model-based ancient Chinese translation method described above, and will not be repeated here.

[0123] In one exemplary embodiment, such as Figure 4 As shown, a large-model-based ancient Chinese translation system is provided, including:

[0124] The standardization processing module 401 is used to acquire the ancient Chinese text paragraph to be translated and perform higher-level semantic standardization processing on the ancient Chinese text paragraph to obtain the ancient Chinese text processing result with structured information; the ancient Chinese text processing result includes character-level information, structure recognition result and dynasty mark; the higher-level semantic standardization processing corresponds to sentence segmentation, structure recognition and dynasty mark;

[0125] The semantic extraction module 402 is used to perform context encoding processing based on the ancient text processing results, integrate character shape information and exegesis information, and combine it with diachronic semantic modeling to obtain the context semantic representation of the corresponding ancient text paragraph;

[0126] Evidence module 403 is used to perform similarity retrieval in the ancient Chinese knowledge graph based on the contextual semantic representation to obtain a set of evidence corresponding to the ancient text paragraph;

[0127] The dual-track translation and annotation module 404 is used to input the contextual semantic representation and evidence set into the dual-track translation and annotation decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks.

[0128] The evidence constraint module 405 is used to constrain and filter the modern Chinese translation track and the modern Chinese annotation track according to the evidence set, so as to obtain candidate translation and annotation results that meet the evidence coverage requirements.

[0129] The text rearrangement module 406 is used to rearrange the text based on the candidate translation and annotation results, and to perform executable processing on the source identification in the annotation track, outputting the translation result containing the modern translation and annotation.

[0130] In one embodiment, the semantic extraction module 402 is further configured to:

[0131] The stroke sequence and component structure of characters in the ancient text paragraph are extracted based on character-level information, and the stroke sequence and component structure are encoded to obtain character shape vectors;

[0132] Based on character-level information, the corresponding historical commentaries and illustrative texts of the ancient text paragraphs are encoded to obtain semantic vectors;

[0133] By fusing the glyph vector and the semantic vector, the corresponding fused representation is obtained;

[0134] The dynasty time-stamped vector is obtained based on the dynasty marker, and the diachronic semantic modeling is performed on the fused representation based on the dynasty time-stamped vector to obtain the diachronic enhanced representation;

[0135] By jointly modeling the diachronic augmentation representation and the structure recognition results, a contextual semantic representation is obtained.

[0136] In one embodiment, the evidence module 403 is further configured to:

[0137] Based on the knowledge graph of classical Chinese, similarity retrieval is performed on the context semantic representation and the node vector representation to obtain candidate nodes;

[0138] Based on the candidate nodes, obtain the allusion nodes, annotation entries and quotation fragments corresponding to the ancient text paragraphs to form an initial set of evidence;

[0139] The evidence in the initial evidence set is weighted according to similarity score, consistency of dynasty timescale, genre matching degree and entity alignment relationship, and then sorted according to the weighting results to obtain the evidence set corresponding to the ancient text passage.

[0140] In one embodiment, a knowledge graph construction module is also included, for:

[0141] We acquire ancient text corpora, annotation data, variant character mapping data, phonological data, and data on people, places, and official titles, and perform standardization processing to obtain an index database of knowledge related to ancient texts.

[0142] Nodes and relational edges were extracted from the ancient Chinese knowledge index database. Nodes include characters, words, meanings, structures, allusions, quotations, figures, place names, official titles, dynasties, and annotation entries. Relational edges include diachronic evolution relationships, cognate relationships, classical reference relationships, example relationship relationships, and compositional relationships.

[0143] Generate node vector representations for nodes, and form an ancient Chinese knowledge graph based on nodes, node vector representations of corresponding node attributes, and relation edges.

[0144] In one embodiment, the dual-track translation module 404 is further configured to:

[0145] Based on contextual semantic representation, modern Chinese translation tracks are generated according to different translation styles;

[0146] Based on the contextual semantics, the sentences are segmented into semantic and syntactic parts to generate annotation tracks, and the source identification corresponding to the evidence set is marked in the annotation tracks to obtain the modern Chinese annotation tracks.

