Exploration on Integrating Accessibility into an AI Course

Full text

Exploration on Integrating Accessibility into an AI Course
Chia-En Tseng
tsengc2@wwu.edu
Western Washington University
Bellingham, Washington, USA
Seoung Ho Jung
jungs7@wwu.edu
Western Washington University
Bellingham, Washington, USA
Yasmine N. Elglaly
elglaly@wwu.edu
Western Washington University
Bellingham, Washington, USA
Yudong Liu
liuy2@wwu.edu
Western Washington University
Bellingham, Washington, USA
Stephanie Ludi
stephanie.ludi@unt.edu
University of North Texas
Denton, TX, USA
ABSTRACT
Understanding accessibility and how it relates to articial intelli-
gence (AI)-based technology is an imperative skill for computing
students from both an ethical and employment standpoint. Un-
fortunately, AI courses do not typically cover accessibility. When
teaching ethics in AI, discussions on bias and fairness cover user
diversity in terms of gender and race, but not disability. To address
the lack of teaching accessibility in AI courses, we conducted a
pilot study to explore what and how accessibility topics can be inte-
grated into an AI course, titled Natural Language Processing (NLP).
We added to the NLP course some general and quick accessibility
topics through means such as a short guest lecture, a programming
assignment and a nal project that connect accessibility and AI. We
gathered student feedback through a pre-survey, a post-survey and
interviews with ve students. The course we looked at took place
synchronously in an online setting using Zoom. In this context,
we observed how implementing accessibility topics remotely into
the course aected students in their knowledge of accessibility and
disability.
The results did not show signicant improvement on students’
understanding on accessibility after completing the course. We
discussed the limitations of our work, and our plans to continue
the exploration on how AI courses can cover accessibility topics, so
that computing students become better equipped and more aware
about inclusive AI-technology development.
CCS CONCEPTS
•Human-centered computing →Accessibility
;
•Social and
professional topics →Computing education.
KEYWORDS
accessibility, articial intelligence, computing education
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specic permission and/or a
fee. Request permissions from permissions@acm.org.
SIGCSE 2022, March 3–5, 2022, Providence, RI, USA
©2022 Association for Computing Machinery.
ACM ISBN 978-1-4503-9070-5/22/03. . . $15.00
https://doi.org/10.1145/3478431.3499399
ACM Reference Format:
Chia-En Tseng, Seoung Ho Jung, Yasmine N. Elglaly, Yudong Liu, and Stephanie
Ludi. 2022. Exploration on Integrating Accessibility into an AI course. In
Proceedings of the 53rd ACM Technical Symposium on Computer Science Edu-
cation V. 1 (SIGCSE 2022), March 3–5, 2022, Providence, RI, USA. ACM, New
York, NY, USA, 7 pages. https://doi.org/10.1145/3478431.3499399
1 INTRODUCTION
In response to the rising need for computing professionals that can
build smart systems, many colleges and universities added Articial
Intelligence (AI), Machine Learning (ML), and Data Science courses
to their curriculum. Unfortunately, many of these courses do not
cover ethics [
9
,
14
]. While smart systems have the potential to
support people with disabilities, several mainstream smart systems
are almost obsolete to people with disabilities [
15
]. This is because
accessibility is usually an afterthought in software development in
general [27], and in AI-systems in particular [32].
Accessibility has been taught in several computing courses [
4
],
such as programming courses [
11
,
19
], and design courses [
24
].
Other researchers created a whole course on accessible comput-
ing [
10
]. However, AI courses did not receive such an attention
on incorporating accessibility into their curriculum despite the
extensive research on teaching ethics in AI courses [
17
]. Ethical
considerations in AI usually look at the diversity of users in terms
of gender and race, but not disability [
25
,
32
], and so teaching ethics
in AI followed the same pattern of excluding accessibility from the
discussion [
6
,
17
]. Hence, we are seeking to establish accessibility
teaching materials for AI courses and to increase student awareness
on accessibility topics in the realm of AI and machine learning.
The research was conducted in the Natural Language Processing
(NLP) course which is an elective course in the computer science
program at Western Washington University. The teaching meth-
ods that were found eective in teaching accessibility included
lectures, projects, and interaction with individuals with disabilities
[
33
]. Hence, we incorporated a guest lecture, a programming as-
signment, and a nal project that cover AI topics with emphasis on
accessibility to improve students’ learning of accessibility. The guest
lecture introduced not only the basic concepts of accessibility, but
also examples of AI systems that are used as assistive technology,
and an example of a non-inclusive AI system that demonstrated the
importance of accessibility. The assignment and the nal project
were designed to contain accessibility features while keeping the
integrity of the NLP learning objectives.
