Construction of English corpus oral instant translation model based on information security and deep learning

Abstract

In order to improve the quality of translation, avoid translation ambiguity, and accurately present the content of the source language, supported by the concept of deep learning and guaranteed by information security, an instant oral translation model is constructed for English corpus. The aim of this study is to enhance the efficiency and accuracy of oral translation systems through the application of deep learning algorithms. Specifically, we employ a sample training mechanism tailored to the unique characteristics of oral translation, allowing for separate training of system interaction and translation data. Furthermore, by redesigning the interaction hardware, this research comprehensively redefines the hardware structure of the translation system, marking a significant step towards improving the usability and performance of such systems. After obtaining and processing effective security sensitive information, language resources are managed by using database management system, which fundamentally improves the level of network information security. The performance of the existing oral automatic translation system (Test Group 1) and the system designed in this paper (Test Group 2) is tested by experiments, and the results are as follows: (1) The translation system designed here has better interactive performance, and it is better than Test Group 1. (2) The adaptive index value of Test Group 1 is 1, and that of Test Group 2 is 0.5, which proves that the adaptive ability of system algorithm of Test Group 2 is better than that of Test Group 1. (3) When comparing the translation speed, the translation time of Test Group 2 is only 70.7 s, while that of Test Group 1 is 130.6 s, so the proposed translation system is obviously superior to that of Test Group 1.

Keywords

information security deep learning English corpus oral instant translation model construction

Introduction

Research background and motivations

With the progress of social sciences in the world, great changes have taken place in translation studies. The innovation of interpretation research methods has promoted the related research and pushed it to a new peak.^1–3 In this process, how to combine information security with deep learning is a crucial link. While occupying an absolute advantage in the study of oral instant translation, English translation corpus has also appeared.⁴ Based on this, scholars in China and other countries propose to use deep learning to improve the accuracy and efficiency of machine translation and improve the adaptability of machines to different languages and dialects. However, with the progress of science and technology, the problem of information security has become increasingly prominent.^5–7 Based on this, this topic intends to make an in-depth study of oral instant translation based on the existing English corpus, and apply deep learning to this research.

Additionally, this paper addresses the aspect of information security, aiming to fortify the translation system with anti-interference capabilities, user privacy protection, and other critical security measures. By examining both the quality of translation and the robustness of information security, this research endeavors to offer valuable insights for the ongoing enhancement and development of the system.

Because the traditional spoken English translation is limited by the transfer of phonetic features, the algorithm can only use the lexical characteristics set by the system to make judgments. This leads to the low accuracy of lexical feature recognition, which cannot meet the needs of oral instant translation.^8–10 In order to solve this problem, this paper intends to use the sample learning mechanism in deep learning and consider the characteristics of real-time speech translation, and proposes a new way of human-computer interaction. On this basis, through the reconstruction of human-computer interaction equipment, the hardware architecture of the system is reconstructed. In order to solve this problem, this paper proposes a semantic recognition technology for human-computer interaction, which can achieve automatic interactive output of machine translation by deeply integrating the characteristics of machine translation and sentence output. Then, the parameters of the software layer are optimized globally with the deep neural network as the center. Finally, this study proposes the development of an instant English corpus translation model, grounded in information security principles and deep learning techniques.

Research objectives

The objective of this paper is to deeply study the construction and optimization of oral instant translation system based on the translation English corpus. By applying deep learning technology, the paper aims to improve the accuracy, fluency, and adaptability of oral translation system to deal with situations with different languages and accents more effectively. Meanwhile, the paper pays attention to information security, and strives to ensure that the translation system has a high degree of anti-interference and user privacy protection mechanism in practical application. By comprehensively considering the quality and safety of translation, this paper aims to provide profound theoretical support and practical guidance for the further development of oral instant translation technology.

Literature review

The exploration of the overarching characteristics of translation encompasses a variety of approaches, including empirical studies that gather concrete data on translation processes, introspective analyses of phonetic thinking which delve into the cognitive aspects of translating speech sounds, the use of Translog software to monitor and record translation keystrokes and eye movements for a deeper understanding of translation strategies, and comprehensive inductive reasoning to synthesize insights from individual instances into general principles. This multifaceted research methodology aims to provide a holistic view of translation practices, uncovering patterns and strategies that translators employ across different texts and contexts.

