An Innovative Use of Interactive Technology in Second Language Writing Instruction and Corpus Linguistic Analysis
Online veröffentlicht: 29. Sept. 2025
Eingereicht: 11. Jan. 2025
Akzeptiert: 24. Apr. 2025
DOI: https://doi.org/10.2478/amns-2025-1122
Schlüsselwörter
© 2025 Yuqing Cui, published by Sciendo.
This work is licensed under the Creative Commons Attribution 4.0 International License.
Writing ability is a very important part of second language proficiency, and the improvement of second language learners’ writing ability is an extremely long process even though they are in the target language environment. Besides, writing has been widely recognized as one of the most important parts of second language learning and teaching.
The main assessment elements of second language writing can be categorized into accuracy, fluency and complexity [1]. In writing assessment, accuracy, fluency and complexity indicators can not only measure the writing performance of second language writers, but also reflect their writing ability [2-3]. A developmental model of writing instruction based on the multidimensional interaction between second-language writing ability and the elements of writing evaluation assessment can provide an in-depth understanding of the micro-level of the language system and reveal the developmental patterns of accuracy, fluency and complexity in language assessment from the inside [4-7].
A corpus can provide a relevant linguistic corpus for analyzing the multidimensional interaction between second-language writing ability and the elements of writing assessment rubrics [8]. A corpus is a linguistic database consisting of a large amount of information about the actual use of a language, dedicated to language research, analysis and description [9-10]. It is created by collecting representative authentic linguistic materials actually used by people on the basis of random sampling [11]. Corpus is a convenient tool for quantitative analysis and study of language, a technological advancement in the means of linguistic research, marking a major shift in the idea of language research, shifting language research from the traditional intuitive empirical method to the quantitative statistical method, and improving the efficiency of language research [12-15].
Literature [16] describes the principle of corpus use in second language teaching and learning, which provides language learners with a large number of situational exposures needed to enhance their understanding of the language in many ways, i.e., a data-driven learning model. Literature [17] discusses the interaction between corpus building and second language learning and provides some suggestions for its development by analyzing practical examples of learner corpus building. Literature [18] outlines the role of linguistic profiling in improving the quality of writing, and investigates the relationship between the quality and development of linguistic writing in terms of the three linguistic structures, namely, lexical complexity, syntactic complexity and textual articulation. Literature [19] emphasizes syntactic complexity as an important factor in second language writing assessment and proposes Biber Tagger, Coh-Metrix and L2 syntactic complexity analyzer to quantify the analysis operations in corpus-based writing research. Literature [20] examined the effectiveness of pre-service teachers’ use of corpora in language learning and teaching, and found that most of the teachers only mastered how to manipulate corpora, but lacked the ability to integrate their linguistic analytic skills in corpora with pedagogical development. Literature [21] proposes a teacher training program based on corpus-based language teaching methodology to solve teachers’ difficulties in using it on the one hand, and to enhance teachers’ ability to utilize corpus resources to teach students effectively on the other hand. Literature [22] shows that formulaic sequences for assessing second language learners’ writing proficiency using a semi-automatic method is the key to improving their writing quality, and that formulaic metrics effectively predict the quality of written texts and provide a solid foundation for corpus-based text analysis.
Aiming at the drawbacks of traditional second language teaching and the advantages of human-computer interaction technology in writing teaching, a learning system based on virtual learning environment human-computer interaction technology is proposed. The human skeletal posture information is collected by Kinect V2.0 sensor, and Euclidean distance and cosine theorem are used to recognize the specific posture of students. The method of creating a dictation recognizer object and registering a listener to handle dictation events completes the design of the voice interaction module. For the data interaction module, the shared memory method is used to realize the transfer and sharing of writing data among Unity3D, MATLAB and PyCharm. Finally, Unity3D virtual engine is used as the development environment to construct the 3D virtual learning environment and teaching system. Using the corpus linguistic analysis method, the writing performance of students receiving the new teaching method is analyzed in terms of articulation ability, semantic similarity, syntactic complexity and writing accuracy to verify the feasibility of the method in this paper.
In the current second language writing teaching, most teachers still follow the “indoctrination” teaching mode, the basic process is: the teacher gives the topic of the composition - the teacher instructs the essentials of writing - students write according to the requirements - students submit their homework -Teachers corrected. This writing teaching mode completely ignores the autonomy of the students, in fact, the students are regarded as the knowledge of the acceptance of tools, is a typical embodiment of behaviorist learning theory. The decrease of students’ autonomy directly leads to the decline of learning interest, which is easy to make students develop the habit of not asking questions and not thinking, and may also produce numbness to second language writing.
