Innovations in Practice

RELC Journal
2023, Vol. 54(2) 508–516
Artificial Intelligence-Based © The Author(s) 2023
Article reuse guidelines:
Content Generator Technology sagepub.com/journals-permissions
DOI: 10.1177/00336882231165060
for Young English-as-a-Foreign- journals.sagepub.com/home/rel

Language Learners’ Reading

Enjoyment

Jang Ho Lee
Department of English Education, Chung-Ang University, Seoul, Korea

Dongkwang Shin
Department of English Language Education, Gwangju National University
of Education, Korea

Wonjun Noh
Department of English Language Education, Gwangju National University
of Education, Korea; Unnam Elementary School, Korea

Abstract
Artificial intelligence has recently seen tremendous growth, and been applied to several fields,
including the second-language domain. In this article, we share an innovative practice based on
one of potential artificial intelligence technologies for second-language teaching and learning – arti-
ficial intelligence-based content generator, which generates texts based on user’s keywords. In
total, 121 young English-as-a-foreign-language learners participated in the study, with half of
them having engaged in the artificial intelligence-based content generator-based activity, and the
other half having received traditional English-as-a-foreign-language reading instruction. We exam-
ined the extent to which the artificial intelligence-based content generator-based activity could
influence the participants’ foreign language enjoyment and interests in reading English books,
and the participants were given the survey addressing these variables, prior to and after the
innovative practice. It was found that the condition based on the artificial intelligence-based con-
tent generator-based activity was more effective in terms of enhancing the target variables, and
that the group which engaged in the artificial intelligence-based content generator-based activity
was largely in favor of artificial intelligence-based content generator technology. Pedagogical impli-
cations for employing this technology in second-language contexts are provided.

Corresponding author:
Dongkwang Shin, Department of English Language Education, Gwangju National University of Education, Buk-
gu Pilmoon-daero 55, Gwangju, Korea.
Email: [email protected]
Lee et al. 509

Keywords
artificial intelligence-based content generation, computer-assisted language learning, English-as-a-
foreign-language learners, enjoyable reading, extensive reading, generative artificial intelligence,
reading motivation

Introduction
Artificial intelligence (AI) technology has been evolving rapidly in recent years. Generative
AI is a branch of artificial intelligence that surpasses simply using AI as a tool for big data
analysis. For example, it can generate images through text input (e.g. OpenAI’s
DALL·E2), video corresponding to the story of the input passage (e.g. Meta’s
Make-A-Video) or music following the given text instructions (e.g. Google’s MusicLM).
OpenAI’s ChatGPT, which is currently attracting the most attention, can generate computer
programming codes according to seed information entered by users; generate various texts
such as dialogues, blogs and novels, and even test items based on these texts. This technology
has recently started to gain attention from the second-language (L2) and Teaching English to
speakers of other languages (TESOL) communities (Shin, 2022).
In this article, we share an innovative practice showcasing the integration of a genera-
tive AI-based tool known as CopyAI – the AI-based content generator (AICG) developed
using GPT-3 (Ghumra, 2022) – into L2 contexts. CopyAI, as the AICG, can generate
texts of varying lengths and genres based on users’ keywords. Some fields, such as cre-
ative writing and literature, have started to adopt and examine the potential of AICG (e.g.
Lee, 2017), which has revealed its value in the respective fields.

The Teaching Context

This innovation was applied in English classes in elementary schools located in Seoul,
Gwangju and Jeollanam-do, South Korea, with 121 Grade five and six students (aged
11–12). The experimental group, which experienced the current innovative practice
and performed the AICG-based activity, consisted of 34 boys and 26 girls, among
whom 48 and 12 were fifth and sixth graders, respectively. The comparison (i.e.
business-as-usual) group consisted of 30 boys and 31 girls, and 47 and 14 were fifth
and sixth graders, respectively. In line with the curriculum of the South Korean ministry
of education, three 40-minute English lessons were given to the participants. Also, by the
time of the study, Grade five and six students had been exposed to approximately 187 and
289 hours of English lessons.

