A Gesture­Based American Sign Language Game for Deaf Children Seungyon Lee, Valerie Henderson, Harley Hamilton , Thad Starner, and Helene Brashear College of Computing Atlanta Area School for the Deaf Georgia Institute of Technology Clarkston, GA 30021 Atlanta, GA 30332 [email protected] ¡ sylee, vlh, thad, brashear ¢ @cc.gatech.edu ABSTRACT to that of hearing subjects [1, 4]. This limited memory We present a system designed to facilitate language develop- processing ability may also be responsible for the slower ment in deaf children. The children interact with a computer acquisition of language due to an inability to receive and game using American Sign Language (ASL). The system process longer utterances efficiently. Our goal was to expose consists of three parts: an ASL (gesture) recognition engine; the children not only to individual vocabulary signs, but an interactive, game-based interface; and an evaluation also to encourage them to formulate those signs into longer system. Using interactive, user-centered design and the concepts and phrases. results of two Wizard-of-Oz studies at Atlanta Area School for the Deaf, we present some unique insights into the spatial organization of interfaces for deaf children. Proposed Solution Our project involves the development of an American Sign Language (ASL) game which uses gesture recognition tech- Author Keywords nology to develop ASL skills in young children. The system Deaf; children; ASL; design; Wizard-of-Oz method; com- is an interactive game with tutoring video (demonstrating puter games; computer aided language learning the correct signs), live video (providing input to the ges- ture recognition system and feedback to the child via the ACM Classification Keywords interface), and an animated character executing the child's H.5.2 [Information interfaces and presentation]: User Inter- instructions. The system concentrates on the practice and face - User-centered Design; K.8.0 [Personal Computing]: correct repetition of ASL phrases and allows the child to Games communicate with the computer via ASL. We target children attending the Atlanta Area School for the Deaf (AASD), INTRODUCTION ages 6–8, who are behind in their language development. Motivation Ninety percent of deaf children are born to hearing parents who do not know sign language [3]. Often these children's only exposure to language is from signing at school. Early childhood is a critical period for language acquisition, and exposure to language is key for linguistic development [7,8]. By twenty-four months of age, hearing children learning a spoken language are combining words in spoken commu- nication [11]. By age eighteen months, deaf children of deaf parents are combining signs to communicate. A third group, deaf children of hearing parents, develop language at a much slower pace, attributable to lack of exposure to language and incomplete language models [5, 9]. The short term memory for sequential linguistic information in deaf children of hearing parents also appears limited compared Figure 1. Screenshot of Game Interface. a) Tutor Video b) Live Camera Feed c) Attention Button d) Animated Character and Environment e) Action Buttons The game involves Iris the cat who can interact with her environment in a number of ways based on what the child signs to her. With the assistance of educators at AASD, we Copyright is held by the author/owner(s). CHI 2005, April 2–7, developed a list of eight age-appropriate ASL phrases (see 2005, Portland, Oregon, USA. Table 1). ACM 1-59593-002-7/05/0004. Playing Time Number When beginning the game, Iris is asleep. The child can press Subject & Self-Ended? an Action Button (Fig. 1e) and watch a short video clip Gender (min:sec) Attempts showing the correct ASL command in the Tutor Window 1 - F 14:06 18 yes (Fig. 1a). Next, the child clicks on the Attention Button 2 - M 4:55 11 yes (Fig. 1c), and Iris wakes up. The child then signs the 3 - F 9:42 21 yes command to Iris. A live video feed allows the child to see 4 - M 18:10 50 no herself (Fig. 1b). If the phrase is correct, Iris carries out the 5 - M 12:22 28 no command or answers the child's question with a pictorial thought bubble in the window (Fig. 1d). If the child does Table 2. Subjects' Pilot Test Data. not sign correctly, a thought bubble with a question mark appears above Iris's head. The system is equipped with a desktop computer, two Eventually, a gesture recognition system will control the monitors, a camera, and a digital video (DV) recording computer's response to the child's signing. However, while device. For input, the mouse from the computer is given the gesture recognition system is under development, we to the child while the keyboard is given to the wizard to began the initial game design via Wizard of Oz (WOz) control the interface actions. This configuration allows for methods. The goal of this