Work the Shell - Looking More Closely at Letter and Word Usage

Time and again I have entreated you, dear readers, with my plea for “A letter! My column, nay, my kingdom for a reader letter!” And lo, the miracle occurred, the heavens parted, the angels sang and a letter arrived:

First off, before I even read the letter, I was intrigued by the closing quote. Latin? Isn't that, like, a dead language? Turns out the quote's a good one though, especially for IT admins in big companies. It roughly translates to “when all else fails, play dead”, and it comes from the Red Green Show, a Canadian comedy. (Thanks Google.)

Now, on to the heart of the letter. Mike's referring to an earlier column where we looked at how to use shell scripts to ascertain letter and word usage, using three books as source material: Dracula, A History of the United States and Pride and Prejudice, all downloaded from Project Gutenberg.

In that series of columns, we ascertained that the ten most common letters in the English language are e, t, a, o, n, i, s, r, h and d. Are they the same if we constrain it to just the first letter of words? Let's find out.

Once we have a corpus of writing and the ability to break it down by words, so that the input stream to the counting script:

is
like
this

it's done like so:

$ cat dracula.txt | tr ' ' '\
' | grep -v '[^[:alpha:]]' | grep -v "^$"

That'll turn Dracula into the world's narrowest book, with one word per line.

Now we simply can add to it to axe all but the first letter by appending cut -c1. The result looks like one of those streams of letters in The Matrix, but that's another story.

So, all that's left is to translate uppercase into lowercase, sort, and then use our friend uniq -c to tally up the results:

tr '[:upper:]' '[:lower:]' | sort | uniq -c | sort -rn | head

And, the resultant top ten are:

20648 t
15787 a
11110 i
10655 w
9906 h
9030 s
7618 o
5720 m
5411 b
4597 f

Quite different! Now, the question is, does it change based on the type of content? Let's do the same command, but this time, let's feed in all three of our books, not just Dracula (though with the rabid <cough cough> popularity of Twilight, maybe Linux Journal would do well to stick with a vampire theme for a few issues?):

34359 t
27053 a
18212 w
18119 h
17854 i
15746 s
13614 o
10076 b
9792 m
7712 f

It's not exactly the same. Isn't that interesting? I'm not sure what to make of it, but as you can see, a good grasp of shell script commands makes finding out this sort of fairly goofy information interesting.

But, we're not quite done, because Mike also wondered about two-letter combinations. It's this sort of query that really shows just how helpful becoming savvy on the command line can be. To calculate that requires only one character to be changed in the command invoked above. Do you know what it is?

It's the cut command. Above, we're specifying that we want only the very first character of each line of input with cut -c1. If we want the first two, we simply can tweak that command flag as appropriate.

But, -c2 won't work, because that'll give us only the second letter of each word (and the most common second letter in the English language is o, followed by h, e, a and n).

Instead, we need to use a letter range, which looks like this: -c1-2. The result of that invocation is:

22100 th
10168 an
9138 to
7508 he
7100 of
5873 i<space>
5517 in
5332 ha
5157 be
4664 wh

There ya go, Mike. The most common two-letter combination in the English language is th, which actually makes some sense, followed as a distant second by an.

I hope it's trivially obvious how you could use this to calculate the most common three-letter combinations (it should be no surprise at all that the is the most common three letter combo, followed by and).

I'll wrap up here, but again, I invite you to send me your letters and queries so we can explore various ways to use shell scripts.