First, students learn some foreign language knowledge in the classroom, but due to the constraints of a fixed writing model, they are not able to apply the new knowledge in the acquisition of writing skills.
Second, the writing process does not fully mobilize students’ thinking system and language system, which stifles their creativity. Generally speaking, the writing of a piece of writing needs to go through the following processes: understanding the problem-content conceptualization-structural design-text output. Teachers in second language writing instruction not only stipulate specific writing outlines, but also limit the number of words to be written. This leads to students’ inability to freely and fully elaborate their own views on the one hand, and exacerbates the homogenization of students’ works on the other.
Thirdly, the teacher’s assessment focuses on summarizing assessment and fails to provide timely guidance for students. Teachers uniformly corrected and evaluated students’ written work, and then uniformly assessed it in the next class. Although students can find out their own shortcomings through teachers’ written evaluation or assessment, they can’t get timely help for the confusion they encounter in the process of writing. Finally, students do not pay much attention to the teacher’s evaluation, nor do they pay attention to checking and correcting mistakes. Usually, students do not pay attention to the errors in writing after they finish their essays, but they pay extra attention to the teacher’s assessment. Some teachers try to break this disadvantage and make partial innovations within the traditional teaching mode, such as letting students correct each other’s compositions and discussing in small groups, etc. However, students still do not show a strong desire to learn second language writing because the essence of the procedural teaching mode has not been changed [23].
The rapid development of human-computer interaction technology promotes students’ learning without the limitation of time and place, and improves the initiative and subject position of students’ learning [24]. In addition, the application of human-computer interaction technology also has a profound impact on traditional writing teaching, which is specifically manifested in the following aspects:
First of all, the interaction technology then has the influence of real-time and synergy. In the network environment, students can get the information of second language writing teaching in time or delayed access through data interaction technology. The so-called delayed access refers to the fact that the access time of students to teaching information is not fixed, as long as the information in the teaching information platform still exists, students can access it at any time through the account login, and the development of the interaction technology also promotes the full interaction of the information, which improves the efficiency of the teacher-student benefit resources to a greater extent.
Secondly, interactive technology has the influence of complementarity and sharing. In traditional writing teaching, teachers and students are in a kind of information unequal position, while with the aid of human-computer interaction, the teacher’s information superiority position no longer exists. Students can freely browse all kinds of materials and information according to their own writing interests and hobbies, and they can also discuss their writing experiences with others through the interactive platform. On the other hand, a large amount of teaching information in the virtual interactive system also provides help for teachers’ preparation and teaching, which makes up for the disadvantage of insufficient teaching resources available for traditional writing teaching to a greater extent.
Finally, a human-computer interaction system also has the influence of dynamics and virtuality. The transmission of teaching information is always in the process of dynamic development, that is to say, teaching information is updated at any time and any place. Virtuality is one of the important features of HCI technology, although many scholars have pointed out that virtual interaction will have a negative impact on students’ learning, but in second language writing teaching, the integration of virtual writing situations can effectively alleviate students’ shyness and nervousness.
In conclusion, taking the theory of second language writing as a guide, integrating interaction technology in second language writing teaching and reconstructing the second language writing teaching mode not only help to overcome the shortcomings of traditional writing teaching, but also help to innovate the concept and way of second language teaching.
In daily communication, every body movement contains huge interactive information, and in the process of interpersonal communication, body language often contains more information than verbal language, so exchanging information with objects in the virtual space through human body postures will make the human-computer interaction more accurate, and make it easier for the virtual objects to correctly understand the interaction intention. In this paper, the human body posture is used to interact with the virtual learning environment, and the Kinect V2.0 sensor collects the human body posture information as input, and when the user makes a specified action and is recognized, the virtual environment will respond and give feedback to the user.
In this paper, Kinect sensor is used to track the human skeleton in real time and calculate the distance and angle between each skeletal joint point to determine the human posture. The real-time acquisition of the human skeleton joints, through the calculation of the distance and angle between the joints to determine the human posture. The Euclidean distance and cosine theorem are used to recognize specific postures, for example, when recognizing the posture of “raising the right hand”, which mainly involves three joints of the right shoulder, the right elbow, and the right wrist, the distances and angles between the three joints are judged, and the posture is recognized successfully when it is within a certain range.
