Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 15, Article number: 15845 (2025) Cite this article
1992 Accesses
Metrics details
This study focuses on the application of virtual reality (VR) technology in English language teaching (ELT), and discusses the effect of combining VR with machine translation (MT) technology. VR technology is introduced into the study to provide innovative teaching methods for English teachers and create an immersive learning environment for students. Based on the latest development of deep learning (DL), a new MT model is proposed in this study, and it is successfully integrated with VR technology to optimize the quality of ELT. The experimental results show that the translation accuracy of the MT model designed in this study reaches 98.5%, the F1 score is stable at around 93%, and the semantic information recall rate is as high as 92%, all of which are better than the traditional model. In the preliminary test, the comparative experiment of 40 English majors further verified the effectiveness of the model in improving translation efficiency and quality. This study shows the great potential of the integration of VR and MT technology in ELT, and also proves its advantages by using experimental data, which provides technical support for ELT and provides reference for future practice.
The dawn of the era of big data has broadened people’s access to knowledge. In this context, the rise of artificial intelligence (AI) has brought great convenience to daily life and provided tangible technical support. VR, as a computer-generated immersive experience, combines human–computer interface, sensing technology and multimedia technology to create an interactive three-dimensional environment¹. With the rapid development of technology, VR has become an innovative tool in education, especially in English teaching. VR technology provides an immersive learning environment for students and promotes the interactivity of language learning. In order to achieve a wider and more efficient application, it is particularly important to combine VR with machine translation (MT) technology. This combination not only solves the language barrier problem, but also further enriches the teaching content and enhances students’ learning experience.
In English language teaching, educators face many challenges. Traditional teaching methods lack authentic contexts, making it difficult for students to apply what they have learned in the classroom to real-life situations. Meanwhile, low student participation, cultural differences, and limited educational resources are also common issues. Exploring new teaching methods and technological means is particularly important to address these challenges. The combination of DL and VR technology provides new possibilities for ELT. By constructing MT models based on DL and VR, an immersive learning environment can be created, providing real-time translation support, thereby stimulating students’ interest in learning and improving their language proficiency. When VR is integrated with DL-based MT applications, it creates a comprehensive language learning environment for students². Through VR, students can experience the culture and life of different countries, while simultaneously obtaining relevant language information in real-time through MT³. This immersive and interactive learning method not only stimulates students’ interest in learning but also significantly improves their learning efficiency. Although the research on VR and MT technology has made some progress at present, there is still a clear gap to effectively integrate these two technologies into ELT. On the one hand, most researches only focus on the single application of VR or MT, and lack of systematic exploration on their integration. On the other hand, existing research often ignores how VR and MT technologies can enhance students’ language learning experience in practical teaching, especially in cross-cultural communication scenarios. In addition, there are few studies on adjusting the combination of VR and MT technology to meet different learning needs according to individual differences of students.
The integration of AI into college English classes is of revolutionary significance. It helps teachers to develop more elaborate and cutting-edge teaching plans, thus improving the quality of ELT. Students also benefit from it, and have a deeper understanding of English concepts, increased participation and a completely new way of learning⁴. By visualizing abstract learning content and combining formal and informal learning methods, AI can also enhance learners’ sense of presence, intuition and attention. Language learning needs a favorable environment and atmosphere, and continuous practice is the key to making progress, which may not be easy for many China students to learn English⁵. The use of VR can create a virtual atmosphere for situational teaching, increase the proportion of students’ training in language learning, and promote the internalization of knowledge into ability. In addition, the application of VR can make English knowledge more three-dimensional and reduce students’ learning difficulty to some extent. With the development of DL, neural network is used to map the source language to the target language translation model, that is, the neural machine translation (NMT) model, which significantly improves the quality of machine translation. These models surpass the traditional statistical machine translation methods in performance and have become the mainstream methods in the industry and academia⁶. The main purpose of this study is to explore the innovative application of VR and MT technology in English teaching, and to solve the language barriers and interaction problems in current teaching. By combining these two technologies, we can improve the teaching quality and provide students with a better learning experience. The innovations include:
This paper integrates VR, MT technology and ELT, subverts the traditional teaching methods and realizes interdisciplinary integration and innovation. It provides students with a richer and more varied learning experience.
Using VR, students can experience the culture and life of different countries in simulated scenes and enhance their interest in English.
The introduction of MT technology, especially the MT model based on DL, has improved students’ English application ability by providing real-time, accurate and context-aware translation support. This goes beyond the traditional MT system, and uses the power of DL to better understand and generate natural language.
By combining VR and MT, teachers can provide personalized teaching according to the individual differences of students. This integration provides a more adaptable and interactive learning environment.
This article begins by introducing the background and significance of the research. It then explores the application of VR and MT technology in ELT. Following that, a detailed introduction to the DL-based MT model integrating VR is provided, and its effectiveness and feasibility are verified through experiments. Finally, in the conclusion section, the main findings and contributions of this article are summarized, and future research directions and suggestions are outlined.
Various languages exhibit distinct word orders, voices, and writing conventions, posing challenges for MT such as inverted word sequences, excessive vagueness, and limited translation precision. Consequently, improving MT quality has become a prominent research focus. The integration of VR/AR technology in language studies presents diverse case studies and rich content, covering vocabulary comprehension to sentence translation and even game design.
Zhang introduced a relative position encoding approach to address the limitations of absolute encoding in expressing positions within attention mechanisms, However, it doesn’t strengthen the attention module’s inherent positional encoding⁷. Wang et al. developed a language-learning VR game that helps learners with accurate word spelling⁸. Hui proposed a neural network-driven intelligent translation assessment technique to tackle semantic obscurity in MT outputs, enhancing both the dynamic assessment of MT quality and translation speed^9,10. Mallek et al. emphasized the importance of integrating educational theory and VR technology to enhance educational experience, which is consistent with the innovative teaching methods discussed in this study¹¹. Al Shloul et al. emphasized the role of activity-based learning in improving students’ performance. The integration of VR technology in ELT can further enhance this effect¹². Shah et al. pointed out that combining educational theory with VR can improve the teaching effect of engineering education and VR laboratory¹³.
Cai et al. developed learning software that transforms the campus into a virtual English realm, fostering student engagement with characters, objects, and media to hone English proficiency and campus understanding¹⁴. Wu et al. combined linguistic big data with language rules to further advance MT¹⁵. Wong et al. recommended a neural MT framework rooted in recurrent neural networks¹⁶, while Wang et al. proposed an attention-mechanism-based neural MT model¹⁷. Shahzad et al. summarized large-scale language models and their problems in learning environment. This is related to the integration of MT technology in ELT¹⁸.
