Evaluation of Intelligent Proofreading Effect of English Translation of Lingnan Hakka Dialect Based on Dependency Syntactic Networks

Dependency syntax refers to the technique of revealing the interdependence between components within a language and its units. Dependency syntax provides a theoretical basis for the understanding of some unfamiliar languages and the processing of natural language through the in-depth analysis of some grammatical components in a sentence. In the process of analysing the language by using dependent syntax, it is not necessary to pay attention to some complicated phrase components in the language, but only need to pay attention to the meaning of the special words themselves and the dependency relationship between the words. An example of a typical dependency syntax is shown in Figure 1.

Figure 1.

Dependency syntax case analysis

As can be seen from Figure 1, the central word of the example sentence is the verb prefer, and analyzing the overall sentence structure, the central verb is dependent on the subject word (nsub) “I” and the direct object structure of the sentence (dobj) flight. while the direct object structure of the sentence flight is dependent on the the definite article (det) “the” and (nmod) “morning”, “Denvor” conjunction “through”. The verb “prefer” exists independently and has no dependency on any word, but in the formed language it is necessary to construct a “root”, which is dependent on the verb “prefer”.

The syntactic structure of dependent syntax is shown in Figure 2.

Figure 2.

Syntactic structure composition diagram

In contrast to constituent syntax, dependent syntactic network analysis is a detailed analysis of sentence structure directly from the main components of the sentence such as subject and predicate. In addition, the results of the dependency syntactic network analysis are more direct in analyzing the relationship between words. As shown in Figure 2, the example of this dependency syntactic network analysis diagram, the direct object of the central verb prefer in the overall sentence structure is flight, and there is a strong dependency relationship between prefer and flight in the dependency syntactic network analysis tree.

The tree analysis of the sentence is shown in Figure 2. From Figure 2, it can be seen that there is a tertiary dependency, but this tertiary dependency between prefer and flight does not exist if analyzed by constituent syntax.

Figure 3.

Dependency syntactic network analysis

When analyzing the English translation of Lingnan Hakka dialect based on the dependency syntactic network, the study of traditional grammatical relations is the basis for exploring the structure of linguistic dependency relations, and it is also the first step in learning the structure of the dependency syntactic network, but in the overall structure of the dependency syntactic network the existence of dependency relations is only binary simple relations. Each binary simple dependency is supported by a dependency and a centralizer, such as the SVP statistic line in the example above, which represents the direct dependency between I and refer, i.e., a dependency predicate is dependent on the centralizer subject. In linguistics, linguists have many definitions for the study of dependency in sentences, among which the definition of dependency syntax may vary from language to language.

The verbal meaning of “Lingnan” is divided into a broader and a narrower sense. Lingnan in the broad sense originally refers to the southern part of the Five Ridges in the south of China, while “Lingnan” in the narrow sense generally refers to Guangdong. Traditionally, the Hakka dialect is distributed over a wide range of areas and is not connected to each other, and in many areas south of the Yangtze River today there is a continuation of the Hakka dialect, including the Guangdong and Guangxi regions, Fujian, Sichuan, Taiwan and other places. It is even distributed abroad and is prevalent in some Southeast Asian regions. Among them, the Hakka dialect of the Lingnan region, due to the intersection of foreign cultures at that time and the later spread of a wider number of people, has a more pronounced structure of dependent syntax, which is analyzed in the following examples.

