Использование приложения для онлайн-обучения LAIX с целью совершенствования навыков устной речи на английском языке у обучающихся средней школы

аспирант кафедры теории и практики иностранных языков института иностранных языков, Российский университет дружбы народов имени Патриса Лумумбы. Сфера научных интересов: английский язык как иностранный, методика преподавания английского языка, полимодальные тексты. Автор пяти опубликованных научных работ. ORCID: Электронная почта:

Postgraduate at the Department of Theory and Practice of Foreign Languages of the Institute of Foreign Languages, Peoples’ Friendship University of Russia named after Patrice Lumumba. Research interests: English as a foreign language, methods of teaching English, polymodal texts. Author of five published scientific works. ORCID: E-mail address:

Introduction. Polycode focuses on the combination of symbols, speech and visual space. Polycode involves more combinations of elements and thus can mobilize the scholar’s sensory system from different aspects. The processing and analysis of polycode have long been an important research direction in the field of educational technology, and a large number of literatures have explored its application and influence from different perspectives. For instance, Perspective of advertising was analysed by D.P. Chigaev who studied methods of polycode in modern advertising texts [2]. By analysing educational and advertising materials in Russian and Chinese, some researchers concentrated on investigating innovative methods of teaching in polycode environment [7]. Numerous scientists examined how the use of polycode text reflected the national linguistic vision of the world [5]. These ideas were expressed in polycode environment both intentionally and implicitly. In addition, components and their correlation in using polycode was studied by some scholars around the world [6; 11]. Internet communication in using polycode texts was another research topic as well [1; 14].

Existing research generally focuses on five key factors, among which a few studies have paid attention to the practice of English learning combined with online learning applications in polycode environment. This field still has the limitation of relatively concentrated research perspectives, and it is urgent to explore from new entry points to enrich its theoretical connotation and application scenarios.

Automatic Speech Recognition. By fusing artificial intelligence and language recognition technology, the LAIX app is a learning tool designed to assist users in becoming more proficient in the English language. The first three letters of LAIX mean learning and artificial intelligence-driven education, leading to a new way of learning. And “X” stands for unknown, meaning unlocking the infinite pos-

Российский университет дружбы народов имени Патриса Лумумбы

sibilities of life. It assesses users' English pronunciation, grammar and fluency through artificial intelligence (AI), helps users identify problems in pronunciation and provides targeted corrective suggestions, thus contributing to users' overall improvement of their English language proficiency. In order to help users practice English in authentic situations and enhance their skills of oral expression and listening, LAIX app offers a variety of scenario simulation dialogues, such as ordering food, shopping, interviews and more. In addition, it skillfully blends English learning courses and games to increase user engagement through voice-overs, punch cards, breakthroughs and other activities. According to Samokhin, online computer games, along with movies and music, make up the material of the polycode methodological complex [9]. Therefore, as an online learning software, LAIX applications essentially contain the feature of poycode.

In recent years, with the continuous development of automatic speech recognition (ASR) technology, its application scope has gradually expanded from traditional fields such as com- puter science, engineering, acoustics, and telecommunications to the field of education, especially demonstrating significant potential in language learning. Researchers in Turkey conducted research in the field of speech recognition in Turkish, examined recent advancements, and analyzed the factors impeded its growth in terms of both labor and material resources [3]. Indian scholars extracted Mel Frequency Resonance Frequency (MFCC) features from recorded Adi speech samples by using a variety of automatic speech recognition models and comparing their minimum Word Error Rates (WERs), all of which ultimately facilitated the creation of various Adi HMI ASR applications [10]. Scholars in China experimented with acoustic models and input features to deal with the problem of degrading the accuracy of Automatic Speech Recognition systems. Furthermore, a language model for grid reconstruction was trained by the researchers, resulting in improvements in the relative word error rate (WER) of 40.7% and 35.7% on the evaluation and development sets respectively [13]. Portuguese and its variants being an un- derstudied language, scholars looked into feature extraction and classification techniques commonly used by automatic speech recognition systems for Portuguese applications [4].

However, current research on the application of ASR technology in enhancing speaking skills mainly focuses on the pronunciation level, such as providing real-time feedback to learners through computer-assisted pronunciation training systems to improve the pronunciation accuracy of students with specific native language backgrounds [8; 12]. This type of research has achieved positive results in pronunciation correction, but its role in promoting sentence fluency has not been fully explored.