In 1996, Mona Baker and Sara Lavisoa started the Translated English Corpus (TEC) project.¹¹ The database has been designed and compiled, and it has become the most widely used corpus in English translation. This kind of corpus was the first in the world. Supported by the British Academy of Sciences, this project was completed in early 1999 and put into production. This corpus was designed by Sartor Ming Luz, a professor at the University of Berlin Trinity. It can be downloaded freely online, and it was mainly responsible for technical maintenance.^12,13 Candel-Mora (2022) introduced the creation and main features of TEC in detail, reviewed the research on oral instant translation based on TEC, and made an overall analysis of the potential application of TEC.¹⁴ Lalrempuii (2021) explained the composition and main features of English corpus, compared TEC with other databases, and finally, based on the analysis of TEC’s development trend, made clear the future development direction of TEC, which had certain enlightenment for oral instant translation based on TEC.¹⁵ Khan (2020) deeply studied modern translation studies, and deeply thought and discussed the characteristics of different languages.¹⁶ Based on this premise, Madahana (2022) analyzed the translated and untranslated texts, made clear the overall characteristics of the translated texts and the translation process, and noticed that the psychological meaning was not publicly displayed in the translation process.¹⁷

Fois (2021) thought that solving the problem of oral interaction was the key to improve the overall performance of the system. The analysis showed that oral interaction can be regarded as a transient interactive learning problem, so it can be solved by adopting deep learning algorithm.¹⁸ Han (2021) mentioned that a complete output translation feature and fusion learning structure can be obtained by using deep learning algorithm for data fusion and combining vocabulary translation training features with sentence translation training features.¹⁹ Yu (2022) explored the construction of an active defense model of cross-language network information security based on machine translation by studying technologies such as Web search engine, machine translation and automatic establishment of corpus database. This model strived for the acquisition, control and use of information by acquiring and utilizing key information in other countries’ networks.²⁰

Current research shows that with the progress of social science, the field of translation studies is constantly making progress, and the study of oral instant translation has gradually become the mainstream. Through deep learning technology, researchers have achieved remarkable research results on the basis of translating English corpus, improving the performance of oral translation system and enhancing its adaptability and accuracy.^21,22 However, there are still some problems to be solved.

Initially, it is crucial for researchers to address the complexities and challenges posed by diverse accents and intricate contexts within oral instant translation systems to enhance their universality. Furthermore, as technology advances, the importance of information security escalates, necessitating reinforced security measures within translation systems to safeguard user privacy and ensure system resilience against interference. Additionally, the practical operability and user experience of the system merit significant attention to guarantee that research outcomes are seamlessly integrated into actual linguistic communication contexts. Addressing these issues is pivotal for advancing the progression of oral instant translation technology.

Research methodology

Design of hardware for oral instant translation platform based on deep learning

In this study, the oral translation system is designed with a distinct hardware module that encompasses an interactive component. At the hardware level, the research introduces a “sandwich” based computation method, which is seamlessly integrated with deep learning algorithms at the system software level. Specifically, the hardware layer is divided into interactive information flow acquisition layer, interactive information processing hardware and interactive perception output layer.^23,24 The hardware structure of the platform is shown in Figure 1.

Figure 1.

Hardware structure of interactive oral instant translation platform.

The hardware parameters and functions of each functional layer will be described in detail below.

Hardware of interactive information flow acquisition layer: it provides a complete set of data input ports for this system, which supports USB, IEEE, XTB, SSB, and other protocols. Using multi-mode control technology, two GP3122 inductance control chips are designed, which realize stable filtering of input signals.

For the processing of interactive information, the system is equipped with two primary conversion processors and an auxiliary processor, all conforming to the technical specifications of 128×128 M. Building upon this framework, an NPU (Neural Processing Unit) neural network processor is incorporated as an interactive data processor, tasked with handling the interactive perception data generated by this study.

