Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

2021 1st International Conference on Artificial Intelligence and Data Analytics (CAIDA)

Investigating the Effect of Chatbot-to-User Questions


2021 1st International Conference on Artificial Intelligence and Data Analytics (CAIDA) | 978-1-6654-1511-8/20/$31.00 ©2021 IEEE | DOI: 10.1109/CAIDA51941.2021.9425208

and Directives on Student Participation

Daniel Bailey Norah Almusharraf


English Language and Culture Department, Applied Linguistics Department,
Konkuk University Glocal Campus, South Korea College of Humanities, Prince Sultan University
268 Chungwon-daero, Dalcheon-dong, Chungju-si, Kingdom of Saudi Arabia, Riyadh, P.O. Box 66833
Chungcheongbuk-do nmusharraf@psu.edu.sa
dbailey056@kku.ac.kr ORCID ID: 0000-0002-6362-4502
ORCID ID: 0000-0003-0278-40

Abstract— Pedagogical chatbots are used to elicit information This study sought to compare the different levels of user
from students. Yet, how the amount of student-generated content output according to the chatbot-to-user question type and
differs according to the type of chatbot-delivered questions is directive by implementing a conversational agent that scaffolds
unknown. To this end, 19 South Korean English majors completed learners through their own storytelling process. Further, a
six chatbot assignments through an in-house developed Facebook comparison to user output between the first and final chatbot
Messenger chatbot. The chatbot activity entailed creating original activity was made to check for previously recognized novelty
stories for class presentations. In addition to directives requesting effects with chatbot participation [2]. Specifically, this paper
plot details, the chatbot used closed-ended button reply questions, centers on the types of questions and directives chatbots deliver
open-ended questions, and fill-in-the-blank template statements to
to students, how those questions and directives influence
help students create stories. Results indicated that button reply
questions allowed for pacing, recall and content assessment and
participation, and how that participation changes over time.
required low levels of critical thinking. Next, open-ended questions With these aims in mind, this paper asks the following questions:
and fill-in-the-blank template statements resulted in similar word 1. What are the rates of participation with a story
count production but different levels of creativity and critical generating chatbot activity?
thinking. Lastly, directives requesting user input resulted in 35%
more output, indicating students took more action when told to do 2. How do differences in chatbot-delivered questions and
something than when asked. Regarding the novelty effect, fewer directives influence participation?
students volunteered to do the sixth and final chatbot activity, but
those who did produced word count on par with their initial 3. How do user responses change over the period of an
chatbot activity. academic semester?

Keywords— Chatbot, conversational agent, chatbot-assisted II. LITERATURE REVIEW


language learning, computer-mediated communication. A. Theoretical Framework: Bloom’s Taxonomy
I. INTRODUCTION Bloom’s Taxonomy [3] was termed after Benjamin Bloom
and initiated some educational objectives and assessments for
With the outstanding expansion of educational technology institutions to utilize [4]. Bloom's revised taxonomy of
throughout the past decade, chatbots are evolving into an educational purposes [5] is a tool that can aid the determining
inexorably prominent alternative method for communicating learners' cognitive development from lower-order thinking, at
with students, and their appropriation is rapidly spreading [1]. the fact and concept levels, to higher-order thinking. The revised
Innovation leads to improved chatbot platforms that can be taxonomy entails six levels of increasing cognitive complexity
trained to respond to user-input with reasonable accuracy, by which learners comprehend content: knowledge,
resulting in user satisfaction and flexibility in functions, understanding, application, analysis, evaluation, and creation.
however, testing methods to mitigate miscommunication and The first two levels reflect lower-order thinking (i.e.,
optimize learner engagement need attention. Further, research remembering and understanding), while the other four reflect
exploring the persuasive power chatbots have on eliciting progressively higher-order thinking, including analyzing,
participation is an opening frontier. evaluating, and creating [6].

