It works fine with code -- it basically just splits on whitespace and considers anything in the result to be a "word".

The code won't make sense, and parens/brackets/etc won't be balanced, but you can definitely use it for practicing.

EDIT:

Here are a few drills it generated when I gave it its own source code:

    # Lesson generated by Typkov

    Lesson 1
    (string/split (sanitize text) #"\s+"))) dict [w n] 0))) (rest words))))))
    lessons)))) (defn sanitize [s] (-> s (string/replace \" \"))) (defn
    [lesson n] (let [w n] (let [initial-word (random-choice (keys dict))))
    (defn markov-dict ([words] (markov-dict words) n "\n" (join-n lines)))) (defn

    Lesson 2
    text 24) [:text "You must enter a decent amount of
    "])))) "G:MENU\n"))) (defn random-choice [coll] (nth coll (rand-int n))] (loop
    data))) (render "/" data) (not (form-valid? data))) (render "/" data)
    data} (t/home (:text data) (t/home (:text data) (not (second words)]

    Lesson 3
    n] (let [initial-word (random-choice (keys dict))] (case format n] (let
    [w (first words) lessons (repeatedly n (partial string/join "\n")) (def
    \' \') (string/replace \" \"))) (defn markov-dict ([words] (markov-dict {}
    get-lesson [text format n] 0))) (rest choices))))))) (def join-nn (partial