Hardware of interactive sensing output layer: The sensing array is composed of four diodes actively triggering IC, which cooperates with TPD2.1x high-speed special transmission interface to form a complete interactive sensing output layer.^25–27

English corpus retrieval model based on information security

In order to convert a large amount of data in the Internet into usable and effective information, it is necessary to go through a series of steps, including extracting the characteristics of things, converting them into data reflecting the attributes of things, and then further converting them into knowledge that can guide business.²⁸ Data, information and knowledge form a complete information chain. In this information chain, information security is particularly critical. With the process of data transformation, there are potential security risks in every link from feature extraction to knowledge formation.^29,30 Therefore, while constructing this information chain, people must pay attention to information security and take appropriate encryption and privacy protection measures to ensure the security of data and knowledge. Only under the premise that information security is fully guaranteed can people make more effective use of the rich data resources in the Internet.³¹ The whole process of information acquisition, processing, and utilization is shown in Figure 2.

Figure 2.

Information processing flow.

In the process of using cross-language network information resources, real-time speech translation technology can be used. In the data preprocessing stage, the real-time conversion between languages is realized, which is inseparable from the support of machine translation engine. This paper proposes a translation retrieval system based on English corpus, which is different from the traditional translation retrieval methods. In addition, people can also search for Chinese-English information on the Internet through the Internet search engine, thus establishing a corpus. Therefore, the biggest difference among them is the method of establishing corpus.³² As a result, a model of English corpus oral-assisted translation search engine system based on information security is established, as shown in Figure 3.

Figure 3.

Model of English corpus oral-assisted translation search engine system based on information security.

Figure 3 provides a detailed analysis of the entire process involved in constructing a spoken instant translation corpus. The primary focus is on capturing network data across various languages and converting these data into secure information for the user.

In this system mode, network robots search a large number of web pages from the Internet, and carry out complex identification and matching, thus establishing a large multilingual parallel translation corpus.³³ The system aims to provide safe and efficient translation reference services for end users, make positive contributions to the construction of network information security in China, and finally form an active defense model of instant translation network information security, as shown in Figure 4.

Figure 4.

Active defense model of instant translation network information security.

First, the open characteristics of the Internet should be made full use of to obtain all kinds of information from networks in different languages. This capture process is similar to the network information capture component described above. The obtained information resources are translated and processed through the voice real-time translation system to obtain the required information. Second, the security analysis of target information mainly includes keyword extraction and keyword information comparison. On this basis, a comparison method based on safety information keyword database is proposed. Effective security information is stored in a sensitive information database, and the final results are presented to users through the provided query analysis user interface. After obtaining and processing the effective confidential information, it is saved as a data table, and this language resource is managed by a DBMS.^34–36

Data interactive output algorithm for instant translation of English corpus

First, this project intends to use the interactive translation data in the oral real-time translation system based on deep learning, and use deep learning and other methods to carry out deep learning. Through this method, the sentences contained in the English translation can have certain semantic features, so that they can better determine the translation needed by the translation.³⁷ The specific learning and training process is as follows: Set the position coordinates of the ith learning particle in the sample as $X_{i} (x_{i 1}, x_{i 2}, \dots, x_{i n})$ , mark the position here as p₁, and mark the optimal positions of all learning particles as p₂. Set the training and learning frequency of individual learning particles to $V_{i} (v_{i 1}, v_{i 2}, \dots, v_{i d})$ . In the process of learning the semantic features of learning particles, the iterative changes of learning particles in the deep network are as follows:

v_{i d}^{k + 1} = v_{i d}^{k} + c_{1} r a n d (p_{1}^{k} - x_{i d}^{k}) + c_{2} r a n d (p_{2}^{k} - x_{i d}^{k})

(1)

x_{i d}^{k + 1} = x_{i d}^{k} + v_{i d}^{k + 1}

(2)

In the above equation, c₁ and c₂, respectively, represent the training depth coefficient, which is mainly used to promote the aggregation of particles in the dominant position of semantic features and globally control the training intensity of learning particles on semantic feature information. Rand represents the dynamic coefficient of [0,1]. The maximum frequency of semantic feature training for learning particles is V_max. When the training frequency of semantic features of learning particles reaches the maximum value V_max, the training frequency will remain constant. In this case, the best training result is obtained by training the semantic features of the particles.