Liang et al. proposed forth an enhanced neural MT approach leveraging generative adversarial networks to address the scarcity of training data for MT models¹⁹. This method’s uniqueness lies in its corpus expansion through data augmentation. Maimaiti built a Chinese-English MT model using kernel ridge regression technology to elevate translation quality, The approach views the translation process as a mapping between source and target language strings, utilizing these mappings for higher accuracy translations²⁰. Jia et al. devised bidirectional attention for language comprehension tasks, but each branch employs independent parameters, resulting in a bulky model that is unable to grasp local semantics effectively^21,22.
Although the above research has achieved important results in its field, there are limitations in the applicability of integrating VR and MT technologies into ELT. For example, Zhang’s research puts forward the relative position coding method, but it is not effective when dealing with complex language structure and context understanding. The language learning VR game developed by Wang et al. is helpful to vocabulary spelling, but it does not fully cultivate students’ overall language ability. Hui’s intelligent translation assessment technology has improved the quality and speed of translation, but it has not fully considered the real-time and interactivity in the actual teaching environment. In addition, most studies do not pay attention to students’ individual differences and learning needs, which are very critical in practical teaching and application of technology. On this basis, this paper puts forward an innovative method combining VR technology with DL-based MT technology. The goal of this improvement is to improve the accuracy of translation, and also to create a brand-new and immersive interactive learning experience for students. With the help of the unique advantages of VR technology and the powerful ability of DL, this paper hopes to bring changes to ELT, break through the limitations of traditional teaching and promote the development of ELT in a more efficient and interactive direction.
VR is a highly integrated entity that combines various disciplines such as contemporary simulation, human–computer interaction, and sensing. Its primary focus is on constructing a three-dimensional environment to offer users a more immersive experience. The application of VR to build a virtual learning environment and create problem-solving scenarios has significantly enhanced the learning enthusiasm and initiative of students²³. VR can concretize and visualize abstract language knowledge points by constructing a realistic and natural language environment, enabling students to access, understand, and apply language knowledge through various perceptual channels. Teachers can also use VR to set language difficulty levels in virtual scenes and configure corresponding brackets at different levels to assist students in making different choices based on their learning needs. Simultaneously, they can help students choose language difficulty levels according to their own knowledge level, facilitating varied levels of learning.
Traditional foreign language teaching classrooms cannot meet the specific “work scenario” teaching practice conditions required for specialized ELT, while VR can build virtual simulation learning environments to provide technical support for the transformation from public English to specialized ELT²⁴. Within the virtual learning environment, teachers offer students a platform and opportunities to freely express their opinions and fully showcase themselves²⁵. Teachers can transform flat language knowledge into three-dimensional real-time interactive scenes, thereby attracting students, improving teaching efficiency, expanding the scope of speech act teaching, and making teaching situations closer to real scenes. In a virtual language environment, students can communicate and interact with the created virtual characters and virtual teachers, making them feel that they are in a real language communication environment. This continuously exercises their English speaking ability and enhances the attractiveness of ELT activities, thereby improving their English learning efficiency.
MT is an emerging technology that integrates and utilizes computer technology and AI technology. The online translation systems or software commonly used on the internet currently belong to the MT system. The principle of MT is that users select example sentences with similar sentence structures to the source language from the corpus, and then convert the source language into the target language²⁶. MT is an intelligent product with high-tech content that uses computers to translate (convert) literature from one language to another. Its main function is to reduce communication barriers caused by differences, making people’s lives and work more convenient²⁷. The current online MT results still exhibit certain shortcomings, especially in the use of servers for the contrastive learning of data from different languages in the full-text scope. This approach aims to obtain grammar and text-related rules between languages, leading to low MT efficiency and accuracy. The prevailing MT technology at present is neural MT technology based on artificial neural networks (ANNs). Figure 1 illustrates the simplest schematic diagram of the ANN structure. A key feature of ANNs is their ability to learn from data, meaning they can automatically acquire optimal weight parameters based on training data.
ANN structure.
Applying VR technology and MT technology to English language teaching can create a more realistic, vivid and interactive learning environment. In this environment, students can have a dialogue with virtual characters, and MT technology can translate the dialogue content into students’ mother tongue or target language in real time, thus helping students better understand and master the English language. Theoretically, the combination of VR and MT has obvious advantages. First of all, the immersive learning environment provided by VR technology can stimulate students’ interest and motivation in learning, so that they can participate in language learning more actively. Secondly, the real-time translation function of MT technology can help students overcome language barriers and understand and master English faster. Finally, this combination can also promote students’ autonomous learning ability and let them explore and discover language rules independently in simulated real scenes.
Anwar et al. studied the subjective quality of experience (QoE) of 360-degree VR video. The research results can be used to improve the ELT experience based on VR²⁸. The QoE challenges and standard requirements in immersive media consumption discussed by Anwar et al. are also applicable to the VR environment in ELT²⁹. From the perspective of technical practice, some key problems need to be solved to realize the combination of VR and MT. The first is data integration, that is, how to effectively integrate interactive data in VR environment with MT model to achieve accurate real-time translation. Secondly, the interface design problem, how to design an intuitive and easy-to-use interface, so that students can easily learn and communicate in the VR environment. Finally, the performance optimization problem, how to ensure that the system can still maintain efficient and stable operation under a large number of interactive data and complex translation tasks.
In order to improve the accuracy of Chinese English MT, the first step is to convert the language to be translated into computer language, that is, to obtain the word vector of each word, extract the contextual information of the current word, and achieve accurate translation. Select word (w_{ji}) within the probability distribution (pleft( {w_{ji} |z_{ji} ,varphi } right)), where the polynomial probability distribution under (topicz_{ji}) is (p), and the Dirichlet distribution of parameter (alpha) is represented by (Dirleft( alpha right)). The probability density function can be obtained as follows:
In the formula, (theta = left( {theta_{1} ,theta_{2} , cdots ,theta_{k} } right),alpha = left( {alpha_{1} ,alpha_{2} , cdots ,alpha_{k} } right),Gamma left( cdot right)) and (Multinomialleft( theta right)) respectively represent the Gamma function and the polynomial distribution when the parameter is (theta). Therefore, the conversion formula for the probability density function is:
Analyzing the aforementioned process, it can be concluded that mining topic information in text first requires determining the proportion of different topics in each document within the document set. Based on probability, specific topics are then sampled from the topic distribution, and the polynomial distribution of the corresponding word list for that topic is sampled to obtain specific word results.
