image courtesy of Kevin McCloskey
Most readers read at a rate within the range of 200-400 words per minute. Multiple bottlenecks, both perceptual and pragmatic, limit reading speed.
In an era when soylent is being sold at Walmart, it’s no surprise that the specter of speed reading would return to entice the current generation of time savers even despite the tarnished reputation and unsupported claims. Proponents of speed reading claim that that reading can be sped up to rates beyond 600 words per minute without sacrificing understanding. Search “speed reading” online and you will find that a lot of people believe that anyone can be transformed into a speed reader. Wikipedia has a quasi-legitimizing entry for “Speed reading”. Stanislas Dehaene, one of the most famous living cognitive scientists, has written optimistically about it in a best-selling book . Fortunately, for the speed-curious, there exists nearly half-a-centuries’ worth of research into the limits of reading speed and whether or not speed reading counts as “reading, only faster” or whether it amounts to little more than good ole’ skimmin’ and guessin’. Below I try to share some of these findings and demonstrate a few simple examples of reading capacity with interactive tasks. While I try to convince you that speed reading is not possible, I invite you to try out these tasks and then judge for yourself whether or not to believe in speed reading.
First, if you want to know where all of the time gets wasted during the reading process, here is my rendition of Mark Seidenberg’s equation for average reading speed:
- Only about seven to eight letters can be read clearly within the field of vision at any given point.
- While reading, your eyes stop to gather information for an average of 200-250 milliseconds (ms):
200-250 ms = 4 to 5 stops-per-second = 240 times-per-minute.
- 240 stops-per-minute x seven letters-per-stop = 1680 letters per minute.
- The average word length is 5 letters.
- 1,680/6 (five letters per word + 1 space)= 280 words per minute.
To some readers, this may sound like a challenge. Can we bump these numbers up?
1. The windows of perception
The number of words you can identify in one glance is limited by the eye and the mind/brain.
Below are 3 rows of text. Each row has a word, then a seven-digit number, then another word. How wide is the area of your vision that can identify words? Try identifying the bolded middle number in each row (for example “8” in “2538217”). Then, while your eyes are on that number, try identifying the words on the right and/or left of the number.
diluted 5743899 emerged
bootleg 5794360 voyages
whiskey 8456752 vectors
Was this difficult? You might have even tried moving away from the screen, and noticed that it does not get easier.
Not all of the area within your gaze is treated equally. The eye is set up so that the images in the center of your gaze are seen with more detail than images in the periphery (can you imagine it the other way?). The retina, the place in the back of your eye where all of the vision receptors are located, gets light from almost 180 degrees. You can hold your fingers to the sides of your head to see how far this span is, but although you might be able to sense your fingers are there, at the edges of your visual span you probably couldn’t tell if they were fingers or hotdogs if you didn’t know in advance. This is because we don’t use most of the information in our vision for identifying things, but simply for noticing objects exist and then dodging or grabbing them. To identify things, including people’s faces and words, you must use the privileged spot towards the middle of the retina called the fovea. The fovea gets light from only 1/36th of the area within your gaze at any given moment. If you want to know how much this, hold your two thumbs away from your chest at an arm’s length and stare at your thumbnails. That’s about it. The density, type (cone receptors), and connectedness of the receptor cells in the fovea allows the cells there to pass on information to the brain that is way more fine-grained than what the rest of the cells in the retina can. This is the kind of information necessary to tell an “n” from an “h” or an “O” from an “G”.
image courtesy of wikimedia commons
Do faster readers have bigger or better foveas than slow readers? No. Sometime before 1975 Mark Jackson and James McClelland found that slow and fast readers do not show different abilities when recognizing 2 letters spaced various distances apart. Since readers of different speeds can do just as well, they have no differences in the width of their perceptual windows.
While the width of the windows for fast readers and slow readers appears the same (is that a pun?), their brains use the information within that window that hits the fovea differently. Identifying two letters is easy. When Jackson and McClelland flashed 8 unrelated letters (for example, VGSFDATU or FGIOVTKA) within this window, faster readers were better able to report which letters they had seen. The difference was not the width, but the ability for fast readers to do fast and reliable coding. The idea suggested by that experiment is that fast and slow readers could be working with the same visual area, but that faster readers are able to use more information from that area. By “use” it, I meant they are better able to quickly and accurately store it in memory. Isn’t that also the goal of reading?