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
864

We collected students’ feedback using pre- and post-survey, in
addition to follow-up interview after the end of the course. Our
ndings indicated that our interventions had small impact on stu-
dents’ knowledge on accessibility. With the qualitative analysis
of the interviews, we were able to plan for further improving the
interventions in future oerings of the course. The contributions of
this work are: (1) Presentation of an AI assignment and a list of AI
projects that cover accessibility and do not require extra preparation
from the AI instructor. (2) Empirical ndings on the eectiveness of
our approach, and discussion on improvement plan of the teaching
materials.
2 RELATED WORK
2.1 Teaching Accessibility in Computing
Teaching accessibility in engineering and design courses has been
a growing topic of discussion both within academia and beyond,
as schools and companies realize the need to assist students in
not only acquiring technical skills but also human-centred ones.
There is an increasing need in awareness for worldwide accessibility
standards of technology, and an increased demand for engineers and
designers who understand diverse social and ethical responsibilities
pertaining to technology innovation.
Though there is such a need in industry, recent papers have
reported a lack of cohesive accessibility teaching methods in higher-
educational institutions for computer science courses, where focus
mainly centres towards courses on human-computer interaction
or on web development as opposed to AI-specic courses [
7
,
8
,
20
].
There have been longitudinal and short-term studies that docu-
mented accessibility teaching interventions such as lectures, project-
based learning, working with individuals who have disabilities, and
media-based learning such as videos and movies [
24
,
28
,
33
]. In our
study, we employed similar teaching interventions, i.e., lecture and
assignments, but in an AI course.
2.2 Teaching Ethics in AI Courses
Studies have been conducted on the teaching of ethics to students
in computer science courses, including ethics in relation to the
elds of AI and machine learning [
6
,
16
,
30
]. However, though stud-
ies mentioned developing courses to teach students about ethical
reasoning, ethical design, and ethical implementation, they do not
seem to mention ethics in terms of accessibility or universal design.
From previous literature we see that there have been studies con-
ducted on teaching accessibility in mainly HCI or web-development
courses with various teaching methods. In addition, we also see that
ethics within AI or machine learning courses are being discussed,
but often not within the context of accessibility. Thus, through this
research we address the gap of specically incorporating accessi-
bility teaching interventions into AI courses.
2.3 Bringing Accessibility to Students
Attention
Studies show that accessibility related interventions result in an in-
crease of students’ awareness and knowledge on the topic [
7
,
8
,
24
].
In order to encourage students to reect more on accessibility, it
is required to have them exposed to accessibility related topics.
Increasing awareness would motivate students to consider accessi-
bility measures in the future which could also increase their level
of knowledge and empathy [
21
]. Higher motivation may provoke
students’ desire to voluntarily acquire knowledge [
19
] and would
be able to develop their sense of empathy. We may increase the
motivation by focusing on students’ personal interest [
1
] or by cre-
ating a concrete connection between accessibility related topics and
industry-related skills [
22
]. Since the main course objectives for AI
courses often do not cover accessibility, we studied how to include
accessibility topics in an AI course through multiple interventions.
3 METHOD
We implemented three interventions to cover accessibility in the
Natural Language Processing (NLP) course, taught in Spring 2021.
This is a 4-credit, 10-week, elective course in a quarter-based system
oered for students in the computer science major. The course was
taught synchronously on Zoom due to COVID-19. The enrollment
capacity was 35 and the class stayed full throughout the quar-
ter. Fundamental concepts, algorithms and applications regarding
natural language processing are covered in this class. Program-
ming assignments and projects are introduced for students to gain
hands-on experience on text processing. Throughout the quarter,
we implemented three interventions: a guest lecture, an accessibility
related assignment, and a nal project that oers the opportunity
for students to choose an accessibility related topic. The teaching
materials can be found on github1.
We collected students’ feedback using pre- and post- surveys,
and follow-up interviews. The study was approved by the Inter-
nal Review Board (IRB). In the following, we will give a detailed
description on each of the three interventions aforementioned.
3.1 Guest Lecture on Accessibility
A guest lecture was given by a faculty member as a domain expert
other than the course instructor during the third week of the quarter.
Through an approximately 15-minute-long presentation, students
were introduced to the basic concepts and the importance of accessi-
bility in computing. The presentation highlighted examples for how
AI and ML can be used in developing accessible technologies for
people with disabilities. Discussed examples included automated
captions, contrast analysis of an image, and text simplication. The
presentation also included examples for serious and fatal problems
that can occur if ML models and datasets excluded people with
disabilities [
32
]. The deck of slides containing the presentation,
references, and suggested readings, was shared with the students
afterwards.
