Conference PaperPDF Available

Developing and Evaluating a Novel Gamified Virtual Learning Environment for ASL



The use of sign language is a highly effective way of communicating with individuals who experience hearing loss. Despite extensive research, many learners find traditional methods of learning sign language, such as web-based question-answer methods, to be unengaging. This has led to the development of new techniques, such as the use of virtual reality (VR) and gamification, which have shown promising results. In this paper, we describe a gamified immersive American Sign Language (ASL) learning environment that uses the latest VR technology to gradually guide learners from numeric to alphabetic ASL. Our hypothesis is that such an environment would be more engaging than traditional web-based methods. An initial user study showed that our system scored highly in some aspects, especially the hedonic factor of novelty. However, there is room for improvement, particularly in the pragmatic factor of dependability. Overall, our findings suggest that the use of VR and gamification can significantly improve engagement in ASL learning.
Developing and Evaluating a Novel Gamified
Virtual Learning Environment for ASL
Jindi Wang1 [0000000209018587], Ioannis Ivrissimtzis1[0000000233801889] ,
Zhaoxing Li1[0000000335603461], Yunzhan Zhou1[0000000316760015] , and Lei
1Department of Computer Science, Durham University, Durham, UK
2Open Lab, School of Computing, Newcastle University, Newcastle upon Tyne, UK
{, ioannis.ivrissimtzis, zhaoxing.li2,
Abstract. The use of sign language is a highly effective way of commu-
nicating with individuals who experience hearing loss. Despite extensive
research, many learners find traditional methods of learning sign lan-
guage, such as web-based question-answer methods, to be unengaging.
This has led to the development of new techniques, such as the use of vir-
tual reality (VR) and gamification, which have shown promising results.
In this paper, we describe a gamified immersive American Sign Language
(ASL) learning environment that uses the latest VR technology to grad-
ually guide learners from numeric to alphabetic ASL. Our hypothesis is
that such an environment would be more engaging than traditional web-
based methods. An initial user study showed that our system scored
highly in some aspects, especially the hedonic factor of novelty. How-
ever, there is room for improvement, particularly in the pragmatic factor
of dependability. Overall, our findings suggest that the use of VR and
gamification can significantly improve engagement in ASL learning.
Keywords: Human Computer Interaction ·ASL Learning ·VR
1 Introduction
Sign language is a visual language that uses hand gestures and facial expressions
to convey meaning. It is primarily used for communication with individuals who
are deaf or hard of hearing or who experience difficulty speaking. Learning sign
language is important for several reasons. Firstly, it enables better communica-
tion and social interaction with the hearing-loss community, thereby promoting
inclusion and understanding. By learning sign language, one can break down
communication barriers and establish meaningful connections with individuals
who might otherwise feel excluded. Secondly, learning sign language has been
shown to have numerous cognitive benefits, including enhancing cognitive devel-
opment and language skills [4]. It is widely acknowledged that learning a second
language has cognitive benefits, and the same is true for sign language. Finally,
2 Jindi Wang et al.
for individuals who experience hearing or speech impairments, sign language can
serve as a crucial mode of communication, allowing them to participate more
fully in society. Despite the importance of learning sign language, traditional
web-based methods of learning have not been able to generate much interest
among learners, partly because of a lack of novelty. Therefore, there is a need
for more engaging and innovative approaches to learning sign language that can
increase user engagement and promote effective learning.
To improve the user experience of ASL learning, we developed a VR-based
learning environment that incorporated a Whack-a-Mole type of game, inspired
by the ASL game Sea Battle used by Bragg et al. for data collection [6]. We
then conducted a user study, utilising a questionnaire proposed by Schrepp et
al. [17], to evaluate the user experience of our system. To the best of our knowl-
edge, there have been no previous user studies focused on the user experience
of ASL learning from numeric to alphabetic in a gamified VR environment.
Hence, our main research question was: “Were users satisfied with the ASL
learning experience from numeric to alphabetic in a gamified VR en-
vironment?”. By conducting this user study, we aimed to gain insight into how
users experienced our system and identify areas where improvements could be
made. Ultimately, we hoped to demonstrate that incorporating gamification and
VR technology into ASL learning can enhance user satisfaction and engagement.
