Here's a poor man's introduction to the concept of a "language model." 

Scrabble uses a langauge model. Let's take a look.

First, a few utility functions (which I keep in text.py) and some examples of how they work:

    >>> from text import ngrams, freq

ngrams(astring, n)  will give you all the substrings of length 'n'. (It also works with objects besides strings, by the way.)

    >>> bigrams = ngrams('abczzz',2)
    >>> bigrams
    ['ab', 'bc', 'cz', 'zz', 'zz']
    >>> fq = freq(bigrams)

freq(sequence) will return an dictionary with frequency counts of all the elements of sequence.

    >>> fq == { 'ab': 1, 'bc': 1, 'cz': 1, 'zz': 2}
    True

Note that the ngrams function doesn't remove duplicates -- that's important.

In Scrabble, every letter has a value. Here's the 

ngrams(open('en').read(),2)
from text import freq # ngrams(open('en').read(),2)
freq( ngrams(open('en').read(),2))
[(v,k) for k,v in freq( ngrams(open('en').read(),2))]
[(v,k) for k,v in freq( ngrams(open('en').read(),2)).items()]
sorted([(v,k) for k,v in freq( ngrams(open('en').read(),2)).items()])
sorted([(v,k) for k,v in freq( ngrams(open('en').read(),2)).items()])[-20:]
for i in sorted([(v,k) for k,v in freq( ngrams(open('en').read(),2)).items()])[-20:]: print i