The Lingnan Hakka dialect is not very different from Putonghua in terms of linguistic structure, but it has obvious features in pronunciation, which are mainly as follows: firstly, in the development of the Lingnan Hakka dialect today, some of the ancient turbid consonants have been gradually eliminated, and most of them are pronounced as clear consonants regardless of their flat or oblique sounds, such as “Peach” (定平) and “Dao” (定仄) are pronounced as clear vowels today, even though one of them is flat and the other is oblique in the records. Secondly, in terms of consonants, the Lingnan Hakka dialect has preserved the full turbid consonants of ancient Chinese to a certain extent, for example, “床” (chuáng) is pronounced as [ch] in Putonghua, while in Hakka it is pronounced more similarly to the [z] sound. In addition, the Lingnan Hakka dialect is extremely rich in nasal consonants, such as [ŋai] for “milk” (nǎi). The Lingnan Hakka dialect has a more complex rhyme system, and maintains the incoming vowel endings in pronunciation, with common endings such as -p, -t, -k, etc. For example, “白” (bái) is pronounced as [bak] in Hakka. In terms of tones, Lingnan Hakka dialect is very different from Mandarin. Lingnan Hakka dialect generally has six to eight tones, which is much richer than Mandarin. In Lingnan Hakka dialect, the pitch of the tones and the changes in the elevation of the tones obviously affect the meaning of the words, just as a word can have different meanings when read in different tones. Finally, in terms of overall pitch, Lingnan Hakka has very varied tones, which to a certain extent show a strong musicality, and changes in pitch are crucial to understanding and expression in Lingnan Hakka.

In the traditional Lingnan Hakka dialect, the analysis of dependent syntax can help us to simplify the understanding of some complex syntactic patterns and unique semantic structures of the dialect to a great extent. For example, in the Lingnan Hakka sentence “Granny cooks rice for Ah Sun”, the dependency relations between sentences can help us understand the interdependence of the components in the sentence: “Granny” as the center word, i.e. subject, is dependent on the verb, i.e. the predicate “cook”, and “rice” is dependent on the predicate verb “to cook” and on the prepositional phrase “give it to Ah Sun to eat”. The word “rice” is dependent on the predicate verb “cook”, and together with the prepositional phrase “畀阿孙食”, it constitutes an extension of the predicate in the whole sentence structure. In this sentence, “畀” (to give) and “阿孙” (grandson) form a preposition-object combination, which is co-dependent on the predicate verb “to cook” and the other verb “to eat”. Another verb “eat” (吃) further modifies “阿孙”. In this example sentence, by analyzing the syntactic network of dependency, we can clearly see the dependency between the words, understand the complex semantic structure, clarify the relationship between the components, so as to understand and grasp the meaning and concept of the sentence as a whole, and preserve the unique expression of Lingnan Hakka dialect through the simple comprehension of the complex syntactic meaning and tone.

Dependency syntactic network analysis is a method of analyzing complex linguistic structures based on dependency syntactic networks, which is applicable to many languages. In the process of dependency syntactic network analysis, the first step is to identify the dependency relationships among the words in various complex sentences, so as to construct syntactic trees. The basic modeling of dependency syntactic network analysis consists of two concepts, namely, dependency relation and syntactic tree. Each of the words in a sentence can be represented as a directed graph (G = (V, E)), where (V) is the set of complex words, which also represents nodes, and (E) is the set of dependency relations. Each dependency (e, in E) can be represented as (e = (h, m)), where (h) refers to the dominant word (head) and (m) refers to the dependent word (modifier), which clearly expresses the dependent syntactic relationships in it. The syntactic tree structure is shown in Figure 4.

Figure 4.

Non-projective dependency trees

In this example, we first set the definition of an edge (arc), which satisfies the Projective condition: suppose that this edge connects A1 and A2, where 1 < 2, and suppose that the central word of the sentence is head (which is one of A1 and A2). If there exists a path from the center word head to any complex word in that tree diagram relative to any of the complex words between 1 and 2, then it means that this line in the tree diagram satisfies the Projective property. On the contrary, if in a dendrogram representing dependency, there is a dependency tree in which there exists a path from the central word head to the word in each branch that satisfies the Objective property, where if there is any branch that does not exist a path from the central word head to the word, then it does not satisfy the Objective property, such as in the simple dependency dendrogram shown in the figure above, which The dendrogram has a branch (side arc) from flight, the center word (head), to was, which indicates that the individual word combinations contain more than one word between them, e.g., the center word (head), flight, has a direct path to which, (flight-was-which), and, conversely, there is no direct path from the center word (head) to the word in the words. On the contrary, there is no branch between the words “this” and “morning” and flight. Therefore, this edge (arc) does not satisfy Projective, i.e., it is non-Projective.

