muflax65ngodyewp.onion/content_muflax/experiments/speedreading.mkd

---
title: Speed Reading
date: 2010-06-23
techne: :rough
toc: true
episteme: :broken
---

Wait, what? Speed reading? Isn't that pseudoscience? Partially, sure. However,
not all of it, and that really surprised me. Yes, speed reading *is* real. This
is my collection of useful hacks.

Hacking your brain for fun and profit
=====================================

Binocular Rivalry
-----------------

While working through [Consciousness Explained][] by Daniel Dennett, I encountered
several experiments that I doubted. So I tried to replicate them. Specifically,
binocular rivalry seemed weird to me. Binocular rivalry occurs when each of
your eyes sees a different thing, typically achieved by just setting up a
barrier between them, and showing different pictures to each, e.g. horizontal
stripes on the left and vertical stripes on the right, or different letters or
faces and so on. What happens is that occasionally one side will dominate over
the other and you will only be conscious of it. Most commonly, at first you
kinda see both sides, then they partially merge, in a very patchy way, and
suddenly your vision _flips_, i.e. one side becomes clear and the other turns
invisible. This process then alternates randomly, unless even a slight
disruption is introduced to one side, like a moving dot, which causes this side
to become dominant immediately.

Allegedly, you may control which side is dominant most of the time, but you can
not be conscious of both reliably. I didn't believe that and tested it by trying
to read two texts simultaneously. In fact, I actually managed to do that! The
main visual problem is focus. It is quite hard to have each eye focus a
different thing, but using e.g. [DXM][], you can actually pull that off. But even
without it, you can try to focus a middle point and just make the letters big
enough so that you can read them even out-of-focus. The real problem comes from
assembling two sources of input into separate sentences; at first, they always
mixed and I couldn't understand anything.

Being Myselves
--------------

I then controlled for that by using series of numbers and suddenly I was able to
read two things at the same time! At first rivalry happened a lot, but soon I
got into mental soft focus and could read (but not parse sentences) just fine. I
then speculated that I might be able to exploit both halves of my brain. In
split-brain patients, who don't have a connection between their left and right
hemisphere, you actually get two independent consciousnesses and I did read up
on people that tried to induce this with normal brains. 

Because the left side of vision (i.e. left in both eyes) is handled by the left
hemisphere and vice versa, you can wear glasses that have either their left half
on each glass blocked by tape or the right side, and so only give visual input
to one hemiphhere. This actually causes a significant effect if your two
hemispheres are currently in disagreement. Some people with depression or
anxiety were able to reduce it or turn it almost off temporarily while wearing
such glasses! So for example, you feel very nervous with your therapist, block
the left side, everything is the same, then instead block the right side,
woosh!, your anxiety is gone. It comes right back when you take the glasses off,
but still, way cool.

So by being able to make each hemisphere dominant at will, you can really fuck
with your mood. Find where your language side is (typically the left side) and
block it - you become more empathetic and reading gets harder. Block the other
side, reading is normal, but relating to content is harder. The effect is
typically not that large because both sides are still internally connected, but
I found it quite noticeable.

Anyway, I tried to improve on binocular reading by separating not only between
eyes, but sides of vision. Let my left half of my left eye read one thing and
the right half of my right eye another. Focus gets really tricky that way, but
it is doable. And lo and behold, I could, albeit slowly, read two things at
once! Parallel processing, bitches!

What's that all got to do with speed reading?!
----------------------------------------------

Binocular reading isn't really practical (for one, you look ridiculous, two,
it's very difficult and slow), but I got interested in *other* ways I could hack
my reading process. If I can read in parallel, can I also read non-linearly?
Start in the middle of a sentence, jump around and still get it? Read really
really fast? Maybe subconsciously?

Now we're getting there! First, let me clarify one thing: speed reading literature
is a complete and total *mess*. Barely anything scientific, vague claims, lots
of lies and false promises, no clear terms, nothing. To remedy this, I'm going
to state *exactly* what I mean and what this is about:

Speed reading involves any technique that makes you read a normal text faster
**without sacrificing comprehension**. No, it's not **skimming**: that only
tries to give you a basic overview of the text. The idea is to be able to
understand the text just as if you had read it "normally", even though the
process of getting there may be very different. There are techniques to organize
your reading better, like first skimming through and getting a feel for the
structure and so on, and they are all useful, but that's *not what this is
about*. We want pure reading speed, nothing more, nothing less.

