You know the feeling. Mid-sentence, you reach for a word you definitely know — you'd recognise it instantly if you read it — and it simply isn't there. You stall. You substitute something clumsier. The conversation moves on without you.
That moment is not a memory failure, and it is certainly not a sign that you're "bad at English." It is a window into one of the most important and least-understood things about language: the mental lexicon. Understand what it is and how it's built, and a great deal about effective language learning — including why we spend so much of our time at Choc Education playing word games that look like pure fun — stops being mysterious.
Your mental lexicon is not a dictionary
The intuitive picture of vocabulary is a dictionary in your head: an alphabetical list of words, each sitting next to its definition, looked up one at a time. It is a comforting picture, and it is almost entirely wrong.
A dictionary is organised by spelling and stores words in isolation. Your mental lexicon is organised by meaning and stores words in a vast, interconnected web. Every word is a node, wired to hundreds of others by threads of meaning, sound, grammar, and association. Dog connects to cat, bark, puppy, leash, loyal, bone, and to the texture of a particular dog you knew as a child. You do not look words up. You traverse to them, jumping from node to node along the threads — which is why one word so reliably summons the next.
We know the lexicon is a web, not a list, because of how it fails and how it leaks:
- Tip-of-the-tongue states — the exact feeling above — show that meaning and word-form are stored somewhat separately. You can have full access to a word's meaning and its first letter and its rhythm, while the form itself stays just out of reach. A list wouldn't fail that way; a network retrieved through multiple routes does.
- Slips of the tongue almost never come out random. When people misspeak, they swap in a neighbour: they say "fork" for "spoon", "Tuesday" for "Thursday", "left" for "right". The error reveals the wiring — the wrong word was sitting right next door.
- Word-association tests show astonishing agreement between native speakers. Say "salt" and most English speakers answer "pepper"; say "dog" and they answer "cat". Those pairings aren't in any dictionary. They live in a shared, conventional network that native speakers all carry — and that, it turns out, is the thing a learner has to acquire.
Co-hyponyms, semantic fields, and "density"
Two ideas make the web concrete, and both matter enormously for learning.
The first is the co-hyponym. Words that share a category are co-hyponyms of each other: rose, tulip, daffodil, lily are co-hyponyms under the umbrella word flower; coffee, tea, juice, water under drink. Native speakers don't store these as a random scatter — they store them as a tight cluster, a semantic field, with all the fine distinctions mapped: which drink is hot, which is sweet, which you'd order at breakfast, which sounds odd to ask for at a bar. That map of fine distinctions is fluency. It's the difference between knowing the word "drizzle" and knowing that drizzle is lighter than rain but wetter than mist.
The second idea is semantic density: how richly a given word is connected into the web. A word you've met once, as a translation, hangs by a single thread — it's reachable only by going back to your first language and translating across. A word that's dense — wired into a category, into examples, into feelings, into the other words it likes to appear beside — can be reached from a hundred directions at once. And density is speed. The more threads lead to a word, the faster you find it under the pressure of real conversation, and the more natural it sounds when it arrives, because you've reached it the way a native speaker does rather than by translating in your head.
Why learners get stuck — the thin lexicon
Here is the core problem with how vocabulary is usually taught. The traditional method — memorise a list, English word beside its Chinese translation, tested by translation — builds exactly the wrong structure. It produces a lexicon of isolated pairs: each English word tied by a single thread to its L1 equivalent, and to almost nothing else.
A learner with a thin, pair-based lexicon can pass a vocabulary quiz and still freeze in conversation, because in conversation there is no quiz prompt to translate from. They have to retrieve straight from meaning, at speed, and their words are reachable only by the slow translation route. They sound "off" not because their words are wrong but because their network is wrong — sparse where a native's is dense, and wired through their first language instead of through English itself. (This is the same failure we describe in Traditional English Teaching vs Modern from a different angle: lists give learners words without giving them language.)
The goal of good vocabulary teaching, then, isn't simply more words. It's a mental lexicon whose shape matches a native speaker's — same clusters, same density, same direct routes from meaning to word. The technical name for the goal is alignment: bringing your network into register with the conventional network that native speakers share.
Why word games do what flashcards can't
This is where games stop being a treat and start being the mechanism. A good word game forces you to travel the web — and travelling the web is precisely how its threads get laid down and thickened.
- Categorising and odd-one-out games make you build and test semantic fields in real time — sorting co-hyponyms, deciding what belongs with what. Every judgement strengthens a cluster.
- Describing and guessing games make you reach a word through its meaning and its neighbours — the native route — instead of through translation. When you have to describe "umbrella" without saying it, you're firing every thread around the node, and firing threads is how you thicken them.
- Association and connection games wire new words straight into the existing web, which is what gives them density — and therefore retrieval speed — from the start.
Compare that to a flashcard, which rehearses one thread (word → translation) over and over. You can drill a flashcard a hundred times and the word will still hang by that single thread. Five minutes of a well-built game can do something the flashcard structurally cannot: give the word a neighbourhood. This is also why games score so high on what researchers call involvement load — the depth of engagement that predicts whether learning sticks. We explain that mechanism fully in How to Improve English Effectively; here it's enough to say that a game makes you need a word, hunt for it, and weigh how it's used, all at once — and the network grows along exactly those lines.
A necessary nuance: letter games vs semantic games
Not all word games are equal, and it would be dishonest to pretend otherwise.
Letter-and-spelling games — Scrabble, and our own gloriously competitive Letter Tree — are genuinely valuable, but for a narrower reason than people assume. They train lexical access and retrieval speed: the sheer quickness of pulling a word out of storage, and the active recall of vocabulary you'd otherwise only ever recognise. One of our teachers credits a childhood of Scrabble for exactly this kind of fast, deep word-store. That's real, and it's worth having.
But spelling games organise words by their letters, and your mental lexicon doesn't — it organises by meaning. So for a learner specifically, semantic games (describing, categorising, distinguishing near-neighbours) do the deeper work, because they build the meaning-based web that fluency actually runs on. Our honest position: Letter Tree is the gateway — it's addictive, it sends people hunting down marvellous rare words, it sharpens recall — but the semantic games are the medicine. The fun gets you in the door; the meaning-work is what rewires the lexicon.
How this shapes what we do
Almost everything distinctive about Choc Education falls out of taking the mental lexicon seriously.
It's why our English events are full of word games rather than small talk, and why those games are chosen, not random — each one targets a specific part of the web. It's why the exercises in our learning ecosystem are designed by pedagogical experts around semantic structure, not auto-generated word lists, and why they're speech-focused: the lexicon you build by speaking is the lexicon you can reach while speaking. And it's why the system analyses what you actually produce — so your teacher can see which parts of your network are dense and which are still hanging by a thread, and steer you straight at the gaps.
You don't need any of this vocabulary — mental lexicon, co-hyponym, semantic density — to benefit from it. But now that you have it, you know why the room is laughing and learning at the same time. The laughter is the network being built.
Come build yours — our word-game events run multiple times a week, and our courses put the same science to work every lesson. Curious about the thinking behind it? See our ethos.