Based on the above operations, a simple analysis algorithm for dependent syntactic relations can be realized, as shown in Figure 5. In this algorithm, there is only one center word root in the initial state column, and all the words are arranged in a queue, and the algorithm loops until it reaches the end state (i.e., only the center word root is left in the state column, and the other queues are empty).

Figure 5.

Dependency syntax analysis algorithm based on transformation

To perform the arithmetic, the following arithmetic steps are constructed (1) X1=∑n=1N(1+voc+an)fn \[{{X}_{1}}=\sum _{n=1}^{N}(1+voc+{{a}_{n}}){{f}_{n}}\]

X is the base model import structure

Voc is the root word in the syntax

an is the nth dependency occurrence rate

fn is the nth syntactic factor in the input process (2) X2=∑n=1N(1+Xn+voc+bn)fn \[{{X}_{2}}=\sum _{n=1}^{N}(1+{{X}_{n+voc}}+bn){{f}_{n}}\] (3) Sn=max[ Sn+∑n=1NEnabM(1+Xn+voc+bn+1+voc+an) ] \[ $${S_n} = \max \left[ {{S_n} + \mathop {\mathop \sum \nolimits^ }\limits_N^{n = 1} _{E\matrix{ n \cr a \cr b \cr } }^M\left( {1 + {X_{n + voc}} + bn + 1 + voc + {a_n}} \right)} \right]$$ \] (4) Sn=n(1+n)x(1+n)x−1(∑aX1(Sa+Sb)+∑bX2(1+Xn+voc+bn)S(Sa+Sb),) \[{{S}_{n}}=\frac{n{{(1+n)}^{x}}}{{{(1+n)}^{x}}-1}\left( {{\sum }_{a}}{{X}_{1}}(Sa+Sb)+{{\sum }_{b}}{{X}_{2(1+{{X}_{n+voc}}+bn)}}{{S}_{(Sa+Sb),}} \right)\] (5) Sn=max[ X1+∑s=1SpS(S1+S2) ]+βRr $${S_n} = \max \left[ {{X_1} + \mathop \sum \limits_{s = 1}^S {p_S}\left( {{S_1} + {S_2}} \right)} \right] + \beta {R_r}$$ (6) p=1−In(11−β)1s \[p=1-\sqrt{\operatorname{In}\left( \frac{1}{1-\beta } \right)\frac{1}{s}}\] (7) WLna=∑n=1W∑a=1L(AnRa+RnAa) \[W{{L}_{\underset{a}{\mathop{n}}\,}}=\sum _{n=1}^{W}\sum _{a=1}^{L}({{A}_{n}}{{R}_{a}}+{{R}_{n}}{{A}_{a}})\] (8) WLn=S(X1−X2) \[W{{L}_{n}}=S\left( {{X}_{1}}-{{X}_{2}} \right)\] S_n represents the number of Stacks entered into the monitoring system, which includes different categories such as root, book, the, me, etc., and WL is the Word List entered. To optimize the system operation, the duplicate items in the two sets are processed as follows

Input function to find duplicate items R, single duplicate Ro, multiple duplicate Rm, βis the operation constant (9) Rm=Ro+11−β∑s=1SXsβs \[{{R}_{m}}={{R}_{o}}+\frac{1}{1-\beta }\sum _{s=1}^{S}{{X}_{s}}{{\beta }_{s}}\] (10) Ro≥0 \[{{R}_{o}}\ge 0\] (11) Rm≥0 \[{{R}_{m}}\ge 0\] (12) βs≥Xs−Ro+m \[{{\beta }_{s}}\ge {{X}_{s}}-{{R}_{o+m}}\] (13) Rmβ=max[ Ro+11−β∑s=1SXSβs ]+βRo $${R_{m\beta }} = \max \left[ {{R_o} + {1 \over {1 - \beta }}\mathop \sum \limits_{s = 1}^S {X_S}{\beta _s}} \right] + \beta {R_o}$$ (14) Ac=∑aX1(Sa+Sb) \[Ac={{\sum }_{a}}{{X}_{1}}(Sa+Sb)\] (15) Re=∑bX2(1+Xn+voc+bn)S(Sa+Sb), $$Re = \mathop \sum \limits_b {X_{2(1 + {X_{n + voc}} + bn)}}{S_{(Sa + Sb)}},$$ The simple analysis algorithm for dependent syntactic relations is shown in Table 1. The results of the algorithm analysis are shown in Table 2, Figure 6 and Figure 7, respectively. The results show that the algorithm maintains high accuracy at any step with an error of no more than 10%.