The three joint points form a triangle, and the relationship between the three sides can be calculated according to Equation (1), Equation (2), and Equation (3):
The angle of ∠
In practice, the lines between the joints cannot be completely horizontal or vertical, so this paper sets an effective error range for it, i.e., ±15°. When the calculation result is in this error range, the user’s pose will also be recognized correctly, but when it is beyond this range, the user’s pose will not be recognized correctly. Similarly, the position calculation also sets an effective error range, and when it is in the error range, the pose will be recognized correctly.
Speech recognition technology can convert the user’s voice input information into text information, and the text information can be used for intelligent Q&A, which is equivalent to the machine can also hear and understand. As of now, automatic speech recognition technology has gradually matured, which involves acoustics, analog recognition, artificial intelligence, deep learning and other disciplines, and with the popularization of 5G technology, the application range of voice interaction is becoming more and more extensive. At present, speech recognition technology is mainly used in the fields of smart home, chat robot, voice navigation, etc. In addition, combining speech recognition technology with natural language processing technology such as machine translation and text matching can be applied in more complex scenes. In this paper, we use Windows Speech API to implement speech recognition in Unity3D, which can easily realize the voice interaction function of smart Q&A. The specific steps are as follows:
Introducing Windows in C# scripts. Speech namespace and create the Dictation Recognizer object. Handle dictation events by registering a listener. The Windows speech API provides users with four types of registered events, mainly including DictationResult, DictationHypothesis, DictationComplete, and DictationError, which handle different dictation events depending on the user’s input. Recognize the speech and display the result in the UGUI interface.
Through the above steps to create a speech recognition script, when running the scene and speak through the microphone can be voice recognition. Combined with the related text matching algorithm to realize the scene of voice intelligent answer, to realize the function of voice interaction.
Since the virtual learning environment is a mapping of the real space, the virtual learning environment in the process of experimental simulation or intelligent question and answer process, it is often necessary to correlate the data input by the user with the data generated by the system processing, including the writing data and the question answer data, through the correlation of the way to make the user in the parameter change, the system according to the user’s input for the output feedback.
In the process of system development, the main use of Unity3D, MATLAB and PyCharm three development platforms, the three development platforms support the development of different languages, such as Unity3D development language for C#, PyCharm using the Python language, and MATLA development language is based on C + + +, so to complete the overall development of the system requires data transfer. The overall development of the system requires data transfer and sharing. In this paper, we use the shared memory method to realize the data transfer and sharing among the three software by saving the data in the same text file, the data sender writes the data, and the data receiver reads the data.
Unity3D virtual engine is used as the development environment to build a three-dimensional virtual learning environment, and the introduction of virtual intelligences makes the interaction process more natural and vivid. According to the requirements and framework of the system, the functional application modules of the system are designed, which mainly include the scene roaming module, the writing and explanation module, the intelligent question and answer module, and the classroom quiz module, etc. The design of the functional structure of the virtual learning system is shown in Figure 1. The functional structure of the virtual learning system is shown in Figure 1.
Virtual learning system functional structure design
Scene roaming function module. With the mouse, keyboard, voice commands and human body movements in the three-dimensional virtual environment roaming, Kinect as a peripheral so that the user has a sense of immersive experience, and improve the interactive performance of the system. Intelligent Q&A module. In the traditional teaching process, Q&A is an indispensable process for students to learn and master new knowledge, and Q&A can consolidate the new knowledge learned by students. The system realizes the intelligent Q&A scenario through the text matching algorithm model based on BERT text preprocessing, and students can complete the Q&A after class through this functional module, which is not restricted by time and space, and they can ask questions and get the answers they want from the knowledge base. Classroom quiz module. After the completion of the entire learning process, the classroom quiz module can test the learning effect of the students, so that the teacher understands the weak points of the students in the learning process, thus helping the teacher to better carry out the subsequent targeted teaching.