Our main contributions are as follows:
1. We successfully created an immersive virtual environment that supports ASL
learning from numeric to alphabetic, incorporating a Whack-a-Mole type of
game. Our system provides a unique and engaging approach to ASL learning,
which we believe can enhance user satisfaction and engagement.
2. Our user study provided initial evidence that our approach has the potential
to improve some aspects of user experience. These findings indicate that
incorporating immersive elements and games into ASL education may be a
promising direction for improving user satisfaction and learning outcomes.
2 Related Work
Sign language recognition: The recognition of sign language through deep
learning and computer vision has been studied by various researchers. Bheda
et al. [3] proposed a method that uses deep convolutional neural networks to
recognize ASL gestures. Kim et al. [12] presented a novel approach that employs
an object detection network for the region of interest (ROI) segmentation to pre-
process input data for sign language recognition. Battistoni et al. [2] described
a method that allows for monitoring the learning progress of ASL alphabet
recognition through CNNs. Jiang et al. [11] proposed a transfer learning-based
approach for identifying fingerspelling in Chinese Sign Language. Camgoz et al.
[7] introduced a transformer-based architecture that jointly learns Continuous
Sign Language Recognition and Translation. Zhang et al. [20] proposed a real-
time on-device hand tracking pipeline called MediaPipe Hands for AR/VR ap-
plications. Goswami et al. [10] created a new dataset and trained a CNN-based
Learning ASL in VR 3
model for recognizing hand gestures in ASL. Finally, Pallavi et al. [13] devel-
oped a deep learning model based on the YOLOv3 architecture, reporting high
recognition rates for the ASL alphabet. These studies demonstrate the potential
of deep learning and computer vision techniques in improving accessibility for
individuals with hearing impairments.
Having reviewed the existing work on sign language recognition, we concluded
that Mediapipe is the most suitable tool for the purposes of this paper, and thus,
we used it for sign language recognition, benefiting from its highly accurate,
real-time detection of hand landmark points. Moreover, as an open-source hand
gesture detection framework from Google, it is well-documented and supported.
Sign language applications: The article discusses various research studies
related to sign language applications. Bantupalli et al. [1] created a vision-based
system to translate sign language into text to improve communication between
signers and non-signers. Schnepp et al. [16] developed an animated sign language
dictionary for caregivers to learn communication with residents who use sign
language. Samonte [15] created an e-tutor system to assist instructors in teaching
sign language. Economou et al. [9] designed a Serious Game to help adults learn
sign language and bridge the communication gap between hearing-impaired and
able-hearing people. Wang et al. [19] designed a sign language game with user-
defined features and found that gamified sign language learning can improve
the user’s learning experience. These studies suggest that dictionary searches
and gamification can improve the learning experience, and influenced the design
choices for our system.
We developed a virtual reality system that offers an immersive and inter-
active learning experience for sign language. To improve the user experience,
we incorporated a quiz and a small game into the system. Given the dearth
of research in this area, we conducted user interviews using a questionnaire to
evaluate users’ satisfaction with ASL learning from numeric to alphabetic in the
system. Our objectives were twofold: to thoroughly evaluate the performance of
our system and to investigate users’ experiences with it.
3 User Interface of VR Environment
This section provides an overview of the main components of our user inter-
face (UI) and highlights the main features of our VR environment. The UI is
comprised of four different modules designed to facilitate effective ASL learning.
1. The Instructions module, which consists of six basic steps, provides users
with an overview of the ASL learning process and guides them through the
initial stages of the programme.
2. The Sign Language Dictionaries module, which enables users to consult
and search for the signs of numbers or letters. This module serves as a
reference tool for users as they progress through the learning process.
3. The Quiz module, which contains question-answer quizzes that allow users
to test their signing skills and self-assess their level of competence. This
4 Jindi Wang et al.
module serves as a valuable feedback mechanism for users and encourages
them to actively engage with the learning material.