But what can be achieved? First, measure your current reading speed. Say, pick a
Wikipedia article, read it, time yourself and then count the words. Average
among most people is about *150-250wpm* (words per minute). Good college
students read at about *300-350wpm*. A fast conventional reader gets up to
*500wpm*, maybe *600wpm* if they are really good. Speed reading, on the other
hand, falls into a range of about *800 to 1500wpm*. For some texts and some
people, this can go even higher, but as a reasonable general limit, 1500wpm is
about it.

Because a normal page in a book has about 350 to 450 words, depending on font
size, people typically read about 30 pages per hour, college students about 50
to 60 and speed readers about 150 to 250.  Those numbers are of course averaged
over a lengthy text and don't have to be constant - a difficult paragraph may
slow you down and a simple one may just fly by.

What about **comprehension**? There are two components to it: **understanding**
the text and **remembering** it. Understanding means being able to follow it,
being able to give a summary of it and so on. Remembering involves still knowing
details, all characters or arguments involved and so on. Basically, if at the
end of the book, you don't sit around confused what the fuck just happened, you
*understood* the text (and didn't read James Joyce). If you can also pretty much
tell someone everything you just read, you also *remember* it. The two are
usually closely connected, but not always.

I am only interested in techniques that *maintain* a high level of
comprehension, typically a retention of 80-90% of the content. Sacrificing
quantity for quality is right out. Nonetheless, it is still true that topics
that are difficult to understand will always be read slowly, no matter the
technique used. Reading about theoretical physics will be slow unless you have
studied it. No speed reading technique will fix this problem. Most texts,
however, fit nicely into your rough skill level and the limiting factor is in
fact your reading, not your understanding.

Related to that, a common objection to speed reading is that it kills the
enjoyment, or that maybe you are just reading too easy texts. "Why hurry
something good?" I don't agree with this sentiment *at all*. If you enjoy
reading so much, why not read at 1 sentence per hour? Why not watch Star Trek at
one scene per day? What, that would be mind-numbingly boring? Why yes it would!
Also, if you increase your throughput, you become able to handle more complex
structures. A series filling thousands of pages is suddenly just as manageable
as a comic book was before. Reading up on moral philosophy by reading works
by/about the 10 or 20 most influential thinkers over the course of a week or two
is doable. Books become what Wikipedia articles were before. So if you don't
like high bandwidth and all the benefits that come with it, this just isn't for
you.

Finally, a note on **subvocalization**. When reading, there are basically 4
different aspects of sound:

1. *Reading out loud*. This is what beginners may do, or what you do when
   reading to someone. It was actually quite common in ancient times and the
   idea that you could read silently was very weird to many Romans.
2. *Reading to yourself internally*. You basically still do the same thing,
   including moving your tongue, but you don't produce a sound. This is often a
   transitional period for early readers (and quite useful - there is some
   evidence, including my own experience, that learning new languages is easier
   when subvocalizing). It will disappear on its own once you become more
   confident.
3. *Subvocalization*. You still *hear* the sound, but you don't feel that you
   produce it. Muscle movement doesn't exist (at least not any you would notice)
   and speed is greatly improved. You often skip words, or only hint at the
   sound. This is the normal mode for most people to be in, even many deaf (who
   often are not 100% deaf), and this is the *inner voice* most of us use to
   think, at least some of the time.
4. *Reading in silence*. Finally, reading without hearing any associated sound.
   No inner voice, but direct meaning, just as you look at a map, for example.
   Because visual processing is, for almost everyone, vastly superior to aural
   processing you can read much faster that way.  Personally, I believe that the
   problem is that to understand an inner or outer voice, your brain has to
   simulate sequential processesing, but the brain is only parallel. This
   makes it all quite slow. Visual processing on the other hand is not - you
   can parse many parts of an image or scene at the same time and only
   coordinate results at the very end. Also, your visual hardware is far more
   optimized and greater in size.

Techniques
==========

The Conventional Approach - How to read fast the normal way
-----------------------------------------------------------

The easiest hack is to just read faster - you do everything you'd normally do,
just faster. As I mentioned, you can go up to about 600wpm that way. When I was
starting out with speed reading, I was already reading at 450wpm. How did I get
that fast?

I could credit reading practice. I do read a lot, especially on the web, but
that's not all that plausible. I know enough people who easily read just as much
as I was reading at 14, and I was reading about 400wpm back then, too. Most
people never seem to go beyond 250-300wpm, no matter how much practice they have
reading texts.

So what *do* I credit? Video games. I'm serious. I played a *lot* of shooters
and racing games and this really improves your ability to react *fast* and react
to inputs from *anywhere* in your field of vision. You are also forced to shift
your attention around a lot and figure out threats as fast as you can. I also
notice that in anyone I know that played a lot of fast games: Their attention
jumps around a lot faster than normal, no matter what they are working on. 