Figure 6.

Algorithm analysis data graph

Figure 7.

Algorithm analysis data graph

In the loop of the above dependent syntactic network algorithm, each step of the loop is automatically analyzed by the system according to the current state, and then the Oracle function is used to perform calculations, and finally the best algorithmic step is selected to perform the operation task. However, when the following situations occur, it may lead to an algorithmic error in the Oracle function, which ultimately leads to an error in the calculation result. If the final parse result is wrong, but it can be proved that the current dependency dendrogram satisfies the Projective nature according to the above algorithm, then at least one operation sequence can be performed in the algorithm. In the following, we take a typical sentence in Lingnan Hakka dialect as an example: “Granny cooks rice for the grandchildren”. We use the dependency syntactic network method to set up a dependency syntactic tree diagram for step-by-step parsing: first, we step-by-step parsed the dependency relationship between the main verb (predicate) and the subject, and the subject center word “阿婆” serves as a dependency of the predicate verb “煮”. In this sentence, the subject center word “阿婆” serves as the dependency of the predicate verb “煮”, and its dependency relation can be expressed as (eA1 = (text{boil}, text{Annie})), where “煮” is the dominant word and “阿婆” is the dependency word. Analyzing the sentence structure, it is clear that “granny” is the executor of the action “cook”. The next step is to analyze the dependency relationship between the object and the verb. The object “rice” is directly dependent on the predicate verb “cook”, and the algorithm is expressed as (e A2= (text{cook}, text{rice})), where the object “rice” is the subject of the predicate verb “cook”. “rice” is the action object of the predicate verb “cook”, i.e., the predicate verb “cook” is executed in order to form the object “rice”. The object of the predicate verb “cook” is the object of the action of the predicate verb “cook”. The next step is to analyze the network dependency structure of the prepositional phrase. The purpose of the prepositional phrase “畀阿孙食” in the sentence structure is to modify the predicate verb “煮”, and the algorithm is expressed as (e A3= (text{cook}, text{畀})), where “畀” is the object of the action of the predicate verb “cook”, i.e., the predicate verb “cook” is executed in order to form the object “meal”. where “give” is the head verb of the prepositional phrase in the sentence structure. In the sentence structure, the preposition “畀” (i.e. the head verb) and the object “阿孙” form a dichotomous dependency, which is represented by the algorithm (e A4 = (text{畀}, text{阿孙})), i.e., it represents the action referred to by the predicate verb “the recipient of the action referred to by the predicate verb. Finally, the dependency between the complements is analyzed. In this example sentence, the complement “food” is not dependent on the subject-centered clause, but on the structure of the prepositional phrase, which is represented by the algorithm (e A5 = (text{畀}, text{food})), and the structure of the sentence “food” as a complement to the action “cook” is not dependent on the subject-centered clause. The sentence structure in which “food” is used as a complement to further explain the action of “Ah Sun” makes the meaning of the sentence clearer. After the complete algorithm operation and analysis of the above departments, finally, the dependency syntactic structure of the whole sentence is clearly shown by constructing a dependency syntactic tree diagram. When the dependent syntax analysis is carried out, the algorithm is keyed into the analysis, and this intelligent analysis method can help people to analyze the dependent syntactic structure of the sentence more clearly, and at the same time, it can also accurately convey the unique syntactic logic of Lingnan Hakka dialect, and translate foreign languages with the systematic thinking of Lingnan Hakka dialect, which can make the communication more precise and provide a powerful support to the grammatical research and translation precision of the Lingnan Hakka dialect. It provides a strong support for the grammatical research of Lingnan Hakka dialect and the precision of translation.