Due to the differences in word order and sentence length between Chinese and English, it is easy to produce word order during translation, which in turn leads to poor MT results. Therefore, a BiLSTM network is used to build the encoder. Using the BiLSTM network to encode the original translated sentences can effectively remember the node information of each memory unit, while avoiding the phenomenon of gradient vanishing during the encoding process of long sentences, captures more semantics, and alleviates the issue of word reordering.
Given the input sequence (x = left{ {x_{1} , cdots ,x_{n} } right}) on the source language side and the input sequence (y = left{ {y_{1} , cdots ,y_{m} } right}) on the target language side, where (x_{i}) and (y_{j}) represent the (i) th word on the source language side and the (j) th word on the target language side, respectively, and (n) and (m) represent the lengths of sequences (x) and (y). The main objective of the neural MT model is to maximize the maximum likelihood estimation function (Pleft( {y|x} right)). Among them, (Pleft( {y|x} right)) calculation is shown in Eq. (6).
Use (X) to represent the source language space and (Y) to represent the target language space. Take (X) as the input sample and use the MT model to map the input sample (X) to (Y). Using the conditional probability distribution (P_{theta } left( {y|x} right)) to represent the MT model, where (theta) represents the model parameters. There are parallel corpora with (N) sentence pairs in standard supervised training, represented by (B = left( {x^{left( n right)} ,y^{left( n right)} } right)) and (n = 1,2, cdots ,N). The MT model obtained by maximizing parallel corpus likelihood learning is as follows:
Maximizing the likelihood of monolingual data samples is a widely adopted technique for handling such data in machine learning, especially in the context of semi-supervised MT. This approach aims to optimize the utilization of available monolingual data, thereby enhancing the performance of the MT system.
The attention mechanism is commonly employed to calculate the similarity between source language words and target end words. Typically, various options exist for computing the alignment coefficients of attention mechanisms, including dot product, concatenation, etc. Drawing from the insights of deep self-attention networks, this article also utilizes a dot product approach, employing a query-key-value structure:
Among them, (Attentionleft( { cdot , cdot , cdot } right)) corresponds to the function of attention mechanism, and (q,k,v) represents the input query, key, and value, respectively.
In order to evaluate the translation performance of the model, this article uses size insensitive BLEU as the assessment indicator of the system’s translation quality. The BLEU script is an internationally recognized MT assessment script, that gauges the effectiveness of the translated text by comparing the similarity ratio of (n) tuples between the translated text and the reference translation, known as the (N – gram) matching rule. BLEU has incorporated corrections in its calculation, and the correction formula is as follows:
Among them, (Countleft( {n – gram} right)) is the number of occurrences of the (n) phrase in the MT translation, (Count_{clip} left( {n – gram} right)) is the minimum value taken between the occurrence of the (n) phrase in the (Countleft( {n – gram} right)) and reference translations, (C) is the translation translation, (C^{prime}) is the reference translation, and in calculation, (n – gram) is generally set to (4 – gram) at the highest.
In DL field, Convolutional Neural Network (CNN) shows excellent feature extraction ability. CNN plays an important role in image recognition, video analysis, drug discovery, board games and many other applications³⁰. It is worth noting that specially designed CNN variants also appear in the field of natural language processing (NLP). When dealing with sequence data or tasks that need context understanding, traditional neural networks face challenges in effectively capturing long-distance dependencies due to structural limitations.
In this case, the recurrent neural network (RNN) intervenes to establish continuity with the previous text, effectively integrate the context information, and enable the network to predict the following text according to the previous details³¹. When combined with VR, this ability can help to better understand and predict the user interaction in the virtual environment and enhance the immersive language learning experience³². RNN has encountered a “long-term dependence problem” in practical application: with the increase of network layers or the extension of sequence length, the gradient gradually decreases in the process of back propagation, which leads to the slow update of network weights. Coping with this challenge is very important for optimizing the integration of RNN and VR in advanced language teaching applications.
In order to overcome this limitation, Long-term and Short-term Memory Network (LSTM) was introduced as an improved version of RNN. On the basis of RNN, LSTM adds three key gating mechanisms (input gate, forgetting gate and output gate) and a cell state. These structures ensure the persistence and selective forgetting of information in the hidden layer, so that neurons can make more effective use of previous information. Figure 2 shows the structural differences between CNN and LSTM models. Among them, CNN mainly extracts local features through convolution layer and pooling layer, while LSTM maintains the continuity of sequence information through gating mechanism.
CNN and LSTM model structures.
In our research, the neural network architecture adopted is DL architecture based on self-attention mechanism-Transformer model. This model is especially suitable for dealing with sequence-to-sequence tasks, such as machine translation. The encoder is responsible for encoding the input sequence into a set of representation vectors rich in context information. These vectors capture the global dependencies in the sequence through the self-attention mechanism. The decoder is based on these representation vectors and the context of the generated output sequence.
The training data mainly comes from the open bilingual corpus, which contains a lot of parallel texts, that is, the corresponding sentences in the source language and the target language. In order to ensure the effect of model training, the data is strictly preprocessed: the cleaning process includes removing repeated sentences, non-standard characters, noise data and irrelevant marks. Word segmentation involves decomposing sentences into word or sub-word units to meet the processing needs of the model. The alignment step ensures that the sentences in the source language and the target language correspond semantically correctly.
Back propagation algorithm and gradient descent optimization strategy are adopted in model training. The model iterates over the training data and gradually learns the mapping relationship from the source language to the target language. Adam optimizer is used to accelerate the convergence process, and dropout technology is introduced to prevent over-fitting.
Taking one-dimensional convolution as an example, assuming the input is a one-dimensional array, its convolution kernel must also be a one-dimensional array. The one-dimensional continuous convolution formula is shown in (1), and the one-dimensional discrete convolution formula is shown in (2); (xleft( k right)) is the input of the convolutional layer mentioned earlier, (hleft( k right)) is the convolutional kernel, and (yleft( k right)) is the output of the convolutional layer:
In the formula, (hleft( k right),xleft( k right)) can be regarded as a variable function for performing convolution operations, (p) is an integral variable, (k) represents the number of digits moved, and (*) represents the convolution operation. Taking one-dimensional discrete convolution as an example, (P) is the upper limit of the convolution.
