T9-QWERTY: The Ultra Efficient Minimal Motion Keyboard Layout

[shaaniqbal]

The Layout

     a      s     d      f       v          j       k     l       ;
    cyk   su   thj   evw  nbz      adf   olx  img   rpq

Select/conflict resolution keys:
m: select second word
i: select third word
o: select fourth word

Finger positions are as standard, except for "nbz" on the v key which can be pressed with the left thumb (right thumb for spacebar).

How to Begin

To begin practicing the layout, you can use the 10FastFingers typing speed test or this alternative test. You will first need a text replacement program.

1. Download and install Phrase Express. During setup you are asked to Select Components. Uncheck the box that says 'Install sample phrase library' if present.
2. Download and install the practice file T9-QWERTY. Choose the "Load as an additional phrase file" option.

Let's try typing "really". Follow the layout and you have ";-f-j-k-k-a". Press each key once just as with normal typing. Some words conflict with others. To resolve this the most frequent word is typed using the normal letters, while to type the second most frequent word you would type "m" at the end of it. For example, "good" and "gold" conflict, but "good" is more frequent than "gold". So to type "gold" you would type "lkkjm".

Conflicts in the practice file are shown below with the leftmost word being the more frequent:

good > gold
round > pound
told > fold

Background

I started off interested in learning stenography using Plover for high speed transcription. I recalled that a dictionary based input was also used on my old Sony Ericsson mobile phone, making it much more efficient to send text messages. I was aware of alternative keyboard layouts, being a touch typist with QWERTY and having also learned the Dvorak layout. I found that the alternative layouts such as Colemak claimed that by reducing motion they could increase speed and comfort. Whilst I was not convinced of the validity of this assumption I realised that a reduced key dictionary based input ported to a normal PC keyboard, an idea which has been considered previously, would reduce motion to a greater degree while still allowing the user to use the familiar QWERTY layout input when required, if used with a text replacement program.

On such a layout each key represents multiple letters and resulting ambiguity is resolved by 'select' keys, with the most frequent word being the default output.

Why the name "T9-QWERTY"?

T9-QWERTY takes the name "T9" from its use of only 9 keys for the letters, and "QWERTY" because of the fact that it can be used alongside QWERTY. Given the low number of words that can be typed on QWERTY's home row, a conflict is unlikely. When you want to type a word that isn't in its dictionary you can fall back on normal QWERTY.

Alternative Names

T9, T9 for PC, T9 on PC, T9-QWERTY, QWERTY-T9, Minimum Motion Keyboard Layout

Layout Variations

     q      w     e      f       v          j       i      o      p
    cyk   su    thj   evw  nbz      adf   olx  img   rpq

This variation places the middle, ring and small fingers on the top letter row which some users may prefer.

Letters from each column of the traditional layouts can be grouped onto the respective homerow keys, to create a derivative reduced key ambiguous layout shown below. Transition to such a layout may be easier as letters stay on the same fingers.

QWERTY:

      a       s      d       f      g         h         j      k     l      ;
    qaz   wsx   edc   rfv   tgb      yhn    ujm   ik    ol     p

Dvorak:

     a     o       e      u       i         d        h        t      n      s
     a     oq     ej    puk   yix      fdb    ghm   ctw   rnv   lsz

Colemak:

      a       r       s      t      d         h       n      e      i      o
    qaz   wrx   fsc   ptv   gdb      jhk    lnm   ue    yi     o

Update Log
-Addition of select keys for simultaneous use as delimiters
-Addition of space to phrase content of words requiring select keys
-Removal of trailing delimiter for select keys
-Removal of apostrophe and semicolon as postfixes

See also

Sentence reconstruction using word ambiguity resolution and Minimum Motion Keyboard
Plover for stenography on a QWERTY keyboard
QWERTH, a 15 key input for Android devices
ASETNIOP, a 10 point input for typing on any surface

Do you have coding skills?

If you are interested in helping me with this project, I would like to hear from you. Plans include implementation of a full English dictionary, a personal dictionary which automatically learns new words typed with QWERTY, an independent program with a single install file, progressive word stems displayed as you type, an autocomplete function, and a context sensitive predictive algorithm to eliminate the need for manual conflict resolution.