The typical example is taking a gaming teenager and having their teacher watch.
Give the teen a computer menu to figure out, or a form to fill out or something
like this, and watch how the teacher desperately tries to keep up, even though
the teacher surely has more reading practice. Still, no chance whatsoever, and
the same goes for all non-gaming teens. But any gamer will have no problem, no
matter the age.

So if you read only 200 or 300wpm, you are not playing enough. Get Quake 3 or
Halo or Starcraft, a big supply of caffeine and *train*. After a while, your
reading speed will pick up, I'm certain of it. Some people, especially those
with ADHD, may be better at this than others. Sometimes, a short attention span
really pays.. oh shiny!

Turning off subvocalization
---------------------------

The most important change to achieve any kind of real speed is getting rid of
the dependency on subvocalization. The rationale is simple: as long as sound is
involved, in any way, even just at the last step of comprehension, you will not
go faster than about 600wpm. Forget it, it's impossible.

However, the idea is *not* to permanently turn off subvocalization. It does have
some useful purposes. It's quite good at understanding names, or unknown words,
or reading anything sound-based, like poetry. However, the vast majority of text
is entirely disconnected from sound. (Some languages maybe more than others.
French and English are already only remotely linked to their actual spelling,
but many Chinese languages have basically *no* written pronunciation.  It also
has never stopped any scholar from reading old languages, whose sounds have been
lost to us.) Ideally, we would like to read visually whenever possible and only
resort to sound when necessary.

So let's cut out the middleman. But how? I'm going to present three techniques that
worked for me, but before I do that, I want to address one common problem. 

It is quite typical to worry how to *suppress* subvocalization. How do you *not*
think in a certain way? The short answer is: you don't because you can't.
Thought suppression never works. You can't *learn* to not think of a cow by
*trying* to not think of a cow. Try it yourself! By giving attention to the idea
of a cow, and you must, otherwise you wouldn't know that you are not thinking of
it, it will always come to mind again. However, certainly *can* not think of a
cow - by not giving cows any kind of attention. The same goes for
subvocalization - the following techniques will simply not care about it and
will in fact make it impossible to use it. It will disappear on its own. 

Chunking
--------

A chunk is the largest unit of information you take in at once. When you learn
reading, your chunk size is "one letter", slowly building up to "one syllable"
to "one word". Unfortunately, most people stop there. The goal is to enlarge
your chunks to multiple words, maybe even whole sentences at once. Chunking is
the whole meat of speed reading. It's the main trick to discover.

Once you read at a very high speed, it really makes a huge difference how large
your chunks are. Here's a little demonstration:

<%= image("fast.gif", "chunk size 1") %>
<%= image("slow.gif", "chunk size 4") %>

Both animations run at the same reading speed of 1000wpm, but the first one
shows every word on its own, while the second one uses groups of 4. If you watch
it for a while, you should be able to read the second one, but the first one is
a lot more difficult. However, notice that it will also get easier once you know
what the sentence is. This is the trick behind chunking: pattern prediction. If
you have a good clue how a sentence is gonna develop, you can read more of it in
one go. This is why this will only work when you know the language well and the
text contains not too many unfamiliar ideas.

Once you go beyond about 10 chunks/second, visual processing starts lagging
behind more and more. After-images, too slow eye movement and so on start
interfering with your reading. This means you can read maybe 600wpm if you read
every word on its own, but increasing your chunk size from just 1 to 2
immediately doubles your speed! The benefit is obvious, so pacing trade-offs
when increasing chunk size are often worth it.

The highest possible chunk size, according to all sources I read, seems to be
about one paragraph, which is about 100 to 150 words long. I suspect that a main
problem here is the size of the area you can keep in focus. Chunking a whole
page at once is probably impossible because you could never get all words to be
sharp and readable.

A very useful technique for training purposes is **Rapid Serial Visual
Presentation**, or RSVP for short. That's quite a big word, when really, it just
means "flashing words really fast", exactly like the two animations before.  

The best RSVP I found is [Eyercize][], even though it has
the stupidest name *evar*. Nonetheless, it's the only speed reading tool I know
with support for fixation points and complete customization. I usually set it to
2-3 fixation points per line, about 5 lines of context and increasingly higher
speeds. I occasionally ignore the marked line and read the upcoming context
instead, though. [Spreeder][] is also nice and maybe
easier to use at first.       

I would also recommend [Look, Ma; No Hands!][], a book that teaches semantic
chunking very well. It's quite short and precise. You get results very fast.