A corpus is a tool for storing, managing and analyzing natural language data, which can be used to study grammar, pragmatics, phonetics and many other fields [25]. The application of corpus in second language writing analysis is of great significance. First of all, corpus can be used to study the correspondence and conversion laws between different languages. By analyzing a large number of parallel corpora, researchers can discover the differences and similarities between the source language and the target language so as to better understand and solve the problems in writing. For example, researchers can use the corpus to study the correspondences in vocabulary, grammar and syntax, as well as the common errors and difficulties in second language writing. Secondly, the corpus can also be used to evaluate the quality of writing. By comparing it with existing high-quality expressions of second language writing, the researcher can assess the level of students’ second language writing and identify areas for improvement. In addition, corpora can be used to assess and optimize the effectiveness of second language writing learning. By comparing the content of writing with that of native speakers, researchers can assess the accuracy and fluency of second language learners’ writing and make suggestions for improvement.
Therefore, this paper designs a teaching experiment and uses corpus linguistic analysis to analyze the effectiveness of the application of virtual learning environment human-computer interaction system in second language writing teaching.
Ninety students majoring in Spanish in a university were selected as the subjects of the experiment, and they were randomly assigned to the experimental group and the control group, with 45 students in each group. Their second language writing level was tested by the pre-test, and it was found that there was no significant difference between the writing level of the students in the experimental group and the control group. The students in the experimental group received the virtual learning system designed in this paper for second language writing assistance teaching, while the control group followed the traditional writing teaching model for writing learning. The study lasted for 10 weeks, and the mid-test and post-test were conducted in the 5th and 10th weeks respectively to reflect the changes in students’ writing level.
Eighty-seven copies of the post-test students’ writing texts were collected, 90 copies of the students’ questionnaires were sent out, and 88 copies were recovered to organize the number of valid questionnaires, which was 84. All statistical data were processed using SPSS23.0 software to do independent samples t-test, correlation test, and statistical description of raw data.
Articulation consists of two parts. The first one is denotative articulation, i.e., the use of pronouns, comparisons, and personification to refer to semantic relations that have appeared in the previous text, which plays a significant role in the compactness and coherence of text construction. In this study, we choose two referential articulation indicators in CohMetrix software: mean argument overlap (MAO) and mean stem overlap (MSO). Thesis element overlap indicates the overlap of semantic consistency of nouns, pronouns and noun phrases in a sentence, and if a large number of personal pronouns and indefinite pronouns are utilized in an article, it may indicate a lack of articulation means and a need to strengthen articulation ability. Root overlap, on the other hand, indicates the overlap of words with the same root in the sentence. In addition, the use of conjunctions is also a reflection of writing articulation ability. The conjunctive features between the experimental group and the control group were mainly examined in terms of the proportion of full-text conjunctions (CNCALL), the proportion of causal conjunctions (CNCCaus), the proportion of transitive conjunctions (CNCADC), the proportion of chronological conjunctions (CNCTemp), and the proportion of progressive conjunctions (CNCAdd).
The posttest pairs of articulation ability in the experimental and control groups are shown in Table 1. The mean number of overlapping thesis elements in the experimental group is 0.527 (P=0.000<0.05), which is a significant increase of 0.235 compared with that in the control group, while the mean number of overlapping lexical roots in the experimental group is 0.546 (P=0.002<0.05), which is a significant increase of 0.197 compared with that of the control group. These two indexes reflect to a certain extent that there is a significant increase in the richness of the means of articulation of the textual organization ability of the experimental group The two indicators reflect, to some extent, the significant increase in the richness of text organization means in the experimental group.
The comparison of connection ability between two groups
| Variable | Experimental group | Control group | p |
|---|---|---|---|
| MAO | 0.527 (0.173) | 0.292 (0.016) | 0.000 |
| MSO | 0.546 (0.215) | 0.349 (0.114) | 0.002 |
| CNCALL (%) | 55.384 (3.392) | 53.333 (3.880) | 0.0024 |
| CNCCaus (%) | 10.603 (1.309) | 9.399 (2.206) | 0.0011 |
| CNCADC (%) | 7.940 (3.683) | 6.436 (2.961) | 0.028 |
| CNCTemp (%) | 28.109 (2.256) | 28.765 (1.973) | 0.257 |
| CNCAdd (%) | 12.044 (3.203) | 8.851 (2.728) | 0.000 |
It was also found that not all the five indicators of conjunctive use were significantly different between the two samples. The proportion of all conjunctions used in the experimental group was significantly higher than that of the control group by 2.051% at independent t-test p < 0.01 (p = 0.0024). The proportion of causal conjunctions was significantly higher than that of the control group by 1.204% (p=0.0011 < 0.05), and the proportion of transitive conjunctions was significantly higher than that of the control group (p=0.028 < 0.05), but the proportion of temporal conjunctions did not differ significantly between the experimental group and the control group (p=0.257 > 0.05).
