There are two neurological 'routes' that you use to identify words.
One of them translates letters into sounds, and then recognizes those sounds as a word. This is called -very creatively - the phonological route. "Sound it out," we were taught as kids. And this is how we read new or nonsense words, like "brillig" and "slithy".
The other route translates the visual input of the word directly into meaning. Called - once again very creatively - the direct route, this is how experienced readers recognize virtually all familiar words. The direct route is also necessary for irregular words like "colonel".
I learned this much in my cognitive neuroscience course this past spring. There is plenty of evidence for these two separate routes - to the point where you can have a dysfunction of one, or the other, but not necessarily both at the same time.
But when I Google "word recognition," I see nothing of this sort, but instead I get results talking about word shape, serial letter recognition, and parallel letter recognition. So what's the deal here?
Turns out that these factors also relate to how we recognize words.
When you look at a word, you can immediately see its shape. Take the word "look" as an example. Its letters have a particular shape: "o" is called neutral, "l" and "k" both rise above the neutral position. Letters like "p" and "j" dip below it. Words that are spelled in all uppercase lose some of their shape. It turns out that you read lowercase words faster than UPPERCASE WORDS, and you read words in AlTeRnAtInG CaSe slowest of all. This lends some evidence to the idea that your brain uses the shape of the word to help determine what the word is. However, if you take out all the letters and just leave a blob that looks like its shape, (as you'd imagine) it's not very easy to identify words. The letters have something to do with it. So...
Serial Letter Recognition:
In English, words are spelled with letters from left to right. Thus, it makes sense that we would recognize words letter-by-letter, from left to right, one at a time. If this were true, then you'd assume that longer words take more time to recognize than shorter words. And this is the case. So this too has something to do with how we recognize words.
Parallel Letter Recognition:
Any literate adult will notice that she doesn't look at every letter of every word when she reads. A reader's eyes will jump from word to word along a page, skipping past function words like "a" and "to" and "the", taking information in gulps. The best hypothesis about how an experienced reader recognizes words involves processing all of the letters in view at the same time and converging upon a word using the total information.
To understand this, visualize a network of words, all connected by the letter positions they share. When you look at the word "humble", all of the word-nodes that begin with "h" followed by "u" become active, and at the same time all of the word-nodes with "m" and "b" in the middle become active, and at the same time all of the word-nodes ending in "e", with an "l" right before it become active. So "humbug" and "bumble" and "tumble" and "humbly" all become active. But the most active word-node will be the one containing all of these letters in this order, and thus you recognize the word as "humble". This in itself may relate to "word shape" in that you identify patterns of adjacent letters and their placement within the word as a way to identify the word.
So now we have two systems of thought as to how we recognize words: the phonological/direct routes, and also the serial/parallel recognition models. How on earth is a neuroscience undergrad supposed to figure out how these work together.
What follows is pure speculation.
Phonology is temporal. What I mean to say is when you hear a word, or sound out a word, you do it in order from beginning to end. So it would make sense (from my limited knowledge) that the phonological route relies more on serial letter processing than parallel letter processing or word shape.
The direct route is not temporal. By that I mean that relating a visual input directly to its meaning doesn't necessarily mean you look at things from "beginning" to "end". So the direct route is free to rely more on the faster and more efficient method of parallel letter processing or word shape instead of being constrained to serial letter processing.
And of course, now that I've posited that wild speculation, I'm thinking of a million ways to verify that experimentally. (Where I'm going to get a population of people with phonological and surface alexia to experiment on - as an undergrad - I have no idea.)