What I Learned Today

An xkcd comic led me to the Simple English Wikipedia. This wikipedia aims to provide simplified versions of articles from the Ordinary English Wikipedia by limiting the vocabulary used, grammar complexity, and sentence length. I admire the motivation behind this resource: to make general knowledge accessible to non-native speakers, youthful readers, or those with disabilities. Yet to an adult native English speaker, the language of these articles can be gratingly unaesthetic (and imprecise). For example, consider this excerpt from the page on Mars:

The planet Mars is made of rock. The ground there is red because of iron oxide (rust) in the rocks and dust. The planet has a small carbon dioxide atmosphere. The temperatures on Mars are colder than on Earth, because it is farther away from the Sun. There is some ice at the north and south poles of Mars, and also frozen carbon dioxide.

This is all factually accurate, but achingly simplistic (especially the “because it is farther away from the Sun” statement — the atmospheric composition is also a critical player). On the other hand, if I had to read wikipedia in, say, French, I would no doubt appreciate the simplicity!

But I experienced even more wincing when reading pages about topics from Computer Science, such as the neural network page, half of which consists of:

What is important in the idea of neural networks is that they are able to learn by themselves, an ability which makes them remarkably distinctive in comparison to normal computers, which cannot do anything for which they are not programmed.

(Technically, they don’t learn by themselves — they require supervision in the form of labeled examples — and any machine learning method exhibits the learning property, not just neural networks, and what is a “normal computer” anyway? A neural network is an algorithm for learning a model, not a special-purpose computer. Finally, even neural networks cannot do anything for which they are not programmed! More accurate: “Neural networks can learn from examples, allowing them to make predictions about objects they may never have seen before (generalize).”)

Or consider this part of the page on Computer Science itself:

A computer is a device which takes orders as fast as you can give them to it and works as fast as it can to solve the orders.

(makes a computer sound like an active agent (e.g., waiter), which it isn’t) or from Computer programming:

The instructions in “machine form” are usually in a .EXE file (which is called an executable, because it can be executed). These machine-instructions will by default open a black “command-prompt” window, but can open games as well as other things.

(Well, am I really surprised that “simple” computer programming has such a strong Windows bias? ;) )

There’s an interesting issue at the heart of this project: how do you talk simply without talking down? (Or worse, misleading the reader!) Clear, simple language has real value even outside of this venue. However, translating all value judgments into the simple words “good” or “bad” not only gives the text a childlike sound but also gives its meaning a childlike interpretation, and important distinctions may be lost.

I actually find this wikipedia harder to read, not easier; the stilted sequences of simple sentences dominate my attention with their awkward rhythms and unanticipated gaps (likewise, you may have found my alliteration distracting :) ). Good writing blends its details in to the background and leaves you room to think about the ideas being presented. But yes, I know: I’m not the target audience for this product. I expect that many people are benefiting from much of the information in the Simple English Wikipedia. Hopefully they also get a chance to dig deeper for the real details on their subjects of most interest or need.

We all grew up being told that the Milky Way was a spiral galaxy, twisting beautiful starry arms through inky space like our photogenic neighbor Andromeda. So it may have come as a surprise to you, as it did to me, to learn that in 2005 the Spitzer Space Telescope confirmed that in fact we live in a barred spiral galaxy instead. This isn’t that strange; apparently up to two-thirds of all spiral galaxies contain a bar.

Even more interesting, our galaxy has four arms (in our current understanding of its structure) and they have each been named. We reside in the Orion arm, a small offshoot between the Carina-Sagittarius and Perseus arms. The main two other arms are Scutum-Crux and Norma.

As a side note, the wikipedia page on barred spiral galaxies includes this rather unfortunate statement:

“Studying the core of the Milky Way, scientists found out that the Milky Way’s bulge was peanut-shaped. This led to the conclusion that all barred spiral galaxies have a peanut shaped bulge.”

I’m guessing that more was involved than simply generalizing from a single example.

Also fun is this list of Ten Things You Don’t Know about the Milky Way Galaxy. I was pleased to discover that I knew all but #8 and #9 — and also pleased that there were two new things to learn.

I am a latecomer, it seems, to some of Edward Tufte’s brilliant ideas. Today I stumbled across the sparkline, a “small, intense, simple dataword” (Tufte) that is best illustrated by example. Sparklines permit you to display a large volume of numeric information in a very tiny space, while conveying the information perhaps even more effectively than if you’d used a large floating figure or a table. Tufte posted a sparkline introduction, which is an excerpt from his book Beautiful Evidence. Many others have followed up with their own sparkline creations, sparkline generators (in perl, HTML, Excel, and even special font encodings), and sparkline critiques.

One tip I particularly liked had to do with aspect ratio; Tufte suggests (after William S. Cleveland, 1993) adjusting the vertical scale so that slopes are about 45 degrees (rather than very flat or very spiky). While this is simply a rule of thumb, to be violated if the situation calls for it, I did find his examples to be compelling; “lumpy” data does seem to be easier to visually process than “spiky” data.

But what really took my breath away was this particular example:

This is cited by Tufte as appearing in Robert Sedgewick’s 1998 “Algorithms in C”. It illustrates several passes of mergesort being applied to a 200-item list. It is absolutely brilliant! Whoever thought of visualizing the values (sort keys) as the angle of the lines was absolutely inspired. This graphic stole my attention as only a true work of art can. I’m still staring at it in fascination.

I’ll have to be on the lookout for places where sparklines could be the right solution in my next technical paper.

Radio telescopes allow us to listen in on distant sources and learn about fascinating objects such as pulsars, quasars, and even (maybe?) extraterrestrial civilizations. Directional antennas for these telescopes have greater sensitivity than omnidirectional antennas, but then they must be pointed in the appropriate direction. However, large telescopes can be prohibitively heavy. Arecibo, which at 300 meters wide is the largest dish in the world, doesn’t even try; it sits in a depression in the ground and lets the Earth’s rotation sweep it around on a daily basis. As a consequence, there are areas of the sky that it cannot study, and everything else can only be imaged for a short time each day. Other facilities such as Green Bank and the Deep Space Network use massive motors and gears to rotate their telescopes to reach other regions in the sky or to focus on a specific target for longer periods of time.

But why move if you don’t have to? Engineers have developed a clever way to simulate a directional antenna from a collection of smaller, stationary, omnidirectional ones. Given multiple antennas in a line, if you shift the phase of each one progressively more, then combine the signals together, the result is the same as if you had rotated a single larger antenna to point to the side. If you have the ability to digitally change the phase shift for each antenna, then you can “point” your array anywhere you like without moving anything physically. This is called “digital beamforming.” (Technically, beamforming permits the manipulation of both the phase and the amplitude of each component antenna’s signal.) The Allen Telescope Array in northern California is an example of an array that uses beamforming (e.g., to listen to the New Horizons spacecraft).

Even more exciting is the recent advance in adaptive digital beamforming. Here, each of the phase (and amplitude) shifts (weights) are modified on the fly to maximize the resulting signal quality. Apparently, some radio transmitters even send “training sequences” to help an adaptive receiver quickly identify the best weights to use.

Thanks to Toby Haynes for his excellent “Primer on Digital Beamforming,” which is both exceedingly accessible (even for those of us without a formal signal processing background) and satisfyingly detailed (with field strength diagrams for different antenna types and component diagrams for beamforming).