Therefore, in most of the indicators, the experimental group’s referential articulation ability and conjunctive articulation ability were higher than that of the control group.
Semantic similarity between sentences is used to define the syntactic structure of the whole text, the similarity of linguistic structure. The higher the semantic similarity, the more repetitive the sentence is, the less syntactic change, and the appropriate semantic similarity is necessary for the coherence of the text articulation language. This dimension examined the semantic similarity between neighboring sentences, the semantic similarity of neighboring paragraphs, and the semantic similarity of all sentences. The analysis found that the independent samples t-test at p < 0.05 indicated that there were significant differences in all three metrics, all of which were manifested in the fact that the experimental group’s semantic similarity was significantly lower than that of the control group. Among them, the semantic similarity of adjacent sentences (LSASS1) was significantly lower than that of the control group by 0.03 units (p=0.002), the similarity of adjacent paragraphs (LSAPP1) was significantly lower than that of the control group by 0.03 units (p=0.001), and the similarity of all sentences (LSAGN) was significantly lower than that of the control group by 0.002 units (p=0.002).
It can be seen that the semantic similarity of the experimental group was significantly lower than that of the control group, i.e., the second language writing sentences of the students in the experimental group were less repetitive and more rich in syntactic variations.
The comparison of semantic similarity between two groups
This paper explores sentence complexity at three levels: sentence, phrase, and T-unit. Synthesizing simple or parallel sentences into complex sentences can make the structure of the sentence or article clearer, the language more vivid, and the logic more rigorous, so the improvement of text construction ability is inextricably linked to the ability of complex sentence construction.
Table 2 shows the post-test comparison of syntactic complexity between the experimental group and the control group, in which a total of 13 indicators in the above 3 dimensions are counted. The results of the independent samples t-test at p < 0.05 show that there is a significant difference between the two groups in all the indicators, and all the indicators of syntactic complexity of the experimental group are significantly higher than those of the control group, indicating that there is a significant improvement in the students’ ability to construct complex sentences through the application of the interactive system to the second language writing classroom.
The comparison of syntactic complexity between two groups
| Dimensions | Variable | Experimental group | Control group | p |
|---|---|---|---|---|
| Unit length | MLC | 9.403 (1.112) | 8.436 (0.747) | 0.025 |
| MLS | 13.785 (3.007) | 10.782 (1.429) | 0.001 | |
| MLT | 12.112 (2.768) | 9.923 (1.375) | 0.000 | |
| Sentence complexity | C/S | 1.428 (0.095) | 1.095 (0.238) | 0.012 |
| Dependent clause usage | C/T | 1.412 (0.084) | 1.206 (0.238) | 0.011 |
| CT/T | 0.180 (0.118) | 0.130 (0.056) | 0.028 | |
| DC/C | 0.299 (0.086) | 0.020 (0.189) | 0.038 | |
| DC/T | 0.295 (0.141) | 0.070 (0.032) | 0.025 | |
| Parallel structure usage | CP/C | 0.293 (0.049) | 0.152 (0.211) | 0.036 |
| CP/T | 0.292 (0.296) | 0.219 (0.140) | 0.038 | |
| Specific phrase structure | CN/C | 0.980 (0.066) | 0.807 (0.298) | 0.007 |
| CN/T | 0.981 (0.482) | 0.675 (0.402) | 0.000 | |
| VP/T | 1.358 (0.308) | 1.192 (0.044) | 0.042 |
At the sentence level, both the mean clause length (MLC), the mean sentence length (MLS) and the mean T-unit length (MLT) can reflect the writer’s ability to build sentences when writing, and the longer the sentence, the better the writer’s knowledge reserve and logical ability. The mean sentence length variable of the experimental group was significantly higher than that of the control group by 0.967 units, 3.003 units and 2.189 units (p=0.025, p=0.001, p=0.000), respectively. Also, the percentage of clauses (C/S) was significantly higher in the experimental group than in the control group by 0.333 units (p=0.012).
