Modelling of pinky/ringer-finger bigrams - General

stevep99
Member

Reputation: 117
From: UK
Registered: 14-Apr-2014
Posts: 978

Website

11-Jan-2018 14:26:00#1 Modelling of pinky/ringer-finger bigrams

There have been some discussions lately about pinky/ring finger combos, e.g. WAR, YOU.

I am thinking to add an additional factor to my layout analysis model to account for this.

Currently the model only considers same-finger bigams, so bigrams like YO or AR not recognised as being a bit more difficult. This is not purely related to thier positions on weaker fingers, because, for example AI is easier than AR, even though they are both pinky/ring combinations. The issue as I see it, is one of using pinky and ring in quick succession on the same hand.

However, I don't want to overcomplicate things, e.g. by considering trigrams or other long sequencies and trying to account of all sorts of possible easy/difficult combinations of fingers. Instead, I'd rather continue to rely on the base value score for each key, and then add on any penalties to reflect difficult bigrams.

Proposal could be as simple as: in addition to exising same-finger bigram penalties, introduce a new same-hand pinky-ring bigram penalty, say at 1.0.

I don't anticipate it would have a significant effect overall.

Any comments/suggestions before I implement?

Last edited by stevep99 (11-Jan-2018 18:33:34)

Using Colemak-DH with Seniply.

Offline

0

DreymaR
Member

Reputation: 214
From: Viken, Norway
Registered: 13-Dec-2006
Posts: 5,361

Website

11-Jan-2018 14:57:28#2 Re: Modelling of pinky/ringer-finger bigrams

Hmmm. Generally, implementing a lot of specific factors isn't a great idea. But what to use instead?

Maybe there's actual research data on the relative strength and independency of fingers? If so, one could rework the penalty system based on that. The bigrams in question are between weak and dependent fingers, which is what this is really about. Bigrams between, say, ring and middle finger probably aren't so great either and this would not be accounted for by your proposed factor but I don't know how much it'd matter of course.

*** Learn Colemak in 2–5 steps with Tarmak! ***
*** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

Offline

0

stevep99
Member

Reputation: 117
From: UK
Registered: 14-Apr-2014
Posts: 978

Website

12-Jan-2018 11:42:24#3 Re: Modelling of pinky/ringer-finger bigrams

There may well be research data on finger strengths, but even if I had it to hand, I don't know how you'd go about applying it to the practice of typing. Since keyboard layout optimization is not well studied, most models, mine included, have to be based on reasoned judgement.

What I'm really talking about is the ability to add penalties for bigrams other than same-finger. The ring-pinky combo is the most obvious candidate, but this could include middle-ring bigrams too. But if included, the penalty for middle-ring would need be a lot less in my view, as they are not too bad.

I also think it would be worth bearing in mind the row-jump for such bigrams, as I do currently for same-finger bigrams. Top-bottom row jumps are worse than top-middle row jumps, and I think this should apply to the neighbouring-finger bigrams (ring-pinky & middle-ring) too.

Last edited by stevep99 (12-Jan-2018 11:44:38)

Using Colemak-DH with Seniply.

Offline

0

DreymaR
Member

Reputation: 214
From: Viken, Norway
Registered: 13-Dec-2006
Posts: 5,361

Website

12-Jan-2018 11:45:01#4 Re: Modelling of pinky/ringer-finger bigrams

These bigrams are a real factor so modeling them is interesting. But I'm uncertain about how to arrive at realistic values.

Last edited by DreymaR (12-Jan-2018 11:45:14)

*** Learn Colemak in 2–5 steps with Tarmak! ***
*** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

Offline

0

stevep99
Member

Reputation: 117
From: UK
Registered: 14-Apr-2014
Posts: 978

Website

12-Jan-2018 14:58:27#5 Re: Modelling of pinky/ringer-finger bigrams

Out interest I have generated the bigram frequencies (percentages) for "neighbour-bigrams" (i.e. ring/pinky and middle/ring). I am not including index/middle obviously because those are very comfortable :)

== Top Neighbour-Finger Bigrams ==
fingers 0-1: AR 0.9739 %
fingers 1-0: WA 0.6321 %
fingers 8-9: YO 0.4908 %
fingers 1-0: RA 0.4367 %
fingers 1-2: RS 0.3652 %
fingers 8-9: IO 0.3286 %
fingers 8-7: IE 0.2960 %
fingers 7-8: E. 0.2488 %
fingers 2-1: FR 0.2093 %
fingers 7-8: EY 0.2081 %

Usual suspects as you'd expect, interesting they are noticeably more common on the left-hand.

Last edited by stevep99 (12-Jan-2018 15:03:13)

Using Colemak-DH with Seniply.

Offline

0

stevep99
Member

Reputation: 117
From: UK
Registered: 14-Apr-2014
Posts: 978

Website

17-Jan-2018 14:30:00#6 Re: Modelling of pinky/ringer-finger bigrams

I have now implemented this feature on my layout analyzer

In the configuration box, the "penalties" section now looks like this:

penalties:
2.5 2.5 3.5 #same-finger
0.5 1.0 1.5 #pinky-ring
0.1 0.2 0.3 #ring-middle

I picked what I thought were reasonable default estimates.
The first row represent penalties for same-finger bigrams, and work the same as before.
The next two lines represent penalties for "neighbour-finger" bigrams (i.e. pinky-ring bigrams and ring-middle bigrams on the same hand)
The columns are the penalties for (respectively) bigram keys on same row, bigram keys differ by 1 row, and bigram keys differ by 2 rows.

TBH it doesn't really make much difference, but it's nice to have this extra factor in the model.

Last edited by stevep99 (17-Jan-2018 14:34:30)

Using Colemak-DH with Seniply.

Offline

0