The non-visual channel is wide enough for notification and too narrow for language

Haptic output is the one channel that does not have to be looked at. It arrives through vibration or force, so it reaches a user whose eyes are on the road, whose view is filled by a headset, or whose hands hold the device. This fifth of ten units argues that this survival under divided attention is real and measurable. What the channel can carry is bounded twice over: the perceiver must keep patterns apart, and the user must have learned what they mean. The result is a small, structured vocabulary, usable in a few roles and unusable as a general language of states.

The channel does work the visual one cannot

Jones and Sarter (2008) give the case in its design form. Their review sets out the properties of vibrotactile perception and the technology that presents it. It identifies the functions touch can serve, including spatial orientation and guidance, attention management and sensory substitution for a user who cannot see the screen, though the hardware difference that improves discrimination is one many users cannot afford to close. That is a list of roles rather than messages. The channel does work the visual channel cannot do when the user cannot look. MacLean (2008) relates human capabilities and hardware to everyday roles that include notification, augmentation of graphical interfaces, expressive control and affective communication. Treating haptics as one capability hides that these roles make different demands.

The benefit is real but conditional

Burke, Prewett, Gray, Yang, Stilson, Coovert, Elliot and Redden (2006) pooled 43 studies comparing visual-auditory and visual-tactile feedback against visual feedback alone. Adding a second modality improved performance overall, but not uniformly. Visual-auditory feedback was most effective for a single task (g = 0.87) and at normal workload (g = 0.71). Visual-tactile feedback was more effective when several tasks ran at once (g = 0.77) and at high workload (g = 0.84). Neither reduced error rates. Touch pays off under pressure and buys speed rather than accuracy.

Two outcomes hide under the word benefit. Hoggan, Brewster and Johnston (2008) found tactile feedback brought finger text entry on a touchscreen close to physical-keyboard performance, and that higher-specification actuators improved it further. Koskinen, Kaaresoja and Laitinen (2008) measured felt quality instead. Piezo actuators gave a slightly more pleasant button click than a vibration motor, though not significantly. A claimed benefit has to name which outcome it means.

Perception caps the channel at a few discriminable parameters

The bound starts in perception. Lederman and Klatzky (2009) give the tutorial account of the systems underlying touch, covering tactile and haptic sensing with texture, shape and object recognition. A designer needs that account to know what the channel can encode at all. Jones and Sarter (2008) say the same from the applied side. The sense is good at detecting contact, change and roughness, and poor at carrying arbitrary symbolic distinctions at speed.

Brewster and Brown (2004) named the practical form of the vocabulary. Tactons, or tactile icons, are structured messages built from the frequency, amplitude, waveform and duration of a pulse plus its body location. They suit displays where vision is overloaded, small or absent. The crucial word is structured. A tacton combines a few dimensions, so a person's ability to tell combinations apart and remember them limits the useful set. Brown, Brewster and Purchase (2006) measured that limit by adding a third parameter to the tacton set. Identification fell to 48 per cent, and recovered to 81 per cent only when one parameter took fewer values.

A measured tactile vocabulary runs to about ten patterns

The size of that set has been measured. Yatani and Truong (2009) built SemFeel, vibration motors on the back of a mobile touchscreen that generate patterns flowing from right to left or top to bottom, telling the user where a finger landed and what the touched object is. Across two user studies participants distinguished ten patterns, linear and circular, at approximately 90 per cent accuracy. Ten is a working vocabulary beside the states a modern interface holds.

Hazra and Shiakolas (2024) tested what happens when the designer fixes the vocabulary in advance. Their system recognises non-verbal input, returns vibrotactile feedback confirming the identified action, then executes it. They compared a fixed set with one the user could define and generate. Users preferred personalised feedback, and success reached 100 per cent after only three successive repetitions. Reported mental demand and effort on the NASA Task Load Index fell. The ceiling is partly a property of the population.

Learning is part of the cost

Learning enters the argument twice, first as motor learning, where haptics teaches the body rather than informing it. Grindlay (2008) had participants learn short rhythmic sequences under audio, haptic and combined training. Haptic guidance significantly benefited recall of timing and velocity. Adding it to audio training produced a 17 per cent reduction in final velocity error, and the combination produced an 18 per cent reduction in early-stage timing error.

The second is the learning a user must do before a pattern means anything. A structure is a convention before it is a signal, so the tacton vocabulary of Brewster and Brown (2004) has to be acquired. Hazra and Shiakolas (2024) put a clock on that cost. Their 100 per cent success came after three successive repetitions, so the user learned the vocabulary within the session. A pattern that must be learned before it can be used is a poor carrier for information that arrives once.

The channel does one job, and composition is not accumulation

The thesis is about tactile patterns, and extending it to force would be a mistake. Bergamasco (1995) argued that where manipulative and explorative procedures must be controlled, every movement of the hand should be recorded and force and tactile stimuli replicated at the hand. Force and tactile feedback then take separate roles for grasping and exploration. That is a requirement of manipulation, not of notification. A hand that must feel an object's resistance needs continuous force, while an interface confirming a tap needs a pattern.

The system described in that paper makes the separation concrete. The Hand Force Feedback device records the hand's movement and replicates contact force at the hand. It treated grasping and exploration as different jobs, and it modelled thermal and indentation stimuli separately. None of that work carries a message.

Adding touch is not adding modes until the interface is rich. Oviatt (1999) catalogued ten myths of multimodal interaction against contrary empirical evidence. People integrate modes in patterned ways, and an architecture must handle parallel, interdependent signals under time pressure. Jaimes and Sebe (2007) survey the same field across input modalities, fusion architectures and application areas. Both treat multimodality as composition: which mode carries which part of the message.

Contested question: is the ceiling in the channel or in the pattern set?

The literature disagrees about where the limit lies. Brewster and Brown (2004) locate it in the message structure. Tactons combine a few perceptual parameters, so the vocabulary is a small set of combinations and the ceiling is set by what those parameters can express. Jones and Sarter (2008) and Lederman and Klatzky (2009) locate it in the perceiver, in what the tactile and haptic sense can encode and discriminate. On either account the bound belongs to the channel and the person.

Hazra and Shiakolas (2024) put it elsewhere. A fixed set fails for some users, while one the user defines is preferred and reaches 100 per cent success after three repetitions, so part of the ceiling is the mismatch between a population average and an individual. Burke and colleagues (2006) complicate all three positions. The benefit of adding a modality depends on task type, workload and the number of concurrent tasks, so a claim about how much the channel can carry is also a claim about a load condition.

What would settle the dispute is a study none of these papers runs. One discrimination task would have the same participants attempt the same large pattern set with and without a competing visual task, before and after training, and with the set fixed for everyone or generated for each person. Yatani and Truong (2009) hold the set fixed and measure ten patterns at approximately 90 per cent accuracy. Hazra and Shiakolas (2024) vary the set but use a small action vocabulary. Neither crosses personalisation with load.

The evidence on each side differs in kind. Jones and Sarter (2008) and Lederman and Klatzky (2009) argue from reviews of perception and of design practice, while Yatani and Truong (2009) measure ten patterns at about 90 per cent accuracy in two studies and Hazra and Shiakolas (2024) measure personalisation in a system comparison. The strongest objection is that the unit's vocabulary bound rests on few measurements, since the largest discrimination result here separates only ten patterns, so the claim that the channel is too narrow for language is extrapolated rather than shown at the scale speech would need. A designer should treat the channel as a small vocabulary for alerts while the dispute is open.

Where the argument is heading

The direction is towards treating the tactile vocabulary as something the system adapts and the user trains, not something the designer fixes. Hazra and Shiakolas (2024) show personalised feedback preferred and three repetitions enough to reach full success. Hoggan, Brewster and Johnston (2008) show the second lever, that actuator quality changes how well a pattern can be felt, with better hardware improving performance further. Jones and Sarter (2008) name the research need that joins the two, a fuller understanding of the channel's properties and of how it interacts with other modalities. This thesis will be tested not by a louder alert but by whether personalisation and better hardware raise discrimination under divided attention. If they do, the boundary moves from the channel to the pattern set.

The sources this unit rests on

• Bergamasco, M. (1995). Haptic interfaces: the study of force and tactile feedback systems. Proceedings 4th IEEE International Workshop on Robot and Human Communication. doi:10.1109/roman.1995.531929 Argues that where manipulation and exploration must be controlled, hand movement should be recorded and force and tactile stimuli replicated. It separates force feedback for manipulation from patterns for notification.
• Brewster, S. A. and Brown, L. M. (2004). Non-visual information display using tactons. CHI '04 Extended Abstracts on Human Factors in Computing Systems. doi:10.1145/985921.985936 Describes tactons, structured tactile messages built from pulse frequency, amplitude, waveform, duration and body location, for displays where vision is overloaded, small or absent.
• Brown, L. M., Brewster, S. A. and Purchase, H. C. (2006). Multidimensional tactons for non-visual information presentation in mobile devices. Proceedings of the 8th conference on Human-computer interaction with mobile devices and services. doi:10.1145/1152215.1152265 Extended the tacton parameter space to a third dimension; identification of three-parameter tactons fell to 48 per cent, recovering to 81 per cent only when one parameter took fewer values.
• Burke, J. L., Prewett, M. S., Gray, A. A., Yang, L., Stilson, F. R. B., Coovert, M. D., Elliot, L. R. and Redden, E. (2006). Comparing the effects of visual-auditory and visual-tactile feedback on user performance. Proceedings of the 8th international conference on Multimodal interfaces. doi:10.1145/1180995.1181017 A meta-analysis of 43 studies found that adding a modality improved performance overall but not error rates, visual-tactile feedback being strongest for concurrent tasks and at high workload. The benefit depends on task and load.
• Grindlay, G. (2008). Haptic Guidance Benefits Musical Motor Learning. 2008 Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems. doi:10.1109/haptics.2008.4479984 A pilot experiment on learning rhythmic drumming sequences found haptic guidance significantly benefited recall of timing and velocity. The channel taught the body.
• Hazra, S. and Shiakolas, P. S. (2024). On the Effects of Personalizing Vibrotactile Feedback to Facilitate User Interaction With a Robotic System. IEEE Access. doi:10.1109/access.2024.3451354 Compared a fixed vibrotactile feedback set with one the user defines, and found personalised feedback preferred, with full success after three repetitions and lower reported effort. A fixed set may not serve a diverse population.
• Hoggan, E., Brewster, S. A. and Johnston, J. (2008). Investigating the effectiveness of tactile feedback for mobile touchscreens. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. doi:10.1145/1357054.1357300 Found tactile feedback brought finger text entry close to physical-keyboard performance, and that better actuators improved it further. The benefit is tied to hardware.
• Jaimes, A. and Sebe, N. (2007). Multimodal human-computer interaction: A survey. Computer Vision and Image Understanding. doi:10.1016/j.cviu.2006.10.019 Surveys multimodal human-computer interaction across input modalities, fusion architectures and application areas.
• Jones, L. A. and Sarter, N. B. (2008). Tactile Displays: Guidance for Their Design and Application. Human Factors: The Journal of the Human Factors and Ergonomics Society. doi:10.1518/001872008x250638 Reviews tactile displays to help practitioners decide when to use touch for information representation, and names the roles touch suits: guidance, attention management and sensory substitution.
• Koskinen, E., Kaaresoja, T. and Laitinen, P. (2008). Feel-good touch: finding the most pleasant tactile feedback for a mobile touch screen button. Proceedings of the 10th international conference on Multimodal Interfaces. doi:10.1145/1452392.1452453 Three experiments sought the most pleasant tactile click for a virtual button; feedback beat no feedback, and piezo actuators rated slightly more pleasant than a vibration motor.
• Lederman, S. J. and Klatzky, R. L. (2009). Haptic perception: A tutorial. Attention, Perception and Psychophysics. doi:10.3758/app.71.7.1439 Tutorials the perceptual and cognitive systems underlying touch, covering tactile and haptic sensing with texture, shape and object recognition. It gives the limits of what the channel can encode.
• MacLean, K. E. (2008). Haptic Interaction Design for Everyday Interfaces. Reviews of Human Factors and Ergonomics. doi:10.1518/155723408x342826 Relates human capabilities and hardware to everyday haptic roles including notification, augmentation of graphical interfaces, expressive control and affective communication. Haptics is role-specific rather than general.
• Oviatt, S. (1999). Ten myths of multimodal interaction. Communications of the ACM. doi:10.1145/319382.319398 Catalogues ten myths of multimodal interaction, arguing that users integrate modes in patterned ways and that architectures must handle parallel signals. Multimodality is composition, not accumulation.
• Yatani, K. and Truong, K. N. (2009). SemFeel: a user interface with semantic tactile feedback for mobile touch-screen devices. Proceedings of the 22nd annual ACM symposium on User Interface Software and Technology. doi:10.1145/1622176.1622198 Presents SemFeel, vibration motors on the back of a mobile touchscreen that tell the user where a finger landed and what the touched object is. Participants distinguished ten patterns at approximately 90 per cent accuracy.