Reason for the Innovation

The reason for the innovation predominantly came from the participants’ lack of engage-
ment in English reading as well as a dearth of English learning enjoyment. That is, the
participants’ teachers (with the third author of this article being one of them) had
noticed that their students were generally not interested in the topics of the passages in
the participants’ English textbook. Given that engaging in reading interesting texts is a
key to the development of English competence (Cho and Krashen, 1994; Day and
Bamford, 2002), the students’ lack of interest in English texts was considered a
510 RELC Journal 54(2)

serious problem. Meanwhile, creating English passages that may suit their students’ inter-
ests was not a possible option due to their heavy workload.
The authors, including the English teacher of the participants, found that AICG has
recently been proposed as a useful technology that could reduce the workload of L2 prac-
titioners by enabling them to generate English texts with ease. In particular, in view of the
finding of a recent study (Shin and Lee, 2023) that the texts generated by the AI could be
perceived to be better or on a par with the one written by the native speaker of English in
terms of the naturalness of story-flow and English expressions, AICG was considered to
have significant potential to contribute to the teacher’s English lessons, if it could gener-
ate English texts that suited the interests and English proficiency levels of his learners.
Thus, the current innovative practice was conducted to address the problem of the stu-
dents’ lack of interest and engagement in English reading.

Implementation of the Innovation

In this innovation, a program called ‘CopyAI’ was employed as an automated text gen-
eration tool. CopyAI was developed by Yacoubian (2020) by using the aforementioned
language model GPT-3. CopyAI provides users with templates (e.g. blog introduction,
email, commercial copy, song lyrics) to create texts of various genres. In addition to
target keywords, a user can input additional conditions for his or her own story, such
as ‘very easy words’ and ‘short sentences’ as shown (see the left-hand image in
Figure 1). The user can also select tones such as ‘witty’, ‘friendly’ or ‘professional’
for the text.
The innovation was conducted in six consecutive sessions (see Figure 2 for a summary
of the procedure). In the first session, the pre-intervention survey was administered to
both the experimental and comparison groups for 15 minutes. The next four sessions dif-
fered for each group. That is, the comparison group was given the traditional reading
lessons based on the participants’ English textbooks, which involved reading the pas-
sages, going over the words and sentences in the passages, and completing the post-
reading comprehension tasks. On the other hand, the experimental group was given
the intervention based on the AICG-based activity, which is elaborated below.

Figure 1. The interface of the text generation menu in CopyAI (left) and the example of the text
generated by CopyAI (right).
Lee et al. 511

Figure 2. The procedure of the innovation (for the experimental group).

In the second session, the participants were briefly introduced to the AICG technology
and the reading activity based on the type of technology they would use in the forthcoming
sessions. The participants in each class were divided into the two groups, and given the
worksheets through which they were asked to decide on the four categories such as
genre, characters, place and events in collaboration for the story they would ask the AI to
generate. For example, regarding the genre, the worksheet included several examples for
the participants to choose: horror, science fiction (SF), comedy, romance, action, mystery,
fantasy and others. Several examples were also provided for the other categories (i.e. char-
acters, place and events). Once the participants in each group finished deciding on the exam-
ples for each category, their options (i.e. keywords) were typed into the AICG programs. In
doing so, in view of the participants’ limited level of English proficiency, the additional
options equipped in the program such as ‘very easy words’, ‘short sentences’ and ‘4
lines’ were also selected (see Figure 1). When the generated text was deemed rather difficult
for the sampled students, the third author revised the texts in terms of the vocabulary and
syntactic structures of the sentences to suit the English proficiency level of the students,
while trying to minimize the modification of the story content.
In the third and fourth sessions, the texts which were first generated by the AI and then
revised by the English teachers were used as reading passages. For each session, the teacher
512 RELC Journal 54(2)

and the students first previewed a list of difficult words and expressions included in the target
passage. Then, the teacher asked the students to predict what the story would be about. Next,
the teacher and students read the passage together and went over difficult words or sentences
in it, if any. Finally, the students were given the reading comprehension task sheet.
In the fifth session, the participants were asked to create their own storybook based on
one of the texts covered in either the third or the fourth session. In doing so, the participants
were also encouraged to add or change any content in the original text, the process of which
provided them with further opportunities to learn new words or practice writing original
sentences. Figure 3 displays the examples of the storybooks that the participants created.
In the sixth session, the post-intervention survey was given to both the experimental
and comparison groups, again for 15 minutes. The participants who volunteered for the
post-intervention interview were individually contacted and interviewed by their English
teachers, with each interview lasting between 5 and 10 minutes.