The use of subordinate clauses also reflects the ability to build complex sentences; the more subordinate clauses there are in a given range of sentences, the more complex and at the same time the more compact the sentence structure is, and the more semantically complete it is. The proportion of clauses in each T-unit, the proportion of complex T-units, the proportion of dependent clauses, and the number of dependent clauses in each T-unit all reflect the amount of dependent clauses used in the construction of a complete sentence, and the more complex the sentence structure, the more dependent clauses are used. All four indicators of dependent clause usage were significantly higher in the experimental group than in the control group at p < 0.05. Among them, the proportion of clauses per T-unit (C/T) was significantly higher than that of the control group by 0.206 units (p=0.011), the proportion of complex T-units (CT/T) was significantly higher than that of the control group by 0.05 units (p=0.028), the proportion of subordinate clauses per T-unit (DC/C) was significantly higher than that of the control group by 0.28 units (p=0.038), the proportion of subordinate clauses per T-unit (DC/ T) was significantly higher than that of the control group by 0.225 units (p=0.025).
The use of parallel structures in constructing complex sentences can increase sentence complexity while enhancing logic, and the greater the number of parallel phrases per clause (CP/C) versus the number of parallel phrases per T-unit (CP/T), the greater the author’s sentence construction ability. These two textual features were significantly higher in the experimental group than in the control group by 0.141 units (p=0.036<0.05) and 0.073 units (p=0.038<0.05), respectively.
In addition to this, the number of complex noun phrases in clauses and T-units and the number of verb phrases in T-units reflect the tightness and semantic coherence of sentences; the more complex noun phrases and verb phrases there are in a sentence, the more concise and tightly structured and semantically clear the sentence is to a certain extent. When p < 0.05, the three indicators of specific phrase structure in the experimental group are significantly different from the control group, in which the proportion of complex nouns per clause (CN/C) is significantly higher than that of the control group by 0.173 units (p = 0.007), the proportion of complex nouns per T-unit (CN/T) is significantly higher than that of the control group by 0.305 units (p = 0.000), and the proportion of verb phrases per T-unit ( VP/T) was significantly higher than the control group by 0.165 units (p=0.042).
In this section, the pre-, mid-, and post-test scores of students’ second language writing proficiency in the experimental group will be compared in order to reflect the changes in writing proficiency over the course of the experiment. Writing accuracy was measured by two dimensions: lexical accuracy and syntactic accuracy.
The changes in vocabulary and syntactic accuracy of second language writing at different time points are shown in Table 3. As can be seen from Table 3, vocabulary accuracy is positively correlated with the time of the experiment, and as the new teaching method continues to advance, the vocabulary accuracy of second language learners’ writing continues to increase and the vocabulary bias rate continues to decrease. Comparatively, the rate of increase in lexical accuracy is uneven, with a large and fast increase from the pre-test (M=0.9375) to the mid-test (M=0.9717) stage, and a small and slow increase from the mid-test (M=0.9717) to the post-test (M=0.9812) stage.
Changes in the vocabulary and syntax of the second language
| Dimensions | Time nodes | N | Mean | SD |
|---|---|---|---|---|
| Vocabulary accuracy | Pre-test | 45 | 0.9375 | 0.0244 |
| Medium test | 45 | 0.9717 | 0.0164 | |
| Post-test | 45 | 0.9812 | 0.0400 | |
| Syntax accuracy | Pre-test | 45 | 0.4840 | 0.2306 |
| Medium test | 45 | 0.6983 | 0.1563 | |
| Post-test | 45 | 0.7718 | 0.1319 |
Syntactic accuracy was positively correlated with language proficiency, with more and more error-free T-units and fewer and fewer writing biases produced in second language learners’ writing as the experiment progressed. Vocabulary accuracy increased at a similarly uneven rate, with a large and fast increase from the pre-test (M=0.4840) to the mid-test (M=0.6983) stage, and a small and slow increase from the mid-test (M=0.6983) to the post-test (M=0.7718) stage.
In order to further investigate the effects of the interactive system teaching experiment on L2 learners’ writing vocabulary and syntactic accuracy, and to test whether the above differences are statistically significant, a one-way ANOVA with post hoc multiple comparisons was conducted.