I'll update this later with more details.

[tp4tissue]

Umm......  is this for regular keyboards, or is it a single key replacing up to 3 letters each.?

[Oobly]

This is very interesting. I can see it being useful for some cases, such as typing in loads of text. It's essentially a multitap mobile phone numpad with prediction for computer I guess.

I don't think I'd use it, but perhaps RSI sufferers could benefit.

About the groupings, a more optimal way to do them is by letter frequency, so the most common letters are on the index and middle fingers, with the ring and pinkies having only rarer letters on, although I would weight the 2 centre keys even less than the pinkies since you have to move your index finger to reach them.

[spiceBar]

The Layout

     a      s     d      f       v          j       k     l       ;
    cyk   su   thj   evw  nbz      adf   olx  img   rpq

Select/conflict resolution keys:
m: select second word
i: select third word
o: select fourth word

Finger positions are as standard, except for "nbz" on the v key which can be pressed with the left thumb (right thumb for spacebar).

How to Begin

To begin practicing the layout, you can use the 10FastFingers typing speed test or this alternative test. You will first need a text replacement program.

1. Download and install Phrase Express. During setup you are asked to Select Components. Uncheck the box that says 'Install sample phrase library' if present.
2. Download and install the practice file T9-QWERTY. Choose the "Load as an additional phrase file" option.

Let's try typing "really". Follow the layout and you have ";-f-j-k-k-a". Press each key once just as with normal typing. Some words conflict with others. To resolve this the most frequent word is typed using the normal letters, while to type the second most frequent word you would type "m" at the end of it. For example, "good" and "gold" conflict, but "good" is more frequent than "gold". So to type "gold" you would type "lkkjm".

Conflicts in the practice file are shown below with the leftmost word being the more frequent:

good > gold
round > pound
told > fold

Background

I started off interested in learning stenography using Plover for high speed transcription. I recalled that a dictionary based input was also used on my old Sony Ericsson mobile phone, making it much more efficient to send text messages. I was aware of alternative keyboard layouts, being a touch typist with QWERTY and having also learned the Dvorak layout. I found that the alternative layouts such as Colemak claimed that by reducing motion they could increase speed and comfort. Whilst I was not convinced of the validity of this assumption I realised that a reduced key dictionary based input ported to a normal PC keyboard, an idea which has been considered previously, would reduce motion to a greater degree while still allowing the user to use the familiar QWERTY layout input when required, if used with a text replacement program.

On such a layout each key represents multiple letters and resulting ambiguity is resolved by 'select' keys, with the most frequent word being the default output.

Why the name "T9-QWERTY"?

T9-QWERTY takes the name "T9" from its use of only 9 keys for the letters, and "QWERTY" because of the fact that it can be used alongside QWERTY. Given the low number of words that can be typed on QWERTY's home row, a conflict is unlikely. When you want to type a word that isn't in its dictionary you can fall back on normal QWERTY.

Alternative Names

T9, T9 for PC, T9 on PC, T9-QWERTY, QWERTY-T9, Minimum Motion Keyboard Layout

Layout Variations

     q      w     e      f       v          j       i      o      p
    cyk   su    thj   evw  nbz      adf   olx  img   rpq

This variation places the middle, ring and small fingers on the top letter row which some users may prefer.

Letters from each column of the traditional layouts can be grouped onto the respective homerow keys, to create a derivative reduced key ambiguous layout shown below. Transition to such a layout may be easier as letters stay on the same fingers.

QWERTY:

      a       s      d       f      g         h         j      k     l      ;
    qaz   wsx   edc   rfv   tgb      yhn    ujm   ik    ol     p

Dvorak:

     a     o       e      u       i         d        h        t      n      s
     a     oq     ej    puk   yix      fdb    ghm   ctw   rnv   lsz

Colemak:

      a       r       s      t      d         h       n      e      i      o
    qaz   wrx   fsc   ptv   gdb      jhk    lnm   ue    yi     o

Update Log
-Addition of select keys for simultaneous use as delimiters
-Addition of space to phrase content of words requiring select keys
-Removal of trailing delimiter for select keys
-Removal of apostrophe and semicolon as postfixes

See also

Sentence reconstruction using word ambiguity resolution and Minimum Motion Keyboard
Plover for stenography on a QWERTY keyboard
QWERTH, a 15 key input for Android devices
ASETNIOP, a 10 point input for typing on any surface

Do you have coding skills?