Evaluation of the Innovation

We adopted the pre-post design, including both the experimental (based on the AICG-based
activity) and comparison conditions (i.e. traditional learning condition based on textbooks).
We focused on two variables: the first variable was Foreign Language Enjoyment (FLE;
Dewaele and MacIntyre, 2014), which has been proposed to be significantly correlated
with several key L2-related variables such as willingness to communicate, foreign language
anxiety and academic achievement (Botes et al., 2022). The other target variable was the
learners’ interests in reading English books (i.e. textbooks and other books).
We first evaluated the value of our innovative practice by examining the two groups’
levels of FLE and interests in reading English books. Table 1 presents the descriptive sta-
tistics related to the experimental and comparison groups’ levels of FLE and interests in
reading English books at the pre-intervention and post-intervention.
Regarding the FLE, both the experimental and control groups showed a highly similar
level of this variable before the intervention, with a moderate-to-high level of FLE.
However, it was found that the means of the two groups’ levels of FLE were significantly
different after the intervention (p < .01).
The result of the other target variable (i.e. interest in reading English books) was
largely similar to that of the FLE. That is, the results of the t-test showed that the

Figure 3. The examples of the storybooks created in the fifth session.

Lee et al. 513

Table 1. The descriptive statistics of the two groups’ responses on the Foreign Language
Enjoyment and interests in reading English books.

Pre-intervention Post-intervention
Target variables Groups Mean (SD) Mean (SD)
Foreign language enjoyment Experimental group (n = 3.68 (.88) 3.84 (.85)
60)
Control group (n = 61) 3.56 (.89) 3.35 (.89)
Interests in reading English Experimental group (n = 3.14 (1.21) 3.47 (1.05)
books 60)
Control group (n = 61) 3.11 (.94) 2.81 (1.10)
SD = standard deviation.

means of the two groups’ responses to their interests in reading English books did not
show a significant difference before the intervention; however, the two groups’ differ-
ences reached a significant contrast after the intervention (p < .01).
We then examined the participants’ perception of the AICG-based activity. It should
be noted that, thus, the results of the survey and interview presented in this section only
concerns the participants in the experimental group, who engaged in the AICG-based
activity. Table 2 summarizes the results of the three post-intervention survey items.
Regarding the first item, approximately 78% of the participants found the reading activity
of generating the text based on AI to be more engaging than its traditional counterpart. The
post-intervention interview revealed that the English-as-a-foreign-language (EFL) learners
appreciated such an activity for various reasons including the AI’s ability to generate
stories based on their own ideas and engaging in communication with their group members
to brainstorm their own ideas and keywords, as can be seen in the following excerpts:

The activity was so much fun! It was so exciting to see how AI generated the story with my
keywords. (Student #6)
It was amazing to see how AI generated the story. (Student #22)
I liked making up stories with my classmates together. (Student #12)

It was also revealed that a majority of the learners (approximately 77%) believed that
they were able to focus more in English lessons compared to more traditional ones, as
their own texts were adopted. One of the participating teachers made a comment on
this issue as follows:

At the beginning, I was worried because my students did not have good attitudes … however,
overall the children responded positively to the contents of the class and engaged with the
activity well … I think they focused better in these activities than ordinary lessons.
(Teacher #4)

Finally, slightly more than 50% of the participants showed more willingness to read
English texts, if the texts were AI-generated ones based on their own ideas. Such texts
seem to be more engaging from the learners’ perspectives than the ones in their textbooks,
as can be seen in the following statement: ‘I definitely want to do another activity [AICG-
based activity] because it is more exciting than doing it with textbooks’ (Student #12).
514 RELC Journal 54(2)

Table 2. The responses of the participants in the experimental group to the items related to the
artificial intelligence-content generator-based activity (n = 60).

Percentage of the responses on

‘Agree’ and ‘Strongly agree’
Questionnaire item combined Mean (SD)
The activity of generating the text based on 78.3% 4.25 (.88)
artificial intelligence (AI) was more engaging
than the traditional reading activity.
It helped me focus on lessons when the texts 76.6% 4.17 (.89)
generated via AI based on my idea were used.
I am willing to do more reading if the target 53.3% 3.57 (1.21)
texts are ones generated via AI based on my
idea.
SD = standard deviation.

Overall, the participants showed favorable attitudes toward the AICG-based activity,
and felt more engaged in such a reading activity than its traditional counterpart, in English
lessons.