Jackson and McClelland followed this up with a study published in 1979, demonstrating that faster readers tended to show better listening comprehension, faster responses when determining whether two letters were the same (for example, A & a or B & d), and increased speed and accuracy at recognizing homophones viewed side-by-side (BOAR-BORE). This again stresses the idea that it’s all about coding, and further clarified the code of the fast reader: knowledge of words, their letters, and their sounds.
If you want to see what it means to code visual information with different levels of skill, try staring at these song lyrics for a few seconds, then look away and try to copy them from memory with a pen or pencil:
1. I thought you would be as big as a whale.
My nets were knit.
2. Eu quero é dar o fora
E quero que você venha comigo.
The task should show how increased knowledge of a language allows us to code the visual information more deeply (quickly and accurately put it into memory). The first (#1) can be coded by readers of English using grammar structure, word meanings, sounds and visually (you might have gotten a few words wrong, but probably got the gist). While most readers of English should be able to read the middle passage (# 2) out loud, they will be slower to move their eyes over it, and they will have difficulty spelling it from memory. Knowledge of the meaning and letter-sound correspondence allows you to code the text as sound. This, in turn, makes it easier to remember what the visual words looked like than if the words were in a different script. For the last section, unless you know Chinese script, you will not be able to remember how to write it (I can’t either), and unless you speak Mandarin, you will have even more trouble. Even in English, some readers will have different levels of vocabulary and/or letter-sound correspondence knowledge which will effect their ability to code what they are able to perceive. Dyslexia due to a difficulty in coding some aspect of letter-sound correspondence is another roadblock to efficient reading.
In the following section I’ve included some examples to help demonstrate how word-sound knowledge is coded.
2. Subvocalization vs phonology
It’s fun to compare reading to listening. According to a 2006 study of spoken communication, conversational speaking rates between familiar speakers are about 216 words per minute (277 milliseconds, per word). This is close to the average amount of time spent looking at a word while reading (220-280 milliseconds). The intuitive, theoretical notion is that written words are coded as sounds (phonological codes) or even motor movements of the mouth at some stage of reading. According to Alan Baddeley’s model of working memory, an articulatory mechanism, akin to repeating words in your head, kicks in to sustain words in memory after 1-2 seconds. It’s hard to distinguish which of these two processes the zealous leaders of speed reading cults wish to eliminate in order to increase reading speed. Is it the initial phonological coding or is it the secondary memory loop which these swindlers detest? Phonological coding is fast and automatic, possible even with massive distraction, so this is not going to be easy to consciously stop.
If you’re curious to see how easy it is to retrieve phonological codes during reading, try this task that Seidenberg gives at the end of his book. Put a pencil between your teeth and repeat the words “coca cola” to yourself out loud (if you don’t have a pencil, try mumbling, sanely, to yourself) while you determine whether the following words rhyme:
This should have been perfectly possible, despite the distraction — further evidence that you can retrieve sound information from words with relative ease. And what happens with ease is hard to turn off.
It is important to mention Dyslexia again here. It could be that attempts to eliminate subvocalization come from reading coaches who are dyslexic themselves, and have claimed to find a better way around phonological coding. Even so, this method of ignoring spelling-sound correspondence has not been shown to help all readers with dyslexia, many of whom might stand to benefit from phonology training despite early diagnoses.
Perhaps attempts to eliminate subvocalization are a means to encourage readers to avoid remembering what they just read, to nudge them to drop what is in their memories and look forward. The extent to which readers rely on Baddeley’s style of phonological loop while reading might vary with the text, as certain texts make you use your short-term, or working, memory more.
Below I try to show situations where remembering what you just read might come in handy (when reading ambiguous phrases).
3. Word presentation speed
Your eyes are constantly moving and stopping, moving and stopping. When they move between two words it is done with a burst of speed in a jump called a saccade. Importantly, you are unable to see during the saccade. This limits how you look at words and so it is another place where reading can be sped up. The average amount of time you spend looking at one word is about 200-280 milliseconds (4-5 per second), although this can vary. This also accounts for roughly 30 milliseconds it takes the eyes to do the moving. But what if you could take away the need to move them?