LSTM contains a storage unit that will update the information it stores as needed. The LSTM unit at time (t) consists of a series of transformation functions, including input function (left( {i_{t} } right)), forget function (left( {f_{t} } right)), output function (left( {o_{t} } right)), memory function (left( {c_{t} } right)), and hidden state function (left( {h_{t} } right)). The expressions for each function are as follows.
Among them, (Theta) represents bitwise multiplication, (b) represents offset vector, and (z_{t}) represents state update function.
In this study, five corpora, namely NUCLE, Supervised, CoNLL, JFLEG and Lang-8, were selected as test data sets. Among them, NUCLE contains 1200 articles, Supervised has 5000 articles, CoNLL has 200 articles, JFLEG has 2000 articles, and Lang-8 has 8000 articles. Their total sentence pairs are 50000, 80000, 3000, 510000 and 210885 respectively. These articles cover six fields: sports, military, economy, education, science and technology and society, which ensures the diversity and universality of data.
In the preprocessing step, firstly, news source details, special symbols and repeated sentences in sentence titles are removed, so as to reduce corpus noise. Then, features are extracted from the preprocessed corpus, and new dependency syntax and part-of-speech tagging techniques are used. To do this, mark the word before the comma as “comma” and the rest as “NULL”. Then, all commas are removed, and all features and labels are converted into the format needed for model training. CRF +  + Toolkit is used to train the Conditional Random Field (CRF) model to further optimize the feature representation.
Software environment: Based on Python programming language, the experiment builds and trains MT model with the help of deep learning frameworks such as TensorFlow and PyTorch. At the same time, NLTK and spaCy are used for text preprocessing and feature extraction.
Hardware configuration: The experiment is run on a server equipped with NVIDIA GTX 1080 Ti GPU, Intel i7 CPU and 32 GB RAM, which ensures high efficiency and speed of model training.
VR environment integration: In this study, the Oculus Rift VR platform and the corresponding controller are used to provide an immersive interactive experience. Customize and develop VR scenes related to teaching content through Unity 3D, such as simulating classrooms and laboratories. After many user tests, the interactivity and user experience are optimized. Under certain VR conditions, the real-time translation accuracy of MT model is tested, and its response speed and translation quality are assessed by simulating real dialogue scenes.
Forty third-year English majors, aged between 19 and 23, with an average age of 21.2, were recruited in the experiment. All participants have passed CET-4 and have no previous experience in VR. Participants were randomly divided into experimental group (20 people) and control group (20 people) according to their language proficiency test scores. The sex ratio of the two groups (experimental group: 8 males and 12 females; Control group: 7 males and 13 females) and English proficiency (the average score of CET-4 in experimental group is 523; There was no significant difference (p > 0.05) in the control group (518 points).
Students in the experimental group wear VR equipment, enter a customized VR learning environment, interact with virtual characters, and realize real-time translation and communication with the help of MT model. The control group used traditional teaching methods, including classroom study, homework and paper textbooks. The experiment lasted for 4 weeks. The experimental group had VR interaction three times a week (45 min each time), while the control group had traditional classroom learning for the same duration.
Students’ learning data and feedback were collected during the experiment. The results of statistical analysis show that the students in the experimental group have improved in participation and academic performance to varying degrees. This shows that the combination of deep learning and VR technology not only improves students’ interest and learning motivation, but also enhances their language ability. Table 1 presents the key comparative information between the experimental group and the control group in detail, including participants’ characteristics, teaching tools, teaching scenarios, intervention duration and assessment methods.
Figure 3 shows the comparison of translation accuracy between the model in this study and the traditional model in various corpora. These corpora represent various language styles, fields and themes, which ensure a comprehensive assessment of the model performance. The results show that the translation accuracy of this model is over 98%, which is better than the traditional model with an accuracy lower than 97%. The high accuracy achieved by our research model can be attributed to the optimization of its deep learning architecture and the customized training for specific corpora. This accuracy is especially important for professional translation scenarios, such as legal and medical documents.
Comparison of translation accuracy between different models.
Figure 4 shows the F1 score changes of the two models during the iterative training. The results show that with the increase of iteration times, the F1 scores of the two models are on the rise, but the F1 scores of the model designed in this study can always be higher than 90% and stable at around 93%, while the F1 scores of the traditional model never exceed 0.9. This research model shows higher performance stability in the iterative training process, and its F1 score keeps at a high level all the time. Although the F1 scores of both models increase with the increase of iteration times, this research model can reach a high performance level in a few iterations. This helps to shorten the training time.
Comparison of F1 values.
Figure 5 shows the comparison of semantic information recall rates between the two models when translating sentences in different corpora. The results show that the semantic information recall rate of this research model is over 90%, which is significantly higher than that of the traditional model. The high recall rate of semantic information shows that this research model can better understand and retain the semantic information of the original sentence in the translation process. The advantages of this research model will be particularly obvious in scenes that need to convey semantic information accurately, such as international business and diplomacy.
Comparison of recall rates.
In order to verify the effectiveness of this model in English teaching, 40 third-year English majors were tested. These students, ranging in age from 19 to 23, come from different language backgrounds and use different mother tongues. They all have a certain level of English, because they have taken courses for English majors. They were equally divided into group A and group B, with 20 students in each group. In each group, students numbered 1–10 are instructed to use the method described in this paper for translation, while students numbered 11–20 do not use this method. The test document is a 578-word English technical document. The final result is assessed based on two dimensions: translation quality and translation efficiency. BLEU score is used to measure translation quality and translation time is used to measure translation efficiency. As shown in Table 2, compared with Group B, the translation efficiency and quality of Group A are obviously improved, which shows that the method designed in this paper can effectively improve students’ translation skills.
Through in-depth analysis of the comparison results of the feature matching degree of sentence topic words between the two models in Fig. 6 on different corpora, we can clearly see that when using the proposed model for translation, the feature matching degree of sentence topic words is as high as 90% or more. In Table 3, this highlights the excellent ability of the model in capturing and preserving the core features of the original topic words. In contrast, traditional models have a feature matching degree of less than 85% in the translation process, indicating certain limitations in handling complex language structures and semantic relationships. This significant difference not only proves the superiority of the model in translation tasks, but also provides strong support for its widespread application in high-precision translation demand scenarios. Whether in professional fields such as legal documents and medical literature, or in important occasions of cross-cultural communication and information transmission such as international business and diplomacy, the model proposed in this article is expected to significantly improve the quality and accuracy of translation, and contribute significantly to the progress and development of related fields.
