john555 Wrote:what is the purpose of this information?
I didn't want to explain this in the original post hoping nobody would ask.
There are a couple of reasons. In the immediate term, because I'm trying to use a modified person-action-object structure on my stories (PAO is a memory sport thing,
see here). I can simplify the story/system for kanji that aren't going to be used as primitives later. These stories will be just as memorable, but will lack an extra hook that would enable it to participate in later kanji, thus saving time and effort. Even at >514 kanji-not-used-as-primitives ('leaf node kanji'), it might make sense to sit down and identify all leaf nodes to maximize the chances of this method's success. (Apparently both Japanese and memory are hobbies of mine. I recognize most people would just buckle down and focus on their primary goal of Japanese.)
---
In the longer term, I'd like to construct the Heisig "dependency graph", with kanji/keywords being nodes and directed edges indicating which are used as primitives for which others, and this number would help get me started towards that. Such a graph would make a neat visualization ("mappr0n" is what kids are calling it nowadays?), but it also enables potentially-interesting things like reordering the kanji from what Heisig presents, i.e., modifying the path that Heisig uses to traverse all 2200 nodes.
This is because if you specify the ordering of the primitives (kanji and non-kanji) and apply a couple of simple hierarchy-preserving rules, i.e., (1) don't introduce a kanji whose primitives you haven't already introduced, and (2) before introducing a new primitive, introduce all kanji that can be built using previous primitives, you get something like RTK. Heisig applied these rules quite loosely in
his RTK1, since the first two lessons are mostly a bunch of primitives to "prime the pump", and sometimes he delays introducing a kanji, i.e., 蜜 could have been introduced much earlier but he wanted to pair it with 密. But the magic of RTK is that for the most part, these rules are followed and you learn kanji by building on ones you already know.
Therefore, you could pick your own ordering of primitives (kanji and Heisig's non-kanji) and apply the rule automatically and get a different sequence of the 2200 jouyou kanji that you could learn, instead of the book's. Why would you want to do something so daft? Because you could use an algorithm to figure out what ordering of primitives would give you your ideal RTK. Suppose you wanted to pack the lower-grade kanji (grade school) earlier in the sequence. Or make more common newspaper words appear earlier in the sequence. Or you had a corpus of literature that you really wanted to read sooner than later. In all these cases, your desiderata would become soft-constraints which a program would use to find a sequence of primitives which in turn build a complete ordering of the jouyou kanji such that your objectives were loosely met. It's a complicated graph optimization problem in the general case, but I think fast simple heuristics will get you 90% there, which is likely good enough, especially since I imagine most people would manually tweak the raw output (smooth out any clumps in the primitives, etc.).
---
I guess I've already made this post about "Ways to improve RTK", so here's another idea. We could introduce new non-kanji primitives for combinations that are found more than once, like 羽+隹 used in 濯 and 曜 (I use someone's idea of calling this combination "futon"), or 臣+又 used in 賢, 腎, and 堅 (I'm calling 臣+又 "Ambassador", in the Firefly-courtesan sense). With the dependency graph, you could automatically find out where a group of primitives are being used multiple times and potentially decide to make new non-kanji primitives out of those. (Sure you could do this by just going through RTK and keeping notes on combinations that you encounter, like the two above, but by then you're not interested in ways to improve RTK
)
One last thing that I think we really ought to fix is the synonym keywords: I just got to "storehouse", which comes after "warehouse", and looking ahead I see "godown". Using
WordNet semantic distance metrics, or just by asking those who've completed RTK, we need to identify clusters of synonyms and change them.
The keywords, and the core ordering of primitives, used in RTK aren't sacred. I suspect those were chosen because they worked for Heisig himself, all those years ago, and aren't going to work for everyone today. The core of the method (kanji building on each other in large subsets) can be used with different keywords, primitives, and orderings. This question is feeling towards this larger goal.
I look forward to reading many (as well as no) sound reasons why any/all of the above is foolish, from PAO down to changing keywords
But like Khatz says, learning is fun. Once you've learned something, it's no longer fun because by then you're back to the grind of living and suffering, except now you complain about taxes and traffic in 日本語
Edited: 2014-08-06, 10:31 am