[0147] In one embodiment, the evidence constraint module 405 is further configured to:

[0148] Generate a constraint lattice based on the evidence set; the constraint lattice contains constraints that must be covered, can be covered, and cannot be covered.

[0149] Based on the constraint grid, the modern Chinese translation track and the modern Chinese annotation track are decoded, and the consistency penalty term is combined to eliminate translations that violate the annotations, resulting in multiple candidate translations. The consistency penalty term includes the penalty weight when the translation and annotation contradict the time and character relationship between the constraint grid.

[0150] Candidate translation annotations are filtered based on evidence coverage to obtain candidate translation annotations that meet the evidence coverage requirements.

[0151] In one embodiment, the text rearrangement module 406 is further configured to:

[0152] The candidate translation and annotation results are scored for literary style, and the literary style score includes the calculation of parallelism, repetition, rhythm, phonetics and rhetoric preservation.

[0153] The candidate translation and annotation results are rearranged based on the textual style scoring results to obtain the textual style optimized translation and annotation results;

[0154] The source identifier is mapped to the knowledge graph node, and the annotation template is rendered according to the preset user view to obtain the translation result containing the modern translation and annotation.

[0155] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps in the above method embodiments.

[0156] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0157] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0158] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for translating classical Chinese texts based on a large model, characterized in that, The method includes: The ancient Chinese text to be translated is obtained, and the ancient Chinese text is subjected to higher-level semantic standardization processing to obtain the ancient Chinese text processing result with structured information; the ancient Chinese text processing result includes character-level information, structure recognition result and dynasty mark; the higher-level semantic standardization processing corresponds to sentence segmentation, structure recognition and dynasty mark; Based on the ancient text processing results, context encoding is performed to fuse glyph information and exegesis information, and combined with diachronic semantic modeling to obtain the context semantic representation of the corresponding ancient text paragraph; Based on the contextual semantic representation, a similarity search is performed in the ancient Chinese knowledge graph to obtain a set of evidence corresponding to the ancient text passage; The contextual semantic representation and the evidence set are input into the translation and annotation dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks. Based on the evidence set, the modern Chinese translation track and the modern Chinese annotation track are constrained and screened to obtain candidate translation and annotation results that meet the evidence coverage requirements; Based on the candidate translation and annotation results, the text is rearranged in terms of style and tone, and the source identifiers in the annotation track are made executable, outputting a translation result that includes the modern translation and annotations.

2. The method according to claim 1, characterized in that, The step of performing context encoding processing based on the ancient text processing results, fusing glyphic information and exegesis information, and combining it with diachronic semantic modeling to obtain the contextual semantic representation of the corresponding ancient text paragraph includes: Based on the character-level information, the stroke sequence and component structure of the characters in the ancient text paragraph are extracted, and the stroke sequence and component structure are encoded to obtain character shape vectors; The corresponding historical commentaries and illustrative texts of the ancient text paragraphs are encoded based on the character-level information to obtain semantic vectors; The glyph vector and the semantic vector are fused to obtain the corresponding fused representation; Based on the dynasty marker, a dynasty time-stamped vector is obtained, and based on the dynasty time-stamped vector, a diachronic semantic model is performed on the fused representation to obtain a diachronic enhanced representation; The contextual semantic representation is obtained by jointly modeling the diachronic augmentation representation and the configuration recognition result.

3. The method according to claim 2, characterized in that, The step of performing a similarity search in the Classical Chinese knowledge graph based on the contextual semantic representation to obtain a set of evidence corresponding to the Classical Chinese passage includes: Based on the ancient Chinese knowledge graph, a similarity search is performed on the context semantic representation and the node vector representation to obtain candidate nodes; Based on the candidate nodes, obtain the allusion nodes, annotation entries and quotations corresponding to the ancient text paragraphs to form an initial set of evidence; The evidence in the initial evidence set is weighted according to similarity score, consistency of dynasty timescale, genre matching degree and entity alignment relationship, and sorted according to the weighting results to obtain the evidence set corresponding to the ancient text passage.