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.
Previous studies have spectacled the usage of chatbots in III. METHODS
learning English. In particular, a study by Chocarro et al. [7]
examined teachers’ attitudes towards chatbots in education by A. Overview
considering the effect of social language, bot proactiveness, and This single sample exploratory study followed a descriptive
users’ characteristics on learner perceptions. The results analysis method to first measure student-generated output
confirmed a positive and substantial influence of the perceived according to chatbot questions and directives to EFL students.
practicality for using chatbots in educational settings. Similarly, Then, how that output changes over time was observed. The
Chen et al. [8] conducted a study with 58 students, which found chatbot used button reply questions, open-ended questions,
chatbots significantly improved the students’ learning close-ended template statements, and directives to elicit the
attainment and found perceived usefulness as the predictor of target language's use (i.e., English).
behavioral intention; however, participants agreed that while
The chatbot activities were administered during the fall
chatbots benefited their learning of Chinese vocabulary, several
semester of 2020 when universities in South Korea were taught
adjustments to functionality, including ease-of-use, need to be
online due to the COVID-19 pandemic. Chatbot logs were
made for further progress.
collected from 19 South Korean (male= 9, female=10) junior
B. Challenges with Chabot-User Interactions and senior English majors attending a multimedia
Although chatbots are a great approach to getting a closer communication course. Participation was defined as content
look into any language that the user hopes to learn and develop generated as measured by word count, which is a useful metric
over time, some argue that they cannot be reliable without for calculating chatbot-user interactions [12], [13]. Overall,
human supervision [9]. Further, another recognized problem students completed six chatbot activities aimed to help them
with current chatbot interactions is the lack of cohesion during create original content for class presentations and small group
the session due to incorrect pair-matching. In EFL, chatbots like discussions. The chatbot would scaffold students through
Mistuku (www.mitsuku.com) and Tutor Mike (www.rong- sessions by asking them questions related to the task. The
chang.com/tutor_mike.htm) have been used as communication chatbot guided students with various questions and directive
channels, often resulting in limited and superficial interactions statements to construct an original story with the main character,
that lack cohesion from one statement to the next. And while the setting, conflict, and resolution. At the end of the chatbot
accuracy of the chatbot responses to user-input continues to session, student responses were returned to them in a single
improve over time, the session topics are random, short-lived, collocated message, which would be used as notes for a class
and often have no point. Too often chatbots go through system presentation.
fails, giving wrong suggestions to the student searching for a B. Materials
specific topic or giving the student unreliable information.
The chatbot activities were created with the ActiveChat.ai
Strategic use of bot-to-user questions is a proposed solution chatbot platform and then integrated with Facebook Messenger.
to superficial, meaningless, chatbot interactions. When All students had Facebook Messenger installed on their phones,
messaging with a chatbot, a combination of question types is and chatbot participation occurred during regular class hours..
recommended so that the student is not either bored with simple
1) Question types and directives
recall questions or overwhelmed with cognitively demanding
questions requiring higher-order thinking skills [10]. Through The purpose of the chatbot questions and directives was to
higher-level questions, learners create original content by collect information from the user. Questions types included
processing beforehand learned material or information [11]. button responses (e.g., yes/no), open-ended questions seeking
Questions igniting higher-order thinking are more practical for specific information (e.g., name, location, time, and weather),
reassuring students to think deeper and inspire them to hunt for open-ended questions asking for plot details, directives
information autonomously. Still, lower-level questions are requesting plot details, and directives telling students to fill out
suitable for assessing learners’ preparation or reviewing content. template statements for plot details that students can model their
responses or replace completely with original details. Table 1
Bailey et al. [23] administered a series of chatbot activities displays examples of questions asked and directives given
for digital storytelling. Students would interact with a Facebook during the “Create Story” activity.
Messenger bot that controlled the pace and plot of a story, asking
students to predict plot details and recall story information. TABLE I. QUESTIONS ASKED ACCORDING TO CHATBOT REQUEST
Results indicated cohesion throughout the chatbot sessions with Example button-reply questions (Recall Questions)
easily corrected instances of miscommunication [23]. Overall, 1 Have you visited my storytelling library and read some of my stories?
the digital storytelling chatbots proved to be appropriate for EFL 2 Do you understand characters, setting and conflict?
reading practice but were limited in their writing practice
Example open-ended questions eliciting specific information (Analyzing
capacity and this was attributed to limited use of bot-to-user Questions)
questions. 1 How old is your character?
To support a chatbot-assisted learning environment 2 Is your character a male, female or something else?
conducive to writing practice, we propose a framework, Open-ended question (Creating Questions)
supported by Bloom’s revised taxonomy, for a type of chatbot 1 How else would you describe your character?
interaction, which functions as a framework for the analysis of 2 What problem in life does your character have?
content emanating from chatbot-to-user questions directives. Fill-in-the-blanks question (Fill in the blank analyzing question)