phase of the project was to design the use of a single desktop machine with two monitors, thus a prototype for the game which the children would find eliminating the need for networking and synchronization engaging and interesting enough to practice their signing. across computers. The DV device records the interface as it By observing how the children use the system, we will is shown to both the participant and the wizard. Additionally enhance usability and enjoyment. the computer creates a log file of concrete events (mouse clicks, button presses, key clicks, wizard actions, etc.). Task Development and Method First Pilot Test Due to a very limited subject population, it was important The first pilot test of our system had three concrete goals. to remove as many errors from the interface and interaction First, we wanted to evaluate whether the game was en- sequence as possible before attempting to run a longitudinal gaging to the children. Second, we needed to confirm study to evaluate language acquisition using the game. We the interchangeability of interpreters and ascertain whether decided to use children slightly older than our targeted age the interpreters were consistent in their evaluation of the (ages 9–11) for short pilot studies because they were capable children's signing. Third, we were concerned with the of giving us more feedback about the interface. We tested “push-to-sign” mechanism and wanted to confirm that the student in groups of 2 or 3, making subtle changes based on children would remember to press the button before and after the previous groups' opinions and feedback. they signed. In addition to giving us crucial user feedback and allowing The test consisted of two girls and one boy from the same us to improve our design, the WOz study provided an ideal class. All children watched a brief video in ASL telling method to gather data for our gesture recognition system. them how to play the game. We informed all participants This diverse sample set allows for testing and development they could stop playing the game whenever they wanted. of our gesture recognition software with a variety of signers, Our facilitator demonstrated the game. The children then signing techniques, skin tones, and physical attributes. This attempted to play the game on their own. Table 2 (subjects data set is invaluable for training of the mathematical models 1–3) shows the amount of time the children spent playing, and performance testing of the gesture recognition system. the total number of attempted phrases, and whether they stopped playing of their own accord or were asked to stop. Glossed ASL English Translation q(YOU LIKE MOUSE) Do you like mice? Apparatus and Conditions q(YOU HUNGRY NOW) Are you hungry now? An ASL interpreter from AASD evaluated the child's signs YOU GO PLAY BALLOON Go play with the balloon. from behind a partition, and the wizard keyed the appropriate IRIS GO-SLEEP NOW Go to sleep now, Iris. response. After finishing the test, we asked participants YOU GO CATCH BUTTERFLY Go catch the butterfly. several questions related to preference, ease of use, and whq(WHO BEST FRIEND satisfaction to evaluate the game and test setting. Who is your best friend? WHO) LOOK-THERE IRIS Look, Iris! A Observations and Conclusions MOUSE OVER-THERE mouse, over there! We observed that the interpreters had very different criteria YOU MAKE FLOWERS Go make the flowers GROW GO-ON grow. for what they considered “correct” signing. They often based their assessment on their personal knowledge of the child's Table 1. Glossed ASL Phrases and English Translations ASL skills. We considered this unacceptable and changed the experimental setup for the next pilot test. WIZARD OF OZ EXPERIMENTS The push-to-sign mechanism was not a problem for the To date, we have performed two pilot studies with five children. After having it demonstrated once or twice, they participants at AASD. rarely forgot to use it. Each child played until he/she indicated that he/she was we report several interesting observations that will heavily finished. Our goal was to engage the child for 20 minutes, influence our future work. yet none of the children met that goal. We decided the game required a smoother flow to encourage the children to attempt the phrases again after an incorrect response. Spatial Aspects of ASL ASL is a spatial language with rich directional expression. In ASL each signer has a “signing space” which is maintained Second Pilot Test in front of the signer. By setting up subsections of the Our previous work on gesture recognition [2] had determi- signing space and indexing the subsections (by referencing ned that a combination of sensors provided better accuracy them in conversation), signers indicate interactions between than only a computer vision approach.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages4 Page
-
File Size-