If you are interested in helping me with this project, I would like to hear from you. Plans include implementation of a full English dictionary, a personal dictionary which automatically learns new words typed with QWERTY, an independent program with a single install file, progressive word stems displayed as you type, an autocomplete function, and a context sensitive predictive algorithm to eliminate the need for manual conflict resolution.

I'll update this later with more details.

I had totally missed this post! I think it would have been OK to also post a reference in the "Keyboard" subforum.

Interesting. I need to think about it. Don't worry about the lack of response of the community. Anything that is not about clacks or colorful keycaps takes a lot of brain power for these guys to digest. Look for example at tp4tissue's answer: he doesn't even understand what you are talking about! You need to add pictures. Like in books for preschoolers.   (I think I'm going to get a lot of love for this)

Thanks for sharing.

[poog]

I heard alternate layouts try to avoid having to use a single finger too many times in a row but with the first layout you showed, or bunching all the most used letters together, things like dad or add could be up to four keystrokes with the index finger right? Is that a problem? And do you find this to be faster than things like plover? Seems pretty cool though

[hoggy]

For dvorak and Colemak, could you change the order of the groups so it's home row key, top row key and then bottom row key? It wouldn't be any harder to learn and might be faster.

[jacobolus]

Ideally you probably want to run some kind of optimization routine which minimizes the number of word conflicts for English, before deciding on which letters to put on which key. Then you probably want your software to predict which word is likeliest based on context.

You’d actually get a better result by trying to place a phoneme / set of phonemes on each key, instead of a set of letters. Some words are homophones, but you can usually pick between them by context.

With the latter approach, you could probably re-use a lot of the code guts of existing speech-to-text engines [in particular, the prediction of words based on context and guessed phoneme].

[shaaniqbal]

Thanks for the responses so far.

[berserkfan]

this doesn't work for me or anyone who needs extensive use of numbers, symbols, etc.

I can imagine it would be perfect for a programming geek though.

Ultimately, the keys on a standard ANSI/ ISO layout are there for a reason. Because people use them. So keyboards have evolved to offer as many commonly used symbols and functions as possible for the greatest number of people.

Like you I consider the modern 104-key layout greatly imperfect. However, I do not believe cutting things down to just one row would solve matters for the vast majority of people, like 99%.

If we want a minimum layout that maximizes the number of keys that can be depressed at any time, I think the four fingers can handle one home row, one on top, one below, as well as two columns between and two on the sides. That would be 24+12=36 keys.

The thumbs are always doing unskilled labor in pressing down the space bar. Methinks the spacebar can be turned into 4 keys minimum. That would give us a grand total of 40.

Within this 40-key space which is already smaller than Minila layouts, it would be nice if we could figure out a way to incorporate at least 120 key functions with minimum use of combo presses to toggle other layers.

I believe the space bar is simply too inefficient. We only need 1 space bar of no more than 2x because we almost always wind up hitting the space bars in exactly the same spot. The other 3 keys in the space bar zone should be like shifts, toggling different layers. If we have 3 function layers and 36 other keys, we would get 108 functions available just by a combo of one thumb-key and one other key.

We could then refine the design a bit further, shuffle the key and numeric arrangements around a bit, create a numpad, etc available by pressing the top L and R little finger keys (ie banish caps lock and numlock to the keys that you would still need to stretch furthest to press normally).

You've inspired me to start looking for a POS matrix to try out some keyboard arrangement experiments. But I fear that once I get hold of such a matrix and get hooked on my ideal keyboard arrangement, all my lovely Model Fs and Ms will be abandoned and looking for new homes!

OR Possibly the two additional columns in between the home row/colums, and the two additional columns on the outside, could be utilitized to offer shift-like functions to toggle additional layers. Having 12 keys to toggle different layers would provide enormous capabilities.

[spiceBar]

this doesn't work for me or anyone who needs extensive use of numbers, symbols, etc.