4. The method according to claim 3, characterized in that, The ancient Chinese knowledge graph was constructed using the following methods: We acquire ancient text corpora, annotation data, variant character mapping data, phonological data, and data on people, places, and official titles, and perform standardization processing to obtain an index database of knowledge related to ancient texts. Nodes and relational edges are extracted from the ancient Chinese knowledge index database. The nodes include characters, words, meanings, structures, allusions, quotations, figures, place names, official titles, dynasties, and annotation entries. The relational edges include diachronic evolution relationships, cognate relationships, classical reference relationships, example relationship relationships, and compositional relationships. Generate the node vector representation for the node, and form the Ancient Chinese knowledge graph based on the node, the node vector representation of the corresponding node attribute, and the relation edges.

5. The method according to claim 1, characterized in that, The process involves inputting the contextual semantic representation and the evidence set into a dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks, including: Based on the aforementioned contextual semantic representation, modern Chinese translation tracks are generated according to different translation styles; Based on the contextual semantics, the sentences are segmented into semantic and syntactic parts to generate annotation tracks, and the source identification corresponding to the evidence set is marked in the annotation tracks to obtain the modern Chinese annotation tracks.

6. The method according to claim 1, characterized in that, The constraint screening of the modern Chinese translation track and the modern Chinese annotation track based on the evidence set to obtain candidate translation and annotation results that meet the evidence coverage requirements includes: A constraint grid is generated based on the evidence set; the constraint grid contains constraints that must be covered, can be covered, and cannot be covered. Based on the constraint grid, the modern Chinese translation track and the modern Chinese annotation track are decoded, and the translation annotations that violate the consistency penalty are eliminated to obtain multiple candidate translation annotation results; the consistency penalty includes the penalty weight when the translation annotations contradict the time and character relationships between them. The candidate translation annotations are filtered based on the evidence coverage rate to obtain candidate translation annotations that meet the evidence coverage requirements.

7. The method according to claim 1, characterized in that, The process of rearranging the text based on the candidate translation and annotation results, and making the source identification in the annotation track executable, outputs a translation result containing the modern translation and annotations, including: The candidate translation and annotation results are scored for literary style to obtain the literary style score; the literary style score includes the calculation of parallelism, repetition, rhythm, phonetics and rhetoric preservation. The candidate translation and annotation results are rearranged based on the textual tone scoring results to obtain the textual tone optimized translation and annotation results; The source identifier is mapped to a knowledge graph node, and the annotation template is rendered according to a preset user view. The translation result containing the modern translation and annotation is obtained by combining the translation and annotation results.

8. A classical Chinese translation system based on a large model, characterized in that, The system includes: The standardization processing module is used to acquire the classical Chinese text paragraphs to be translated and perform higher-level semantic standardization processing on the classical Chinese text paragraphs to obtain classical Chinese text processing results with structured information; the classical Chinese text processing results include character-level information, structure recognition results, and dynasty markers; the higher-level semantic standardization processing corresponds to sentence segmentation, structure recognition, and dynasty markers; The semantic extraction module is used to perform context encoding processing based on the ancient text processing results, integrate glyph information and exegesis information, and combine it with diachronic semantic modeling to obtain the context semantic representation of the corresponding ancient text paragraph. The evidence module is used to perform similarity retrieval in the ancient Chinese knowledge graph based on the contextual semantic representation to obtain a set of evidence corresponding to the ancient text passage; The translation and annotation dual-track module is used to input the contextual semantic representation and the evidence set into the translation and annotation dual-track decoder to generate multiple modern Chinese translation tracks and modern Chinese annotation tracks. The evidence constraint module is used to constrain and filter the modern Chinese translation track and the modern Chinese annotation track according to the evidence set, so as to obtain candidate translation and annotation results that meet the evidence coverage requirements. The text rearrangement module is used to rearrange the text based on the candidate translation and annotation results, and to perform executable processing on the source identification in the annotation track, and output the translation result containing the modern translation and annotation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images