Lambda the Ultimate

a picture is worth a thousand words (if not millions).

So you can replace a million word program, using a corpus of C programs in front of me and "wc" that's about a 344000 lines of C or assuming about 680 lines per c file... You're claiming you could replace about 500 c modules by one, say, UML diagram!??

And then EXECUTE IT!?

Nope! You way way way over estimate the usable information content of a picture.

Perhaps we could take a snapshot picture of that C program and then write a computer program with character recognition?

UML is just another language with its own built-in symbols, syntax and grammer - and not a very dense one at that. It's aim is not so much information density as it is a communication shortcut within a specific domain.

The real problem with the phrase "a picture is worth a thousand words" is that it's usually used to judge the relative merits of visual images relative to language. I'd retort back that "a single word is worth a thousand images". And then I'd go even further and say that "a single word is worth thousand of other words". Indeed, one can not understand an image or words without reference to a history of countless other sensory experiences.

What you say has no meaning without a clear operational definition of "thought". Thought you'd like to know that. (Sorry, couldn't resist).
And it is by no means apparent that the brain stores sensory input. And what is "cognitive, etc."?

Neither language nor thought are "pictures". If you believe they are then here's a thought experiment:

Think of a cat. What room is it in? Is it standing, sitting or running? What color is the cat and what is the color of it's eyes? Is it purring or meowing? How tall is it? What is it standing, sitting or running upon? etc.

Why is my statement misinterpreted to refer to UML!? Oh brother, give me a break. A PICTURE takes up kilobytes or megabytes of storage -- information. If you were to describe the picture completely with words (or, for instance, logic expressions), in the same amount of detail -- a lossless compression -- you would surely use thousands and thousands of statements of fact about the picture and entities in them! (I'm talking photographs of, say, plants, animals, etc.)

And it is by no means apparent that the brain stores sensory input.

In spite of the fact that you can picture your house or apartment and describe it with much detail -- hundreds or thousands of facts, depending on how detailed you can picture it.

Perhaps there is a difference in the way the brain stores sensory representations of prototypical nouns, and the way it stores crisp memories of concrete nouns.

ASL seems to be built in four dimensions, X-Y-Z and time. It seems to be more typical to establish the context (such as I, you, they) first and then indicate a message about the actor identified in that context. As long as we continue to use two dimensional displays, VPLs will be unable to use the depth and time dimensions - though we can trick the eye into thinking that it is seeing depth (via shading and proportions) and time can be represented via direction (left to right, top to bottom, or with arrows).

Anyhow, I'm not sure that CAD offers any more help than the typical IO with lines in terms of establishing context and relationships. With a VPL still end up with a language that has to have a set of rules governing actors and their relationships. And although you might start with transparent symbols, eventually they will morph and become less intuitive - becoming opaque in the process. And if the Saphir-Whorf hypothesis is not correct, a VPL will be no more powerful than a text based one (though this is probably just a twist on all PLs are Turing Complete).

Non-contrived sign languages have surprising analogues to the grammar of film. Film uses the 4 dimensions of space-time (even if 3D space is simulated with 2D), in addition to cinematic techniques, like mise en scene and camera movement. In sign, while you can't simultaneously show complex sets as film can, you can progressively add "props" and actors in the 3D space, and even refer back them long after you set them up.

"Camera movement" is also common to sign language and film. You might represent two people, one chasing the other, with each hand, (a wide-shot, if you will), and then "cut" to medium shot of the pursuer-- the other person is "out of frame"-- and you use your full body to sign what they are doing or saying. You can then cut back to a wide, or through re-orientation of your body, cut to the other person.

There's a great short film in ASL called Vital Signs (http://www.youtube.com/watch?v=NzybkJx0Lo8) that perfectly illustrates these techniques.

I haven't found any academic research that looks into the relation between sign languages and film grammar, but given that there's some evidence that signing pre-dated spoken language, I suspect that we comprehend films so effortlessly due to visual communication being one of the things our brains have been doing for a very long time.

As for VPLs, I agree that sign languages offer clues as to how they might work well, but we're still stuck in Church-Turing Thesis land, so we can only hope to improve the ease of programming, not add power.

I know when I type, I don't think of individual characters at a time. For that matter, I already have in my mind what the entire idea is that I'm expressing - the words just flow out - limited in volume by my inability to type fast using two fingers.

In trying to learn how to sign words using the alphabet, my current lack of proficiency means that I have to consciously think of each letter and how it's represent with my hand. I am assuming that with constant use and drilling that eventually one can think of words and ideas and they will flow out the hands as if I was typing them on the keyboard.

So, I'd agree that unless it becomes second nature - you don't have to think about the mechanics of translating from one language to another - that you really won't be able to distill input from a language and use that same language for output. [Edit Note: from a PL standpoint, it's similar to thinking first how you would do something in Java when programming in Haskell].