3.2 Accessibility-related Assignment
As the programming assignment one (Assignment zero is a small
warm-up programming assignment that focused on the introduc-
tions of the NLP toolkits including NLTK [
5
] and spaCy [
18
], and
the development environment Google Colab
2
), it was introduced in
late of week 2 and made due in the middle of week 4. The description
of the assignment goes as follows:
•Title: Text Classication
1https://github.com/Teaching-Accessibility/Accessibility-AI
2https://colab.research.google.com/
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
865

•
Description: For this assignment, we’ll be building a text clas-
sier. The goal of our text classier will be to distinguish
between words that are simple and words that are complex.
Example simple words are heard,sat,feet,shops, and town,
and example complex words are abdicate,detained,liaison,
and vintners. Distinguishing between simple and complex
words is the rst step in a larger NLP task called text simpli-
cation, which aims to replace complex words with simpler
synonyms. Text simplication is potentially useful for re-
writing texts so that they can be more easily understood
by younger readers, people learning English as a second
language, or people with learning disabilities [13, 23, 26].
•
Data and Assessment: Students were provided with training
and development data that has been manually labeled. Stu-
dents were also given a test set without labels. Students were
tasked to build a classier to predict the labels on the test
set.
•Deliverables:
–Implementations for the functions in the skeleton code.
–Model’s output for the test set
A more detailed description on the complex word identication task
was given along with the specic data examples and the data format
explanation in the subsequent sections in the assignment release.
Students were asked to follow the instructions to implement and
evaluate three non-ML based baseline models, two ML models and
one ML model of their own with the requirement of outperforming
the baseline models. Error Analysis on the built models was oered
as a bonus question in the end.
3.3 Final Project
The nal project was in a larger scale and had a wider scope than
the programming assignments and accounted for almost one third
of the nal grade. It was introduced later in the course. Students
were allowed to work in pairs, and had four weeks to complete
the project. Unlike the aforementioned programming assignment
on text classication, students were given a list of topics for their
nal project and they can select the one they prefer. They were also
allowed to propose one not in the list.
The list of projects was compiled in such a way that that every
project had an equivalent level of diculty with respect to NLP
skills and knowledge. We also tried to keep the applied domain
of the projects to be diverse, e.g., some projects were related to
news, others were related to music, etc. We wanted to observe how
many groups would choose accessibility related topics over non
accessibility related topics. The nal project topics that students
chose go as follows (projects with bold font are accessibility related):
•Tweet clustering and topic modeling
•Haiku generator
•Toxic comment classication
•Text complexity assessment on Shakespeare plays
- Au-
tomatic classication of a text according to its level of com-
plexity can enable suitable recommendations and selections
of materials to a learner. This can particularly help children
with dyslexia.
•Question answering system using BERT
•An autocomplete system for the Scottish Gaelic lan-
guage
- An autocomplete tool can make it easier to type text
by providing suggestions based on the characters already
typed. This particularly helps people who nd typing di-
cult and people who may be susceptible to spelling mistakes.
•
Sentiment analysis on vaccinations over the course of the
pandemic
•Neural language models for song lyrics generator
•
News generator that simulates the local student newspaper
•Neural language models for children’s stories generator
•News article summarization
- Summarization is the task
of producing a shorter version of one or several documents
that preserves most of the input’s meaning. This could po-
tentially help people with reading disabilities.
•A summarization system of BBC political news arti-
cles - Same as above.
•Language model-based Poem Generation
•Click-bait detection from news headline
•Automatic image caption generator
- Automatic image
captioning is the task where, given a photograph, the system
generates a caption that describes the contents of the im-
age. One of its important applications is creating alt-text for
images which is a particularly important for screen reader
users.
•Aspect-based opinion mining from movie reviews
- A
system that automatically extracts or highlights important
and relevant information for targets (or aspects of targets)
can help improve comprehension for people with dyslexia
or people who are deaf and hard of hearing.
•Fine-Tuning DialoGPT with movie dialogue data
•
Topic extraction and sentiment analysis of tweets on the
pandemic and Native American communities
•Text simplication on climate change related articles
- Text Simplication is the task of reducing the complexity of
the vocabulary and sentence structure of text while retaining
its original meaning, with the goal of improving readability
and understanding. Simplication has a variety of important
societal applications, for example increasing accessibility for
those with cognitive disabilities such as aphasia, dyslexia,
and autism, or for non-native speakers and children with
reading diculties
•MCU (Marvel Cinematic Universe) story analysis project
•Lyrics-based genre classication on Indie music
Among the 21 project topics that students ended up presenting, 7
of them (marked in bold) were systems that can be useful for people
with disabilities. These seven projects were either text simplica-
tion, text complexity prediction, or information extraction projects.