Comparison of feature matching degree.
It is not difficult to find from Table 4 that the model proposed in this paper is superior to other advanced models in translation accuracy, F1 score and semantic information recall. This shows that this model can better capture and retain the semantic information of sentences while maintaining the accuracy of translation.
It is not difficult to find from Table 4 that the model proposed in this paper is significantly higher than other models in feature matching, which shows that it can better retain the core features of the original sentence in the translation process. The translation speed and memory occupation of this model are also at a high level, but considering its superior performance, these are acceptable.
Table 5 shows the learning progress of students with different learning backgrounds in VR + MT environment. The results show that intermediate learners have achieved the greatest learning progress and the highest learning efficiency in VR + MT environment.
Intermediate learners may be more adapted to VR + MT, a new learning method, and can make full use of its advantages to improve their English. VR + MT environment provides personalized learning paths for students with different learning backgrounds, which helps to meet the different needs of students. In the future, we can further study how to optimize the VR + MT environment to better adapt to students with different learning backgrounds.
The experimental results of this study strongly prove that it is feasible to integrate VR and MT technology into ELT. These achievements reflect the possibility of technology integration, and also show significant educational advantages. The machine translation model designed in this study has a translation accuracy as high as 98.5%, far exceeding the traditional model. Such a high accuracy is crucial to ensure the completeness and correctness of the translated content. The F1 score is stable at about 93%, which further proves that the model is robust and reliable. The high F1 score shows that the model performs well in accuracy and recall, which is an important index to evaluate the overall quality of machine translation. The semantic information recall rate of 92% also shows that the model can accurately capture and retain the core meaning of the original text.
A comparative experiment involving 40 English majors further verifies the effectiveness of the model in improving translation efficiency and quality. Students who use VR + MT method get higher BLEU score and shorter translation time than those who don’t use this method. The average translation time of group A students using this tool is 0:48:39, and the BLEU score is 78.83. The average translation time of group B students who also used this tool was 0:51:18, and the BLEU score was 81.36. These data clearly show that VR + MT method is helpful to achieve faster and more accurate translation, which is very important to improve students’ language ability.
In addition, using the model proposed in this study, the feature matching degree of sentence subject words is over 90%. Such a high feature matching rate highlights that the model can accurately capture and retain the core features of the original sentence, which is of great significance to maintaining the contextual coherence of the translated text. In contrast, the feature matching degree of the traditional model is less than 85%, which shows that it has limitations in dealing with complex language structures and semantic relationships.
This study puts forward several strategies on how to effectively integrate VR and MT into English teaching practice. On the one hand, teachers can use VR to create an immersive learning environment, simulate real scenes, and improve students’ participation and learning motivation. VR enables students to interact with virtual characters, experience different cultural situations and greatly enrich the language learning experience. On the other hand, the integration of machine translation technology can provide real-time translation support, help students overcome language barriers and communicate more effectively, which is especially beneficial to students who have difficulties in learning a second language and communicating in real time.
Research shows that intermediate learners may benefit the most from VR + MT environment. Intermediate learners show the greatest learning progress and the highest learning efficiency. This discovery means that VR and MT technology may be effective in narrowing the gap between junior and advanced learners, and can provide customized learning experiences to meet their specific needs.
Theoretically, this study adds new contents to the related literature on the application of VR and MT in the field of education. By demonstrating the feasibility and effectiveness of integrating technology into English teaching, it lays a foundation for future research in this field. From a practical point of view, the research results provide valuable insights for educators and policy makers who expect to improve the quality and efficiency of language teaching. The VR + MT method proposed in this study can be used as an example to develop innovative teaching methods and improve students’ learning results with the help of cutting-edge technology.
However, while enjoying the convenience brought by VR and MT technologies, we must also face up to data privacy and security issues. These technologies will involve the collection and processing of a large number of sensitive user data in the process of use. Once these data are leaked, abused or attacked by security, it may bring serious losses and troubles to users.
In order to meet these challenges, a series of strict security measures must be taken. First, the collected user data should be encrypted. Secondly, strict access control mechanism should be established. At the same time, it is necessary to establish and improve the relevant legal and policy framework, and clarify the use norms and accountability mechanisms of technology.
User education is also an indispensable part. Educators should use various ways to raise users’ awareness of data protection, so that they can consciously protect their privacy in the process of using technology.
Although this study has achieved some promising results, there are still several limitations. The model has limitations in storage space, test data and reasoning time, and needs to be further optimized. At the same time, there are relatively few experimental data, and more extensive tests are needed to fully verify the performance of the model.
Future work can focus on the following directions. On the one hand, it plans to integrate more advanced deep learning technology, expand bilingual corpus and improve the accuracy of machine translation. On the other hand, we will study ways to optimize the VR + MT environment to better adapt to students with different learning backgrounds. In addition, it is also intended to explore the application of this integrated technology in teaching scenarios of languages other than English.
This study explores the application of VR in ELT and shows its great potential in teaching reform. The introduction of VR has built an immersive English learning environment for college students and promoted students’ free learning experience. This study demonstrates the unique advantages of VR technology in promoting ELT reform. VR also shows great potential in improving the overall teaching effect and reducing the teaching cost.
The results show that the quality of college English translation teaching has been improved by organically combining human intervention with machine advantages. The innovative MT model based on DL proposed in this study achieves 98.5% translation accuracy, and the F1 score is stable at around 93%, and it shows high stability in practical application. This trend indicates that the informationization of English translation teaching in colleges and universities will steadily move towards a higher level.
However, this study also has some limitations. Limited by storage space, test data and reasoning time, the proposed model still needs further optimization. Future research can focus on improving the accuracy of MT and expanding and enhancing bilingual corpus in various ways.
The datasets used and/or analyzed during the current study are available from the corresponding author Lynette P. Rue on reasonable request via e-mail philipwuhu@163.com.
Deep learning
Machine translation
Virtual reality
English language teaching
Quality of experience
Zhang, Y. application of intelligent virtual reality technology in college art creation and design teaching. J. Internet Technol. 6, 22 (2021).
Google Scholar 
Gong, W. An innovative english teaching system based on computer aided technology and corpus management. Int. J. Emerg. Technol. Learn. (IJET) 14(14), 69 (2019).
Article  Google Scholar 
Shen, H. & Chen, X. Virtual reality-based internet plus smart classroom. J. Internet Technol. 2, 23 (2022).
Google Scholar 
Qian, J. Research on artificial intelligence technology of virtual reality teaching method in digital media art creation. J. Internet Technol. 1, 23 (2022).
