Major system mnemonics with a smaller vocabulary

In the previous post we constructed a mnemonic phrase using parts of speech and simple grammars, but the generated phrases used a lot of abstract and esoteric words. To make phrases more concrete, we can use a smaller list of words, such as this one on Wikipedia:

(def nouns
  '[hose hat hen home arrow whale shoe cow hoof pie
    sauce seed sun sumo sierra soil sewage sky sofa soap
    daisy tattoo tuna dome diary tail dish dog dove tuba
    nose net onion enemy winery nail nacho neck knife honeybee
    mouse meadow moon mummy emery mole match mug movie map
    rice road urine rum aurora railway roach rag roof rope
    louse lady lion lime lorry lily leech leg lava lip
    cheese cheetah chin gem shrew chilli cha-cha chick chef jeep
    goose cat coin game crow clay cage cake cave cube
    vase video fan fairy fool veggie fig fife vibe
    boss bead pony puma berry bell pouch bike beef pipe])

We can define verbs and adjectives similarly.

Some of the words in Wikipedia's list aren't in the CMU pronunciation dictionary, so we'll define their encodings manually:

(def manual-encodings
  {"shrew"   "64"
   "chilli"  "65"
   "cha-cha" "66"
   "swishy"  "06"
   "sappy"   "09"
   "nudgy"   "26"
   "naggy"   "27"
   "haram"   "43"})

Now we can assemble our dictionary:

(defn encode-as [part-of-speech]
  (fn [word]
    (let [word (name word)]
      [part-of-speech
       (or (major-phoneme-encode word)
           (manual-encodings word))
       word])))

(def dictionary
  (reverse (concat
             (map (encode-as :noun) nouns)
             (map (encode-as :verb) verbs)
             (map (encode-as :adjective) adjectives))))

I've reversed the list so generate-mnemonic prefers words for double digits.

Let's see what using this smaller vocabulary generates for the grammatical pattern [:noun :verb :noun]:

There are fewer matches generated for each sequence, but there are lots more memorable images. Given some imagination and personification, only a few phrases are hard to conjure a mental picture for.

To create additional matches, we can try other grammatical patterns, such as [:adjective :noun]:

or [:adjective :noun :verb]:

Longer sequences of digits require longer grammar patterns. You can guess at the necessary pattern length given that a single word only encodes one or two digits. To encode the first ten digits of the golden ratio (1.618033988), we need at least five parts of speech:

While outlandish imagery is a helpful mnemonic, the longer the phrase the more strained the image becomes. It can be difficult to generate something outlandish but memorable without crossing the line into forgettable nonsense, turning a sequence of random digits into merely a sequence of random words.