We note here that automated text simplication systems are used
as reading assistance tools that were shown to be helpful for people
with various disabilities, e.g., people with dyslexia [
13
,
29
] and deaf
and hard of hearing individuals [
2
]. Similarly, text summarization
can be used as a substitute for skimming, and it was shown to be
helpful for blind people [
31
], and people with dyslexia among other
groups of users with disability [12].
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
866

3.4 Survey
A pre-survey and a post-survey were implemented to compare stu-
dents’ knowledge and awareness on accessibility prior to the inter-
ventions and after the interventions. The questions were designed
to determine the participants’ perception and existing knowledge
on accessibility similar to [
24
]. The survey consisted of two open-
ended questions asking students about accessibility and disability,
and 4 yes/no questions about students’ experiences on accessibility
education and interacting with people with disabilities. The survey
also included 8 questions on students’ knowledge of technical chal-
lenges faced by the various groups of people with disabilities, such
as people with visual impairment, people with learning disabilities,
and people with motor impairments.
The pre-survey was administered during the third week of the
quarter, and was closed before the guest lecture. The post survey
was administered during the last week of classes.
3.4.1 Participants. Twenty-six students (19 male, 6 female, 1 gen-
derqueer) completed the pre-survey and 15 students (10 male, 4
female, 1 genderqueer) completed the post-survey. We had a total
of 12 students (8 male, 3 female, 1 genderqueer), of an average age
of 21.7 (
𝜎
= 1.886) completed both the pre- and post- survey. The
class standing among these 12 students is senior status.
3.5 Interview
After the end of the course, we conducted a semi-structured inter-
view to further understand students’ perspective on their accessibil-
ity learning in the NLP course. The interview included questions on
students’ learning experience taking the natural language process-
ing course, their knowledge regarding accessibility technologies,
and their interest in further accessibility education in the future.
Interview sessions were all conducted remotely using the Zoom
video-conferencing application. On average, the interview lasted for
thirty minutes. Each interview was audio-recorded and transcribed
for analysis. Each participant in this study was able to enter a rae
for the chance to receive a $50 electronic gift-card.
3.5.1 Participants. Among the students who participated in the
post-survey, 5 of the students volunteered to participate in the
follow-up interview which provided more detailed information on
students’ learning experience. Three of the interviewees were male,
two were female, of an average age of 21.4 (
𝜎
= 1.356) and all were
in the computer science major registered in the NLP class during
the Spring quarter of 2021. Four students were computer science
Senior students, while one was a computer science Junior student.
4 RESULTS
4.1 Survey Results
For the combined survey data, we coded the students’ responses to
the survey questions and analyzed the data using Wilcoxon Signed-
Rank Test. In addition, we qualitatively coded the responses to the
survey open-ended questions.
Using the Wilcoxon Signed-Rank Test, we did not nd a sig-
nicant dierence between pre- and post- results on most ques-
tions. However, we did nd a signicant dierence in students’
self-reported learning on technology challenges faced by blind peo-
ple (p= 0.049), and by older individuals (p= 0.023).
Ten students out of the 12 who responded to the pre- and post-
survey reported that they had interacted with people with dis-
abilities, mostly as colleagues or family members. Seven students
reported in the pre-survey that they had some education on ac-
cessibility. From the responses to the open-ended we received in
both the pre- and post-survey, students mentioned that disability is
something that makes it challenging for an individual to accomplish
certain tasks. We also noticed a tendency of students dening dis-
ability and those who have one as a subject that needs comparison
with those who do not. For example, a student said:
“Disability [is] something that someone has which
impacts everyday tasks... there are things [people with
disabilities] might not be able to do quite as well as
others”. (Student 5)
Students often described accessibility as creating something that
is usable for everyone, regardless of whether someone has a dis-
ability or not. For instance, a student described accessibility as:
“Accommodation for [people] with disabilities, often
beneting everyone in the long run”. (Student 6)
By comparing the pre- and post-survey responses, we found that
students answered very similarly in both instances.
4.2 Interview Findings
Two researchers qualitatively analyzed the interviews responses
using inductive coding. Four themes were identied in the students
responses: Accessibility Education, Self-learning Accessibility, Per-
sonal Relationships with People with Disabilities, and Personal
Interests.
1. Accessibility Education
Students did not see a strong focus on accessibility topics within
the current computer science curriculum, but are willing to learn
more on the subject. One student gave suggestions on how accessi-
bility topics can be incorporated into the general computer science
curriculum.
