Intelligent Error-Correction from Marginal Keypresses

One of the most frustrating things about using a conventional keyboard is the experience of typing a word, missing a letter, and having the system assume I’m going for a different word to the one I’m trying to type (and thus replace the nearly-right word with a completely wrong one). Very often I’ll go for the letter with my finger and touch the key, but not quite press hard enough to activate the key. I’m human and so inconsistent in how hard I hit even the same key each time. I can’t be the only person for whom this is a reality (If I am, don’t tell me!).

This happens a lot on my Mac, where I go to type a word like “beam” and miss the final ‘m’, and so it corrects me to “be a” (and I suppose it could also go for “beat” depending on the context). Obviously, it’s the computer making its best guess when my input is non-English/nonsense, but more often than not it guesses wrong.

It strikes me that the TextBlade has more information at its disposal that might enable it to make fewer errors in this respect. I’ve read about how it has a spectrum of input where a threshold is determined to qualify a keypress (and how over time it might change its threshold to compensate for long fingernails, for example). Is there any possibility that the TextBlade could use its input data to make more intelligent suggestions for which word I mean when I miss a letter?

It seems that this could be possible through measurement of touch input, taking note of when a particular activation makes it only to 80% of the threshold, and thus suggesting the letter I nearly hit when I get to the end of the word. I feel like this could be combined with some kind of timing analysis, to find a longer-than-usual gap in my typing and thus determine where in the word the additional letter needs to go. In my example above this would enable it to decide between the missing character being a space in the case of “be a” and a letter at the end of the word in the case of “beam”.

@waytools Is this stuff possible/feasible with modern processing power? Does the TextBlade do anything like this already?

I’m a TREG user so I have been using the TB for many months. I can’t speak for WayTools but I don’t think what you are describing is possible with the TB. The TB uses its intelligence to try to figure out what letter you are typing based on your touches and presses to the device. It then feeds this information to the computer where the OS of that device makes the spelling corrections you are describing. I will tell you that WayTools has done an amazing job with trying to figure out the correct letter to send to your computer but in the end it does not send words but rather characters.

Interesting that you mentioned sometimes not getting a key to press down. That kind of error probably represents 40% or so of all my typos! In my case, it feels like my finger just locks up as I hit the key. Very weird.

Anyway, I think you may have a good idea there and maybe the basic tech required already exists in the TB. It does sense touch so it would be interesting if that could be used to influence the correction process. Like in your example of “beat”. You successfully hit “bea” but only tap the “t” without pushing it down.

It seems logical that it could sense the timing (to help judge if contact with the “t” is likely a continuation of a word) and what force you used (I suspect that, even without pushing the key down, you strike more forcefully than if you just were making incidental contact) and factor all that in for corrections. I like the idea! Could even turn out to be brilliant.

@bobgillen So jealous! Would love to be in TREG, anything to take the bulk out of my current setup. And yes, it may well be beyond the sandbox of a bluetooth keyboard to have a purview over whole words/sentences, but in the fullness of time with AI improving every part of our lives then perhaps.

@dabigkahuna Yep, it’s definitely the main error I make on a normal keyboard. That’s a good description of the experience, the force just isn’t there. And darn, maybe I shouldn’t have mentioned it! I’ll register a patent right away…

1 Like