On this basis, this project plans to adopt the deep learning method to realize the effective fusion of multi-source information, as shown in Figure 5.

Figure 5.

Output translation features and fusion learning structure.

The training process for fusion learning is divided into three distinct components: the integrated training for sentence segmentation, the adjustment training for sentence sequencing, and the holistic training of fusion features. First, the obtained translated sentence segmentation samples are used to further optimize the accuracy of sentence structure through secondary feature learning. Second, in the training stage of sentence order adjustment, according to the given sequence of training sentences, the known training sentences are sorted and adjusted to get the optimal collocation order. On this basis, by integrating the input and output signals, the sample information independent of training parameters is extracted to ensure the integrity of the output. On this basis, the interactive deep neural network based on deep learning is studied, and thresholds with different feature numbers are constructed to achieve the optimal matching between the number of output translations and the number of interactions.³⁸ The interactive deep learning neural network is described as follows:

i_{t} = σ (W_{t x} x_{t} + W_{i m} m_{t - 1})

(3)

f_{t} = σ (W_{f x} x_{t} + W_{f m} m_{t - 1})

(4)

o_{t} = σ (W_{o x} x_{t} + W_{o m} m_{t - 1})

(5)

c_{t} = f_{t} Θ C_{t - 1} + i_{t} Θ h (W_{c x} x_{t} + W_{c m} m_{t - 1})

(6)

m_{t} = o_{t} Θ c_{t}

(7)

p_{t} = s o f t m a x (m_{t})

(8)

In the above equation, W represents the interaction matrix, which corresponds to the interaction training parameters of the deep learning neural network. $σ$ and h represent nonlinear functions. $Θ$ represents the training iterative calculation process. Equation (8) uses the softmax function to obtain the optimal translation quantity, thus forming the matching quantity p_t of the interactive quantity.

Experimental design and performance evaluation

Datasets collection

Taking 1.5 million sentences as samples, this paper establishes a system that can improve the efficiency of translation interaction. The hidden cell size of this system is set to 1000, and the word dimension is set to 620. This paper is based on real data, and the automatic oral English translation system is the control group, namely the Test Group 1. On this basis, the system designed in this subject is named as Test Group 2. In the same experimental environment, the translation effects of the two Test Groups are compared.

Experimental environment and parameter setting

Taking spoken English as an example, words, short sentences and long sentences are used as test types. The time of this test is 60 minutes, which is mainly used to detect the number of words in sentences and the vocabulary used in longer sentences, as shown in Table 1.

Table 1.

Test sample parameters and related parameter configuration.

Sample type	Word	Short sentence	Long sentence
Word quantity	1	2–9	10–20
Test time	60 min
Interaction frequency	40kbit/s
Initial translation rate	20kbit/s

The test indicators listed in Table 1 all come from the practice norms of oral English and the design of historical interpretation system. Hence, the established test parameters can be viewed as the threshold for discriminating between the performance standards of the two conversion systems. During the preliminary test screening, the actual translation performance of the reference system surpasses the predefined benchmarks. This approach ensures that the performance of any system is not constrained, thereby guaranteeing the objectivity and accuracy of the test outcomes.

Performance evaluation

Interactive performance test of oral instant translation

The data in Table 1 are input into Test Group 1 and Test Group 2, respectively. This method randomly sends word data, short sentence data and sentence data according to the maximum capacity that the tested system can handle. If it exceeds the system capacity, stop the test immediately, and record the test value corresponding to the termination time as the maximum value. The test result is shown in Figure 6.

Figure 6.

Interactive test results of different translation systems.