A student mentioned that in general, they did not see accessi-
bility topics as a topic strongly focused on within the computer
science program. They mentioned the possibility for implementing
accessibility topics throughout the core computer science curricu-
lum as a way for students to deepen their exposure and knowledge
on the subject. In addition, they also mentioned how knowing ac-
cessibility topics can assist them in technology development during
the student’s computer science senior project.
“I think in general, we don’t really touch on acces-
sibility at all. It’s kind of something we put on the
back-burner... I think accessibility topics need to be
sprinkled in throughout ... [so] you’re getting kind
of a general sense. I think once you get into more
senior projects, then we denitely should have had
something about accessibility.” (Student 2)
Another student mentioned how computer science courses di-
rectly related to accessibility are often elective courses. They would
be interested to register for an elective on accessibility but they
would choose to nish up their core courses rst.
“I am not registered in [accessibility courses] right
now. There are some other core major courses that
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
867

I think I need to take earlier on. However, I would
be interested in taking [accessibility courses] in the
future.” (Student 4)
In general, students responded positively when asked if they would
consider taking accessibility courses in the future, but they also
remarked that accessibility topics are not currently a strong focus
within computer science courses.
2. Self-Learning Accessibility
Students reported on self-learning accessibility topics. Students
spoke about encountering and self-learning accessibility topics
through various means such as self-driven research on accessible
technology and summer coding programs.
One student remarked learning about iOS development, brows-
ing through iOS documentation, and working on iOS development
as a way to learn more deeply about current usability and accessi-
bility topics.
“I rst learned about accessibility through iOS devel-
opment. I’ve been interested in Apple, iOS documen-
tation, and iOS development for a long time. Apple
really has this big focus on accessibility... and so I
learned a lot about accessibility technology just from
diving deep and looking into Apple’s design and de-
veloped systems.” (Student 3)
Another student mentioned learning about accessibility from
a high school summer program. They mentioned much of their
knowledge of accessibility came from that program.
“I think what I know about accessibility mostly came
from my experience with a summer program in high
school called Cohesive Code.”(Student 5)
3. Personal Relationships with People with Disabilities
Some students reported on knowing people with disabilities in
their own personal lives. Students mentioned how knowing close
friends or family members can challenge them to think more about
disability and accessibility.
A student who knew of close friends and family members who
are blind, hard of hearing, or deaf tends to focus on those disabilities
more when thinking about topics that are accessibility-related.
“Some of the people closest to me with disabilities are
blind or hard of hearing or deaf. So I tend to focus
a little bit more on those disabilities when thinking
about accessibility.” (Student 1)
A student mentioned having a parent who has a medical con-
dition that prompts them to use more accessible technology. The
student remarks that they are uncertain if they would classify their
parent as having a disability even though their parent uses assistive
technology.
“I do not personally have a disability. I wear glasses,
but I can see up close, so it’s not an issue. However,
my parents have bad eyesight and so my mom always
uses really big text, but I don’t know if I would classify
that as an actual disability.” (Student 2)
4. Personal Interests
Students comment on choosing nal project topics based on inter-
est regardless of whether a topic was accessibility-related. Students
were found to have chosen topics for their nal project based on
personal interest that may or may not connect with topics of acces-
sibility.
A student mentioned selecting a poem generator as their nal
project choice. When we asked why they chose the particular topic,
they mentioned how it seemed like an interesting project to do and
that they did not give particular thought about whether to choose
an accessibility related or non-accessibility related nal project.
“Our nal project was basically trying to generate
poems given a whole big dataset. For me personally, I
didn’t think about accessibility at all... I think we were
just going after what sounds cool and what would be
a cool thing to do.” (Student 2)
Another student mentioned working on an accessibility related
auto-complete and auto-correct tool for typing Gaelic. They men-
tioned that they chose this topic because they were interested in the
language. Despite not focusing too much on explicitly picking an
accessibility-related topic for the nal project topic, they ended up
choosing this topic which connects well to language accessibility.
“I’ve been learning Scottish Gaelic. [Gaelic] is ex-
tremely hard to type, so if it’s extremely hard to type,
it means that the language will be used less online... I
wouldn’t say it’s like a super obvious connection with
disability [and] accessibility, but I’d say it’s denitely
accessibility in the sense of making technology more
accessible to those who speak a dierent language.”
(Student 3)
5 DISCUSSION
5.1 Teaching Accessibility Resources for AI
While keeping the learning objectives of the NLP course seamless, it
was also required to keep the interventions explicit enough to help
students remember and solidify accessibility topics within their
consciousness. As NLP is a very broad sub-eld of AI and only ten
weeks were given to cover a wide range of topics, the NLP course in
question focused solely on the data in text format and aimed to cover
the core NLP techniques and algorithms. The reason that we focused
on natural language processing in text format only is twofold: 1.