4. The Whack-a-Mole Game module, which is to increase user motivation
and engagement with the learning process. This module presents users with a
fun and interactive way to practice their ASL skills, reinforcing their learning
and providing a welcome break from more traditional learning methods.
Together, these four modules work in concert to provide users with a com-
prehensive and engaging VR-based ASL learning experience. By incorporating
elements of gamification and interactivity into our VR environment, we hope to
improve user satisfaction and facilitate more effective ASL learning outcomes.
We separated the scene of the immersive environment into two parts. Adopt-
ing the concept of a simple to complex learning process, the first part is for
learning the numeric ASL, something that is considered a relatively easy task.
The second part of the scene is for the more challenging task of learning the
alphabetic ASL, excluding J and Z, which require dynamic gesturing.
Fig. 1(a) shows the initial view of the user when entering the VR environment,
which includes the Instructions and Sign Language Dictionary interfaces.
Fig. 1(b) shows the Quiz and Whack-a-Mole Game interfaces of numerical
ASL learning, which are located to the left of the numerical ASL dictionary.
Fig. 1(c) shows the Quiz and Whack-a-Mole Game interfaces of alphabetic
ASL learning, which are located to the right of the alphabetic ASL dictionary.
The scene was developed in Unity 2020.3.32f1, and user interaction was fa-
cilitated through eye tracking using HTC Vive Pro. After 3 seconds of fixed
attention, users can click or select objects in the scene. An integrated camera
was used to acquire images; openCV (version 3.4.2) [5] was used for image pro-
cessing on a PC. Hand gestures were detected using Mediapipe, which extracted
a feature vector of 21 points corresponding to landmarks on the detected hand.
An MLP consisting of 3 fully connected layers was implemented in Python 3.6
[14] and Tensorflow 2.6.0 [8] for gesture recognition. The classifier was trained
on a standard PC with an RTX3080 GPU, achieving recognition accuracy rates
above 90%, deemed sufficient to ensure a smooth user experience in our study.
4 User Study Design
In order to evaluate the immersive environment design, we adopted the user sur-
vey scheme proposed by Schrepp et al. [17], which is commonly used to evaluate
user experience in human-computer interaction systems. It consists of six eval-
uation factors, called scales:Attractiveness,Efficiency,Perspicuity,De-
pendability,Stimulation,Novelty. Each scale is further divided into four or
six items, as shown in Table 1. We evaluated the proposed VR environment, on
all scales and items, on a 7-point Likert scale ranging from -3 (fully agree with
a negative term) to +3 (fully agree with a positive term), and studied the user
feedback against the benchmark proposed in [18]. In that paper, the authors
analysed a large database of questionnaire responses and derived the benchmark
intervals shown in Table 2. These intervals correspond to the distribution:
Learning ASL in VR 5
(b) (c)
Fig. 1. The implemented immersive virtual environment. (a) Left: the numeric ASL
sign language dictionary. Centre: Instructions interface. Right: the A-Y except for J
sign language dictionary. (b) The numeric ASL learning quiz (left) and game (right).
(c) The alphabetic ASL learning quiz (left) and game (right).
Table 1. Summary of the user experience questionnaire.
Attractiveness Perspicuity
A1: annoying / enjoyable P1: not understandable / understandable
A2: good / bad P2: easy to learn / difficult to learn
A3: unlikable / pleasing P3: complicated / easy
A4: unpleasant / pleasant P4: clear / confusing
A5: attractive / unattractive
A6: friendly / unfriendly
Efficiency Dependability
E1: fast / slow D1: unpredictable / predictable
E2: inefficient / efficient D2: obstructive / supportive
E3: impractical / practical D3: secure / not secure
E4: organized / cluttered D4: meets expectations / does not meet expectations
Stimulation Novelty
S1: valuable / inferior N1: creative / dull
S2: boring / exciting N2: inventive / conventional
S3: not interesting / interesting N3: usual / leading edge
S4: motivating / demotivating N4: conservative / innovative
Excellent: In the range of the 10% best results.
Good: 10% of results better, 75% of results worse.