I can imagine it would be perfect for a programming geek though.

Ultimately, the keys on a standard ANSI/ ISO layout are there for a reason. Because people use them. So keyboards have evolved to offer as many commonly used symbols and functions as possible for the greatest number of people.

Like you I consider the modern 104-key layout greatly imperfect. However, I do not believe cutting things down to just one row would solve matters for the vast majority of people, like 99%.

If we want a minimum layout that maximizes the number of keys that can be depressed at any time, I think the four fingers can handle one home row, one on top, one below, as well as two columns between and two on the sides. That would be 24+12=36 keys.

The thumbs are always doing unskilled labor in pressing down the space bar. Methinks the spacebar can be turned into 4 keys minimum. That would give us a grand total of 40.

Within this 40-key space which is already smaller than Minila layouts, it would be nice if we could figure out a way to incorporate at least 120 key functions with minimum use of combo presses to toggle other layers.

I believe the space bar is simply too inefficient. We only need 1 space bar of no more than 2x because we almost always wind up hitting the space bars in exactly the same spot. The other 3 keys in the space bar zone should be like shifts, toggling different layers. If we have 3 function layers and 36 other keys, we would get 108 functions available just by a combo of one thumb-key and one other key.

We could then refine the design a bit further, shuffle the key and numeric arrangements around a bit, create a numpad, etc available by pressing the top L and R little finger keys (ie banish caps lock and numlock to the keys that you would still need to stretch furthest to press normally).

You've inspired me to start looking for a POS matrix to try out some keyboard arrangement experiments. But I fear that once I get hold of such a matrix and get hooked on my ideal keyboard arrangement, all my lovely Model Fs and Ms will be abandoned and looking for new homes!

OR Possibly the two additional columns in between the home row/colums, and the two additional columns on the outside, could be utilitized to offer shift-like functions to toggle additional layers. Having 12 keys to toggle different layers would provide enormous capabilities.

I think it's great that you shuffle ideas and think about new ways of doing things.

One of the things you say, however, goes against this. You say: "Ultimately, the keys on a standard ANSI/ ISO layout are there for a reason. Because people use them."

This is a great justification for inertia, and unfortunately it used to be correct.

Not anymore. Now, we have much more computing power available. We can put a whole computer inside a keyboard. Or we can dedicate one percent of the power of the computer the keyboard is attached to, and do some things differently.

The T9 idea exploits this idea. Autocomplete in most current IDEs does as well.

Our keyboards come from an era of computing power starvation. In those days, they would have added a physical key for a new function, because overloading an existing key with the new function would have meant a lot of additional circuits in the keyboard, and it was simply cheaper to just add a key.

Nowadays, we are able to have a 60% keyboard mimic a much larger one by overloading keys. This is not very complex, but all this work is done inside the keyboard.

Just an idea: a modern keyboard could have a multilingual dictionary built-in. It could be possible to switch this keyboard in a "I'm typing text" mode in which the keyboard would assume that you are in a word processor or in an editor, and would offer word autocomplete. The keyboard could finish your words after the first letters have been typed, and if the words has been incorrectly guessed the keyboard could send backspaces to show you its next guess, or let you write it yourself.

That may sound overkill, but it's similar to what T9 does.

Going full T9 is maybe a little bit too much, but who knows. I had a friend who was able to type SMS incredibly fast with just one hand on his Nokia phone. It does not take long for anyone to be quite efficient with T9. It is certainly faster and easier to learn than to learn typing on a standard keyboard.

So while I'm not really convinced by the OP's idea, I think that we have certainly not explored much of what an "intelligent" keyboard could offer.

[jacobolus]

I recommend anyone interested in these topics read Jef Raskin’s book The Humane Interface. You may not agree with it, but it discusses several important considerations having to do with multi-function keys, etc.

In particular, I think people overestimate the advantage of having things be done “automatically”, and neglect the disadvantage: the system becomes less predictable and actually requires more deliberate action and more explicit correction of mistakes.

When a system behaves entirely predictably, humans are very good at memorizing a sequence of actions within that system, and performing them very efficiently. Thus, when we type on a keyboard, we start chunking our finger movements at the level of words, and not consciously thinking about each letter anymore.

