American Sign Language Vocabulary: Computer Aided Instruction for Non-signers Valerie Henderson-Summet, Kimberly Weaver, Tracy L. Westeyn, and Thad Starner Georgia Institute of Technology, Atlanta, GA School of Interactive Computing, College of Computing {valerie, kimberly.weaver}@gatech.edu, {turtle, thad}@cc.gatech.edu ABSTRACT In this paper we present the results of a study designed to evaluate the computer-based methods of learning American Sign Language (ASL). We describe a method including an initial instruction session along with receptive and generative language tests which were administered after a week-long retention interval. We show a strong correlations (ρ=.62, ρ=.57) between the initial session’s instruction and the receptive and generative levels of vocabulary signing. Based on the results of our experiment, we establish a baseline for further exploration of ASL vocabulary acquisition and identify further paths for language based instruction. Categories and Subject Descriptors K.3.1 [Computing Milieux]: Computers and Education— Computer-assisted instruction General Terms Languages, Human factors, Experimentation Keywords Sign Language, computer assisted language learning 1. INTRODUCTION Communication between deaf individuals and hearing individuals can be very difficult. For people who are born deaf, English is often a second language with the first language being American Sign Language (ASL). Ninety percent of deaf children are born to hearing parents who may not know sign language [3]. Often these children’s only exposure to language is signing at school, and they are often not immersed in language in the same way as hearing children of hearing parents (or deaf children of deaf parents) who are immersed in language from birth. Without this immersion, deaf children of hearing parents may miss the critical period for language acquisition. Strong and Prinz [8] Copyright is held by the author/owner(s). ASSETS’08 October 13–15, 2008, Halifax, Nova Scotia, Canada. ACM 978-1-59593-976-0/08/10. have shown that there is a strong relationship between early ASL proficiency and later English literacy for deaf children. In order to address early sign language acquisition, we must consider the two sides to the problem. One strategy to combat this problem is to enhance young children’s signing skills. The other strategy is to increase their immersion in ASL by teaching their parents ASL. There are many projects which address children learning ASL (e.g., [5, 4]). Thus, our work concentrates on the task of teaching parents basic level ASL vocabulary. Knowledge of vocabulary is a crucial first step for many parents who wish to communicate with their deaf children. While vocabulary knowledge alone does not constitute fluency or command of ASL, Marchman and Bates have shown that a knowledge of approximately 150 words is enough to increase the rate at which new words and grammatical skills are acquired [6]. Further, research has shown that some level of ASL exposure for deaf children at home is vastly superior to no language exposure [7]. This means that even “survival level” vocabulary is a worthwhile endeavor for families with deaf children. 2. METHODOLOGY We have designed an experiment and Flash-based Web interface to evaluate test–and–train methods of ASL instruction. Participants completed two sessions which occurred approximately a week apart to learn 80 basic ASL vocabulary signs. The words were chosen by an ASL linguist from the MacArthur-Bates Communicative Development Inventory (MCDI) [2] as words suitable for communication with a child from infancy. Participants were asked to provide demographic information including handedness, as the web application changed the handedness of the signed video accordingly. Subjects completed 5 trials with mandatory 5 minutes breaks in between each trial. Each trial consisted of 80 videos of ASL vocabulary words presented in random order. After each video, the participant was presented with multiple choice answers consisting of the correct answer, 3 incorrect answers, and “I don’t know”. After participants selected an answer, they would be told whether their answer was correct or incorrect, and the correct answer was indicated. Subjects spent an average of 51.68 minutes (SD=6.2 minutes) on language instruction (not including breaks) ranging from 44.72 minutes to 64.58 minutes. Participants were scheduled for a second session approximately a week later. During session 2, participants were given receptive and generative language tests. During the receptive test, participants were shown videos of 40 words and asked to write down the English translations. During the generative test, the participant attempted to sign 40 English words. The participants had 10 seconds to provide each word or sign. The words for both the receptive and generative tests were randomly selected from the original 80 word vocabulary list. Half the participants performed the generative test first, and the other half the receptive test. The generative test was analyzed for correctness using a 3 point scale. A score of 1 indicated the subject either made no attempt at the