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.
1 Now it is your turn. Fill in the blank. "I am going to tell you a story
about ________________" Research question 1 revealed that students wrote more when
2 Now it's your turn. Fill in the blank with your own answer. "I went to
___________ to meet
asked more questions. Now, question 2 asks how participation
__________________________________________." differs according to question type, fill-in-the-blank statements,
Chatbot-to-user directives and directives. Table 3 displays ANOVA results for word count
1 Now it is your turn. Write the 4th line of your story. I'm not going to concerning the bot's different types of questions to the students.
give you any help. Do this line by yourself.
Stark differences were revealed between the content
2 Now it is your turn. Write a line in your story that introduces conflict. produced according to question-type. Directives resulted in the
a.
Note: a Close-ended button reply statement led to an open-ended response by students with most student-generated content (M=34.55, SD=19.41). When
considerable output.
told, not asked, students committed to more work resulting in
C. Data Analysis more product. Button response questions (BRQ) and specific
information questions (SIQ) overwhelmingly resulted in
The statistical package for social science (SPSS; Version 25)
answers with a single word or short phrase but rarely a complete
was used to calculate descriptive statistics. For research question
sentence. These questions often inquired towards background
1, data were sourced from 19 chatbot session logs, the first
information about characters (e.g., age and name) and settings
“Create Story” activity. The participation rate was measured as
(e.g., coffee shop). When given the opportunity, students
total bot messages, non-question button replies, bot-to-user
avoided writing complete sentences and instead wrote a single
questions and directives, and user-to-bot response. For research
word or number. With an average output of approximately 19
question 2, student responses to the corresponding question
words per question (SD=16.94), open-ended questions resulted
types were tagged as button response questions (BRQ), specific
in substantially more content generated by students than button
information questions (SIQ), open-ended questions (OEQ),
response questions and questions asking for specific
directive statements (DS), and template style questions (TSQ).
information. Further, open-ended questions were answered in
Next, analysis of variance (ANOVA) followed by Bonferroni
complete sentences. Template style questions shared similar
post hoc analyses were used to check for significant differences
output levels as open-ended ones with approximately 17 words
between the amount of content produced by the different chatbot
per question (SD =13.02). Students also wrote in complete
questions and directives. For research question 3, 10 chatbot
sentences; however, less word variation was recognized because
session logs from activity six, the second “Create Story”
students often copied the given template instead of creating
activity, were coded following steps taken for the initial 19
original content.
session logs. Then pre/post-analysis of word count for the
different question types and directives were measured between TABLE III. ANOVA RESULTS FOR WORD COUNT FOR DIFFERENT
the second chatbot activity (i.e., first create story activity) in QUESTION TYPES
week three to the sixth chatbot activity (i.e., second create story
activity) in week 14. N M SD SE
BRQ 247 1.23 0.48 0.03
IV. RESULTS SIQ 149 3.32 3.83 0.31
For research question one, rates of participation were OEQ 113 19.21 16.94 1.59
measured. Table 2 displays descriptive statistics for DS 105 34.55 19.41 1.89
participation metrics emanating from the initial “Create Story” TSQ 96 16.93 13.02 1.33
chatbot activity (n=19). Overall, approximately 35% of
Total 997 8.63 14.61 0.46
messages were sent from students accounting for 21% of the
total word count produced during the session. Chatbots sent 65% b.
Note: BRQ, button response questions; SIQ, specific information question; OEQ, open-ended
questions; DS, directive statements; TSQ, template style question.
of the messages resulting in the remaining 79% of the content.
Further, the bot sent 709 total questions averaging 37 questions Research question 3 addresses possible novelty effects
per activity. Of the 709 questions, 477 required written, open- resulting in decreased participation levels between the first and
ended, user responses, and 244 entailed closed-ended choices. last “Create Story” activity. The first bot activity resulted in all
Non-question button choices were used for pacing the flow of 19 students completing the activity. This number dropped by the
information to the student. In contrast to the research leading up last chatbot activity, which resulted in only 10 of the 19 students
to the current stud [23], students produced more words and sent participating. Once engaged with the activity, the amount of
more messages, and this was attributed to the increased writing by the student was on par with the first “Create Story”
frequency of questions asked and directives given. activity. Students wrote similar amounts of content, which was
consistent for the questions and directives being addressed but
TABLE II. DESCRIPTIVE STATISTICS FOR PARTICIPATION METRICS not for the fill-in-the-blank statements. Table 5 shows
(N=19) descriptive statistics of content generation between the pre and
post-activity. For open-ended questions and directive
Message
M SD statements, participation trended upward at the end of the
s course; however, differences were not statistically significant.
Bot Messages 1854 17.81 9.42
Questions 709 15.29 7.89
The opposite trend was observed for template style questions,
User Messages 1009 8.62 14.38 which showed a 25% drop in words per question.
Open-ended questions 477 16.83 17.58
To test for significance in change between the first and
Button Questions 244 1.21 0.41
second “Create Story” activity, a series of independent t-tests