Above average: 25% of results better, 50% of results worse.
Below average: 50% of results better, 25% of results worse.
Bad: In the range of the 25% worst results.
We conducted the user study obtaining feedback from 15 participants, 8
males and 7 females, aged between 19 and 21 years old, who had little or no
prior experience with ASL or any other sign language. At the start of the session,
6 Jindi Wang et al.
Table 2. Benchmark intervals for the user experience scales.
Attractiveness Perspicuity Efficiency Dep endability Stimulation Novelty
Excellent 1.75 1.78 1.90 1.65 1.55 1.40
Good [1.52, 1.75) [1.47, 1.78) [1.56, 1.90) [1.48, 1.65) [1.31, 1.55) [1.05, 1.40)
Above average [1.17, 1.52) [0.98, 1.47) [1.08, 1.56) [1.14, 1.48) [0.99, 1.31) [0.71, 1.05)
Below average [0.70, 1.17) [0.54, 0.98) [0.64, 1.08) [0.78, 1.14) [0.50, 0.99) [0.30, 0.71)
Bad <0.70 <0.54 <0.64 <0.78 <0.50 <0.30
participants had the freedom to explore the system and consult the Instructions
module. Then, each participant followed a six stages learning process:
1. Learn numeric ASL for 3 minutes from corresponding dictionary module.
2. Improve numeric ASL comprehension for 3 minutes in numeric quiz module.
3. 30 seconds on numeric ASL game module.
4. Learn alphabetic ASL from corresponding dictionary module for 3 minutes.
5. Improve alphabetic ASL literacy for 3 minutes in alphabetic quiz module.
6. 30 seconds on alphabetic ASL game module.
5 Result Analysis
Fig. 2 shows the average scores for the six scales, denoted by ‘x’, plotted over a
colour code of the corresponding benchmark interval. For each scale, the min-
imum and the maximum of the average scores on its individual items are also
shown. In Fig. 3, the box plots show the minimum, first quartile, median, third
quartile, and maximum, for each individual item of each scale.
Attractiveness Perspicuity Efficiency Dependability Stimulation Novelty
2.5 Excellent
Above Average
Below Average
Fig. 2. Benchmark intervals for the six scales
Attractiveness: The mean value of the user scores is 0.39 (SD = 1.24),
placing it in the “Bad” category, indicating that their overall impression of the VR
Learning ASL in VR 7
A1 A2 A3 A4 A5 A6
(a) Attractiveness
P1 P2 P3 P4
(b) Perspicuity
E1 E2 E3 E4
(c) Efficiency
D1 D2 D3 D4
(d) Dependability
S1 S2 S3 S4
(e) Stimulation
N1 N2 N3 N4
(f) Novelty
Fig. 3. Box-plots of the scores for each item of the six scales.
environment was not favourable, and the system requires further improvements.
Notably, the average score for item A5, shown in Fig. 3(a), is slightly below 0,
which suggests that the users did not find the system particularly appealing. This
may be because the learning environment relies on 2D user interfaces, whereas
incorporating 3D elements may be more visually engaging for users. Therefore,
we plan to integrate 3D user interfaces in future iterations of the ASL learning
environment, aiming at enhancing its attractiveness.
Perspicuity: The average score is 1.40 (SD = 1.28), placing it in the “Above
average” category, indicating that users perceive the VR environment as clear
and understandable, facilitating their ASL learning experience. However, it seems
that some of the users may have encountered some problems when using the en-
8 Jindi Wang et al.
vironment, possibly due to their unfamiliarity with VR devices, and they may
require some initial training.
Efficiency: In the “Below average” category, the average score is 0.87 (SD
= 1.27). We note that, while the average score over the whole scale is slightly
below average, analysis of individual item scores shows that our VR environment
adequately fulfills some users’ requirements. In particular, users found the system
easy to use (as reflected by item E1) and believed that they could practice ASL
effectively in the scenario (as reflected by item E3), see Fig. 3(c).