Google Scholar 
Pan, W. English machine translation model based on an improved self-attention technology. Sci. Programm. 14, 2021 (2021).
Google Scholar 
Lv, N. & Gong, J. The application of virtual reality technology in the efficiency optimisation of students’ online interactive learning. Int. J. Contin. Eng. Edu. Life-long Learn. 1, 32 (2022).
Google Scholar 
Zhang, X. Re-conceptualizing techno-linguistics: using technology for english language teaching?. Int. J. Emerg. Technol. Learn. (iJET) 14(17), 162 (2019).
Article  Google Scholar 
Wang, L. & Huang, R. English teaching platform based on online education technology and licensure and master’s program technology. Int. J. Emerg. Technol. Learn. (iJET) 14(16), 154 (2019).
Article  Google Scholar 
Hui, Q. Design and application of the foreign language electronic teaching diaries based on the virtual reality technology. Basic Clin. Pharmacol. Toxicol. S1, 125 (2019).
Google Scholar 
Lee, H., Lee, S. H., Nan, D. & Kim, J. H. Predicting user satisfaction of mobile healthcare services using machine learning: Confronting the COVID-19 pandemic. J. Organ. End User Comput. (JOEUC) 34(6), 1–17 (2022).
Article  Google Scholar 
Mallek, F. et al. A review on cultivating effective learning: Synthesizing educational theories and virtual reality for enhanced educational experiences. PeerJ Comput. Sci. 10, e2000 (2024).
Article  PubMed  PubMed Central  Google Scholar 
Al Shloul, T., Mazhar, T. Iqbal, M., et al. Role of activity-based learning and ChatGPT on students’ performance in education. Computers and Education: Artificial Intelligence, 100219 (2024).
Shah, S. F. A. et al. Integrating educational theories with virtual reality: Enhancing engineering education and VR laboratories. Soc. Sci. Human. Open 10, 101207 (2024).
Google Scholar 
Cai, Y. & Zhao, T. Performance analysis of distance teaching classroom based on machine learning and virtual reality. J. Intell. Fuzzy Syst. 40(2), 2157–2167 (2021).
Google Scholar 
Wu, J., Ding, Z. & Yu, H. Application of AI technology in english writing teaching on coronavirus paper in multimedia environment. Basic Clin. Pharmacol. Toxicol. S1, 127 (2020).
Google Scholar 
Wong, C. H. S., Tsang, K. C. K. & Chiu, W. K. Using augmented reality as a powerful and innovative technology to increase enthusiasm and enhance student learning in higher education chemistry courses. J. Chem. Educ. 11, 98 (2021).
Google Scholar 
Wang, X. Building a parallel corpus for english translation teaching based on computer-aided translation software. Comput. Aided Des. Appl. 18(S4), 175–185 (2021).
Article  Google Scholar 
Shahzad, T. et al. A comprehensive review of large language models: Issues and solutions in learning environments. Discover Sustain. 6(1), 27 (2025).
Article  Google Scholar 
Liang, X. & Pang, J. An innovative english teaching mode based on massive open online course and google collaboration platform. Int. J. Emerg. Technol. Learn. (iJET) 14(15), 182 (2019).
Article  Google Scholar 
Maimaiti, K. Modelling and analysis of innovative path of English teaching mode under the background of big data. Int. J. Continu. Eng. Edu. Life-long Learn. 29(4), 306–320 (2019).
Google Scholar 
Jia, Y., Carl, M. & Wang, X. Post-editing neural machine translation versus phrase-based machine translation for English-Chinese. Mach. Transl. 33(1–2), 1–21 (2019).
Google Scholar 
Wu, J. & Zhang, K. Machine learning algorithms for big data applications with policy implementation. J. Organ. End User Comput. (JOEUC) 34(3), 1–13 (2022).
Google Scholar 
Wang, C., Cai, S. J. & Shi, B. X. Visual topic semantic enhanced machine translation for multi-modal data efficiency. J. Comput. Sci. Technol. 38(6), 1223 (2023).
Article  Google Scholar 
Nonaka, K., Yamanouchi, K. & Tomohiro, I. A compression-based multiple subword segmentation for neural machine translation. Electronics 11(7), 1014 (2022).
Article  Google Scholar 
Chatzikoumi, E. How to evaluate machine translation: A review of automated and human metrics. Nat. Lang. Eng. 26(2), 1–25 (2019).
Google Scholar 
Liu, H. I. & Chen, W. L. Re-transformer: A self-attention based model for machine translation. Procedia Comput. Sci. 189(8), 3–10 (2021).
Article  Google Scholar 
Bhadwal, N., Agrawal, P. & Madaan, V. A Machine Translation system from Hindi to Sanskrit language using rule based approach. Scalable Comput. 21(3), 543–554 (2020).
Google Scholar 
Anwar, M. S. et al. A moving metaverse: QoE challenges and standards requirements for immersive media consumption in autonomous vehicles. Appl. Soft Comput. 159, 111577 (2024).
Article  Google Scholar 
Anwar, M. S. et al. Subjective QoE of 360-degree virtual reality videos and machine learning predictions. IEEE Access 8, 148084–148099 (2020).
Article  Google Scholar 
Li, Y., Li, X. & Yang, Y. A diverse data augmentation strategy for low-resource neural machine translation. Information (Switzerland) 11(5), 255 (2020).
CAS  Google Scholar 
Zhang, W., Li, X. & Yang, Y. Keeping models consistent between pretraining and translation for low-resource neural machine translation. Future Internet 12(12), 215 (2020).
Article  Google Scholar 
Park, C., Yang, Y. & Park, K. Decoding strategies for improving low-resource machine translation. Electronics 9(10), 1562 (2020).
Article  Google Scholar 
Download references
Anhui Province Philosophy and Social Science Planning Youth Project: Research on the Inheritance and Translation of Anhui Folk Literature Intangible Cultural Heritage, AHSKQ2020D190.
School of Foreign Studies, Anhui Polytechnic University, Wuhu, 241000, China
Tao Su
Education Program Supervision in Mathematics in the Department of Education, Trece Martirez City, 4109, Cavite, Philippines
Lynette P. Rue
You can also search for this author inPubMed Google Scholar
You can also search for this author inPubMed Google Scholar
Tao Su: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing—original draft preparation, funding acquisition. Lynette P. Rue: writing—review and editing, visualization, supervision, project administration.
Correspondence to Lynette P. Rue.