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.
were performed. While the mean count differences were robust, V. DISCUSSION
the differences were not statistically significant, attributed to Several findings emanated from this study. In particular,
high variation in student responses. findings contributed to missing attention concerning chatbot-
delivered questions and directives in chatbot dialogues [14].
TABLE IV. BONFERRONI POST HOC ANALYSIS OF STUDY QUESTION
TYPE RESPONSES Student content varied according to whether they were
answering a particular question type or responding to a directive.
MD SE p Low High Evidence for both lower and higher levels of critical thinking
BRQ SIQ -2.08 0.99 0.53 -4.99 0.82 were apparent in the students’ interactions in the form of shorter
OEQ -17.98 1.08 0.00 -21.16 -14.80
and more extended responses. Open-ended questions and
directives required analyzing and creating original content,
DS -33.32 1.11 0.00 -36.58 -30.05
therefore, relying on deeper critical thinking levels. Contrarily,
TSQ -15.69 1.14 0.00 -19.06 -12.32 button-response questions and questions asking for specific
SIQ BRQ 2.08 0.99 0.53 -0.82 4.99 details required recall of specific information, therefore, relying
OEQ -15.90 1.19 0.00 -19.39 -12.40 on lower levels of critical thinking and a faster pace in the
DS -31.24 1.21 0.00 -34.81 -27.67 activity. Fill-in-the-blank template style questions resulted in
TSQ -13.61 1.25 0.00 -17.28 -9.95 similar levels of words per question as open-ended questions.
OEQ BRQ 17.98 1.08 0.00 14.80 21.16
Low word variation with fill-in-the-blank template style
questions was attributed to students borrowing language directly
SIQ 15.90 1.19 0.00 12.40 19.39
from the template.
DS -15.34 1.29 0.00 -19.14 -11.54
DS BRQ 33.32 1.11 0.00 30.05 36.58 In response to research question 1: What are the rates of
SIQ 31.24 1.21 0.00 27.67 34.81
participation with a story generating chatbot activity?
participation resulted in increased levels of writing by students
OEQ 15.34 1.29 0.00 11.54 19.14
compared to past chatbot studies that relied on less variation and
TSQ 17.63 1.34 0.00 13.67 21.58 frequency in chatbot-to-user questioning behavior [15]. Chatbot
TSQ BRQ 15.69 1.14 0.00 12.32 19.06 activities requiring students to create original stories resulted in
SIQ 13.61 1.25 0.00 9.95 17.28 a substantial amount of user output, and this was partly
DS -17.63 1.34 0.00 -21.58 -13.67 attributed to questions asking students to analyze, evaluate, and
c.
create content. There was a combination of both long and short
Note: BRQ, button response questions; SIQ, specific information question; OEQ, open-ended
