At a Different Pace: Evaluating Whether Users Prefer Timing Parameters in American Sign Language Animations to Differ from Human Signers' Timing
Sedeeq Al-khazraji, Becca Dingman, Sooyeon Lee, Matt Huenerfauth · 2021 · Proceedings of the 23rd International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '21) · doi:10.1145/3441852.3471214
Summary
This paper investigates whether deaf and hard-of-hearing (DHH) ASL signers actually prefer sign language animations with timing parameters that match human signers, or whether they prefer some form of exaggeration for additional clarity. Adding ASL animations to websites can improve information accessibility for DHH users who may have lower English literacy, and generating these animations from script representations allows content to be easily updated. However, prior research had universally assumed that the ideal animation timing should replicate human signing as closely as possible — an assumption this paper challenges through two empirical studies. The researchers examined five specific speed and timing parameters: sign duration (how long each sign takes), transition time (time between signs), differential signing rate (dynamic speed variation within a passage), pause length, and pausing frequency. Using Sign Smith Studio to generate animations from three ASL passages of approximately 75 words each, the team created stimuli that systematically varied each parameter across five levels centred on typical human values. In Study 1 (N=20), participants viewed five animations side-by-side for each parameter and rated each on a five-point quality scale, identifying the top two preferred values. In Study 2 (N=20, mostly different participants), the two finalists for each parameter were compared directly in pairs, with one always being the typical human value. Both studies were conducted remotely via video conferencing with a Deaf native ASL signer as the researcher.
Key findings
The results revealed a nuanced pattern that challenges the assumption that human-like timing is always preferred. For the three speed-related parameters, participants significantly preferred animations that differed from human timing: they preferred faster sign durations (0.81 seconds average vs. the human-typical 1.28 seconds), slower transitions between signs (0.5 seconds vs. the human-typical 0.23 seconds), and less dynamic variation in differential signing rate (exponent of 0.75 vs. the human-typical 1.0). However, for the two pause-related parameters, participants showed no significant preference between the human value and the alternative — pause length and pausing frequency similar to human signers were rated equivalently to slightly different values. The authors speculate that the preference for faster signs with slower transitions helps viewers more clearly distinguish when a sign is being performed versus when the avatar is transitioning between signs, improving understandability. The preference for less extreme speed dynamics may relate to current animation technology limitations, where rapid speed-ups and slow-downs can appear jumpy or unnatural. The acceptance of human-like pausing may reflect that pauses carry important grammatical and prosodic meaning in ASL, making deviations confusing, and that pausing is technically easier for avatars to perform naturally since they simply stand still.
Relevance
This study provides concrete, empirically derived guidance for anyone creating ASL animation technology. The specific preferred parameter values (sign duration: 0.81s, transition time: 0.5s, differential rate exponent: 0.75, with human-like pausing) can be directly applied by animation artists, rule-based systems, and AI models generating ASL content. For accessibility practitioners, the broader finding is equally important: designers should not assume that replicating human behaviour as closely as possible is the ideal goal for assistive technologies. User preferences may diverge from human norms for legitimate reasons related to understandability, the limitations of current technology, or the different perceptual context of viewing an animation versus a human. The study also demonstrates the value of conducting empirical research with DHH users rather than making assumptions about their preferences. This principle extends beyond sign language animation to any accessibility technology where designers might be tempted to simply mimic non-disabled patterns rather than investigating what disabled users actually prefer.
Tags: American Sign Language · sign language animation · deaf and hard of hearing · timing parameters · avatar · accessibility · user preferences · natural language generation