If you have autocorrection and word completion that sometimes completes with the word you want, and sometimes completes with a different word, then it throws a giant wrench into that system. Personally, I find cutesy autocomplete and autocorrect type systems which are designed to perform their function without being explicitly called on can make flashy demos that sell well, but actually end up being a gigantic pain in the ass in practice, causing a lot more frustration and trouble and not really yielding efficiency improvements. On the other hand, autocomplete that needs to be explicitly invoked via a separate key can be extremely helpful, e.g. when trying to type long function names in a code editor.

[shaaniqbal]

When a system behaves entirely predictably, humans are very good at memorizing a sequence of actions within that system, and performing them very efficiently. Thus, when we type on a keyboard, we start chunking our finger movements at the level of words, and not consciously thinking about each letter anymore.

T9 (word disambiguation) by itself is quite a predictable system. The same sequence of letters always produces the same word. When you throw in context aware prediction and autocomplete is when things get less predictable.

[jacobolus]

T9 (word disambiguation) by itself is quite a predictable system. The same sequence of letters always produces the same word. When you throw in context aware prediction and autocomplete is when things get less predictable.

It’s not predictable by my meaning of the word. Basically, I can’t predict, before I type a particular sequence of letters, what word will be produced, unless I’ve already memorized the result. I have a memory of English spelling which allows me to use a traditional keyboard in a highly efficient way, and I don’t have to *ever* consider what the result will be when I type a particular sequence of letters. On your new system, I need to guess, and when I’m wrong, I need to go back and fix my guess. Even once I’ve spent hundreds of hours memorizing the results of various key sequences, I still have to spend conscious effort wondering whether any of the upcoming words I want to type might not be among the set I’ve memorized, or whether I’m mis-remembering what word was first for a particular key sequence.

It’s more cognitive effort, slower typing, more mistakes.

(Note, this is different from Proword’s setup in Word Perfect, where he himself has manually created all of the shortcuts, and they only are applied when he explicitly requests one. I don’t have the desire to create a system like that for myself, but I think it’s an impressive achievement, and can understand the attraction.)

At the point where someone is willing to invest that kind of effort, I’d personally just point them at existing stenography keyboards/systems.

* * *

Anyway, don’t let me dissuade you from working on this though! I think it’s an interesting project, despite my misgivings about it.

[shaaniqbal]

For it to be predictable you could rote memorise every single conflict word if you wanted. Apart from that you spell words exactly as you normally do. Conflict words are arranged in order of frequency within a corpus. That ordering stays the same. So when you've written a conflict word a number of times you end up learning to type the select key at the end of it. 

[jacobolus]

I think you’re missing my point.

On a traditional keyboard, I don’t need to look at my hands or the screen, and I don’t need to think about anything except the words I want to type. Once I’ve made the relevant gesture for each word, which I am 100% certain will produce the output I’m looking for, no matter what the word is, even if I just made it up on the spot, then I don’t need to worry any further about what the computer is going to do with my input.

But I admit I might be atypical. On those cell phones with just a numeric keypad, I always turned off the feature for predicting my words, because I found that whenever they got it wrong, which was often, it caused me incredible frustration. I turn off all autocorrect features in my text processing programs on the computer.

[shaaniqbal]

To type new words, ideally you'd type them in QWERTY and they would automatically be added to your dictionary.

The best way to arrange letters I think is as you said, to try and reduce conflicts so less memorisation is required. The problem with the standard letter arrangement used on T9 with "dumphones" is that letters are arranged alphabetically which causes more conflicts.

I think autocorrect could also have a lot of potential if it is done right.

[berserkfan]

Hello There SpiceBar and Jacobolus

I did not mean to sound conservative. I don’t use quite a number of key functions such as the caret ^ or the tilde. I am assuming that since they haven’t been replaced, they are (or were) useful to substantial numbers of people. EG programming geeks use them all the time, and probably ^ should be where $ is because $ is easier to type without stretching.

So on my keyboard, I will assume that all symbols available on a 104 MUST be available on my 40-key layout. Even if I don’t use it. Because somebody might.