In the case of dynamic testing, the interactive weights of Test Group 1 and Test Group 2 shown in Figure 6 are compared, and the results show that the weight coefficient of Test 2 is significantly greater than that of Test 1 in the same situation when dealing with translated materials. On this basis, this paper puts forward a method based on interactive learning, which can obtain higher interaction weight in a large training value and a long training time, thus indicating that the model has high interaction ability. The results show that the translation system designed by Test Group 2 has higher expressive ability and higher sentence characteristics than the translation of Test Group 1. This is mainly due to the interactive semantic recognition analysis based on deep learning, the fusion of translation vocabulary and sentence output characteristics based on deep learning, and the automatic interactive output of translation based on deep learning.

Translation accuracy test

The instant random translation accuracy of Test Group 1 and Test Group 2 is tested according to the parameter values set in Table 1. PATOTS is used for testing, which is the main tool for evaluating the performance of oral English translation system. The frequency unit of data transmission is ms, and the set amount of test data (words: 1800, short sentences: 1000, and long sentences: 500) is completed within 60 minutes, and the test accuracy results are recorded and analyzed. The specific results are shown in Figure 7.

Figure 7.

Translation accuracy test results of different translation systems.

From the above data analysis, the overall test results can be summarized into three dimensions, namely, the correct rate, the interaction index and the adaptability index. At three levels, the interaction index reflects the system’s ability to deal with algorithms and information flow, and the lower the feedback value, the better the effect; the adaptability index reflects the response ability of the system under the condition of high-intensity calculation. Among them, the number of accurate translations in Test Group 1 is 1742-974-487, and the number of accurate translations in Test Group 2 is 1798-998-498. After one-on-one comparative analysis, the results show that the accuracy of the translation used in Test Group 2 is obviously higher than that in Test Group 1. As far as the interaction index is concerned, the value of Test Group 1 is 1, and the value of Test Group 2 is 0.5. The experimental results show that the interactive effect of the translation system proposed in this paper is better than that of the first group. As for the adaptability index, Test Group 1 is 1, and Test Group 2 is 0.5. Similarly, it shows that the algorithm adaptability of Test Group 2 is better than that of Test Group 1.

Translation speed test

In order to verify the translation efficiency of the system, the translation speed of Test Groups 1 and 2 is tested according to the above settings, and the test results are shown in Figure 8.

Figure 8.

Translation speed test results of different translation systems.

In Figure 8, in the process of translating 8000 sentences, it only takes 70.7 seconds for Test Group 2 and 130.6 seconds for Test Group 1.

The findings indicate that Test Group 2 has effectively addressed challenges in practical applications within a brief period. Overall, the translation system developed by Test Group 2 fulfills the criteria for instantaneous spoken English translation, demonstrating commendable accuracy, stability, and adaptability.

Discussion

Based on the interpretation corpus, Hammou (2020) analyzed the multimodal corpus and processed it. However, this method still needs a lot of manual processing and proofreading. In the future construction of multimodal interpretation corpus, multimodal retrieval technology based on artificial intelligence (AI) is still a very meaningful topic.³⁹ On this basis, Alshemali (2020) compared several methods by modifying the sentence length. This project was based on the recurrent neural network to realize the efficient output of safety information. On this basis, a method based on word alignment is proposed, which realizes the segmentation of long sentences, and makes linear adjustment according to the word order of sentences, and obtains the optimal order adjustment scheme. The results show that the evaluation value of this algorithm greatly exceeds the existing algorithms, and further verify the effectiveness and superiority of this algorithm in error correction.⁴⁰ In short, from the current research, the development of AI is very rapid, and the understanding and processing of natural language has also developed rapidly. Deep learning is a new research idea, which provides a new idea for the processing of language information.

Conclusion

Research contribution

In recent years, AI technology has developed rapidly, especially in the understanding and processing of natural language. In recent years, with the development of deep learning technology, its application in the field of language processing has attracted more and more attention. English is a worldwide language, which plays an important role in promoting information exchange and cultural exchange among countries. This makes the instant translation of spoken English an urgent problem. This project intends to use deep learning method to train samples, and consider the characteristics of oral instant translation, and use features to train system interaction and translation data. On this basis, this project takes English corpus as the research object, reconstructs the interactive hardware and its hardware architecture, and establishes an instant oral translation model based on information security and deep learning.