Table 4 shows the results of the statistical test for writing vocabulary accuracy. The results of the one-way ANOVA showed that there was a significant difference between the groups in vocabulary accuracy in the pre-test, mid-test, and post-test (F(2, 132) = 22.871, p < 0.01). To further analyze which level groups differed from each other, post hoc multiple comparisons revealed that there were significant differences between the pretest and the midtest, the pretest and the posttest, and the midtest and the posttest (p<0.05). Vocabulary accuracy was significantly greater in the mid-test than in the pre-test, and significantly greater in the post-test than in the mid-test and pre-test, but the difference between the pre-test and the mid-test was greater than the difference between the mid-test and the post-test. This suggests that syntactic accuracy is effective in distinguishing between second language learners’ writing levels and that lexical accuracy develops faster from the pretest to the midtest level stage and slows down the growth rate from the midtest to the posttest stage.
Statistical test results of writing vocabulary accuracy
| (a) Intergroup differences in vocabulary accuracy | |||||
|---|---|---|---|---|---|
| Sum of squares | df | Mean square | F | Sig. | |
| Intergroup | 0.027 | 2 | 0.013 | 22.871 | 0.000 |
| Within group | 0.068 | 132 | 0.001 | ||
| Toal | 0.095 | 134 | |||
| (b) Results of multiple comparisons of vocabulary accuracy | ||||
|---|---|---|---|---|
| Dependent variable | Independent variable Ⅰ (experimental time) | Independent variable Ⅱ (experimental time) | Difference of mean | Sig. |
| Vocabulary accuracy | Pre-test | Medium test | 0.0342 | 0.001 |
| Pre-test | Post-test | 0.0437 | 0.000 | |
| Medium test | Post-test | 0.0095 | 0.021 | |
Table 5 shows the results of the statistical tests for writing syntactic accuracy. The results of the one-way ANOVA showed that there was a significant between-group difference in syntactic accuracy between the pre-test, mid-test, and post-test (F(2, 132) =, p<0.01). To further analyze which level groups differed from each other, post hoc multiple comparisons revealed that there were significant differences between the pretest and the middle test, the pretest and the posttest, and the middle test and the posttest (p<0.05). The syntactic accuracy of the mid-test was significantly greater than that of the pre-test (p=0.002), and that of the post-test was significantly greater than that of the mid-test and the pre-test (p=0.005, p<0.001), but the difference between the pre-test and the mid-test was greater than that between the mid-test and the post-test. This indicates a faster rate of development from the pre-test to the mid-test stage and a slower rate of growth from the intermediate to the high level stage.
Statistical test results of writing syntax accuracy
| (a) Intergroup differences in syntax accuracy | |||||
|---|---|---|---|---|---|
| Sum of squares | df | Mean square | F | Sig. | |
| Intergroup | 1.551 | 2 | 0.728 | 25.832 | 0.000 |
| Within group | 3.278 | 132 | 0.038 | ||
| Toal | 4.829 | 134 | |||
| (b) Results of multiple comparisons of syntax accuracy | ||||
|---|---|---|---|---|
| Dependent variable | Independent variable Ⅰ (experimental time) | Independent variable Ⅱ (experimental time) | Difference of mean | Sig. |
| Vocabulary accuracy | Pre-test | Medium test | 0.2143 | 0.002 |
| Pre-test | Post-test | 0.2878 | 0.000 | |
| Medium test | Post-test | 0.0735 | 0.005 | |
A corpus linguistic analysis of the effectiveness of the application of the virtual learning system proposed in this paper is divided into four aspects. The analysis of articulation ability in second language writing found that the mean values of MAO and MSO in the experimental group were significantly reduced by 0.527 and 0.235, respectively, compared with the control group, and the proportion of all conjunctions used in the experimental group was significantly higher than that in the control group by 2.051%. It indicates that the students in the experimental group have better articulation skills in writing. The semantic similarity of neighboring sentences, the similarity of adjacent paragraphs and the similarity of all sentences in the experimental group were 0.06, 0.05 and 0.05, respectively, which were significantly lower than those of the control group. Among the 13 indicators reflecting sentence complexity, the experimental group performed better than the control group, indicating that the new teaching method helps to improve the syntactic complexity of second language writing. This paper also compares the writing accuracy of students in the experimental group in the pre-middle and post-tests, and finds that the pre-middle and post-test means of vocabulary accuracy are 0.9375, 0.9717 and 0.9812 respectively, and the pre-middle and post-test means of syntactic accuracy are 0.4840, 0.6983 and 0.7718 respectively. It can be seen that with the progress of the experiment, the accuracy of writing at each stage is higher than that of the previous stage and that the growth of accuracy in vocabulary and syntax is characterized by an imbalance.