Quote:
Can the complexity of a go problem be measured in terms of such concepts rather than just the raw variation tree?
That seems to me like the approach of a numbers guy. Experience tells me that people on one side of a dividing line are very rarely willing to contemplate looking at things from the point of view of the other side. But let me risk that, and give a words guy view of it.
You probably read that paragraph and understood it almost instantly. We all do that all the time, on both sides of the divide. But if I ask you now to close your eyes and repeat the words back to me, I'd expect almost everyone to fail miserably. I would, and I wrote it. But it goes further. Tell me whether the word 'try' appeared in it. I'd expect you to be rather unsure (you might remember 'risk', or was it 'try'? Or did try come somewhere else?
But did you understand it? Yes. Well, more or less. I'm certain you got the gist of it, but you may have put more weight on a portion of it than I intended, and you almost certainly made associations that I wouldn't hope for (e.g. is this guy a plonker on his hobby horse again?). And despite that, I don't think we would normally say we have mis-communicated with each other.
Now join me in the assumption that a tsumego problem can be likened to a paragraph of text. After reading it, you will not have a single point of view. You will have the gist of it with lots of associations (it's probably dead, but there's a ko if I want it, and it's not all that big so I might not play it in a real game - in fact it's so weird I might never see it in a real game; but can I grab it anyway in sente?) I suggest that something like that is the likely (and best) form of answer in real life.
But what about the apparent perfection reading it out move by move? Well, I think everyone accepts that in most of the really interesting cases we can't do that. The human brain doesn't work that way. Even bots have their limits on that way of working. Yet, on behalf of the human brain, take the word 'set'. We meet that with very high frequency in all sorts of sentences, and in virtually every case we read and understand it instantly. But "easily?" Jellies set, dogs set, I set the table, you set an exam, I dance in a set, I set to my partner, she's one of the smart set, the doc sets the bone, the carpenter adjusts the set of a saw, a rock star performs his set, Nadal wins the final set, the custard is set off, we set off on our journey, we are all set, they are setting themselves up for a set-to, does a badger live in a set or a sett, set the alarm at a set time every day at sun set. And so on and so forth. This is a level of complexity far, far, far, far greater than the level of complexity of a typical tsumego problem, where you get typically a choice of just one to three or four moves at each branch.
Yet we humans cope instantly with a sentence with 'set' in it (and bear in mind that almost every other word sentence will also have a similar range of complexity). By 'cope' I don't just mean understand, for we instantly also dredge up a huge range of useful associations.
So, it seems to me that we should be looking for ways to simulate that process in 'reading' tsumego. It can be done. I have witnessed numerous occasions where go players (both pro and amateur) are presented with a problem they have not seen before, but they solve it instantly. They are clearly not solving it by brute force. They are seeing 'simple' components like 'set'. By the same token, there have been cases where even a pro is stumped by a short problem simply because he's never seen anything like it before. I'd likewise expect most readers to be puzzled if I say 'latitudinarian.'
The keys to profound learning of tsumego techniques seem to be like native-language learning. Foreign-language learning is a very different and inferior beast, but most adults actually follow that route in go.
The main characteristics of NLL are, initially, massive repetition and not worrying about mistakes. We further teach children a small sub-set of the basic language and provide lots of extraneous clues such as hand gestures.
Has this been done for go? I think not. Adults don't like repetitive work, so don't do it or buy books that contain it. They hate making mistakes, and many hate even more having their mistakes pointed out. Many hate having tsumegos with hints. The books that are provided are, it is true, a sub-set of tsumego, but totally the wrong sort - under the stones, flower six and double kos.
So, making the probably wild assumption that a decent number of adults can be persuade to try the NLL approach, the first task in tsumego seems to be to establish what the basic go language is. My contention is that that cannot be done properly by adults making lists based on their own adult experience. I'd argue that the only way to do it is to get the probabilities for each type of tsumego technique from real games. This has been done to an extent in traditional teaching (so, for example, we learn about nakade moves and L shapes). But there are two problems with traditional teaching. One is that it stops after a very few basic techniques, then jumps straight to under-the-stones and the like. The other is that there is next to nothing that allows us to practise basic techniques over and over again. There are some honourable exceptions. Magazines like Go World often have a few pages of easy problems of the same type, and the Meijin Inseki wrote a whole book of just this type (most of it sadly lost). But, for the most part, go teaching is treated like adult food: beef today, chicken tomorrow, fish on Friday, etc. But in go we adults are really still toddlers - we need to be suckled on milk, milk, milk.
If we can be taught this way, we don't just absorb the techniques themselves but also the context and probabilities of their occurrence. Once that is achieved, we can 'read' problems instantly and reliably (enough), just as we read threads like this.
Being clever humans, we can then even become adept enough to read the go equivalent of sentences like "Pas de l'yeux Rhône que nous."
ADDENDUM
A grandson of mine is learning to read using phonics. I've been quite impressed with the results, and so looked the topic up. Here is one brief description.
Quote:
How is phonics taught?
Words are made up of just 44 sounds in English. You may have heard your child or their teacher use particular words that form the core of understanding phonics. Here's a quick explanation of some of the key concepts.
Phoneme - the smallest unit of sound as it is spoken.
Grapheme - a written symbol that represents a sound (phoneme) that's either one letter or a sequence of letters
Digraph - two letters that work together to make the same sound (ch, sh, ph)
Trigraph - three letters that work together to make the same sound (igh, ore, ear)
Split digraph (sometimes called 'magic e') - two letters that work together to make the same sound, separated by another letter in the same word. This enables children to understand the difference in vowel sounds between, for example, grip/gripe, rag/rage, tap/tape.
Rather than memorising words individually, children are taught a code which helps them to work out how to read an estimated 95% of the English language.
What I take away from this is that: (1) for even the task of reading English, a task vastly more complex than for tsume go, the number of basic overarching concepts can be very small; (2) the results apply to 95% of what we meet in real life.