Dependability: In the “Bad” category, the average score is 0.28 (SD =
0.89). That means that the VR environment’s dependability needs significant
improvement. Despite the low overall average score, some users still believed
that on individual items, particularly D2 and D4, the system adequately fulfilled
their requirements, see Fig. 3(d).
Stimulation: In the “Below average” category with an average score of 0.87
(SD = 1.43). Even though the score is slightly lower than average, the large
variance indicates that some users find the learning environment stimulating. As
shown in Fig. 3(e), the first quartile of all items is non-negative, indicating that
a majority of users have a consistently favourable outlook regarding this scale.
Novelty: In the “Good” category with an average value of 1.25 (SD = 1.08).
Again, the first quartile of all items is non-negative, see Fig. 3(f), indicating a
consistently favourable view from a majority of users. They perceive the VR
environment as a novel and innovative way of learning ASL.
6 Conclusion
We have developed a VR system for learning numeric and alphabetic ASL and
conducted a questionnaire-based user study to evaluate the user experience of
learning ASL in the system. We found that to some extent it satisfied some user
satisfaction factors, however, the system needs further development to enhance
user experience, especially on the factors of attractiveness and dependability.
There are several limitations to our ASL learning system, which have been
discussed for each scale of user experience separately. The identified shortcomings
include a lack of animated hints; an interface that requires users to actively
press a start button to commence an action; difficulty in moving around the
VR scene; a relatively large number of incorrect judgments of correct signs, i.e.,
many false negatives; user expectations for a more creatively designed system;
and an overall perception that the learning task was too easy. Additionally, the
user study included 15 only participants, primarily between the ages of 19 and
21, and there was a complete lack of research on users in other age groups.
To address these limitations, we plan to revise the content, design, and im-
plementation of the system as follows: add more interactive elements; implement
automatic settings; create a follow-through user interface; develop a more robust
sign recognition model; and include more sophisticated sign language learning
material. We also plan to recruit a larger and more diverse group of participants
for a follow-up user study.
Learning ASL in VR 9
1. Bantupalli, K., Xie, Y.: American sign language recognition using deep learning
and computer vision. In: 2018 IEEE International Conference on Big Data (Big
Data). pp. 4896–4899. IEEE (2018)
2. Battistoni, P., Di Gregorio, M., Sebillo, M., Vitiello, G.: Ai at the edge for sign
language learning support. In: 2019 IEEE International Conference on Humanized
Computing and Communication (HCC). pp. 16–23. IEEE (2019)
3. Bheda, V., Radpour, D.: Using deep convolutional networks for gesture recognition
in american sign language. arXiv preprint arXiv:1710.06836 (2017)
4. Bialystok, E., et al.: Bilingualism in development: Language, literacy, and cogni-
tion. Cambridge University Press (2001)
5. Bradski, G., Kaehler, A.: Opencv. Dr. Dobb’s journal of software tools 3, 120
6. Bragg, D., Caselli, N., Gallagher, J.W., Goldberg, M., Oka, C.J., Thies, W.: Asl
sea battle: Gamifying sign language data collection. In: Proceedings of the 2021
CHI conference on human factors in computing systems. pp. 1–13 (2021)
7. Camgoz, N.C., Koller, O., Hadfield, S., Bowden, R.: Sign language transformers:
Joint end-to-end sign language recognition and translation. In: Proceedings of the
IEEE/CVF conference on computer vision and pattern recognition. pp. 10023–
10033 (2020)
8. Dillon, J.V., Langmore, I., Tran, D., Brevdo, E., Vasudevan, S., Moore, D., Patton,
B., Alemi, A., Hoffman, M., Saurous, R.A.: Tensorflow distributions. arXiv preprint
arXiv:1711.10604 (2017)
9. Economou, D., Russi, M.G., Doumanis, I., Mentzelopoulos, M., Bouki, V., Fer-
guson, J.: Using serious games for learning british sign language combining video,
enhanced interactivity, and VR technology. Journal of Universal Computer Science
26(8), 996–1016 (2020)
10. Goswami, T., Javaji, S.R.: Cnn model for american sign language recognition. In:
ICCCE 2020, pp. 55–61. Springer (2021)
11. Jiang, X., Hu, B., Chandra Satapathy, S., Wang, S.H., Zhang, Y.D.: Fingerspelling
identification for chinese sign language via alexnet-based transfer learning and
adam optimizer. Scientific Programming 2020 (2020)
12. Kim, S., Ji, Y., Lee, K.B.: An effective sign language learning with object detection