It is a common practice in the discipline; and 2. Students are not
expected to have prior knowledge on Signal Processing or Computer
Vision. As most of the current accessibility related applications that
use NLP techniques involve speech processing including a screen
reader, or image alt-text generation that has the image recognition
as an integral part of the task, most students are not technically
ready to take on such a task when they are in the NLP course.
The assignment and the nal project were focused on text-based
topics since non-text-based topics require prior knowledge on more
sophisticated tools, such as signal processing or computer vision,
which are not prerequisites for the NLP course. Finding relevant
materials and applications that can directly combine the NLP topics
covered in class and accessibility topics seamlessly was one of the
major challenges faced. Resources that we found inspiring include
a Workshop on NLP
3
for Improving Text Accessibility and Text
Simplication [3].
3https://aclanthology.org/volumes/W13-15/
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
868

We note that only the guest lecture was delivered by a Human
Computer Interaction professor, while the assignment and the nal
project were carried out by an AI professor. This implies that with
the creation of “Accessibility for AI” teaching materials, AI profes-
sors can nd it easy to integrate accessibility into their teaching.
5.2 Eectiveness of the Accessibility
Interventions
The survey and interview results showed that the interventions had
little impact on students’ knowledge on accessibility. The Wilcoxon
Signed-Rank Test on the combined pre- and post-survey data re-
ported only one factor, i.e., knowledge on technology barriers, that
showed a signicant dierence. The interviews revealed that stu-
dents were looking for more opportunities to learn about acces-
sibility. In general, the interventions did not signicantly aect
students’ education on accessibility.
Similar to previous work, we found that students overlooked
the connection between assistive technology and accessibility, such
as Braille in [
19
], and reading assistance in our work. This may
indicate that there is an actual need to integrate human-centered
perspective in non-design computing courses, so use cases become
a natural thought to students when working on their dierent
programming assignments.
Based on our ndings, we suggest the following to improve the
integration of accessibility into AI courses:
•
Make the connection between AI topics and accessibility
more explicit.
•
Reinforce learning through repetition and consistency of
presented accessibility materials.
•
Encourage students’ engagement with AI-accessibility topics
by involving students in discussions and reective writing.
•
Mix accessibility assignments and projects with topics that
are more appealing to students or trend-adjusting, such as
social media and gaming.
5.3 Drawing Student Interest
Students tend to show higher motivation towards topics that are
closely related to their personal interests [
1
] or that will acquire
them advantages for their future careers. First of all, implementing
accessibility to trend following topics would be able to draw higher
motivation. From the nal projects, we learned that 5 student groups
chose topics that mainly concerns social media and 3 groups chose
topics that are related to pop culture. Similarly, the interview results
showed that students with an interest in the accessibility eld
are willing to take future courses that mainly covers accessibility.
Secondly, two students from the interview mentioned that they
would not take accessibility classes because it does not seem to be
related to the career that they are pursuing in the future. In other
words, if they could nd an advantage in learning accessibility for
their career, they would more likely be interested in the topic. In
order to achieve this objective, we may provide the students with
resources that demonstrate the importance of accessibility in the
technology industry and motivate them to reconsider the leverage
of accessibility.
5.4 Limitations of the Work
We acknowledge that small sample size and a single iteration of the
course brought limitations to our research. Twelve out of 35 stu-
dents participated in both the pre- and post-survey, and 5 students
participated in the interview. Especially, there was a signicant
decrease of student participation in the post-survey (15 students)
compared to the pre-survey (26 students). Conducting the research
over multiple iterations of the course would provide comparable
data, allowing us to better identify determining factors for students’
accessibility learning.
6 CONCLUSION AND FUTURE WORK
Presenting accessibility in AI courses provides computing students
with ethical point of view which is more and more demanded in
the current society, especially with the increase in AI-systems that
exclude people with disabilities [
32
]. We emphasize that teaching
ethics in AI courses should cover user diversity in terms of disability
as well as the other axes of diversity, e.g., race and gender. In this
study, we incorporated accessibility topics into an NLP course while
keeping the original course material seamless. We created teaching
materials that tie together AI topics and accessibility. We identied
a list of AI-accessibility projects that are of moderate diculty
for students, and do not require extra preparation work for AI
instructors. Despite the fact that we did not observe a statistically
signicant change in students’ knowledge on accessibility, multiple
students showed interest in learning more about accessibility and
were aware of the importance of this topic. More work is needed
in order to improve the eectiveness of our presented teaching
materials. We plan on improving the current materials by making
the connection to accessibility more clear and the project topics
more engaging. In the future, we will explore how accessibility can
be covered in other AI courses, such as computer vision and signal
processing.
