[LCWO LOGO]  

Login

User name:
Password:


Language
Български Português brasileiro
Bosanski Català
繁體中文 Česky
Dansk Deutsch
English Español
Suomi Français
Ελληνικά Hrvatski
Magyar Italiano
日本語 Bahasa Melayu
Nederlands Norsk
Polski Português
Română Русский
සිංහල Slovenščina
Srpski Svenska
ภาษาไทย Türkçe
Українська 简体中文

Who is online? (21)


LCWO Discussion Forum [Atom LCWO Forum Feed]

This is a simple discussion forum for LCWO users. Feel free to use it for any kind of discussion related to this website.

Thread: Improvement on Koch Method

Back to the Forum

AuthorText


Posted: 2013-10-14 15:15
The difficulty I am encountering with the Koch Method is as follows.

As you progress in the alphabet, the set of characters in use grows ever larger. This means that each new try contains a progressively smaller set of "new" characters, and is more and more occupied by characters you know well.

Therefore you have to do more and more tries before you'll have to copy a given new character a certain number of times.

The result of all this is I find myself an expert at the early characters, but struggling at the most recent ones... despite many, many tries!

As a solution I am currently experimenting with custom code groups, where the set of characters "in play" is limited to the "most recent" ones.

This solution is not perfect. One of the skills of Morse is being able to recognize the right character from among the full set of characters. It will be therefore necessary to go back to the original Koch Method sequence after I have proficiency with a given custom code group.


Posted: 2013-10-14 16:44
Would it be an idea for you to do the following:
- Start at lesson one, and go on until you are at lesson number five.
- When you can copy the first five letters, than go on with the next five letters.
- And when you can copy them also, than you add the first five letter and the second five letters together, and start copying them until you can copy them at more than 90%.
You go on with the rest of the fourty lessons the same way: every time you go for the next five letter/signs untill you can copy them well enough, and then add them to the rest of the letters/signs.
And maybe it is reccomended to start at 12 wpm for all the fourty lessons.
Then do the same thing from the beginning: first five, and then a nother five, adding them together, but with a higher speed: 13 wpm, and you repeat all the lessons until you are at,let say, 15 wpm or higher.
I don't know if this would be a good thing, but you can Always try if it suits for you.


Posted: 2013-10-14 18:57
I noticed this too. I don't think it is specific to the Koch method in general, but rather a naive implementation (with all characters equally likely). An adaptive system that favors the letters missed most often would probably be better. But LCWO doesn't do that, at least not in a way I find useful, so I implement it myself.

To make it work on LCWO, I don't find that hard-and-fast rules are useful, so what I do is simply this: as long as I'm making progress, I keep going. I keep a spreadsheet with the dates I reached a new speed, and the little line climbing from 1 towards 40 is a big incentive so I go with it. Eventually, I hit a plateau where it is hard to get a new letter. That means it is time for review.

If I think I'm generally held back by my overall letter recognition, I'll go practice at higher speeds--that works well to ensure that I'm recognizing letters by their overall sound. I keep my progress at higher speeds on the chart as well to provide incentive to work as hard on those as on my basic goal (which for me happens to be to finish the course at 12 WPM, but could be different for you). But if I think it is that I don't know the more recent letters well enough, which happens pretty often, then I start doing code groups with the most troublesome letters. If I can't do that at full speed (which would show I'm right about the problem), I'll toss some out until I can. When I'm doing well on those, I add more troublesome letters, and keep doing that until I have those letters brought up to the level of the previous ones. Sometimes I bring the troublesome letters up to 15 WPM so that I'm sure I've really beat on them. Then I go back to trying to add letters on the regular progression.

I can't say if that's the best way, but review works for me. Usually when I come back after review I immediately add several new letters, which I think shows the method is working.

I still consider this "doing the Koch method," it is just a specific method of review I find useful.


Posted: 2013-10-15 00:28
hithere:
Therefore you have to do more and more tries before you'll have to copy a given new character a certain number of times.

