I’ve mentioned our free AT-node website in other posts (like this one with a neat infographic), without really demonstrating what it is, so in this post, I want to give you a quick introduction to using the AT-node website.
As a brief refresher, AT-node is a website that organizes the available research evidence on text entry rates (typing speeds) for people with physical disabilities. Sajay Arthanat and I did a systematic review of the literature a couple of years ago and wrote about it in several papers (see kpronline.com/pubs). After all that work, we wanted to put the data in a form where anyone could use it. The result is AT-node. It includes data from all published studies where participants had physical disabilities and typing speed was measured. It covers a wide range of assistive technologies used to provide an alternative interface to the computer, such as speech recognition, modified keyboards, on-screen keyboards, switch scanning, brain-computer interface, etc.
Using AT-node, you can explore the data set by diagnosis, interface, and body site. For example, you can see all reported typing speeds for people who have cervical spinal cord injuries when using speech recognition vs. using the regular keyboard. The idea is to help inform decision-making (e.g., what interface might work best for me?) and provide rough expectations for future performance (e.g., how fast am I likely to type with this interface?).
Searching with AT-node
With that in mind, let’s see how AT-node works. Perhaps the best way is to show it in action, as in the video below.
The video shows you how to do a couple of user profile searches. The starting point is the AT-node home page. Then with a couple of clicks you specify the diagnosis and interface you’re interested in, and receive a report on all the matching data, with graphs, downloadable tables, and citations with abstracts for all the matching studies.
You can re-create the example searches from the video using the following links:
1. Cervical spinal cord injury using physical keyboard
2. Cervical spinal cord injury using either physical keyboard or speech recognition
You can also see all of the data using this link: https://kpr.pythonanywhere.com/q
How you might use this information
Here’s an example of how you might use this type of info. Remember YN, the college student from our last post? He needed an alternative access interface to type more effectively and accommodate his physical disabilities. The team did a great job measuring typing speed with the different candidate interfaces, to inform YN’s decision. YN started out at 4.4 wpm, and increased to 7.6 wpm with some AT enhancements to his keyboard.
You could use AT-node in this situation to bring in some additional evidence: for people like YN, what sorts of typing speeds have been reported, and how do those vary by interface?
To address those questions, you could do two related searches with AT-node, as described below.
Data for YN’s diagnosis, across all interfaces
First, do a user profile search for diagnosis of acquired brain injury, across all interfaces. The report shows that the average text entry rate for the 24 matching cases is 6.2 wpm, with a wide range from 1.2 to 18.6, reflecting the diversity of this population. Three different interfaces are represented: speech recognition (ASR), physical keyboard, and switch scanning, as shown in the graph below.
Speech recognition comes in with the fastest speed, but using speech is not an option for YN. The reported average for physical keyboard is 7.2 wpm, very similar to YN’s 7.6 wpm using his AT enhancements. This is considerably faster than the 2.9 wpm reported for switch scanning, which tends to confirm that using the physical keyboard is an appropriate choice for YN.
Data for YN’s diagnosis, for physical keyboard only
Next, hone in on the physical keyboard interface. This doesn’t add too much to our understanding, since we already know that YN is right in the average range for physical keyboard typing. But it’s somewhat interesting to look at the results by body site, as shown below.
Here we see that the fastest folks in this population are those who are using multiple fingers on both hands (averaging 11 wpm). Those using a single finger on one hand, as YN does, come in at 5.6 wpm. Thus YN is 36% faster than the most comparable cases in the research dataset (others with acquired brain injury who type with a single finger). This also tends to add confidence that YN’s typing method is on the right track.
Admittedly, the most important thing is how YN himself is doing with the setup. But the additional evidence from the literature helps provide a context for YN’s results, which can give a nice reality check and possibly lead to further enhancements for YN.
And when it’s so easy to get the additional data, why not?
What would you use AT-node for? What additional features would you want to see?