The authors declare no competing interests.
The studies involving human participants were reviewed and approved by School of Foreign Studies, Anhui Polytechnic University Ethics Committee (Approval Number: 2022.6520002). The participants provided their written informed consent to participate in this study. All methods were performed in accordance with relevant guidelines and regulations.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Su, T., Rue, L.P. The analysis of English language teaching with machine translation based on virtual reality technology. Sci Rep 15, 15845 (2025). https://doi.org/10.1038/s41598-025-00592-4
Download citation
Received: 03 October 2024
Accepted: 29 April 2025
Published: 06 May 2025
DOI: https://doi.org/10.1038/s41598-025-00592-4
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Scientific Reports (Sci Rep)
ISSN 2045-2322 (online)
© 2025 Springer Nature Limited
Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

source

Private schools are a small part of the total U.S. K-12 school market, but with the expansion of private school choice programs in many states, the sector is on track to grow in enrollment and value to education companies.
The most recent federal data available, from 2021, found 29,730 K-12 private schools in the U.S. In the 2019-20 school year, over 61 percent of private schools were prekindergarten, elementary, and middle schools, compared with 71 percent of public schools.
U.S. private schools enrolled 5.47 million students in 2021, according to the National Center for Education Statistics. That’s 10 percent of all K-12 students.
In recent decades, the share of students in private schools has decreased. From 1995 to 2001, between 11.4 and 11.7 percent of all students attended private schools. That fell, starting in the early 2000s, and in recent years the percentage has flattened out at around 10 percent on average. Federal data lag several years behind year-to-year changes in enrollment.
In recent years, school choice programs have expanded in many states, potentially leading to a significant increase in the percentage of students served by private schools that aren’t yet reflected in federal estimates.
Private schools enroll the largest share of students in the northeast (13 percent) and the lowest in the west (8.3 percent), according to 2021 federal data.

Private schools are generally smaller than public schools, with over 44 percent of them enrolling fewer than 300 students, according to 2021 data. More than 60 percent of public schools enroll more than 500 students.

Almost 75 percent of private schools in the U.S. are religious, and about half of those religious schools were Catholic in the 2020-21 school year, the most recent year with federal data available.

However, since 2009, Catholic school enrollment has decreased from 39.4 percent to 33.2 percent of all private school enrollment. Nonsectarian (non-religious) schools now enroll 25.5 percent of private school students, up from 22.8 percent in 2009.

Average U.S. private school tuition was $12,790 in the 2020-21 school year, according to federal data.
Tuition varies widely between schools of different religious backgrounds. Catholic schools charged the least tuition, on average, at $9,720 per year in 2020-21, compared with nonsectarian schools, which charge $19,590, and other religious schools, which charge $10,910.
By comparison, in public schools in fiscal year 2021, the U.S. Census Bureau reported that average per-pupil spending was $14,347.
About half of Catholic schools are administered by a diocese rather than an individual parish, and the diocese may hold more centralized purchasing power (more akin to a school district) than with individual, independent private schools.
Some states have programs, open to all families, that allow families to use public funds for private school tuition or, in some cases, other education-related expenses.
Students in these private school choice programs represent a small fraction of the nation’s total K-12 enrollment, almost 2 percent in the 2023-24 school year, but the numbers signing up for new state programs have sometimes exceeded projections.
And, in recent years the number of private school choice programs have grown rapidly. In 2023, lawmakers in 42 states introduced bills to establish or dramatically expand school choice programs.

Private school choice programs differ widely in how much funding they provide, how they’re set up, where the funding comes from, and how the money can be used. Some states have several different kinds of programs.
See this glossary from Education Week to learn more:
Education Savings Account (ESA)
Education savings accounts provide public per-pupil funds—often a percentage of per-student state funding—to families with children who don’t attend public schools that they can use to pay for private school tuition or other education expenses, such as tutoring and homeschooling supplies. Some states restrict ESAs or specific ESA programs within the state to students with disabilities, students attending schools with poor performance, and/or students from low-income families. Recently, more states have begun adopting universal ESAs, which all families can access regardless of income, disability status, or any other qualifying factor. ESA funds are generally given directly to families, often in the form of debit cards with restrictions on how the money can be spent. While ESAs and vouchers are often used interchangeably, what sets ESAs apart from vouchers are that they can be used for a wide array of education expenses, not just private school tuition. (See EdWeek’s 2023 explainer on ESAs.)
Voucher
School vouchers describe public funds that families can use at private schools of their choice, including those that are religious, to subsidize the cost of student tuition. Many vouchers are restricted to students with disabilities, students attending poor-performing schools, and students from low-income families, but some states have vouchers that are available to any student. (See EdWeek’s 2017 explainer on vouchers.)
Tax-Credit Scholarship
Tax-credit scholarship programs provide scholarships to families that they can use at private schools of their choice, including those that are religious. The scholarships most commonly come from state-authorized nonprofit organizations, which issue the scholarships out of donations that they receive from businesses or individual taxpayers who receive tax credits for those donations. Eligibility can be limited based on family income, disability status, or other factors, or it can be universal. (See EdWeek’s 2024 explainer on tax-credit scholarships.)
Tax-Credit Education Savings Account
Tax-Credit ESAs are a less common form of ESA through which families receive a designated, per-pupil amount from a state-authorized nonprofit organization that administers the account. Families can use the funds to cover any educational expense, including private school tuition, tutoring, or homeschooling costs. Businesses and individual taxpayers receive tax credits for donations to those nonprofit organizations. (See EdWeek’s 2024 explainer on tax-credit education savings accounts.)
Direct Tax Credit
Some states offer tax credits directly to parents to defray the cost of private school tuition or home-school expenses. Such credits are still among the rarer forms of private school choice, but they have become gradually more common as Oklahoma and Idaho most recently have adopted new tax-credit programs. States’ existing tax-credit programs have varying levels of generosity. Some states offer tax deductions instead of direct credits to defray private-school tuition costs. EdWeek doesn’t track these deduction programs, as they tend to cover a smaller portion of private-school costs than other forms of private-school choice. (See EdWeek’s 2024 explainer on states’ use of tax credits to fund private school choice.)

Universal school choice
Private school choice programs that are open to all families regardless of disability status, income, location, or public school performance. Universal policies have become more popular in recent years.
School choice
State and, to a lesser extent, federal policies and programs that allow families to send students to schools that they wouldn’t be assigned to attend in the traditional public school system. This can include charter schools; magnet schools; traditional public schools outside of a family’s assigned school zone, district, or town; homeschooling; and private schools, including those that are religiously affiliated.