Yes, that's possibly the greatest weakness of most implementations of the Koch method. This site does try to weight the newer characters more, but it isn't really enough...


Posted: 2013-10-15 05:53
Hmm, I didn't notice it weights the recent letters more. It must be a subtle effect. Anyway, I don't think that's the right approach, because I find some letters I acquire immediately, some I have trouble with long after I'm past them and have reviewed them. I think an adaptive weighting toward the most often missed letters, probably with exponential decay of old data, would be best.

Admittedly, there would be some implementation subtleties due to the fact that you can hop around among the lessons and run at all different speeds.


Posted: 2013-10-15 12:52
The MorseMachine does seem to emphasize those characters which give you most trouble.

Unfortunately, it is only able to train you the 40 characters learned in the lessons, nothing more.

I would wish for all characters in the MorseMachine on the one hand, and the option to emphasize "difficult" characters in all other training modes on the other hand.

73 -- Georg


Posted: 2013-10-15 14:18
The Morse machine doesn't really do it right anyway--the weighting of a letter doesn't increase if you miss it, for starters.


Posted: 2013-10-15 16:19
Dustin,

try the following:

* Open MorseMachine.
* Set Lesson to 3 (K, M, U).
* Reset the Machine.
Then, start practising. Each time you hear a "K", type first "N", then "K" (deliberate "miss"). Each time you hear "M" or "U", type the correct letter.

You will see that the error bar for "M" and "U" decreases, while the error bar for "K" does not decrease.

As an effect, you will soon be confronted with a lot more "K" than other characters. I think it is working as it should - apart from the fact that I cannot practise letters like ä, ö, ü and ß with it.

73 -- Georg


Posted: 2013-10-15 16:49
Sigh. I know how it works--I went through all the letters with it. I just don't agree that is how it should work. It should most heavily weight the letters most often missed *recently*. That means if you start missing a letter, that letter's weight should increase. I find that to be the only effective review method.


Posted: 2013-10-15 17:20
Dustin, how do you define "a letters weight increases"?

Let's assume I do the above test, and I get "K", "M" and "U" 5 times each. Let's further assume I am perfect with "M" and "U", but I always miss "K" on my first try, and enter:

5 times "M" correctly
5 times "U" correctly
5 times "K" incorrectly
5 times "K" correctly,

then I would expect "K" to be weighted more from now on. This is exactly what happens: The probability to get a "K" instead of "M" or "U" increases.

After some time, if I from now on enter "K" always correctly, this will average out, and the chance to get a "K" will slowly drop to the same probability the other characters have.

Eventually, all characters will have the same probability again (exactly only after an infinite number of tries, the decay time does depend on how many tries you have made in total, so far).

Make a mistake (recent event), and it will be remembered again for some time (probability increases).

One could argue about which decay function would be better/best, but at least there is a memory function, and you can even interfere with the memory by manually adjusting the error bars (mouse click on them).


Posted: 2013-10-15 23:00
Your point is about relative vs. absolute weights; when I wrote that I assumed some knowledge of the how the LCWO Morse Machine works that I didn't make explicit. What you describe only works near when a letter is first introduced--soon enough the weights have decreased sufficiently that either they do not decrease anymore or the amount is too small to be measurable. With the current algorithm, there really isn't any way for it to adapt over the course of learning the whole alphabet. I repeatedly hit a situation where I had to adjust my learning of an old letter because it was "closer" to a new letter than; for example, I might at first start hitting 'Z' for 'G' because G is a prefix of Z but up until that time I have not needed to wait to hear all the characters of Z; my brain makes a decision after hearing the first three elements. That isn't a conscious decision, obviously, I try not to do that, but it happens anyway as an apparently inevitable by-product of learning to recognize the letter by reflex.

That means that when I add G (say) I must also alter my recognition of Z slightly. But Z was so long ago that it's weight is minimum or near minimum, and the current algorithm is incapable of recognizing that my learning of Z has taken a small setback and Z needs to have its weight increased.

