So, worried that I was being overly-negative, I thought I'd offer something of substance. If you wanted to separate the Markov bits from the string bits, you co

So, worried that I was being overly-negative, I thought I'd offer something of substance. If you wanted to separate the Markov bits from the string bits, you could do it thusly: (defn markov-chain [coll] (let [adjacencies (partition 2 1 coll) initial-chain {::start [(first coll)] (last coll) [::end]}] (reduce (fn [chain [predecessor successor]] (assoc chain predecessor (conj (get chain predecessor []) successor))) initial-chain adjacencies))) (defn successor [chain item] (rand-nth (get chain item))) (defn follow-chain [chain] (->> ::start (iterate (partial successor chain)) (remove #{::start ::end}) (take-while some?))) (defn sentences [text] (clojure.string/split text #"\.")) (defn words [sentence] (remove empty? (clojure.string/split sentence #"\s+"))) (defn markov-text [text] (->> text sentences (map words) (map markov-chain) (apply merge-with into))) (defn render-sentence [words] (str (clojure.string/join " " words) \.)) (defn sentence [chain] (render-sentence (follow-chain chain))) I'm not saying this is in any way strictly better, but it leaves you with abstractions that can be reused, and text-specific stuff in one place that could be a separate namespace. This is what I'd look for in a code-review of a refactoring like this - not just neatness, but making abstractions explicit and separate, naming as many things as possible, etc etc.