sign or an incorrect attempt. A score of 2 indicated that some aspects of the sign were correct. A score of 3 indicated an entirely correct sign comprehensible to someone familiar or fluent in ASL 3. RESULTS AND DISCUSSION During the multiple choice sessions, we collected 100 data points for each of the 80 words (20 participants x 5 trials). During session 2, we collected 40 generative data points and 40 receptive data points per participant. Because the 80 words were randomly divided between generative and receptive, some words have more receptive data points then generative, or vice versa. As a simple measure of mastery of a sign across subjects, we summed the results of the multiple choice trials T1 –T5 . For example, the aggregate data points for the correct choice of the English word “school” (which appeared 5 times per subject) were T1 =2, T2 =11, T3 =16, T4 =18, and T5 =19. Thus, school was correctly recognized 66 out of 100 times. For the generative test during session 2, we averaged the number of signs which were correct and intelligible (i.e., those rated as a 3). on a per word basis. Collectively, our participants generated 278 words perfectly and recalled 524 words after less then one hour of computer based instruction. 3.1 Adaptive Instruction Our participants learned certain signs much more effectively then others. Some signs are highly iconic, such as book which mimics opening a book with two hands while other signs such as “grandmother,”“grandfather,”“mother,” and “father” are highly related in handshape, sign-space, and motion. Previous work has shown that adaptive presentation in second language learning can be highly beneficial for the learner [1]. In this manner, more time is spent on the signs which are hard for participants to learn, and less time is spent on the signs which have already been learned. Instances in our data (such as the sign for “banana” which was recalled correctly 100% of the time and generated correctly 98% of the time) can make a strong case for adaptive instruction. In the future, we hope to continue work on the modeling of ASL instruction. In particular, we hope to improve upon the baseline established in this paper and develop a more adaptive sequencing of instruction which would generate higher retention rates. 4. CONCLUSIONS We have introduced a computer aided training and testing algorithm plus software to assist adults with the acquisition of key vocabulary words in American Sign Language. We selected an 80 word subset from the MCDI that conveyed distinct and effective content for use during adult-child interactions. Our study consisted of 20 participants that were trained with five trials of 80 multiple choice selection tasks during the first session. After a week hiatus, the participants were asked to perform 40 generative tasks and 40 receptive tasks using the vocabulary from the previous session. We found a strong correlation between the number of times a word was correctly identified in session 1 (the training) and the number of correct responses to the receptive tasks of session 2. Likewise we found a strong correlation between the number of correct responses to the generative tasks of session 2. Thus, performance in the training session may lead to predictions in performance for both receptive and generative tasks given a particular vocabulary set. In addition, the results suggest that adaptive presentation of vocabulary during training may further increase optimal learning and retention of ASL. 5. ACKNOWLEDGMENTS This work was supported by Deptarment of Education Grant #R324A070196. The authors would like to thank Don Schoner, Dr. Jay Summet, and Dr. Harley Hamilton. 6. REFERENCES [1] R. C. Atkinson. Optimizing the learning of a second language vocabulary. Journal of Experimental Psychology, 96:124–129, 1972. [2] L. Fenson, V. A. Marchman, D. J. Thal, P. S. Dale, J. S. Reznick, and E. Bates. MacArthur-Bates Communicative Development Inventories. Brookes Publishing, 2004. [3] Gallaudet Research Institute. Regions regional and national summary report of data from the 1999-2000 annual survey of deaf and hard of hearing children and youth. Technical report, Gallaudet University, Washington, D. C., January 2001. [4] V. L. Hanson and C. A. Padden. Handson: A multi-media program for bilingual language instruction of deaf children. In Proc. of Computing Applications to Assist Persons with Disabilities, pages 5–6, 1992. [5] V. Henderson, S. Lee, H. Brashear, H. Hamilton, T. Starner, and S. Hamilton. Development of an American Sign Language game for deaf children. In Proc. of Interaction Design and Children, pages 70–79, June 2005. [6] V. Marchman and E. Bates. Continuity in lexical and morphological development: A test of the critical mass hypothesis. Journal of Child Language, 21:339–366, 1994. [7] J. Singleton and E. Newport. When learners surpass their models: The acquisition of American Sign Language from inconsistent input. Cognitive Psychology, 49:370–407, 2004. [8] M. Strong and P. Prinz. A study of the relationship between American Sign Language and English literacy. Journal of Deaf Studies and Deaf Education, 2(1):37–46, 1997.