Here is a paragraph from Harry Potter and the Sorcerer’s Stone played at 3000 wpm:
Here it is played at 3000 wpm:
Here it is played at 1000 wpm:
Here it is played at 800 wpm:
Here it is played at 400 wpm:
Here it is played at 200 wpm:
Which one did you prefer?
The technique for displaying one word at-a-time is called rapid serial visual presentation or RSVP. This method has allowed readers to strictly control the rate of word presentation. Back in 1986, Michael Masson tested to see if people could understand sentences presented fast (100 ms or 600 wpm) via RSVP. He found that at speeds of 400-600 words per minute via RSVP, people were able to use context to predict the final words in sentences such as “stamp” in “He mailed the letter without a stamp,” while they were unable to predict final words such as “stamp” in scrambled sentences like “Without he mailed letter a the stamp.” If context had an effect, the logic was that people are “doing comprehension” at these fast rates. After a series of experiments where subjects were asked to name or respond to the final words in the Ok vs Scrambled sentences, Masson concluded that context did have an effect at high presentation rates: readers could use the context to help them identify upcoming words, and they could do this faster than they could move their eyes during reading.
As mentioned above, it takes the eyes roughly 280 milliseconds (250 for fixation + 30 for saccade), on average, to begin reading one word and then get to the next one. Masson was saying that comprehension processes within that 280 millisecond window might take less time if they weren’t bogged down by the muscle movement in the eye. Does this mean that reading comprehension can happen at faster-than-normal (280 ms) presentation rates? Here, please consider that these “comprehension processes” being referred to in the above experiment are merely one part of what goes on during comprehension which includes coding information and the remembering it later. Eggs, for instance, are a part of the cake process, but you wouldn’t eat raw eggs and say you were having cake.
While Masson did not test comprehension in terms of recall for the material, the experiment shows that when words are presented rapidly, some early comprehension processes are able to cope with the speed. This example however, was for single, unrelated sentences with breaks in between. What happens when you try to use RSVP to string sentences together?
In another series of experiments, researchers tested RSVP reading at average (200 wpm), high (400 wpm), and really high (600 wpm) presentation rates. Readers who participated read paragraphs borrowed from an old experiment by John Brandsford and Marcia Johnson. One example is below.
You can try to guess what it is about. The paragraphs go like this:
The procedure is quite simple. First you arrange everything into different groups. One pile may be enough if you don’t have much to do. Then you have to go somewhere else if you do not have a machine. You put them into the machine and turn it on. It is better not to put too many in at once. Then you sit and wait. You have to stay there in case anything goes wrong. Then you put everything in another machine and watch it go around. When it stops, you take the things home and arrange them again. Then they can be put away in their usual places. Soon they will all be used again and you have to do it all over. The whole thing can be a pain.
In case you couldn’t tell, the paragraph describes the process of doing laundry! When RSVP and normal readers were given the same amount of time to read these passages, they performed equally well at getting the gist. When the word “Laundry” was inserted, readers in the RSVP condition were able to use this word to their advantage, more so than the regular readers. This shows how RSVP might provide a good gist based on single words, similarly to how speed readers might make educated guesses and inferences to make sense of a paragraph using only a few words, skipping the rest.
When else does RSVP not work?
Logical difficulties with RSVP arise when you read so-called garden path sentences:
“The raft floated down the river sank.”
or otherwise ambiguous sentences like this from a Time magazine headline:
“The FBI director responded to the President’s angry tweet storm calling for him to be jailed.”
In the ambiguous case, it is hard to tell who is calling whom to be jailed. When connecting sentences, ambiguity from the text or from the reader (because they blinked or spaced out) can lead to errors that require looking back at the text. As a result, RSVP hinders comprehension when larger texts like paragraphs. When paragraphs are cranked up to 600 wpm, you won’t have time to reanalyze garden-path sentences or resolve ambiguities.
An argument for skimming
You’ve gotta get that reading done, so go ahead and skim around. Soon you will find something you want to actually read.
For more info check out:
For info on speed reading, check out Rayner, K., Schotter, E. R., Masson, M. E., Potter, M. C., & Treiman, R. (2016). So much to read, so little time: How do we read, and can speed reading help?. Psychological Science in the Public Interest, 17(1), 4-34.
For books on reading:
Check out The Psychology of Reading by Keith Rayner, Alexander Pollatseck, Jane Ashby and Charles Clifton Jr.
Here is a link to Seidenberg’s chapter on speed reading: