Typing with a tongue computer interface

It’s possible to type with your tongue using tiny sensors worn in the mouth. This post presents text entry rate results from a recent study, including two people with cervical spinal cord injuries.

Image shows the 18 sensors embedded into the device, which is worn like an orthodontic retainer. Also shows how letters are assigned to each sensor to allow for typing with the tongue.

Image shows the 18 sensors embedded into the device, which is worn like an orthodontic retainer. Also shows how letters are assigned to each sensor to allow for typing with the tongue.
For about 30 years, researchers have experimented with different ways of typing with your tongue. The Tongue Touch Keypad from the 1990’s used a tiny keyboard embedded in an orthodontic-style retainer. A newer approach continues to use the orthodontic-style retainer, but now embeds inductive sensors that are activated by moving a small magnet attached to the tongue. How well can you type with this sort of tongue computer interface? Read on for results from a recent study.

As part of our AT-node project, we regularly search the literature on the use of alternative access interfaces by people with disabilities. This study popped up in our latest search: Error-free text typing performance of an inductive intra-oral tongue computer interface for severely disabled individuals by Lotte N.S. Andreasen Struijk and colleagues, published in the IEEE Transactions on Neural Systems and Rehabilitation Engineering, November 2017.

The goal of the study was to measure text entry rate for people using a particular tongue computer interface, including people with tetraplegia, who are part of the target user population for this typing method.

The ITCI — Intra-oral Inductive Tongue Computer Interface

The paper gives a good description of the ITCI, and I’ve included Figures 2, 3, and 4 from the paper below, which will hopefully clarify how it works. Basically, the retainer holds 18 sensors at the roof of your mouth. You attach a magnet to your tongue; this can be glued for a temporary trial, or, as in this study, attached more permanently using a tongue piercing. As you move your tongue, you activate different sensors in the retainer.

The study authors set up two methods of typing using this basic set up:

  1. Keypad mode: assigns a group of several letters to each of 10 sensors, much like a telephone keypad (see Fig 2 below). To type a letter, you first activate the sensor that contains your letter. Then you specify which specific letter you want, from the group you just selected. This involves a query procedure where you continue activating the same sensor (holding your tongue in the same position), while the system cycles through each letter in the group. When your letter is displayed, you deactivate the sensor to confirm the selection. Fig 3 below illustrates and describes the typing procedure.
  2. Mousepad mode: uses the sensors to move the mouse cursor and generate clicks (see Fig 2). Each of 8 cursor directions is assigned to a sensor, and any of 4 sensors can generate a click. You type by using the mouse cursor to control a typical Windows on-screen keyboard.

Figures 2, 3, and 4 from the Andreasen Struijk paper. Figure 2 shows the palate-worn interface with its 18 sensors, as well as how letters and mouse functions are assigned to each sensor for use during Keypad typing or Mousepad typing. Figure 3 describes the typing procedure during Keypad mode: with 4 letters in the chosen group, the system cycles through each letter in the group. When your letter is displayed, you deactivate the sensor to confirm the selection. Figure 4 shows a drawing of a human head with the interface worn in the mouth and controlled by the tongue. Signals are sent wirelessly to a PC laptop with a USB interface.

This interface is available at tks-technology as their Itongue product.


The study initially included 4 people with tetraplegia and 2 people with typical motor function. The 2 typical individuals already had tongue piercings several years prior to the study. The 4 people with tetraplegia agreed to get a tongue piercing, after trying the ITCI for a day using a glued tongue magnet. Two of those individuals did not complete the study, however: one didn’t like the tongue piercing and had it removed, and the other had significant difficulty typing with the ITCI and discontinued after trying for 2 days.

So the text entry rate data comes from 2 people with tetraplegia (P1 with a C1-2 spinal cord disease, and P2 with C5 spinal cord injury), and 2 people with typical motor function.

Text entry rate measurements

As you can imagine, this method of typing takes a bit of practice. Fortunately, the study design allowed participants to gain a fair amount of experience. Participants used the ITCI for about 8 hours prior to the final typing measure. This was divided about evenly between the Keypad and Mousepad modes, and was spread over 5 sessions on 5 different days.