questions; DS, directive statements; TSQ, template style question. answers depending on whether students were addressing
questions or directives requiring lower or higher critical thinking
levels. A similar pattern of results was obtained in Rahman and
TABLE V. MEAN SCORE COMPARISON BETWEEN PRE AND POST Abdul Manaf’s [6] study that indicated lower-order thinking is
“CREATE STORY” ACTIVITIES a precondition for higher-order thinking to analyze, evaluate, or
generate information. Button reply options served the purpose
Question Pre and
Type Post
N M SD SE of both guiding low-level thinking skill questions and pacing
through information flow, but perhaps most importantly,
Pre 45 3.87 4.74 0.71 provided the foundation for deeper and more critical level
SIQ
Post 41 4.00 4.33 0.68 thinking later on in the activity.
Pre 33 18.58 16.43 2.86
OEQ Further, button reply options quickened the story's pace,
Post 31 20.13 16.413 2.95 which Tallyn et al. [16] recognized as a valuable attribute to
Pre 28 34.54 20.36 3.85 button reply questions. The next finding concerning
DS participation pertained to communication breakdown and added
Post 31 37.32 19.17 3.44
cohesion throughout the chatbot session. Giving control of the
Pre 32 20.56 16.04 2.84
TSQ conversation to the chatbot system to amend instances of
Post 25 15.20 12.75 2.55 miscommunication through the use of a closed-system dialog
d.
Note: BRQ, button response questions; SIQ, specific information question; OEQ, open- chatbot mitigated communication breakdown recognized in past
ended questions; DS, directive statements; TSQ, template style question. studies [10], [17]. Further, having the activity relate to a single
topic, such as creating a story, added the missing cohesion in
TABLE VI. INDEPENDENCE T-TEST FOR PRE AND POST RESPONSES TO chatbot-user interactions recognized in past chatbot research [2].
DIFFERENT QUESTION TYPES
Research question 2: How do differences in the types of
p t df p MD SE chatbot-delivered questions and directives influence
SIQ .533 -.136 84 0.890 -.130 .980 participation? observed the different influence types of
OEQ .545 -.378 62 0.710 -1.55 4.11 questions and directives have on word count. A key finding
DS .413 -.541 57 0.590 -2.79 5.15 emerged from our study indicated that directives resulted in the
TSQ .486 1.367 55 0.177 5.36 3.92 most extended messages and provided the most details for the
created stories in comparison to button reply questions, open-
ended questions, and fill-in-the-blank template style questions.
More words equated to more ideas, cohesion between ideas, and