based roi segmentation. In: 2018 Second IEEE International Conference on Robotic
Computing (IRC). pp. 330–333. IEEE (2018)
13. Pallavi, P., Sarvamangala, D.: Recognition of sign language using deep neural net-
work. International Journal of Advanced Research in Computer Science 12, 92–97
14. Python, W.: Python. Python Releases for Windows 24 (2021)
15. Samonte, M.J.C.: An assistive technology using FSL, speech recognition, gamifica-
tion and online handwritten character recognition in learning statistics for students
with hearing and speech impairment. In: Proceedings of the 2020 The 6th Inter-
national Conference on Frontiers of Educational Technologies. pp. 92–97 (2020)
16. Schnepp, J., Wolfe, R., Brionez, G., Baowidan, S., Johnson, R., McDonald, J.:
Human-centered design for a sign language learning application. In: Proceedings
of the 13th ACM International Conference on PErvasive Technologies Related to
Assistive Environments. pp. 1–5 (2020)
17. Schrepp, M., Hinderks, A., Thomaschewski, J.: Applying the user experience ques-
tionnaire (UEQ) in different evaluation scenarios. In: International Conference of
Design, User Experience, and Usability. pp. 383–392. Springer (2014)
10 Jindi Wang et al.
18. Schrepp, M., Thomaschewski, J., Hinderks, A.: Construction of a benchmark for
the user experience questionnaire (UEQ). International Journal of Interactive Mul-
timedia and Artificial Intelligence 4(4), 40–44 (2017)
19. Wang, J., Ivrissimtzis, I., Li, Z., Zhou, Y., Shi, L.: User-defined hand gesture in-
terface to improve user experience of learning american sign language. In: Interna-
tional Conference on Intelligent Tutoring Systems. pp. 479–490. Springer (2023)
20. Zhang, F., Bazarevsky, V., Vakunov, A., Tkachenka, A., Sung, G., Chang, C.L.,
Grundmann, M.: Mediapipe hands: On-device real-time hand tracking. arXiv
preprint arXiv:2006.10214 (2020)
... In the past years, face-to-face teaching of sign languages has been severely affected by the COVID-19 restrictions, with alternative online approaches filling in the gap. The majority of the latest approaches to teaching sign languages [10,11,20] employ website-based tools, while approaches based on immersive virtual reality (VR) technology [29] are more sparse in the literature. The following are some of the few studies on the VR-based approach. ...
Full-text available
Sign languages enable effective communication between deaf and hearing people. Despite years of extensive pedagogical research, learning sign language online comes with a number of difficulties that might be frustrating for some students. Indeed, most of the existing approaches rely heavily on learning resources uploaded on websites, assuming that users will frequently consult them; however, this approach may feel tedious and uninspiring. To address this issue, several researchers have started looking into learning sign language in a game-based environment. However, the majority of the existing work still relies on website-based designs, with only a very few proposed systems providing an immersive virtual environment, and there are no user studies comparing website-based and immersive virtual environments. In this paper, we present an immersive environment for learning numbers 0–9 in American Sign Language (ASL). Our hypothesis is that an immersive virtual environment can provide users with a better learning experience and that users will show a higher level of engagement compared to website-based learning. We conducted a questionnaire-based user survey, and our initial findings suggest that users prefer to learn in an immersive virtual environment.KeywordsHuman Computer Interaction (HCI)Interaction DesignEmpirical Studies in Interaction DesignASL Learning
Conference Paper
Full-text available
Sign language can make possible effective communication between hearing and deaf-mute people. Despite years of extensive pedagogical research, learning sign language remains a formidable task, with the majority of the current systems relying extensively on online learning resources, presuming that users would regularly access them; yet, this approach can feel monotonous and repetitious. Recently, gamification has been proposed as a solution to the problem, however, the research focus is on game design, rather than user experience design. In this work, we present a system for user-defined interaction for learning static American Sign Language (ASL), supporting gesture recognition for user experience design, and enabling users to actively learn through involvement with user-defined gestures, rather than just passively absorbing knowledge. Early findings from a questionnaire-based survey show that users are more motivated to learn static ASL through user-defined interactions.