The full paper reports various performance measures for all of the sessions, and for subtasks within each session. For our purposes, the main take-home metric is how well participants could type a short paragraph, at the end of the study — specifically, how many correct characters could they type per minute. I’ve included Fig 10 from the paper below. Remember that participants 1 and 2 have tetraplegia, and participants 5 and 6 do not.

Figure 10 from the Andreasen Struijk paper. Bar graphs show text entry rate in correct characters per minute for each participant with each of two texts. Left bar graph has results for Keypad mode, for participants 1, 2, 5, and 6: 5, 6.5, 9, and 6. Right bar graph has results for Mousepad mode, for participants 1, 2, 5, and 6: 7.5, 8, 16, and 14.

Since Text1 and Text2 are each short paragraphs, the overall text entry rate for each participant can be represented by averaging the results across texts. When we eyeball this from Fig 10, we get the following for participants 1, 2, 5 and 6, respectively: 5, 6.5, 9, and 6 for Keypad mode, and 7.5, 8, 16, and 14 for Mousepad mode, in correct characters per minute. To convert the characters per minute to a more common words per minute unit, we divide by 5 (representing the typical number of letters in an English word).

The following table sums up the results for each participant, showing text entry rate in words per minute (wpm):

Participant Spinal cord injury Keypad (wpm) Mousepad (wpm)
1 C1-2 1.0 1.5
2 C5 1.3 1.6
5 None 1.8 3.2
6 None 1.2 2.8

Some thoughts about the results

The Mousepad mode may have allowed faster typing than the Keypad mode, although the study is not conclusive on that. (In fact, the Keypad mode showed some advantages during the practice tasks, so the effect of mode appears to be inconsistent).

It’s hard to know what to make of the differences between subjects. Text entry rates for the two with tetraplegia were very similar. Rates for the typical participants were faster than those with tetraplegia, which the study authors attribute to their familiarity with tongue piercing. So it’s possible that the individuals with tetraplegia would continue to improve over time, but we just don’t know from the results of this study.

Some context for the results

The ITCI system does allow people to type, using only their tongue, which is a pretty amazing technical achievement when you think about it. Text entry rates were not especially fast, however. For example, the Mousepad mode text entry rate for the participants with tetraplegia averaged 1.55 words per minute (with Keypad mode only at 1.15 wpm). It’s interesting to note that the original Tongue Touch Keypad yielded similar but somewhat faster results, averaging about 3 wpm across 4 people with tetraplegia, with a range of 0.6 to 5.0, back in 1993.

Using the AT-node search tool*, we can compare these results to others reported for people with spinal cord injuries at C5 and above. Across all types of alternative typing interfaces for this population, the average is 11.3 wpm. The fastest interface is speech recognition, at about 20 wpm, followed by physical keyboard at 12.5 wpm. Those two interfaces might be usable by the participant 2 who had a C5 SCI, but probably not by participant 1 whose spinal cord was affected at the C1-2 level. That individual might be a candidate for interfaces such as sip-puff Morse code or switch scanning, however, both of which would likely yield faster typing than 1.55 wpm.

Results from an AT-node search on text entry rates for individuals with spinal cord injuries at C5 or above. Text entry rates are shown for 7 different interfaces, in words per minute: speech recognition 20, physical keyboard 12.6, 2-switch Morse code 12.4, cursor on-screen keyboard 5.3, other 3.9, scanning 1.1, brain computer interface 0.6.
AT-node search results for people with spinal cord injuries at C5 and above.

*To do this search yourself: visit AT-node, choose User Profile Search. Click Use Advanced Search, then choose the following: All interfaces, SCI high cervical diagnosis, All body sites. Click Search.

Note that these ITCI data have not yet been added to AT-node, but we plan to do that soon.

What do you think about the prospects of tongue typing?

Leave a Reply

Your email address will not be published. Required fields are marked *