Based on the above research background and existing gaps, this paper aims to explore the synergistic impact of the LAIX automatic speech recognition system on learners' pronunciation accuracy and sentence fluency during the English learning process from the perspective of comprehensively improving oral English ability. By recording and analyzing the pronunciation scores of learners at the word and sentence levels, this study attempts to evaluate the actual effect of the ASR system in a real learning environment, thereby providing new empirical evidence and theoretical supplements for the educational application combining polycode text and intelligent voice technology.

Objective. This study takes the widely used learning software LAIX as a case, focusing on evaluating the improvement effect of its built-in automatic speech recognition (ASR) system on learners' speaking skills (especially pronunciation and fluency), with the aim of providing empirical evidence for technology-assisted language teaching practice.

Methodology. Fifty students from two classes in a middle school in China participated in the study. These students were from grade 9, including 28 boys and 22 girls, aged 16 or 17.

Before the experiment, the researchers conducted a questionnaire survey on the participants to systematically estimate the current English speaking level of the participants. The survey results showed that 46 students had studied English for 9 years or more, accounting for 92% of the total number. 41 participants rated their English speaking ability as poor, and 5 rated their English speaking level as medium. The questionnaire survey results showed that 47 people rarely spoke English or never spoke English after school.

Procedures. This study consists of four steps: (1) conducting an English proficiency test, (2) recording word and sentence pronunciation (pre-test), (3) conducting an experiment based on LAIX app learning (4) recording word and sentence pronunciation grades after experiment (post-test).

Test and Evaluation. The 50 participants first completed the 15-minute standard English test (https://www. of EF Education First (EF). To evaluate the English proficiency of the participants, this test is made up of 20 questions (100 points), including 10 reading questions and 10 listening questions. The beginner level refers to scores from 0 to 60 points, the intermediate level refers to scores from 61 to 85 points, and the advanced level refers to scores from 86 to 100 points. Results in table 1 clearly shows that 46 pupils have a beginner's level of English listening, while the remaining have an intermediate level. English reading level of 6 pupils is intermediate, while that of 44 is beginner.

Based on the EF test results, the LAIX beginner course on greetings was selected to test the participants’ current English speaking level, which contained 28 words and 5 sentences. The full text of the course is as follows:

• A:    Good morning (sentence 1).

• B:    Good morning. It’s a beautiful day (sentence 2).

• A:    Yeah, it is. There's not a cloud in the sky (sentence 3).

• B:    Yeah, it's perfect. Have a good day (sentence 4).

A: Thanks. You, too (sentence 5).

Following the participants' completion of the required reading of each word, the researchers recorded the app's 15-word scores. Participants then read the five phrases once more and noted the score that the software automatically assigned to each one. Tables 2 and 3 display the app's word pronunciation and sentence scores for the pupils.

50 students completed a four-week oral English experiment during the summer vacation. The researchers assigned 50 participants 30 minutes of daily oral check-in exercises on the LAIX app for 28 days. The exercises were all from the elementary courses on the LAIX app, which focused on daily communication. Fun voiceovers, course breakouts, learning punch cards and more were all included in these online learning programs. Participants received the lessons they needed to practice in the WeChat group every day. They shared a screenshot of the successful checkin to the WeChat group after completing the 30-minute exercise.

Analysis and results. Participants were re-scheduled to do the previous word and sentence pronunciation tests again after finishing the 28-day systematic study. To make data accurate to some extent, the content of these words and sentences was the same as that of the previous test. However, four participants failed to complete their daily training tasks on time during the learning period of the system for various reasons. Therefore, the total number of participants whose scores could be counted in the end was only 46. The word and sentence pronunciation scores are shown in Tables 4 and 5 respectively.