You may say "well, just reset the machine frequently." That will work for a bit, but it doesn't scale with the number of letters involved. By the time you're more than half-way through the course it would take a great many runs to get the probabilities to adjust usefully after a reset, because the number of trials required is *per letter*. Knocking down a half-dozen letters you know to concentrate on a handful you don't is feasible. Knocking down 25 you know to concentrate on 5 you need to work on is a complete waste of time. I simply abandoned the Morse Machine as not polished enough and use the Code Groups page, which does an excellent job if you're willing to tell it which letters need work by hand.

Basically, it's just an algorithms issue; the Code Machine algorithm in use isn't sophisticated for me to use for at least the two reasons mentioned.

There is a third issue which may or may not be an algorithm design issue. After I'd more or less gone through the Morse Machine I felt I'd gone backwards in some ways and should have done Koch alone (with Code Groups and/or Word Training as an additional review tool). The Morse Machine gives you all the time in the world to recognize a letter. It would be better, and in the spirit of Koch, if it had a timeout proportional to the speed setting after which getting it correct will not decrease its weight. Since it does not have that feature, I felt like using it was in some ways going back to the Farnsworth method without the Koch progression, something I think is misguided.

It's an interesting tool, but as implemented here I didn't find it useful. Well, choice is why you have multiple tools. I'm sort of curious if the original Cunningham Morse Machine adapted in a more sophisticated way, but haven't been curious enough to download it and see.


Posted: 2013-10-16 00:08
Yes, I agree Dustin.

So, the MorseMachine would need a few addons for more usefulness:

1. Option for a custom character set including special characters (ä, ö, ü, ß, and maybe self defined characters like in the G4FON software),

2. Maybe a customizable decay function with customizable memory parameters,

3. Customizable "importance" for each character, which will increase or decrease its probability,

4. A time penalty, which will increase character probability when a long time is needed to type the correct answer (and decrease the probability if it is typed quickly and correctly).

I guess I didn't really run into the "newest characters appear too seldom" problem for three reasons:

1. I used G4FON whenever I had the impression that I really needed to practise "problem characters" a lot,

2. Increasing character probabilities by clicking on the respective error bars in the MorseMachine worked well enough for me,

3. I didn't really spend very much time on each new character, but practised sets of 3 to 5 completely new characters in G4FON instead, and went to MorseMachine and word training rather soon after finishing all LCWO characters, with all 40+ characters being more or less equally "problematic" for me.

Probably 3. is completely against the idea behind the Koch method - but it seems to work for me. :)

Nevertheless, I would still be very glad if the custom character set could please be implemented for all training methods (word training, morse machine, etc.).

73 -- Georg


Posted: 2013-10-16 03:29
Interesting. I didn't notice that you could re-adjust the bars manually. That is a big improvement over what I thought its limitations were. With the same work as using the Code Groups tool, you could do about everything except the timeout. I still think I'd need that, but it's clearly better than I realized, if manual.

I guess your method is different than Koch, but I don't know that it's clearly worse. I almost end up doing something similar when I do reviews and decide I need to hammer on, say, the last six letters. The overriding key idea of the Koch method seems to be never practicing slow. I don't see why fiddling with the progression would really invalidate that, as long as you are learning the sound at speed.


Posted: 2013-10-23 08:52
Interesting suggestion! I ran into the same problem recently when I started to learn the numbers. Well, I actually have two problems, I do believe I count dits and dahs for the numbers, and they come up too rarely in a new lesson. I created a code group with just the numbers and interpunctuation signs to specifically train these. We'll see how that works.


Posted: 2013-10-24 06:21
I can count at 12 WPM. I try not to, but when I think it is a problem I use code groups to practice (I might choose the most recent letters or perhaps some problem letters) at a higher speed, say 15WPM or better. 15WPM is a pretty good speed for me to focus on the whole letter sound. That way, I'm not only hammering the recent letters I haven't learned well or the problem letters, I'm doing so in a way that requires the proper learning.