To make chunking possible, you have to watch out for the right **font size**.
You can only see about 5 degrees sharp enough to read. If you hold out your arm
and make a piece sign, then your index and middle finger are about 5 degrees
apart. So it is crucial to get enough words into focus.

I was often reading texts at very high font sizes, like 30pt or so. Hey, I have
bad eye sight and sit quite far away from my monitors. But I found that this
makes it really hard to read fast, so I fixed my setup. I moved my monitors a
lot closer and decreased my font size. *A lot*. 

In my experience, a font size of 12pt (assuming normal DPI) is the *largest* you
want to use. I currently read websites at 10pt, which seems to be the best
compromise between readability and strain on the eyes. (I also find it hard to
read Japanese below 10pt. There just aren't enough pixels left.) 

At those sizes, the font used matters a lot. I've always been very fond of the
Microsoft fonts, even though I haven't run any of their systems for years. The
Google Droid font is also very nice. Regardless, experiment and use something
that is clean and very easy to read.

Speaking of font size, column width matters just as much. It's no use if you see
a lot of text, but the current paragraph fits into one huge line across your >20
inch display. Ideally, you would get a whole sentence into your focus at once.
The maximum line length therefore should be about 100 characters or 20 words.  

If you are a console hacker, then I'd also recommend checking out bitmap fonts.
They really shine at such sizes. Remember that you can only fix bugs in code
that you see. The more lines fit on your screen, the better you can debug. 

Faster pacing
-------------

This is where the "speed" in "speed reading" comes from. Chunking is the
requirement, but it on its own won't make you faster. The human body, and that
includes the brain, is very efficient at avoiding work. It is a ruthless
optimizer and always do what is easiest *right now*. Being good at conserving
energy is the reason we are still here, but also why any kind of exercise is so
hard. If you *somehow* can get away with spending less, you will do so. That's
why you only get muscle growth if you push yourself hard enough to make it
absolutely necessary. You will never get stronger just by jogging, and you will
never get faster by reading at a comfortable pace. 

To get faster, we need a kind of setup that makes it easier to process text
faster than anything else. The best way to do this is to externally enforce a
high speed. For digital texts, you can use the RSVP again. I prefer short
sprints, so take a text of at most 5000 words, which is about a longish blog
article. Take your current reading speed and multiply by 2. Use a chunk size you
are comfortable with, or about 3-4 words when in doubt. Try to keep up, but
never slow down. If you missed too much, try again, at the same speed. You may
change the chunk size, but never decrease the words per minute!

Don't worry that you will not get everything at first. In fact, don't worry if
the text makes no sense at all. Concentrate and try to get as much as possible.
At first, you may only make out a word here and there. Soon, you make out
groups. Whole chunks. The occasional sentence. Then some meaning returns. That's
when you do the next step - you go *faster*.

Because the computer shows the text for you, you can't cheat. You can't fall
back and read a sentence again or slow down in any way. You must either pay
attention and use your eyes maximally efficient or you won't understand the
text.

Here's an idea for an exercise and how I did it. Because I read at 400wpm, I set
my speed to 800wpm. I would read like that for about 5 minutes. Then I increased
my speed to 1000wpm, again for 5 minutes. Then go back a bit, to 900wpm, which
will now seem much easier. Continue alternating between "can kinda keep up" and
"can barely make anything out" for a total of about 30 minutes, maybe an hour at
most.

The principle behind this is a bit like high-intensity interval training where
you run as fast as you can for 20 seconds, then jog for 10 seconds and repeat
this in total for 5-10 times. The idea is not to be able to always run as fast
as during those sprints, but by putting this huge, but short pressure on your
muscles, to greatly increase your normal speed. 

It is perfectly normal and actually good to be confused and not understand
anything during this exercise. :) This speed is far too fast for your internal
voice to keep up and your brain is under huge pressure to make any sense of what
you are reading asap. Once you go down to a more normal rate, you will actually
overshoot and read faster than you thought would be necessary. Voilà, you read a
bit faster! The brain gets used to this high speed and soon comprehension
returns. In fact, I found that I got bored now if I would read at 400 or 500wpm,
even after just one week! Be warned that this may annoy any non-speed-reading
observer. ;)

As material I used minor blogs I enjoyed reading, but didn't care too much about
if I missed anything, and novels I had already read or that were kinda
predictable. That way, if you go blank from time to time, you can find back
into the text easily. And it's a great chance to read Twilight without any
guilt! (You can find usable novels in .txt or .pdf form in certain bays or,
for older texts, in the Gutenberg archive, for example.)