Future Pedagogical Directions

In this article, we shared an innovative practice based on the AICG that could be used
in L2 contexts and examined its pedagogical effects on young EFL learners. It was
found that this technology could enhance the level of young learners’ enjoyment of
learning English (Dewaele and MacIntyre, 2014) and their interest in reading
English books, both of which could contribute to their English learning in several
ways. Especially regarding the FLE, it is expected that building positive emotions
(i.e. enjoyment) would lead to positive learning outcomes (Botes et al., 2022), and
we believe that doing so would be particularly important for young learners, who
are at an earlier stage of English learning and would engage in such a process in the
longer term. On the other hand, in the case of the comparison group, it is possible
that the students were not in favor of the passages included in the textbooks, which
may have weakened their motivation to engage in reading English texts (Oldfather,
2002).
Our investigation has pedagogical implications for both learners and teachers. For the
former, it is expected that AICG could be a viable addition to the extensive reading enter-
prise (Cho and Krashen, 1994; Day and Bamford, 2002), by generating an unlimited
amount of new English texts for learners to read. With AICG, when learners find a par-
ticular theme to be interesting, they can access the AICG application, and generate ori-
ginal passages on the same theme. In this process of using AICG, learners are also
expected to develop their information literacy (Association of College and Research
Libraries, 2016) – an ability to identify, access, evaluate and use the type of information
(i.e. English reading texts that suit one’s interests in our context) one requires. For tea-
chers, it is expected that the AICG would significantly reduce the workload of English
teachers, by providing a wide range of original English texts that could be used in
reading lessons, and for assessment purposes.
Lee et al. 515

Based on our experiences, we suggest the following practical tips for teaching practi-
tioners who would like to employ this technology. First, regarding the ‘themes’ option in
CopyAI, it is recommended to input only the main characters. We have found that when
you put information (or details) in addition to the main characters, the overall content of
the text could sometimes deviate significantly from what the user intended. Second, it is
recommended that the user inputs as many details as possible regarding the difficulty (or
level) of texts into the target tool (e.g. kindergarten-level vocabulary, elementary words,
short sentences, five to six lines) if such texts are to be used for beginner-level students.
We have found that including only one or two can sometimes generate rather linguistic-
ally difficult texts. Finally, when a user inputs additional conditions for his or her own
story, it is recommended to include information related to the difficulty of the text
first, followed by the content of the story. When this order was switched, information
related to the difficulty of the texts could be recognized as content. For example, when
the event-related information (e.g. ‘chase’) was entered first, followed by the above-
mentioned ‘kindergarten level vocabulary’, the text on the chasing scene in the kindergar-
ten (as the venue of the story) was generated. Along with the tips mentioned above, it is
recommended that teachers experiment with the tool extensively to find the best combin-
ation of the options in the target AICG tool that would work best for their students.

Funding
The authors received no financial support for the research, authorship, and/or publication of this
article.

ORCID iDs
Jang Ho Lee https://orcid.org/0000-0003-2767-3881
Dongkwang Shin https://orcid.org/0000-0002-5583-0189

References
Association of College and Research Libraries (2016) Framework for information literacy
for higher education. Association of College and Research Libraries. Available at: http://
www.ala.org/acrl/sites/ala.org.acrl/files/content/issues/infolit/framework1.pdf (accessed 13
October 2022).
Botes E, Dewaele JM and Greiff S (2022) Taking stock: A meta-analysis of the effects of Foreign
Language Enjoyment. Studies in Second Language Learning and Teaching 12(2): 205–232.
Cho K-S and Krashen SD (1994) Acquisition of vocabulary from the Sweet Valley Kids series:
Adult ESL acquisition. Journal of Reading 37(8): 662–667.
Day RR and Bamford J (2002) Top ten principles for teaching extensive reading. Reading in a
Foreign Language 14(2): 136–141.
Dewaele JM and MacIntyre PD (2014) The two faces of Janus? Anxiety and enjoyment in the
foreign language classroom. Studies in Second Language Learning and Teaching 4(2):
237–274.
Ghumra F (2022) OpenAI GPT-3, the most powerful language model: An overview. e-Infochips.
Available at: https://www.einfochips.com/blog/openai-gpt-3-the-most-powerful-language-
model-an-overview/ (accessed 13 October 2022).
Lee J (2017) A study on creative writing by artificial intelligence and the destiny of the Author. The
Journal of Korean Fiction Research 68: 103–131.
516 RELC Journal 54(2)

Oldfather P (2002) Students’ experiences when not initially motivated for literacy learning.
Reading and Writing Quarterly 18: 231–256.
Shin D (2022) What is the limit for using AI in English education?: Feasibility of using Automated
Text Generation and Automated Item Generation. Journal of the Korea English Education
Society 21(3): 147–163.
Shin D and Lee JH (2023). Integrating AI-based Text Generation Technology into EFL Teaching
[Manuscript submitted for publication].
Yacoubian P (2020) CopyAI. CopyAI Inc. Available at: https://www.copy.ai/ (accessed 20 April
2022).