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.
this finding indicated directives are responsible for more research. These findings add to the growing body of evidence
profound levels of critical thinking than actual questions. supporting the use of chatbot activities in flipped, blended, and
online learning environments [21], [22]. In the EFL context,
Directives, in the form of the chatbot telling the student what mainly, chatbots can be used to preview vocabulary and key
to do, resulted in the most action taken. This result ties well with phrases that can be later practiced with a live partner [10]. In
previous studies wherein imperative statements resulted in the other academic domains, the possibilities are only constrained
most word count in responses, and longer messages resulted in by one’s imagination. Chatbots are, in essence, nothing more
more detailed stories [14], [18]. This result of directives having than an interactive website that builds itself around the user,
a more considerable impact on work than questions has through mapping reference points, knowledge level, and
significance in EFL because content generation equates to target participation patterns.
language awareness and consequential second language
acquisition [19]. Further, directives' greater use can compensate In classroom settings, the chatbot activity's content has
for the richness lacking in current chatbot-human added value when it can be utilized in real-world settings. In
communication [20]. comparison to the Bailey et al.’s [23] preceding study on
storytelling chatbots, students produced significantly more
Both open-ended and fill-in-the-blank question styles had content, which was directly attributed to the increase in
similar word count levels; however, students borrowed language
questions and question types. With storytelling, students
from the fill-in-the-blank template statements resulting in less answered hypothetical (e.g., what would you do if this happened
creative storylines than if students were responsible for to you?) and predictive (e.g., what would you do next) style
producing an initial response without an example (i.e., help). questions but did not create content for their own story,
Open-ended questions seeking specific information resulted in producing no original content to present in the classroom after
short phrases (e.g., my house) or single words (e.g., daytime) finishing the chatbot activity. In contrast, students answered a
with a few students writing simple sentences (e.g., in my house). variety of questions to help create an original story for
Consequently, the story's pace moved much faster and required presentation purposes.
less cognitive load with button-reply responses and short answer
open-ended questions. To this end, directives, open-ended Chatbots that ask various questions types and directives
questions and directives are recommended to be placed at the contribute to the current set of chatbot activity in EFL that have
end of chatbot activities, while short answer and button-reply shown benefits to learning [10], [24], [25], [26]. A combination
questions should be used earlier in the chatbot session in attempt of question types is recommended for pedagogical chatbot
to grab the user’s interest. systems [10]. The student is neither bored with simple recall
questions nor overwhelmed with open-ended questions and
Research question 3: How do user responses change over directives. Closed-ended button reply questions can then be
an academic semester? compared participation levels between followed by open-ended questions that require recall
the second and sixth chatbot activity over the course of the information like names, locations, or specific information
semester. Both activities were identical in design and guided requiring one-word to single phrase answers. Once engaged,
students through the process of writing an original story with chatbot activities can use fill-in-the-blank questions to scaffold
characters, settings, conflict, and resolution. Findings echo Yin the behavior of writing complete sentences. Open-ended
and Satar’s [13] chatbot study that found lower proficient EFL questions that require students to generate original ideas should
students engaged less with the chatbot system. In the current be introduced. In the second half of the chatbot activity,
study, all students were asked to participate through Facebook directives requesting students to produce original content should
Messenger. In five cases, students interacted with the chatbot be given. Button reply questions can be used more liberally. In
with less than five messages, out of 50+ interaction contrast, general open-ended questions and directives should be
opportunities. In four other cases, students never responded at used strategically, with each use contributing in a significant
all. This activity was administered during class time, and their way to the final product.
instructor requested students to participate. The class was
conducted through the video conferencing platform zoom Fryer [2] states that “teachers should be aware that learners’
during the fourteenth week of the semester, and the motivation to interact with chatbots may decrease as the novelty
observational data recognized participation drop-off by several effects diminish.”, therefore, a variety of chatbot activities need
of the same students who chose to either underperform or not to be explored and categorized. In essence, the chatbot activity
participate in the sixth and final chatbot activity. Students that can be viewed as an interactive worksheet, and just as worksheet
chose not to participate in the chatbot activity were many of the activities are specific to learning objectives, so are chatbot
same students who displayed low or no participation in other activities. In the future, chatbot activities will become more and
multimedia activities during the semester, indicating student more idiosyncratic to the learners’ needs as designing activities
characteristics, such as those related to L2 proficiency, self- becomes more manageable for teachers to use.
efficacy, or intrinsic motivation may have influenced the
novelty effect rather than the value chatbots have towards VI. CONCLUSIONS
language learning. Future research is needed to better In this study, it was recognized that students produced
understand the phenomenon of continued chatbot use. original stories with the assistance of an in-house developed
A. Pedagogical Implications and Future Research chatbot. A few significant takeaways emanated from this study.
Moving forward, continuing awareness should be placed on how
This study was exploratory in nature, however, resulted in questions and directives elicit varying levels of user-generated
findings that apply to pedagogical chatbot systems and future output in chatbot learning activities. Content written by students