Arguably I should just do the course at 15 WPM instead of 12 WPM. I actually do, though, it's just slower. I'm about 3/4 of the way through at 12 WPM and 1/2 way at 15 WPM. So it works out.

There may be better ways, that's just what I found useful.


Posted: 2013-10-24 10:37
Dustin, I think you are on the right way.

I would suggest not to overemphasize the counting issue (if that's an issue at all). In effect, at any speed that I master I *could* count. If I don't that's because of no time (when speed is higher) or no need due to proficiency (that comes with practice).
Of course when the signal is low or there is QRM/N I *will* count just to make sure in case of doubt.
However instead of "counting I would call this "careful listening".

So I guess beeing more than halfway through the course (congrats!) at 12wpm is good, and the counting that you fear will go away once you completed the 40 lessons and increase the speed.

Btw when I am working on problem characters I go down in speed to reach 0% errors, then slowly walk up again, trying to keep error rate low.

73

Gerd.



Posted: 2013-10-24 18:35
What i did was, i started to use the morsemachine, and bit by bit i went to a higher speed.
I started at 12 wpm with the morsemachine, about 5 months ago.
Now i am using the morsemachine with a speed up to 34 wpm, and yes, every now and then i make a mistake, but not as much as in the beginning.
So, maybe it is helpfull to use the morsemachine also, to speed up the wpm.
The higher the speed you can copy, the faster you stop counting the dits and dahs.
It works for me.
I can tell a big difference when i go back to the lessons; i can copy at a higher speed than in the beginning.


Posted: 2013-10-24 18:53
Yeah, the trouble with advice like "don't count" is that it just makes you more aware of the possibility, and once you're aware of the possibility it will be harder to avoid. I can still count at well over 30wpm speed...


Posted: 2013-10-24 20:58
The point is that you have to STOP counting, and i know how difficult that is.
In stead of counting the dits and dahs, you have to learn the SOUND of each caracter.
Once you've learned that, the counting won't be a problem anymore, and in stead of that, you will only listen to the sound of each caracter.
If you are still having the possibilaty to count at a speed of 30wpm, how fast do you count then? :)
I would think that after two caracters you lose track ?
But one way or another, you have to learn somehow to stop counting, and start listening.
And like i said: it is difficult, but not impossible.


Posted: 2013-10-24 21:25
The counting issue is an issue because it is my natural way of learning and may or may not apply to any other human being on the planet. The Koch method isn't, it is completely alien to me. When I started (I wasn't using LCWO then) I never went at less than 20 WPM, and continued with that until I had forced my brain to accept a different way of learning. When I got to the point where I could consciously tell what I was doing and whether it was correct or not, I slowed down to my actual target speed. But I always pay attention, and will work at faster speeds whenever necessary to keep my brain working in the right mode.

That is simply the only way this works well for me, whether it makes sense to anyone else or not, and I'm not about to change it because I already know all about the alternative.


Posted: 2013-10-25 15:46
Well, then all is for the best in the best of all possible worlds

(Pangloss in Candide, Voltaire)


Posted: 2013-10-25 18:14
PC1LH44:
The point is that you have to STOP counting


In general "Don't do it this way" is, for me, exactly the wrong way to phrase advice, but I can't think of a positive way to phrase it that works for me. The closest I've been given is to learn the "sound" or "tune" or "rhythm" of each character, but even that doesn't really work, because the "sound" or "tune" or "rhythm" changes with speed.

PC1LH44:
If you are still having the possibilaty to count at a speed of 30wpm, how fast do you count then? :)


Dits get too fast to count before dahs, of course. By 40wpm I most definitely can't count either dits or dahs.

PC1LH44:
I would think that after two caracters you lose track ?


Yep, which is why I might sometimes catch a callsign at over 30wpm, but I won't catch a phrase.