I understand what Spicebar is saying about autocomplete. It works for most people most of the time. It would be a big disaster for people who use a vast range of words in different languages though. I freely use words from different languages (when not communicating with Americans) and also words from different occupations (eg legal, economics or medical terminology). Typically the autocomplete function on my smartphone cooks up the wrong word and I have to correct it. I would strongly not prefer an intelligent keyboard because such keyboards will default to the ‘lowest common denominator’ intelligence of the world’s largest English language market, the United States, and would totally not be suitable for me.

Jacobolus is fully right in his statement about overestimating the advantage of having things done automatically. I have enormous difficulty with text predictors, and I suspect people smarter than I with bigger vocabularies will have even more problems. Lots of medical terms are already left out of text predictors simply because they are too niche.

I suspect that in unique locations such as Dubai, Hong Kong and Singapore, where multilingualism is the norm and people will invoke different languages in the same text and expect other people to understand, are very ill-served with autocomplete functions.

In Singapore all GPS available on the market come with either US-accented English or China-accented Mandarin. Due to the small size of the local market, there are no local accents. This means both the English and the Mandarin automated GPS reader will botch up all local names in the most ridiculous ways, and older people are very uncomfortable with using GPS due to the accents.

Now to address Shaaniqbal. This is just a guess, but your name suggests a connection with Pakistan or Afghanistan. Now these are multilingual places where many urban areas are ethnically mixed. Somebody using an Urdu or Pashto text predictor will not be likely to stay entirely within Urdu or Pashto. It’s is certain that they will invoke Arabic, Dari, Sindhi, Hindi and Uzbek words, in addition to the usual English slang that makes it worldwide. However, not every user of Urdu will use the same non-Urdu words (Eg a more educated guy might use more English; a more religious guy might use more Arabic, a guy in the north will use more Pashto and a guy in the Southwest will use more Baluchi; an anime fan would put Japanese and Korean words in his emails and SMSes to his friends). So you’re never going to get a common dictionary created. You’ll get a procrustean bed that won’t fit most Urdu speakers. (Heck, how do you translate procrustean bed in most languages? Use the English phrase? In fact, now that I’ve mentioned it, there is no Chinese equivalence for this Greek word so when using Chinese I would probably select a different phrase with different contexts to avoid having to interrupt the flow of language with a Greek word which requires switching language mode) But considering that Urdu has more native speakers than German, and as many speakers overall as Portuguese/ Brazilian, it’s not a small niche language. Somebody eg Samsung is going to try and make money off these 200million Urdu speakers. So alternatively you wind up with a vast dictionary with 10+ options for the first few letters that you type.

Now the problem with a vast dictionary, is that you have to reach for the mouse or arrow keys and scroll through your many options. That’s very true of Chinese Hanyu Pinyin input, where sixty or more characters share the first two letters so you can’t even use the numeric keypad to select from the first 10 options.

There are two types of system at this point. One will bump up a character progressively the more you use it, so it gradually rises to the front of such a list. My experience with a system is that you have to keep looking for the character where it wasn’t last time. Extremely irritating because when I use Chinese, sometimes I am talking about literature and sometimes about economics/ politics and so not always the same characters get used more often.

Another system keeps all characters in their places. Again, infuriating to me. Because I often use technical vocabulary, and a character that is more obscure may get used a lot, requiring scrolling through every single time.

Actually, the Chinese characters for keyboard also don’t get used much outside the usage for ‘key’, so on a Nokia phone I had to search 3 screens down to find the character. I would not like to conduct a geekhack conversation about keyboards in Chinese because it’s such a pain to type keyboard – or else I would prefer to pick up the pen and tablet and just write the character.

Result? I never use text prediction. The majority of my friends and relatives don’t use text prediction. It’s more of a hassle than benefit. It only benefits people in a situation where they use a relatively limited range of words repetitively in quick succession, such as a programming geek typing goto ifthen else.

Over the next year I intend to pursue my project of a POS matrix keyboard with about 40 keys. I can't rely on prediction, but I can rely on the notion of programmability. If Soarer says teensy can accept 6 function layers, I'll go for all 6 and maximize the numbers of uses.