Private school choice
Policies and programs that direct state and other public funds to private schools, including religious options, where families can choose to enroll their children.
Public school choice
Policies and programs that allow families to attend public schools other than the school to which a child would normally be assigned. These schools include charter schools, magnet schools, as well as traditional public schools where families proactively decide to enroll their children. This EdWeek tracker is focused on private school choice and does not include data on public school choice programs.
Magnet schools
Public schools with a specific focus, such as STEM, performing arts, or career and technical education, that are free to attend and open to all students in a district. Some magnet schools are also open to students outside of a designated district or state and require students to apply to attend.
Charter schools
Schools that receive public funding but typically operate independently of local school districts, with private nonprofits most commonly running them and less often for-profit entities. Districts or state authorizing bodies create contracts, or “charters,” with organizations that want to open charter schools, often for a designated period of time. Charters are tuition-free and are often open to all students in a district or an even broader metro area. However, they tend to have caps on enrollment and decide enrollment based on a lottery system. Because charter schools are a form of public school choice, they are not included in EdWeek’s private school choice tracker. (See EdWeek’s 2018 explainer on charter schools.)
Inter- and intra-district choice
Policies and programs that allow students to attend public schools other than those to which they would normally be assigned. Those schools can be located in the student’s home district but outside of their traditional school zone (intra-district choice) or outside of their home district (inter-district choice). These policies are sometimes referred to as open enrollment.

The money available per student in states’ different private school choice programs varies widely. Oklahoma has the most generous tax-credit scholarship program, offering up to $7,500 to families whose children attend private school. Most tax-credit scholarship programs give between $1,500 to $2,500 per student. Education savings accounts, however, tend to offer families similar amounts of money to the per-pupil amount the state spends on a public school student.
According to EdChoice, a nonprofit that advocates for private school choice, Florida has the most students on education savings accounts — 136,000. Indiana has over 69,000 students in their voucher program, the country’s largest. The largest tax-credit scholarship program is in Pennsylvania, where over 54,000 students are in the program.

The different things parents can pay for using education savings accounts vary state to state, but many states allow expenses for private school tuition, therapies like speech-language or behavioral therapy, fees for tests, tutoring, textbooks, computer hardware, and uniforms.
Arizona, which created the first ESA program in 2011, also allows other expenses for materials like books, educational discs, backpacks, furniture like desks and chairs, and tickets for educational outings, like to museums and plays. A few states allow expenses like summer and after-school programs and workforce credentials.
According to EdWeek Market Brief reporting in 2023, the biggest needs for private schools are teacher recruitment, financial planning, and students’ social-emotional learning.
Myra McGovern, vice president of media for the National Association of Independent Schools, said in EdWeek Market Brief reporting in 2023 that companies should think about private schools like small colleges, and that they should try to get to know the school’s philosophy before trying to sell to them to find the right angle.
ISC Research, which collects data on international schools, defines international private schools as those that teach wholly or partly in English to at least some students in countries where English is not an official language, or schools that offer a curriculum that’s not the host country’s national curriculum in countries where English is an official language.
International private schools have also grown in recent years and now enroll an estimated 7 million students globally, in almost 14,500 schools, according to ISC Research.
The countries with the most international schools are:

Asia has 57 percent of all international private schools, enrolling 4.7 million students.
Much of the growth in international private schools is in medium-fee schools, which appeal to families from a broader range of incomes. Medium-fee schools grew 17 percent from 2018 to 2023, compared with the market’s overall enrollment growth of 10 percent.
The most popular curriculum in international private schools is the Cambridge curriculum, with 35.4 percent market share, followed by the U.K. and IB curricula, which both have 28.1 percent of the market, and a U.S. curriculum at 19.1 percent. But many schools offer more than one curriculum.

source

Actions attack funding and accreditation but also seek to increase affordability and retention at Black colleges
Donald Trump signed executive orders on Wednesday targeting universities as his administration seeks to reshape higher-education institutions and continues to crack down on diversity and inclusion efforts.
The actions address foreign gifts to universities, directing the federal government to “enforce laws on the books” related to the disclosure of large donations, and college accreditation, which the president has referred to as his “secret weapon” to upend US universities. While reading the orders to Trump, the White House staff secretary Will Scharf said that the third-party groups that accredit universities have relied on “woke ideology” rather than merit.
Linda McMahon, the education secretary, added during the signing in the Oval Office: “We should be looking at those who have real merit to get in, and we have to look harder at those universities that aren’t enforcing that.”
Trump’s administration has been engaged in an all-out attack on US universities since the president took office in January, seeking to dramatically alter institutions he has claimed have been taken over by “Marxist maniacs and lunatics”. The federal government has sought to cut billions in funding from universities unless they comply with administration demands; banned diversity initiatives; and detained international students in retaliation for their activism.
This week, more than 150 US university presidents signed a statement condemning the Trump administration’s “unprecedented government overreach and political interference” in higher education. Meanwhile, Harvard University – which Scharf mentioned by name when introducing the order related to foreign gifts – has sued the government in response to the threatened funding cuts.
The president has referred to accreditation as a “secret weapon” in his fight against universities.
“I will fire the radical-left accreditors that have allowed our colleges to become dominated by Marxist maniacs and lunatics,” he said last summer. “We will then accept applications for new accreditors who will impose real standards on colleges once again and once for all.”
According to a statement from the White House, the order directs McMahon to hold accreditors accountable with “denial, monitoring, suspension, or termination of accreditation recognition, for accreditors’ poor performance or violations of federal civil rights law”. It also orders administration officials to investigate “unlawful discrimination” in higher education.
The White House alleges accreditors have imposed “discriminatory diversity, equity, and inclusion (DEI)-based standards”, which it describes as a violation of federal law and an abuse of their authority.
While signing orders on Wednesday that Scharf said would direct schools out of the “whole sort of diversity, equity and inclusion cult”, the president said that the US was “getting out of that … after being in that jungle for a long time”.
Despite his condemnation of diversity and inclusion efforts, Trump also signed an order establishing a White House initiative on historically black colleges and universities to promote “excellence and innovation”. The order facilitates the creation of a presidential advisory board on HBCUs and seeks to address funding barriers and increase affordability and retention rates.
The president also signed orders related to workforce development and artificial intelligence education to ensure the future workforce is “adequately trained in AI tools”, Scharf said.

source