Yet more results


Here are yet more results. As before, the number is the alpha parameter of a stable distribution model of paragraph lengths. I've refined my estimation procedure somewhat (but two decimal places of accuracy is still almost certainly excessive).

 1.01 j d salinger         catcher in the rye
 1.24 c s lewis            last battle
 1.33 c s lewis            dawn treader
 1.33 c s lewis            lion witch wardrobe
 1.35 john steinbeck       mice and men
 1.37 charles dickens      christmas carol
 1.40 charles dickens      oliver twist
 1.40 lewis carol          alice
 1.43 mark twain           huckleberry finn
 1.43 jane austen          pride and prejudice
 1.45 charles dickens      two cities
 1.47 mark twain           yankee in king arthurs
 1.49 scott fitzgerald     the great gatsby
 1.50 jane austen          emma
 1.51 jane austen          sense and sensibility
 1.52 lewis carol          looking glass
 1.55 thomas hardy         far from the madding crowd
 1.56 lewis carol          sylvie
 1.58 jules verne          around the world
 1.58 h g wells            men in the moon
 1.60 orson scott card     seventh son
 1.60 jules verne          20000 leagues under the sea
 1.62 orson scott card     xenocide
 1.65 orson scott card     enders game
 1.69 paul harrison        overclocking
 1.71 cory doctorow        down and out
 1.73 h g wells            invisible man
 1.76 thomas hardy         tess of the dUrbervilles
 1.76 michael crichton     jurassic park
 1.79 j r r tolkien        lord of the rings 1
 1.80 h g wells            war of the worlds
 1.80 robert m pirsig      motorcycle maintenance
 1.81 j r r tolkien        lord of the rings 3
 1.84 g k chesterton       all things considered
 1.84 margaret mitchell    gone with the wind
 1.85 thomas jefferson     state of the union
 1.85 h g wells            doctor moreau
 1.87 cory doctorow        the super man and the bugout
 1.91 j r r tolkien        lord of the rings 2
 1.96 charles darwin       origin of species
 2.00 cory doctorow        craphound
 2.00 cory doctorow        drm talk
 2.00 cory doctorow        return to pleasure island
 2.00 michelle dawson      misbehaviour of behaviourists
 2.00 michelle dawson      no autistics allowed
 2.00 paul harrison        finity
 2.00 richard stalmann     should be free
 2.00 temple grandin       inside view
 2.00 temple grandin       my experiences
 2.00 temple grandin       thinking in pictures

It's pretty clear that authors produce documents with moderately consistent alpha. Sci-fi/fantasy authors generally seem to have greater alpha than authors of "literature". I had suspected the sci-fi/fantasy genre might be linked to autism in some way, and this would seem to be some confirmation of that. It is a genre that has failed to be classified by any more obvious metric, such as the presence of spaceships or dragons, and yet seems clearly distinct from the mainstream.

As I said earlier, I was expecting alpha values to be lower for autistic people. This would not seem to be the case. I can think of reasons why this might be so, but bear in mind I am inventing a model to fit the data, not making a prediction...

My guess would be that although autistic people write paragraphs of more even length, the topics they discuss are more peculiar than normal. If paragraph length is a measure of excitement, people towards the autistic end of the spectrum may get less excited about unusual ideas -- unusual ideas don't blow them away and set them to babbling. This would then allow them to explore more unusual ideas without freaking out.

For example, the sentences "I went to Chadstone to do some shopping and have a look around Borders, but they didn't have any good new stuff." and "He was ambassador to 1975, but back home in 1898, in New Jerusalem, Utah, they all thought he was ambassador to France." contain the exact same number of words.

A computational model of this i am not sure about. Either a difference in priors or a difference in search strategy. I would be quite upset if it were a difference in priors, but this may turn out to be the case.

Theory aside, assuming this alpha value is a useful measurement, what are the implications?

I can't help but imagine an AUTOEDITOR program:

Author: I have written a new fantasy novel. It has elves and magic and stuff. Will you publish it?


Author: Now hang on, that's not very nice....



... though of course discrimination can go either way.

The other worry about a measure like this (again, assuming it's a useful measurement) is that dysfunctional forms of autism are clearly just the extreme end of a spectrum. Autism isn't just an unfortunate disease confined to a few individuals, but rather an omnipresent parameter that influences the whole structure of society, intimately caught up in institutions such as churches and monasteries, academia, geeks vs jocks, muggles vs wizards, and so on.

It might even be the case that dysfunctional autism represents the failure to converge of an algorithmic extreme that, when it does converge, is very useful in social-evolutionary terms. There are plenty of examples of evolution making such bets. See for example plants that sow many seeds, only a few of which survive.

In any case, a genetically based caste system as in a bee-hive or ant-nest. Why not? There are clear evolutionary advantages to such a thing. The expression of a certain phenotype dependant on the presence of certain recessive genes, thus ensuring their expression at the correct rate even if that phenotype does not itself tend to breed or survive at the same rate. How would society deal with such a thing? Well, obviously, our current society couldn't without some ugly upheavals. It's just not a nice scenario.