The core of building a modern multilingual intelligent translation system lies in creating a rich and comprehensive dialect corpus. In the process of developing intelligent English translation of Lingnan Hakka dialect, the difficulty that still needs to be faced is the construction of Lingnan Hakka dialect corpus, the cost of which is still being raised, and the construction of highly intelligent corpus requires large-scale high-quality parallelisms, which are being accumulated and sorted out at the moment, and the slow process of constructing high-quality corpus of Lingnan Hakka dialect leads to the limited data for the training of the intelligent translation model. Intelligent English translation based on the existing Lingnan Hakka dialect corpus still needs to rely on manual collection and some traditional methods of collation, because of its narrow coverage, so it can not go to a comprehensive reflection of the diversity and complexity of the Lingnan Hakka dialect.

The construction of Lingnan Hakka dialect corpus is crucial to the development of intelligent English translation of Lingnan Hakka dialect, and the process of its construction is complicated and critical, involving a large number of linguistic and phonetic data collection and data organization. At present, the construction of Lingnan Hakka dialect corpus is mainly in the stage of collecting Lingnan Hakka dialect text and special pronunciation audio data. The text collection of Lingnan dialect mainly includes the collection of some literary works with Lingnan Hakka dialect as the main material, as well as some oral histories and folktales from local villagers, etc. The special audio collection of Lingnan Hakka dialect focuses on the recording of natural dialogues and some reading materials of Lingnan Hakka dialect speakers in different regions and at different age stages.

In the process of corpus data organization, firstly, we should carry out detailed word division, labeling and annotation of the collected materials, and then we should construct grammatical structures, analyze the dependent syntactic structures, and carry out detailed organization and summarization of the phonological system of the vocabulary list. In addition, it is also necessary to add rich contextual features and cultural background information according to the language habits of Lingnan Hakka dialect. The core purpose of constructing the Lingnan Hakka dialect corpus is to enhance the practicality of the corpus, and focusing on the organization of linguistic and cultural details is the most convenient way to achieve its purpose. At present, although the construction of Lingnan Hakka Corpus has made some progress based on the efforts of all parties in collecting and organizing information, the scale and coverage of the Lingnan Hakka Corpus still need to be further expanded in order to satisfy the needs of research and intelligent translation of the Lingnan Hakka dialect.

In the process of intelligent English translation of Lingnan Hakka dialect, the factors of local cultural characteristics and the unique meanings of some vocabularies in the context should not be neglected. Lingnan Hakka dialect contains a large number of local cultural connotations, which need to be dealt with through the contextual understanding and some cultural translation strategies in the process of language translation. In the process of upgrading the Lingnan Hakka dialect translation system, we need to update the following two models: the context-aware model, the principle of which is to make the system training model have the ability to perceive the context and differentiate between different words and sentences in the dialect, so as to translate more accurately. Cultural Translation Database Model: By constructing a database that covers the entire Hakka cultural background in Lingnan, it can help the AI system to make accurate cultural conversions when translating. The intelligent translation system for Lingnan Hakka dialect relies on multi-modal learning by combining various data types such as dialectal speech, text, and images to achieve more accurate recognition and translation of Lingnan Hakka dialect. At the same time, the cross-language modeling of the intelligent translation system will also help the system to have stronger adaptability and learning ability when dealing with complex language environments.

The intelligent English translation system for Lingnan Hakka dialect mainly relies on NLP and deep learning technologies. In the intelligent English translation of Lingnan Hakka dialect, the detailed operation steps are shown in Figure 8.

Figure 8.