Future works and research limitations

This study has led to numerous insights, yet there remains significant room for enhancement. The corpus selected for this experiment is somewhat limited in scope and exhibits certain constraints, highlighting the need for an expansion to cover a broader range of translation domains. This expansion is crucial not only for overcoming the limitations inherent to the current corpus but also for ensuring that the translation models are trained on a diverse set of data, which can improve their generalizability and effectiveness across various contexts.

Additionally, the creation of a comprehensive translation evaluation system is paramount. Such a system would not only assess translation accuracy but also evaluate other critical aspects like fluency, context relevance, and cultural appropriateness. By identifying and correcting errors, this system would contribute significantly to enhancing the quality of translations. Moreover, it would provide valuable feedback for further refinement of the translation algorithms, leading to improvements in their performance.

Expanding the market potential of the translation service is another key area of focus. By addressing the current gaps in translation accuracy and broadening the applicability of the translation system to include more languages and dialects, the service can appeal to a wider audience. This, in turn, can open up new market opportunities and foster greater acceptance and use of automated translation tools in professional and personal settings alike.

In summary, by expanding the corpus and developing a comprehensive evaluation system, this research can pave the way for more accurate, reliable, and widely applicable translation technologies. These improvements are essential for realizing the full potential of automated translation services in bridging language barriers and facilitating global communication.

Statements and declarations

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Luo

. Universals in machine translation? A corpus-based study of Chinese-English translations by WeChat Translate. Int J Corpus Linguist 2022; 27(1): 31–58.

Laviosa

Falco

. Using corpora in translation pedagogy. In: New perspectives on corpus translation studies. Berlin, Germany: Springer Nature, 2021, pp. 3–27.

Dayter

. Variation in non-fluencies in a corpus of simultaneous interpreting vs. non-interpreted English. Perspectives 2021; 29(4): 489–506.

Liu

. Corpus design of Chinese medicine English vocabulary translation teaching system based on Python. J Info Know Mgmt 2022; 21(Supp02): 2240022.

Wang

Guzmán

, et al. Putting words into the system’s mouth: a targeted attack on neural machine translation using monolingual data poisoning. arXiv preprint arXiv:2107.05243 2021.

Alfuraih

. The undergraduate learner translator corpus: a new resource for translation studies and computational linguistics. Lang Resour Eval 2020; 54(3): 801–830.

Orenha-Ottaiano

Garcia

Eugênia

, et al. Corpus-based methodology for an online multilingual collocations dictionary: first steps. Electronic lexicography in the 21st century (eLex 2021) Post-editing lexicography. Brno, Czech Republic: Lexical Computing CZ s.r.o., 2021, p. 97.

Cho

Kim

Cho

, et al. Kosp2e: Korean speech to english translation corpus. arXiv preprint arXiv:2107.02875 2021.

Almahasees

Meqdadi

Albudairi

. Evaluation of google translate in rendering English COVID-19 texts into Arabic. J Lang Ling Studie 2021; 17(4): 2065–2080.

10.

. Research on business English translation framework based on speech recognition and wireless communication. Mobile Inf Syst 2021; 2021: 1–11.

11.

. Exploring processing patterns of Chinese-English sight translation: an eye-tracking study. Babel 2020; 66(6): 999–1024.

12.

Yan

Research on English Chinese translation system for tourism based on globish. In: EAI International Conference, BigIoT-EDU. Cham: Springer Nature Switzerland, 2022, 7, pp. 14–25.

13.

Yang

. The application of Artificial Intelligence in English teaching Int J Front. Sociol 2020; 2(3): 76–81.

14.

Candel-Mora

MÁ

. Big data to assess genre-specific features of the machine translation output of online travel reviews in Spanish. Quaderns de Filologia-Estudis Lingüístics 2022; 27: 49–69.

15.

Lalrempuii

Soni

Pakray

. An improved English-to-Mizo neural machine translation. ACM Trans Asian Low-Resour Lang Inf Process 2021; 20(4): 1–21.

16.