Word Scores. The research results shows that participants demonstrated continuous score improvements in the pronunciation of the same word throughout all five experimental stages, indicating that the LAIX application had a general promoting effect on pronunciation ability in a continuous training environment. It also indicated differentiated manifestations among learners of different founda-

ИСПОЛЬЗОВАНИЕ ПРИЛОЖЕНИЯ ДЛЯ ОНЛАЙН-ОБУЧЕНИЯ LAIX С ЦЕЛЬЮ СОВЕРШЕНСТВОВАНИЯ НАВЫКОВ УСТНОЙ РЕЧИ НА АНГЛИЙСКОМ ЯЗЫКЕ У ОБУЧАЮЩИХСЯ СРЕДНЕЙ ШКОЛЫ

Table 1

Results of standard English test (n = 50)

EF set level	Beginner (0–60)	Intermediate (61–85)	Advanced (86–100)
Reading	n = 44	n = 6	n = 0
Listening	n = 46	n = 4	n = 0

Table 2

Grades of words pronunciation (n = 50)

Results Words	Number of students (n / %)
	≤ 60	61–70	71–80	81–90	91–100
good	0	0	7 (14)	2 (48)	9 (38)
morning	0	1 (2)	8 (16)	2 (50)	16 (32)
it’s	17 (34)	1 (30)	9 (18)	6 (12)	3 (6)
a	0	0	5 (10)	24 (48)	21 (42)
beautiful	4 (8)	9 (18)	12 (24)	17 (34)	8 (16)
day	0	3 (6)	16 (32)	24 (48)	10 (20)
There’s	19 (38)	13 (26)	13 (26)	3 (6)	2 (4)
not	0	0	18 (36)	12 (24)	20 (40)
cloud	0	7 (14)	14 (28)	19 (38)	10 (20)
sky	0	3 (6)	19 (38)	22 (44)	6 (12)
perfect	4 (8)	8 (16)	14 (28)	18 (36)	7 (14)
have	0	2 (4)	23 (46)	14 (28)	11 (22)
thanks	8 (16)	14 (28)	23 (46)	3 (6)	2 (4)
you	0	0	21 (42)	24 (48)	5 (10)
too	9 (18)	16 (32)	17 (34)	4 (8)	4 (8)

Table 3

Grades of sentences pronunciation (n = 50)

Results Sentences Number of students (n / %) ≤ 60 61–70 71–80 81–90 91–100 1 0 1 (2) 7 (14) 26 (52) 16 (32) 2 0 0 6 (32) 23 (46) 11 (22) 3 0 9 (18) 18 (36) 14 (28) 9 (18) 4 0 6 (12) 13 (26) 17 (34) 14 (28) 5 5 (10) 9 (18) 20 (40) 12 (24) 4 (8) tion levels. Improvement was most significant among the group of participants whose word pronunciation scores were below 60 before the test. Meanwhile, LAIX also showed a stable supportive effect on pronunciation improvement for learners at an intermediate level (61–90 points).

From the perspective of the learning mechanism, LAIX could quickly identify specific phonemes that par- ticipants had difficulty pronouncing and provide them with immediate and targeted correct pronunciation feedback. This function was particularly effective for the speech acquisition of intermediate and low-level learners. However, for participants who were already in the high-scoring stage (above 90 points) before the test, the extent of improvement was relatively limited. It showed that the pro- nunciation precision requirements for high-scoring words were higher, and the system's judgment of phoneme accuracy was stricter. In addition, these words might contain phonemes that were difficult to pronounce, such as friction sounds, which posed a challenge to learners' ability to control their pronunciation.

Sentence Scores. Data before and after experiment revealed that improve-

Table 4

Grades of words pronunciation (n = 46)

Results Words	Number of students (n / %)
	≤60	61–70	71–80	81–90	91–100
good	0	0	5 (11)	18 (39)	23 (50)
morning	0	0	3 (6)	22 (48)	21 (46)
it’s	11 (24)	9 (19)	12 (26)	11 (24)	3 (7)
a	0	0	1 (2)	17 (37)	28 (61)
beautiful	0	6 (13)	11 (24)	17 (37)	12 (26)
day	0	0	12 (26)	18 (39)	16 (35)
There’s	9 (19)	11 (24)	13 (28)	10 (22)	3 (7)
not	0	0	12 (26)	14 (30)	20 (44)
cloud	0	3 (6)	11 (24)	21 (46)	11 (24)
sky	0	0	14 (30)	21 (46)	11 (24)
perfect	0	4 (9)	11 (24)	23 (50)	8 (26)
have	0	0	17 (37)	17 (37)	12 (26)
thanks	3 (6)	9 (20)	25 (54)	5 (11)	4 (9)
you	0	0	15 (33)	24 (52)	7 (15)
too	4 (9)	7 (15)	26 (56)	4 (9)	5 (11)

Table 5

Grades of sentences pronunciation (n = 46)