REFERENCES
[1]
Mary Ainley and John Ainley. 2011. Student engagement with science in early
adolescence: The contribution of enjoyment to students’ continuing interest in
learning about science. Contemporary Educational Psychology 36, 1 (2011), 4–12.
[2]
Oliver Alonzo, Matthew Seita, Abraham Glasser, and Matt Huenerfauth. 2020.
Automatic TextSimplication Tools for Deaf and Hard of Hearing Adults: Benets
of Lexical Simplication and Providing Users with Autonomy. In Proceedings of
the 2020 CHI Conference on Human Factors in Computing Systems. 1–13.
[3]
Oliver Alonzo, Jessica Trussell, Becca Dingman, and Matt Huenerfauth. 2021.
Comparison of Methods for Evaluating Complexity of Simplied Texts among
Deaf and Hard-of-Hearing Adults at Dierent Literacy Levels. Association for
Computing Machinery, New York, NY, USA. https://doi.org/10.1145/3411764.
3445038
[4]
Catherine M Baker, Yasmine N Elglaly, and Kristen Shinohara. 2020. A systematic
analysis of accessibility in computing education research. In Proceedings of the
51st ACM Technical Symposium on Computer Science Education. 107–113.
[5]
Steven Bird, Ewan Klein, and Edward Loper. 2009. Natural language processing
with Python: analyzing text with the natural language toolkit. " O’Reilly Media,
Inc.".
[6]
Emanuelle Burton, Judy Goldsmith, Sven Koenig, Benjamin Kuipers, Nicholas
Mattei, and Toby Walsh. 2017. Ethical considerations in articial intelligence
courses. AI magazine 38, 2 (2017), 22–34.
[7]
Paula Conn. 2019. A Systematic Analysis of Accessibility Education Within Com-
puting Disciplines. Rochester Institute of Technology.
[8]
Paula Conn, Taylor Gotfrid, Qiwen Zhao, Rachel Celestine, Vaishnavi Mande,
Kristen Shinohara, Stephanie Ludi, and Matt Huenerfauth. 2020. Understand-
ing the motivations of nal-year computing undergraduates for considering
accessibility. ACM Transactions on Computing Education (TOCE) 20, 2 (2020),
1–22.
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
869

[9]
Eric Eaton, Sven Koenig, Claudia Schulz, Francesco Maurelli, John Lee, Joshua
Eckroth, Mark Crowley, Richard G Freedman, Rogelio E Cardona-Rivera, Tiago
Machado, et al
.
2018. Blue sky ideas in articial intelligence education from the
EAAI 2017 new and future AI educator program. AI Matters 3, 4 (2018), 23–31.
[10]
Yasmine N Elglaly. 2020. Teaching Accessibility to Software Engineering Stu-
dents. In Proceedings of the 51st ACM Technical Symposium on Computer Science
Education. 121–127.
[11]
Yasmine N Elglaly, Weishi Shi, Samuel Malachowsky, Qi Yu, and Daniel E Krutz.
2020. Presenting and evaluating the impact of experiential learning in comput-
ing accessibility education. In 2020 IEEE/ACM 42nd International Conference on
Software Engineering: Software Engineering Education and Training (ICSE-SEET).
IEEE, 49–60.
[12]
Richard Foulds and Cynthia Camacho. 2003. Text summarization contributions
to universal access. Technology and Disability 15, 4 (2003), 223–229.
[13]
Núria Gala and Johannes Ziegler. 2016. Reducing lexical complexity as a tool
to increase text accessibility for children with dyslexia. In Proceedings of the
Workshop on Computational Linguistics for Linguistic Complexity (CL4LC). 59–66.
[14]
Natalie Garrett, Nathan Beard, and Casey Fiesler. 2020. More Than" If Time
Allows" The Role of Ethics in AI Education. In Proceedings of the AAAI/ACM
Conference on AI, Ethics, and Society. 272–278.
[15]
Abraham Glasser, Kesavan Kushalnagar, and Raja Kushalnagar. 2017. Deaf, hard
of hearing, and hearing perspectives on using automatic speech recognition in
conversation. In Proceedings of the 19th International ACM SIGACCESS Conference
on Computers and Accessibility. 427–432.
[16]
Judy Goldsmith and Emanuelle Burton. 2017. Why teaching ethics to AI practi-
tioners is important. In Workshops at the Thirty-First AAAI Conference on Articial
Intelligence.