[shaaniqbal]

I've set up a GitHub repository so anyone can contribute their work for this project. Thanks to Peter Kamb, developer of the One Hand Keyboard for the suggestion.

https://github.com/shaaniqbal/T9-QWERTY/

While Phrase Express is brilliant for demoing how it works, I hope we can have a standalone app (like Plover or One Hand Keyboard).

Things to work on:

- A standalone program
- Create a letter grouping optimization tool where we can test out different layouts and the number of conflicts.
- Dictionary generation tool
- Context sensitive prediction
- Partial word prediction
- Autocomplete
- Autocorrect
- Personal dictionary that learns what you type with QWERTY (equivalent to multitap on a phone)
- Whatever else you or I can think of


Let's make this project a success!

[shaaniqbal]

I heard alternate layouts try to avoid having to use a single finger too many times in a row but with the first layout you showed, or bunching all the most used letters together, things like dad or add could be up to four keystrokes with the index finger right? Is that a problem?

What could be implemented is a key repeater, say G or H, that repeats whatever the last key was. So you avoid same finger combos.

Or maybe holding a key for anything over a user specified amount of milliseconds could count as a double press. Not sure how well that would work in practice though.

[shaaniqbal]

You’d actually get a better result by trying to place a phoneme / set of phonemes on each key, instead of a set of letters. Some words are homophones, but you can usually pick between them by context.

With the latter approach, you could probably re-use a lot of the code guts of existing speech-to-text engines [in particular, the prediction of words based on context and guessed phoneme].

I like this idea jacobulous. T9-Steno sort of.

[shaaniqbal]

Umm......  is this for regular keyboards, or is it a single key replacing up to 3 letters each.?

It's for regular keyboards.

[shaaniqbal]

For dvorak and Colemak, could you change the order of the groups so it's home row key, top row key and then bottom row key? It wouldn't be any harder to learn and might be faster.

How would it make a difference?

[shaaniqbal]

You've inspired me to start looking for a POS matrix to try out some keyboard arrangement experiments. But I fear that once I get hold of such a matrix and get hooked on my ideal keyboard arrangement, all my lovely Model Fs and Ms will be abandoned and looking for new homes!

OR Possibly the two additional columns in between the home row/colums, and the two additional columns on the outside, could be utilitized to offer shift-like functions to toggle additional layers. Having 12 keys to toggle different layers would provide enormous capabilities.

Thank you for the response, beserkfan. I'm glad you've been inspired to go on and do your own thing.

[shaaniqbal]

So while I'm not really convinced by the OP's idea, I think that we have certainly not explored much of what an "intelligent" keyboard could offer.

Definitely! One of the other ideas I've had is "spacebarless typing" where spaces get added automatically. And I really want to see a really intelligent autocorrect program for PC.

[shaaniqbal]

This is very interesting. I can see it being useful for some cases, such as typing in loads of text. It's essentially a multitap mobile phone numpad with prediction for computer I guess.

I don't think I'd use it, but perhaps RSI sufferers could benefit.

About the groupings, a more optimal way to do them is by letter frequency, so the most common letters are on the index and middle fingers, with the ring and pinkies having only rarer letters on, although I would weight the 2 centre keys even less than the pinkies since you have to move your index finger to reach them.

Thanks Oobly. I have tried to arrange them by letter frequency but maybe you can come up with a better arrangement? I'd be interested to see what you come up with. Bear in mind you want to keep conflicts down.

I used Wikipedia's letter frequencies:

ewv = 12.702%+2.360%+0.978% = 16.04%
thj = 9.056%+6.094%+0.153% = 15.303%
adf = 8.167%+4.253%+2.228% = 14.648%
olx = 7.507%+4.025%+0.150% = 11.682%
img = 6.966%+2.406%+2.015% = 11.387%
su = 6.327%+2.758% = 9.085%
nbz = 6.749%+1.492%+0.074% = 8.315%
rpq = 5.987%+1.929%+0.095% = 8.011%
cyk = 2.782%+1.974%+0.772% = 5.528%

[shaaniqbal]

I've just noticed that the link for the T9-QWERTY pxp file was down all this time. Surprised no one commented on it. Here's the reuploaded file:

http://www.mediafire.com/download/j69bn01v7o8chzn/T9-QWERTY.pxp

Let me know if there are any problems, thanks.