Results Sentences Number of students (n / %) ≤60 61–70 71–80 81–90 91–100 1 0 0 7 (15) 23 (50) 16 (35) 2 0 0 16 (34) 20 (44) 10 (22) 3 0 4 (9) 20 (44) 13 (28) 9 (19) 4 0 4 (9) 14 (31) 14 (30) 14 (30) 5 2 (5) 6 (13) 19 (41) 14 (30) 5 (11) ments in participants’ word pronunciation scores positively influenced their sentence pronunciation scores to a certain extent, indicating that foundational pronunciation skills exert a positive transfer effect on overall spoken language performance. However, the chart data showed that the overall improvement in sentence pronunciation scores was significantly lower than that of word pronunciation scores. This gap was particularly pronounced among learners who had already achieved sentence pronunciation scores of 70 points or higher before the experiment. This phenomenon is closely related to the scoring mechanism used by LAIX applica- tions. Sentence pronunciation assessment not only focuses on the accuracy of individual phonemes but also incorporates fluency as an important indicator. Its scoring criteria cover more complex dimensions such as the differences in pronunciation rules between words and meaning units and the coherence of speech flow. In contrast, the scoring of word pronunciation is mainly based on the absolute phonetic standard, focusing on the accuracy of pronunciation in an isolated environment.

In summary, LAIX delivered clear results in word-level pronunciation training, particularly excelling at helping intermediate and beginner learners overcome pronunciation challenges. However, if educators would like technology-supported pronunciation instruction to achieve effective improvement at the sentence level, training design must systematically incorporate specialized support for sentence fluency rather than focusing solely on word-level pronunciation accuracy.

Discussion. The discussion of this study needs to take into account the limitations brought about by the experimental environment. As the experiment was conducted outside the classroom and the process lacked direct supervision, this might have a certain impact on the integrity of

ИСПОЛЬЗОВАНИЕ ПРИЛОЖЕНИЯ ДЛЯ ОНЛАЙН-ОБУЧЕНИЯ LAIX С ЦЕЛЬЮ СОВЕРШЕНСТВОВАНИЯ НАВЫКОВ УСТНОЙ РЕЧИ НА АНГЛИЙСКОМ ЯЗЫКЕ У ОБУЧАЮЩИХСЯ СРЕДНЕЙ ШКОЛЫ the data. For instance, four learners dropped out halfway due to insufficient initiative, resulting in incomplete data collection. Furthermore, the subjective motivation of learners becomes a key variable influencing the experimental results. For participants with stronger motivation, the validity of their data was higher; However, for participants with insufficient motivation, the validity of their final test results may be questionable. These factors provide important directions for improvement for subsequent related research.

Conclusion . This study employed empirical methods to investigate the impact of the LAIX application on the speaking skills of English learners in secondary school. Experimental results indicate that LAIX significantly enhances learners’ word pronunciation. This effectiveness stems primarily from its built-in automatic speech recognition system, which instantly diagnoses pronunciation issues and provides correct demonstrations.

This enables learners to continuously make targeted corrections through interactive shadowing and level-based exercises. The app’s real-time scoring and three-color visual feedback mechanism effectively guides learners to focus on specific pronunciation challenges, thereby optimizing learning efficiency. This study also reveals the ability transfer path from word accuracy to sentence fluency. Learners need to master the accurate pronunciation of words and further pay attention to the coherence in the speech flow in order to achieve a higher score in sentence pronunciation.

This study innovatively reveals the intrinsic path of technology-assisted oral language learning by analyzing its feedback mechanism and scoring criteria. The attribution analysis of the scoring differences between words and sentences provides detailed empirical evidence for understanding the advantages and limitations of ASR technology in current educational applications. This study confirms the operability of integrating intelligent voice technology into mobile learning platforms and its positive teaching value. It provides decision-making references for educators to screen and utilize technological tools in the context of blended teaching. It also provides a new direction for the optimization of language learning technology products. The development of future online language learning software should not only consolidate the advantages of word pronunciation training but also enhance the intelligent feedback function for fluency at the sentence level and discourse coherence.

Acknowledgement. I would like to thank my supervisor, Ivan Sergeevich Samokhin. His unwavering support and thought-provoking advice gave me help during PhD program. Furthermore, I would like to appreciate the kindness from my college, Gerges Sarah Nady Nekhela, who spared no effort in sharing a successful experience of academic publication.