>> Thank you for showing us your brain!
And showing us how to understand more
about ours. Well... Coming up, we
have... The third of four talks. In this
block. I am pleased to introduce...
Da-da-da da! I'm just waiting for it to
show. I'm pleased to introduce my friend
Adam Parrish, who will be speaking on
the topic of Scrabble sucks! Towards
higher order word games.

 >> Hi, everybody.

 >> Hi, Parrish.

 >> How is it going?

 >> Good afternoon, sir.

 >> Okay, so... For the purposes of this
talk, I'm going to assume you know what
Scrabble is. If you don't, I give you
permission to look it up on your phone
on Google or whatever. I'm going to
assume that you know what Google is as
well. This is me. I am a computer
programmer and experimental poet and a
game designer. The title of my talk
today is Scrabble sucks, but the thing
with Scrabble is that lots of people
have fun playing it, and there's an
amazing culture surrounding it in the
competitive scene or whatever. Obviously
people think it's fun so I want to
revise the statement down a little bit
and say instead this. So what I'm going
to talk about in this talk is... Sort of
what I think is wrong with Scrabble and
some games that I've designed that work
differently and I hope better. So there
will be a little bit of computer
programming along the way.

First of all, I want to explain what
evil Scrabble visited upon me that made
me hate it so much. And that is that
once I was visiting my family in Utah
over the holidays and we decided to play
a board game. We picked Scrabble, and
this is basically how the game went. My
mom played the word North, which is
fine, 24 points, my sister played
fireman, which is great, 26 points, and
I look at my rack and think -- it's my
term to play. And I don't want to be an
insufferable smart-ass, but what can I
do? So I end up playing qoph, qi, and pe
and hm. This is perfectly legal
Scrabble, and it's worth 66 points, but
is it worth the contention and strife it
caused? Not one, but four weird words in
one turn. This had a very deleterious
effect on our fun. Everybody thought I
was engaging in ostentatious brain show
boating, and I kind of was. But it
wasn't fun for anybody, so eventually we
decided to stop playing Scrabble and
play something else. So basically this
is why I hate Scrabble.

(applause)

Now, a nicer way to phrase this might
be... Competitive Scrabble playing
requires a lot of arcane knowledge. You
have to memorize a lot of words, both
tiny and large, so when you're playing
with people who don't have that
hermetic, monk-like knowledge of all the
things you need to know to play
competitive Scrabble, it can lead to
hurt feelings, that aren't fun. Other
games aren't like this. If I win at a
game of soccer or Street Fighter or
something, nobody thinks I'm a
smart-ass. It's something particular to
Scrabble. So it's easy to chalk these
problems up to the culture of Scrabble,
or my personal inability to keep my
smart-assery in check for more than 30
seconds at a time. But there are
important structural problems with the
game that are important to mention. It's
based on unigram frequency, which means
moving parts whose value and commonality
are determined by frequency of letters
in the English language. More As and As
are worth less, one Z and the Z is worth
a lot. Letters are put into play by
drawing them randomly one at a time.
Another important aspect. Now, there are
advantages to this model. One is it's
easy for people to understand. We get
that E is a very common letter so there
should be fewer of them and they're
worth less, but Z is a very uncommon
letter so you should get more points for
playing them. The model is very familiar
and used over and over in many word
games. One issue is that our intuitive
understanding of how words are put
together doesn't stop at just knowing
the frequency of letters. We also know
things like the sequence ING is common,
specifically, but the ends of words --
the letter K is frequently preceded by
C, and re is a sequence of letters you
can put in front of a word to give you
another word. Scrabble doesn't reward
that knowledge. It rewards having a big
arcane vocabulary. So my
hypothesis/hunch/things I'm operating on
as an assumption is that if valuable
letters in a game are associated with
higher scores and there are fewer
valuable letters and those letters are
drawn at random, the most valuable plays
are going to be shorter words with rarer
letters, which tend to be those
smart-ass vocabulary test words that we
were talking about earlier, that stop
you from being able to have fun playing
a game with your family.