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.
was proportional to the level in which the question appeared on considering the effect of social language, bot proactiveness, and users’
Bloom’s revised taxonomy, with less time spent on recall characteristics,” Educational Studies, pp. 1–19, 2021.
questions and the most time dedicated to questions asking [8] H.-L. Chen, G. V. Widarso, and H. Sutrisno, “A ChatBot for Learning
Chinese: Learning Achievement and Technology Acceptance,” Journal
students to use creative thinking skills. Chatbot-to-user of Educational Computing Research, vol. 58, no. 6, pp. 1161–1189, 2020.
directives resulted in the most content produced. These [9] A. Veglis and T. A. Maniou, “Chatbots on the Rise: A New Narrative in
directives explicitly asked students to create original story Journalism,” Studies in Media and Communication, vol. 7, no. 1, p. 1,
content and therefore were placed at the tip of Bloom’s 2019.
taxonomy. [10] Y. Goda, M. Yamada, H. Matsukawa, K. Hata, and S. Yasunami,
“Conversation with a Chatbot before an Online EFL Group Discussion
Several limitations were recognized with this study, and the Effects on Critical Thinking,” The Journal of Information and
including a small sample-size of junior and senior South Korean Systems in Education, vol. 13, no. 1, pp. 1–7, 2014.
university English majors. Therefore, further research can [11] J. Pereira, “Leveraging chatbots to improve self-guided learning through
examine the factors affecting students’ engagement by conversational quizzes,” in Proceedings of the fourth international
comparing chatbots with other activities and with different types conference on technological ecosystems for enhancing multiculturality ,
pp. 911–918.
of students. A larger sample of participants needs to be gathered
[12] S. H. Qashoa, “Effects of teacher question types and syntactic structures
in future studies in order to provide more valuable and widely- on ESL classroom interactions,” The International Journal of Social
utilized insights into chatbot technology in the blended and fully Science, vol. 7, no. 1, pp. 52–62, 2013.
online learning environment. Comparisons between questions [13] Q. Yin and M. Satar, “English as a foreign language learner interactions
should be observed in L1 settings, with different age groups and with chatbots: Negotiation for meaning,” International Online Journal of
geographic locations. Further, consistency in findings with Education and Teaching (IOJET), pp. 390–410, 2020.
educational research is imperative before making any [14] R. G. Athreya, A. C. Ngonga Ngomo, and R. Usbeck, “Enhancing
procedure-base conclusion. Community Interactions with Data-Driven Chatbots--The DBpedia
Chatbot,” in Companion Proceedings of the The Web Conference 2018,
Further work is certainly required to disentangle these pp. 143–146.
complexities in exploring functional websites for additional [15] D. R. Bailey. Chatbots as conversational agents in the contxt of language
evidence to enhance the existing knowledge base. In addition, learning. Proceedings of The Fourth Industrial Revoloution and
research is required to confirm these original findings regarding Education 2020, pp. 32- 120.
chatbots’ usefulness. Cultivating chatbots' performance and [16] E. Tallyn, H. Fried, R. Gianni, A. Isard, and C. Speed, “The Ethnobot:
Gathering Ethnographies in the Age of IoT,” in proceedings of the 2018
usefulness is an essential determinant for users when CHI Conference on Human Factors in Computing Systems, 2018, pp.1–
considering adopting chatbot technology [7]. Merging relevant 31, doi: 10.1145/3173574.3174178.
data from all the stored documents may empower the system to [17] S. Ghose and J. J. Barua, "Toward the implementation of a topic specific
yield answers to complex questions. Features in communication, dialogue based natural language chatbot as an undergraduate advisor,"
such as the quality of user responses, apply across different bot 2013 International Conference on Informatics, Electronics and Vision
designs because it is not the bot's sophistication in question but (ICIEV), Dhaka, Bangladesh, 2013, pp. 1-5, doi:
10.1109/ICIEV.2013.6572650.
instead bot communication strategies, including the analysis of
[18] N. Y. Kim, Y. S. Cha, and H. undefined Kim, “Future English learning:
bot-to-user question types and commands. Chatbots and artificial intelligence,” Multimedia-Assisted Language
ACKNOWLEDGEMENT Learning, vol. 22, no. 3, pp. 32–53, 2019.
[19] M. Swain, “The output hypothesis and beyond: Mediating acquisition
The authors would like to thank Prince Sultan University through collaborative dialogue,” in Sociocultural theory and second
(PSU) for funding this research project under grant number language learning, Oxford University Press, pp. 97–114.
IRB-CH-2021-6-1 and in affiliation with the Applied [20] J. Hill, W. R. Ford, and I. G. Farreras, “Real conversations with artificial
Linguistics Research Lab. intelligence: A comparison between human–human online conversations
and human–chatbot conversations,” Computers in human behavior, vol.
49, pp. 245–250, 2015.
REFERENCES
[21] W. Huang, K. F. Hew, and D. E. Gonda, “Designing and evaluating three
[1] D. Laurillard and A. Masulis, Rethinking university teaching: A chatbot-enhanced activities for a flipped graduate course,” International
conversational framework for the effective use of learning technologies. Journal of Mechanical Engineering and Robotics Research, vol. 8, no. 5,
Routledge, 2013. pp. 1–6, 2019.
[2] L. K. Fryer, M. Ainley, A. Thompson, A. Gibson, and Z. Sherlock, [22] J. Jia, “CSIEC: A Computer Assisted English Learning Chatbot Based on
“Stimulating and sustaining interest in a language course: An Textual Knowledge and Reasoning,” Knowledge-Based Systems, vol. 22,
experimental comparison of Chatbot and Human task no. 4, pp. 249–255, 2009.
partners,” Computers in Human Behavior, vol. 75, pp. 461–468, 2017.
[23] D. R. Bailey, A. Southam, & J. Costley (2021). Digital storytelling with
[3] B. S. Bloom (1956), Taxonomy of educational objectives. Allyn & Bacon. chatbots: Mapping L2 participation and perception patterns. Interactive
[4] D. R. Krathwohl, “A Revision of Blooms Taxonomy: An Technology and Smart Education, online journal: DOI 10.1108/ITSE-08-
Overview,” Theory Into Practice, vol. 41, no. 4, pp. 212–218, 2002. 2020-0170
[5] B. S. Bloom, L. W. Anderson, and D. R. Krathwohl, A taxonomy for [24] D. Coniam, “Evaluating the language resources of chatbots for their
learning, teaching, and assessing a revision of Blooms Taxonomy of potential in English as a second language,” ReCALL, vol. 20, no. 1, pp.
educational objectives. New York: Longman, 2001. 98–116, 2008.
[6] S. A. Rahman and N. F. A. Manaf, “A Critical Analysis of Bloom’s [25] L. Fryer and R. Carpenter, “Emerging Technologies Bots as Language
Taxonomy in Teaching Creative and Critical Thinking Skills in Malaysia Learning Tools,” Language Learning & Technology, vol. 10, no. 3, pp.
through English Literature,” English Language Teaching, vol. 10, no. 9, 8–14, 2006.
p. 245, 2017. [26] Höhn, Artificial Companion for Second Language Conversation.
[7] R. Chocarro, M. Cortiñas, and G. Marcos-Matás, “Teachers’ attitudes Springer International Publishing.
towards chatbots in education: a technology acceptance model approach

Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 15:14:49 UTC from IEEE Xplore. Restrictions apply.

You might also like