Intelligent English translation process of Lingnan Hakka dialect

As can be seen from Figure 8, a comprehensive corpus containing contextual understanding and cultural translation of Lingnan Hakka dialect needs to be constructed first. Then, through the preprocessing step, the pronunciation correction and the sub-sentence and phonetic labeling are performed to ensure that the intelligent Lingnan Hakka dialect translation system can correctly identify and understand the features of the Lingnan Hakka dialect. Finally, in the model training phase, a deep learning algorithm is used to train the translation model so that the intelligent system can accurately handle the unique grammar and vocabulary of the dialect. In the actual translation, the intelligent system will translate the Hakka dialect into English by analyzing the dependent syntactic structure through the automatic translation tool, and ensure the accuracy and naturalness of the English translation through the contextual analysis and cultural adaptation of artificial intelligence.

Intelligent translation proofreading database, as the main storage location of translation data and the means of connecting diversified languages, plays a decisive role in the data structure of Lingnan Hakka dialect intelligent dialect translation system and the correctness of translation. By using the object class ADO.NET, the core Lingnan Hakka dialect intelligent translation proofreading database table structure is designed, as shown in Table 3-Table 5. Among them, the statement relationship table between each language structure determines the corresponding words, complex texts, and sentences to be extracted from the statements generated by the intelligent translation system. The translation decision table is used to detect whether there is a direct translation relationship between the source language of the natural translation system and the target language of the Lingnan Hakka dialect intelligent translation system, i.e., to analyze whether the Lingnan Hakka dialect intelligent translation system supports the translation processing between the languages, and the structure of the table is important for strengthening the expandability, plasticity and flexibility of the Lingnan Hakka dialect intelligent translation system, and for increasing the number of intelligent translations of complex dialect language varieties in the future. The data table structure lays a material foundation for strengthening the expandability, plasticity and flexibility of the Lingnan Hakka dialect intelligent translation system, and for increasing the number of intelligent translations of complex dialects and language types in the future, and provides the function of storing the classic phrases and cultural and historical traditions of Lingnan Hakka dialect as well as storing the initial texts and corrected pairs of phrases of the natural source language and the target language of the intelligent translation.

The user side of Lingnan Hakka dialect intelligent translation and proofreading system is located in the front end of the whole Lingnan Hakka dialect translation and proofreading system. The main interface of the application program of this translation and proofreading system consists of the continuation relationship between the view and the view group objects, and it is written in Java language through the running of the application program. The main interface function of this intelligent English translation and correction system is to complete the transmission of language text and media information and the jumping function of the main page of English translation and correction by applying the interrelationship between the view and view group objects, and the main principle of the system is to complete the system response and processing by using the NLP and the deep learning technology.

The proofreading of the translation effect of the intelligent system for the Lingnan Hakka dialect is the key link to ensure the translation quality of the intelligent translation system. After the Lingnan Hakka dialect intelligent translation system completes the preliminary search and analysis of the comprehensive corpus constructed and makes reference to the initial text storing the natural source language and the target language of the intelligent translation, the correction system evaluates and corrects the output translation results, and the proofreading stage audits and optimizes the resulting translation by the various means described above. The intelligent translation correction system analyzes the translation results to automatically detect grammatical errors in the translation results, conducts a comprehensive analysis in conjunction with the grammatical structure of the dependent syntax, eliminates unnatural linguistic expressions as well as errors in cultural connotations and natural contexts, provided that appropriate expressions in cultural connotations and natural contexts were typed into the language and writing system during the previous construction of the corpus, and corrects potential errors in the translated structure output by the intelligent translation system. potential misunderstandings of semantic comprehension in the translation structure output by the system. The final manual proofreading is carried out, which mainly emphasizes in-depth analysis of the culture and some special expressions in the regional context, so as to correct the subtle differences that the intelligent language system is unable to accurately analyze and deal with, as well as the unique expression habits of the Lingnan Hakka dialect. The final output of the translation should reach the standards of accurate semantic expression, natural emotion and proper transmission of Lingnan Hakka culture after several rounds of proofreading. The intelligent translation correction system for Lingnan Hakka dialect not only improves the translation efficiency, but also improves the translation quality through the above correction steps.