[17]
Thilo Hagendor. 2020. The ethics of AI ethics: An evaluation of guidelines.
Minds and Machines 30, 1 (2020), 99–120.
[18]
Matthew Honnibal, Ines Montani, Soe Van Landeghem, and Adriane Boyd.
2020. spaCy: Industrial-strength Natural Language Processing in Python. https:
//doi.org/10.5281/zenodo.1212303
[19]
Lin Jia, Yasmine N Elglaly, Catherine M Baker, and Kristen Shinohara. 2021.
Infusing Accessibility into Programming Courses. In Extended Abstracts of the
2021 CHI Conference on Human Factors in Computing Systems. 1–6.
[20]
Saba Kawas, Laura Vonessen, and Andrew J Ko. 2019. Teaching accessibility: A
design exploration of faculty professional development at scale. In Proceedings of
the 50th ACM Technical Symposium on Computer Science Education. 983–989.
[21]
Claire Kearney-Volpe, Devorah Kletenik, Kate Sonka, Deborah Sturm, and Amy
Hurst. 2019. Evaluating instructor strategy and student learning through digital
accessibility course enhancements. In The 21st International ACM SIGACCESS
Conference on Computers and Accessibility. 377–388.
[22]
Megan Lawrence and Mary Bellard. 2017. Teach Access: Preparing Computing
Students for Industry. In Proceedings of the 2017 ACM SIGCSE Te chnical Symposium
on Computer Science Education. 700–700.
[23]
John SY Lee and Chak Yan Yeung. 2018. Personalizing lexical simplication.
In Proceedings of the 27th International Conference on Computational Linguistics.
224–232.
[24]
Stephanie Ludi, Matt Huenerfauth, Vicki Hanson, Nidhi Rajendra Palan, and
Paula Garcia. 2018. Teaching inclusive thinking to undergraduate students in
computing programs. In Proceedings of the 49th ACM Technical Symposium on
Computer Science Education. 717–722.
[25]
Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram
Galstyan. 2021. A survey on bias and fairness in machine learning. ACM Com-
puting Surveys (CSUR) 54, 6 (2021), 1–35.
[26]
Gustavo H Paetzold and Lucia Specia. 2017. A survey on lexical simplication.
Journal of Articial Intelligence Research 60 (2017), 549–593.
[27]
Rohan Patel, Pedro Breton, Catherine M Baker, Yasmine N Elglaly, and Kristen
Shinohara. 2020. Why Software is Not Accessible: Technology Professionals’
Perspectives and Challenges. In Extended Abstracts of the 2020 CHI Conference on
Human Factors in Computing Systems. 1–9.
[28]
Cynthia Putnam, Maria Dahman, Emma Rose, Jinghui Cheng, and Glenn Bradford.
2016. Best practices for teaching accessibility in university classrooms: cultivating
awareness, understanding, and appreciation for diverse users. ACM Transactions
on Accessible Computing (TACCESS) 8, 4 (2016), 1–26.
[29]
Luz Rello, Ricardo Baeza-Yates, Stefan Bott, and Horacio Saggion. 2013. Simplify
or help? Text simplication strategies for people with dyslexia. In Proceedings of
the 10th International Cross-Disciplinary Conference on Web Accessibility. 1–10.
[30]
Jerey Saltz, Michael Skirpan, Casey Fiesler, Micha Gorelick, Tom Yeh, Robert
Heckman, Neil Dewar, and Nathan Beard. 2019. Integrating ethics within machine
learning courses. ACM Transactions on Computing Education (TOCE) 19, 4 (2019),
1–26.
[31]
K Mona Teja, S Mohan Sai, PV Sai Kushagra, et al
.
2018. Smart Summarizer
for Blind People. In 2018 3rd International Conference on Inventive Computation
Technologies (ICICT). IEEE, 15–18.
[32]
Meredith Whittaker, Meryl Alper, Cynthia L Bennett, Sara Hendren, Liz Kaziunas,
Mara Mills, Meredith Ringel Morris, Joy Rankin, Emily Rogers, Marcel Salas, et al
.
2019. Disability, bias, and AI. AI Now Institute (2019).
[33]
Qiwen Zhao, Vaishnavi Mande, Paula Conn, Sedeeq Al-khazraji, Kristen Shino-
hara, Stephanie Ludi, and Matt Huenerfauth. 2020. Comparison of methods for
teaching accessibility in university computing courses. In The 22nd International
ACM SIGACCESS Conference on Computers and Accessibility. 1–12.
Virtual Session: Instructional Strategies
SIGCSE ’22, March 3–5, 2022, Providence RI, USA
870