Posted: 2013-10-25 18:33
@Rick, do you think that you have improved your cw since you registered here ?
Some people just need more time, to learn it, than others.
Nothing wrong with that.


Posted: 2013-10-25 23:59
Yes, but not through using Koch. I wasted about two years trying to make Koch work for me using this site, and G4FON, and JustLearnMorseCode. For me, Koch was no good. The "Convert text to CW" has, however, been very useful for making training recordings, and the callsign and word training functions also help.


Posted: 2013-10-26 06:00
Interesting. It just goes to show that no teaching method is right for everyone. I can't imagine Koch not being the best way to get to a reasonable speed, but clearly it isn't for Rick.


Posted: 2013-10-26 23:44
In hindsight it should have been obvious that the Koch approach was not working for me far sooner. With hindsight, I'd say if you're giving it time every day and you're getting through at least one lesson a week then keep going, but if it takes a month to get through one lesson then it's definitely not working, so try something else.


Posted: 2013-10-30 15:33
What speed were you doing Koch at?


Posted: 2013-10-31 12:58
I tried various character speeds from 30 down to 12. However, changing the speed doesn't address the un-equal exposure to different characters that's the flaw with most Koch implementations under discussion here.


Posted: 2013-10-31 14:26
The unequal exposure is fixable manually. I find that the most important aspect of how I train is how I review, and I review recent letters far more than I review the old ones.

I think there is a good reason for review being key. It seems that the first time I add a letter it goes into short-term memory. It is in review that I start moving it into permanent memory, and to be effective at that I have to review the letters not in permanent memory the most.


Posted: 2013-10-31 21:24
That kind of review could be handled by the program noting where the errors were and tailoring its output accordingly, but not many do. The problem I had with the unequal exposure was related to the progressive nature of Koch training. When a new character came along I found I had to un-learn the way I was recognising some of the previous ones. In the end, just trying to learn them all at once in one go resulted in more progress in a month than Koch had achieved in two years. Even so, I still have more trouble with M and K than seems reasonable given they were the first two I learned...


Posted: 2013-11-02 06:01
I agree it would be nice if the software would give the most weight to the letters most often missed, I only meant that you can still get a similar effect by how you use the tools we do have.

I also have had to alter my recognition of previously learned letters after learning new letters. I observed this particularly when a letter was a prefix of another letter (like 'a' is a prefix of 'r'), because I realized that my brain always tries to optimize by making a decision as soon as the letter is unambiguous among the set already known. So several times I had to learn not to type a letter until I heard the space between characters.

However, this was never a big problem, it just required a bit of review. I don't know why it would give you so much trouble, but if it does then it does. I don't see any reason to believe that one learning method is best for everyone, and even the worst-sounding methods have people who like them and claims (at least) of people who went on to achieve impressive speeds. It seems likely to me that the Koch method is the best method for the most people, but that doesn't matter if it doesn't work for you.

It would be nice to have some data about different morse learning methods, but I don't think anyone ever cared enough to study it even when Morse was commercially and militarily important. Seems like it would be interesting, but it won't ever be done.


Posted: 2013-11-02 10:15
I've read some articles about the teaching of Morse back when it was of major commercial importance, and the three main points I've gathered are, first, that a small but significant proportion of those being trained failed to make the grade, second, that sending and receiving were usually taught at the same time, and third, that something like Farnsworth timing was usually used right from the start, because commercial operators needed to get to 20wpm or so before they could be employed handling messages, so learning slow characters was not worthwhile.

Oh, and the training was usually "full-time"; none of this "only half an hour a day" stuff.


Posted: 2013-11-02 15:28
The only place I've found much information about professional telegraphy training is here:

http://www.tasrt.ca/TASRT/

In any case, the goals of professional training are very different than for individual amateurs. The data I would like to see is just not worth the effort in a professional setting.

Back to the Forum

You must be logged in to post a message.

$Id: forum.php 62 2015-01-12 17:34:44Z fabian $