Khan

Abid

. A novel natural language processing (NLP)–based machine translation model for English to Pakistan sign language translation. Cognit Comput 2020; 12: 748–765.

17.

Madahana

Khoza-Shangase

Moroe

, et al. A proposed artificial intelligence-based real-time speech-to-text to sign language translator for South African official languages for the COVID-19 era and beyond: in pursuit of solutions for the hearing impaired. S Afr J Commun Disord 2022; 69(2): 915.

18.

Fois

. Translator training, English language teaching and corpora: scenarios and applications. Internat J Lang Stud 2021; 15(4): 59–78.

19.

Han

Yin

. Application of web embedded system and machine learning in English corpus vocabulary recognition. Microprocess Microsyst 2021; 80: 103634.

20.

. English translation model based on intelligent recognition and deep learning. Wireless Commun Mobile Comput 2022; 2022: 125–139.

21.

Zhou

Chen

, et al. Sign-to-speech translation using machine-learning-assisted stretchable sensor arrays. Nat Electron 2020; 3(9): 571–578.

22.

Popel

Tomkova

Tomek

, et al. Transforming machine translation: a deep learning system reaches news translation quality comparable to human professionals. Nat Commun 2020; 11(1): 4381.

23.

Vartiainen

Tedre

Valtonen

. Learning machine learning with very young children: who is teaching whom? Intern J Child-Comp Inter 2020; 25: 100182.

24.

Mujahid

Awan

Yasin

, et al. Real-time hand gesture recognition based on deep learning YOLOv3 model. Appl Sci 2021; 11(9): 4164.

25.

Wang

Zhou

Jiang

. A survey of word embeddings based on deep learning. Computing 2020; 102: 717–740.

26.

Yang

Wang

Chu

. A survey of deep learning techniques for neural machine translation. ar**v preprint ar**v:2002.07526 2020.

27.

Yang

, et al. Review the state-of-the-art technologies of semantic segmentation based on deep learning. Neurocomputing 2022; 493: 626–646.

28.

Dong

Wang

Abbas

. A survey on deep learning and its applications. Comput Sci Rev 2021; 40: 100379.

29.

Sharma

Jindal

. Machine learning and deep learning applications-a vision. Glob Trans Proc 2021; 2(1): 24–28.

30.

Wen

Zhang

, et al. AI enabled sign language recognition and VR space bidirectional communication using triboelectric smart glove. Nat Commun 2021; 12(1): 5378.

31.

Rastgoo

Kiani

Escalera

. Sign language recognition: a deep survey. Expert Syst Appl 2021; 164: 113794.

32.

Wan

, et al. Deep learning models for real-time human activity recognition with smartphones. Mob Netw Appl 2020; 25: 743–755.

33.

Michelsanti

Tan

Zhang

, et al. An overview of deep-learning-based audio-visual speech enhancement and separation. IEEE/ACM Trans Audio Speech Lang Process 2021; 29: 1368–1396.

34.

Qin

, et al. Deep learning enabled semantic communication systems. IEEE Trans Signal Process 2021; 69: 2663–2675.

35.

Rank

Pfahringer

Kempfert

, et al. Deep-learning-based real-time prediction of acute kidney injury outperforms human predictive performance. NPJ Digit Med 2020; 3(1): 139.

36.

Torfi

Shirvani

Keneshloo

, et al. Natural language processing advancements by deep learning: a survey. ar**v preprint ar**v:2003.01200 2020.

37.

Jain

Kumar

Garg

. Sarcasm detection in mash-up language using soft-attention based bi-directional LSTM and feature-rich CNN. Appl Soft Comput 2020; 91: 106198.

38.

Yilmaz

Toklu

. A deep learning analysis on question classification task using Word2vec representations. Neural Comput Appl 2020; 32: 2909–2928.

39.

Hammou

Lahcen

Mouline

. Towards a real-time processing framework based on improved distributed recurrent neural network variants with fastText for social big data analytics. Inf Process Manag 2020; 57(1): 102122.

40.

Alshemali

Kalita

. Improving the reliability of deep neural networks in NLP: a review. Knowl Base Syst 2020; 191: 105210.