It took me about a week to read 800wpm that way without missing anything. After
two weeks, I could keep up 1000wpm almost all the time, and 1200wpm if I really
concentrated. You don't have to do this all day, but try to do at least 20
minutes daily.

Reading nonlinearly
-------------------

Finally, it's time to fully exploit the parallel processing and to do more
aggressive pattern prediction. It's time to throw away the chains of
<del>oppression, comrade!</del> intended text flow that the author gave us and
to read in any order and any direction that gets to the meaning faster.

Reading nonlinearly just means you read text the same way you look around. You
jump to the points that look most interesting, figure out the context around
them, then jump to the next spot. But if you read everything sequentially, you
can't do that! At least, you'd have to go back and start reading the current
sentence you're in. 

Imagine your vision would work sequentially - like normal reading. You go into a
room and move your eyes to the upper left, start moving them to the right, line
by line, until you have scanned the whole room. Sure, you would *see* everything
eventually, but it would be *way* stupid and inefficient. Instead, you first
have a quick look around, maybe 2 or 3 unconscious eye movements, to figure out
if anyone is in the room and where the interesting stuff is. Nothing unusual on
the floor or ceiling, so you skip those areas altogether. But you saw something
like a face over there, so you concentrate more on this point until you
recognize who it is (and in what mood they are). This takes maybe a second or so
in total, and you may have only actually looked at 5% of the scene, but you sure
know everything that matters. So why not read that way?

A good exercise I found was to enforce a time limit per page. I set up a
timer[^pororo] to give me a little beep every 20 seconds, following which I
would *have* to turn the page, no matter how far I was. This would equal a
reading speed of about 800wpm for a small paperback. You do this for maybe 5
minutes, then go faster. Go to 15 seconds, then 10, then 5. Finally, go back to
20 again. It will now be far easier.

Sometimes, it was no problem to read a page very fast, but soon I could tell on
first sight if I would be able to make it or not, *before* being conscious of
any content. If I recognized the page as hard, I would scan it rapidly first,
working out the structure and main phrases on it. This would take only a few
seconds, but reduce the difficulty of the page drastically. I could then clarify
the missing pieces, reading them far faster than before. Like with vision, you
first establish where core ideas (=people) or interesting words (=colors) are,
then concentrate on them exclusively.

Instead of going for whole pages, you can also train to read multiple lines at
once. At first, start with 1 second per line, for maybe a minute. Read any way
you want, but after 1 second, move on to the next line. It helps to trace the
lines with your finger or a pen to enforce a consistent speed. Then do 2 lines
simultaneously in 1 second, again for a minute. Then 4. Then whole blocks of
texts, ideally whole paragraphs. Such huge blocks are very nice for skimming and
getting a feel for the book, where everything is and what the main ideas will
be, but it's a bit troublesome for normal reading. Still, it took me about 2
weeks to get used to reading about 2-3 lines at once. I now have a far broader
pattern in which my eyes move over the page, not clinging to every word, but
rather "painting" the page in a zig-zag pattern with a brush about 2-3 lines
thick.

Another good exercise is to read *backwards*. You start at the end of the line
and right to the beginning, i.e. for an English text, you read right-to-left.
Once you got a bit of practice at that, you can alternate and read in a zig-zag
pattern. The advantage is two-fold: you save a lot of eye movement and you get
used to understanding sentences out of order.

Combined with a harsh time limit, I found that this exercise greatly improved my
ability to jump in the middle of a paragraph, figure out what's going on and
assemble meaning by moving into all directions, not just left-to-right.[^ltr]

[^ltr]: 
    It may help if you are used to multiple languages that have a different
    word order or writing direction. German and Japanese, for example, build up
    quite large word stacks and you may end up with a sentence that keeps on
    piling up modifiers and objects without revealing the crucial verb or target
    at the end, so maybe this practice makes it easier for me to adapt to
    backwards reading than for others. Also, Japanese is read both left-to-right
    and up-to-down (and then right-to-left), depending on context, so I'm
    already used to changing directions.

Once you go beyond a certain speed, it stops being uniform. I noticed that I can
read consistently at 300-400wpm using my previous techniques, but when speed
reading I vary between 700wpm to 1200wpm from page to page. Especially dialogue
really slows me down. This also means that each book has its own speed, so
measuring reading speeds in "words per minute" stops being useful. "Bits of
information per minute" would be better, but how do you calculate that?

[^pororo]:
    I wrote my own timer for such purposes. You can check it out on
    [Github][Pororo]. Basically, you set a timer for each level
    of the task, like a 23s timer for the page and a 200page timer for the book.
    Alternatively, I used the metronome function of my mp3 player, especially
    when reading on the train or when waiting for something.