Full-text available
One in every six persons in the UK suffers a hearing loss, either as a condition they have been born with, or they developed during their life. Nine hundred thousand people in the UK are severely or profoundly deaf. Based on a study by Action on Hearing Loss UK in 2013 only 17 percent of this population, can use the British Sign Language (BSL). That leaves a massive proportion of people with a hearing impediment who do not use sign language struggling in social interaction and suffering from emotional distress. It also leaves even a larger proportion of Hearing people who cannot communicate with those of the deaf community. This paper presents a Serious Game (SG) that aims to close the communication gap between able hearing people and people with hearing impairment by providing a tool that facilitates BSL learning targeting the adult population. The paper presents the theoretical framework supporting adult learning based on which a SG game using Virtual Reality (VR) technology has been developed. The paper explains the experimental framework of the study. It presents the creation of the research instruments to facilitate the study comprising of a SG that integrates video and conventional video-based educational material. It reports and analyses the study results that demonstrate the advantage of the SG in effectively supporting users learning a set of BSL signs. It also presents qualitative outcomes that inform the further development of the game to serve learning needs. The paper closes with conclusions, directions for further development of this educational resource, and future studies.
Full-text available
As an important component of universal sign language and the basis of other sign language learning, finger sign language is of great significance. This paper proposed a novel fingerspelling identification method for Chinese Sign Language via AlexNet-based transfer learning and Adam optimizer, which tested four different configurations of transfer learning. Besides, in the experiment, Adam algorithm was compared with stochastic gradient descent with momentum (SGDM) and root mean square propagation (RMSProp) algorithms, and comparison of using data augmentation (DA) against not using DA was executed to pursue higher performance. Finally, the best accuracy of 91.48% and average accuracy of 89.48 ± 1.16% were yielded by configuration M1 (replacing the last FCL8) with Adam algorithm and using 181x DA, which indicates that our method can identify Chinese finger sign language effectively and stably. Meanwhile, the proposed method is superior to other five state-of-the-art approaches.
Conference Paper
Full-text available
Prior work on Sign Language Translation has shown that having a mid-level sign gloss representation (effectively recognizing the individual signs) improves the translation performance drastically. In fact, the current state-of-the-art in translation requires gloss level tokenization in order to work. We introduce a novel transformer based architecture that jointly learns Continuous Sign Language Recognition and Translation while being trainable in an end-to-end manner. This is achieved by using a Connectionist Temporal Classification (CTC) loss to bind the recognition and translation problems into a single unified architecture. This joint approach does not require any ground-truth timing information, simultaneously solving two co-dependant sequence-to-sequence learning problems and leads to significant performance gains. We evaluate the recognition and translation performances of our approaches on the challenging RWTH-PHOENIX-Weather-2014T (PHOENIX14T) dataset. We report state-of-the-art sign language recognition and translation results achieved by our Sign Language Transformers. Our translation networks outperform both sign video to spoken language and gloss to spoken language translation models, in some cases more than doubling the performance (9.58 vs. 21.80 BLEU-4 Score). We also share new baseline translation results using transformer networks for several other text-to-text sign language translation tasks.
This paper proposes a model based on convolutional neural network for hand gesture recognition and classification. The dataset uses 26 different hand gestures, which map to English alphabets A–Z. Standard dataset called Hand Gesture Recognition available in Kaggle website has been considered in this paper. The dataset contains 27,455 images (size 28 * 28) of hand gestures made by different people. Deep learning technique is used based on CNN which automatically learns and extracts features for classifying each gesture. The paper does comparative study with four recent works. The proposed model reports 99% test accuracy.