So it doesn't have to be this way. I
don't think that people should have to
memorize long lists of words or have
extensive vocabularies in order to enjoy
and be competitive at word games. As
a linguist and poet, I'm sensitive to
the fact that everybody is a fluent
speaker of their own idiom, has an
intuitive understanding of how words are
put together, and everybody wants to
play with that and have fun with it and
come away without the feeling that
they're unintelligent. It breaks my
heart that somebody could play with
words, something we all understand and
like, and come away thinking they're not
smart. So here are some of my
experiments in game design that I think
might address some of those problems,
that specifically don't use unigram
frequency as their model.

This first game is called Rewordable. I
designed this with my friends Adam Simon
and Tim Szetela. The basic idea of this
game is -- what if the pieces of the
game, instead of unigrams were bigrams
and trigrams. This is a deck of 160
words -- some of them are single
letters, but most of them are bigram or
trigram sequences of two or three
letters that were selected based on
their frequency in English. So the idea
is that because the units are larger,
players are going to be able to form
longer, more satisfying words. We're
still kind of working on the rules of
this, but I think it kind of goes a ways
to accomplishing the goals I was talking
about before. We're going to do a
Kickstarter for this soon, so keep your
eyes out for that. A second game is
called Characterror. That I made a few
years ago. You play as this sort of
ASCII ship on the right with the letters
trailing behind it. The idea is that you
select one of the slots on the left and
then you kind of fire the letters from
the right into the slot to make words.
And then you can press a button that
will score the word. Kind of bank it. So
the trick of this is that the letters to
the right there are generated with a
Markov chain, instead of just being
selected...

 >> Woo!

 >> Instead of just being selected
randomly, based on frequency or
something like that. So now some of you
are probably saying... Wait, Markov?
What does that even mean? Probably very
few of you are saying that. But I just
kind of wanted to explain, give my very
brief explanation of how a Markov chain
works. So let's start with a corpus
consisting of a single string, the word
condescendences. Which is what this talk
consists of. If you find all the unique
bigram sequences of two letters in this
word we end up with a list like this.
But what a Markov chain does is in
addition to having a list of bigram, it
finds for each bigram what letters
follow each bigram and builds a
probability table. We see that the
bigram de half the time is followed by s
and half the time is followed by n.

If we were to continue and make a
probability chart like that for all of
the letters, it would look something
like this. So this is a statistical
model of what the text in that corpus
looks like. Now, Markov chains are
famously used for generating amusing
text. So if we recursively feed our
prediction back into the prediction
making thing, we end up with new
sequences of letters that
statistically resemble but are not
identical to the original source text,
so that's like this. Condendescencesces.
This is generated by the chain
condescendences. Imagine if we had a
Markov model of all the words in the
English language. Then we would be able
to create a way of giving people letters
in word games that will be useful to
them, based on what's already in play.
So another game I made to sort of play
with this is a game called Lexcavator.
This is kind of a cross between Boggle
and Mr. Driller. You select words and
that allows you to let your guy go
further down into the infinite well of
words here. So to ensure that the words
you find are interesting and fun, the
board is generated actually with Markov
chains sort of moving in two directions,
and I kind of made like a tiny diagram
of how that works.

So, like, each one of these slots, going
from left to right, is going to be
filled in -- that character is going to
be filled in with a Markov chain based
on the probability that it would come
after some word either directly above or
diagonally above, across the columns.
And that would just get repeated for
each one of these things. So what that
ends up with doing -- what that ends up
doing is giving you sort of more sure
ways to precede downwards, and also
longer words that are slightly, like,
not quite as arcane. So the question,
then, arises with the remaining 30
seconds... Is does this actually work?
And I did a corpus analysis of Scrabble
games, versus Lexcavator games. This is
a histogram which shows word frequency,
word commonality in English versus word
frequency in both games. So you can see,
I pointed out sort of on these graphs
where Lexcavator has more words than
Scrabble, and you can see they're kind
of bunched up here to the left, which
means basically yes. In Lexcavator, you
have words that are more common, instead
of in Scrabble, where the words are more
spread out across the whole thing.
Anyway, that's all the time I have.

(applause)