I Had 10,000 English Words in My Head. Why Couldn’t I Speak?

You know the words. Reading feels like automatic doors. Then you open your mouth. The doors jam. You want to say, “Wire this into the pipeline.” What comes out is, “Put this new code into the old process.” The other person nods. You nod. The conversation moves on. You also move on, but part of you stays behind, staring at the sentence you did not say.

The usual diagnosis: not enough vocabulary. So you download another list. You learn apocryphal, quintessential, obfuscate. You finish the list. The next conversation still has a pause in it. The pause comes from retrieval, chunks, timing, and the mess of real-time speech.

The 10,000-word illusion

A 10,000-word vocabulary can mean many things. It can mean you recognize 10,000 words when you read. It can mean you can match them to Chinese definitions. It can mean you have seen them in Anki. It does not mean you can pull them out in a conversation at 150 words per minute.

Receptive vocabulary: words you understand when you hear or read them. Productive vocabulary: words you can use without hunting. The gap between them is where most learners live. You can read a novel and enjoy it. You can listen to a podcast and follow the argument. Then you try to tell a friend about the podcast, and the argument collapses into “it was about... you know... how people... yeah.”

Word counts also cheat. If run, runs, ran, running are four entries, your 10,000 is smaller than you think. If you count word families, 10,000 is strong. It sits near the lower end of an educated native speaker’s reading vocabulary. But daily speech does not run on rare words. It runs on a few thousand high-frequency words arranged into tens of thousands of chunks. A native speaker does not wake up and decide to use commensurate. She says, “That works.” She says, “That’s fair.” She says, “Let me get back to you.”

Spelling bee lists are not conversation scripts. You can spell antidisestablishmentarianism. You can define floccinaucinihilipilification. Then you need to tell your coworker, “I’ll send you the notes after lunch,” and your tongue trips over after lunch. The list gave you trophies. The conversation asks for plumbing.

Knowing a word vs using it

A word is a machine with ports. To use it while speaking, you need several layers at once.

Pronunciation and stress. You can spell development, but if you stress the wrong syllable, the listener blinks.

Inflections. Past tense, plural, third-person -s. In writing, you can check. In speech, the clock is running.

Multiple meanings. Get, take, make, run, hold, turn. Each one is a forest. You know run. You may not know run a meeting, run a test, run into a friend, run out of time.

Collocations. You know decision. The natural verb is make, not do. You know question. You can ask a question, raise a question, pose a question. The noun is the same. The sentence changes.

Syntactic frames. A word comes with slots. Wire A into B. Hand something over to someone. Blame A on B. Provide someone with something. If you only memorize “wire = 电线,” you will never say, “Wire this into the pipeline.” Once you learn the frame, the sentence is ready-made.

Register. Commence belongs in a ceremony. Start belongs in a kitchen. You can write “commence operations.” You cannot announce “Dinner commences” without sounding like a robot in a cape.

Pragmatics. Could you...? Would you mind...? I was wondering if...? These are not advanced vocabulary. They are social handles. A large vocabulary with no handles makes you sound like a dictionary with a bad temper.

Retrieval speed. This is the one that kills conversation. You know the word. You need three seconds to find it. In a conversation, three seconds is a hole. The other person fills it. The topic moves. Your word arrives after the train has left.

This is the receptive/productive gap. You have storage. You have poor retrieval. You have a tip-of-the-tongue phenomenon that lasts all day. You can read wire into and understand it instantly. You can hear handover and nod. When you need to produce it, your brain opens an empty drawer.

Conversation is real-time multitasking

People imagine speaking as a simple loop: think of meaning, choose word, choose word, add grammar. The real loop is a circus.

You listen, segment the stream into words, and catch the tone. At the same time, you decide what to say, retrieve words, choose collocations, order information, manage grammar, adjust pronunciation, monitor mistakes, watch the listener’s face, prepare a repair, and decide whether to keep going or stop.

Every task takes working memory. Slow retrieval eats the budget. Shaky listening eats it too. Translating from Chinese eats whatever is left. Soon there is no budget for the actual thought. You sound stupid. The cause is not intelligence. The cause is a marathon while juggling.

Native speakers look relaxed because they drop in prefabricated walls instead of building every sentence from bricks. Follow up. Circle back. Touch base. Run through. Wrap up. Sort out. Figure out. Come up with. Get rid of. These are not fancy words. They are bearings. Without them, the machine grinds.

If you only study single words, you have a warehouse full of loose parts. When an order comes in, you are still looking for screws.

The strong-listener paradox

A strong listener can make any vocabulary level fluent. Partly.

If communication means “the other person roughly understands you,” then a strong listener can do a lot. AI can guess. A patient native speaker can fill gaps. A parent understands a toddler. A good teacher recasts your broken sentence into a clean one. You say, “Put this code into old process.” The AI says, “Wire this into the pipeline.” You see the gap. You learn.

But if fluent means independent, natural, accurate, fast, and appropriate across situations, the strong listener does not solve your problem. The strong listener lowers the bar. The bar is still there when you walk into a meeting with ordinary people, a noisy restaurant, a phone call with bad audio, a group conversation where everyone talks over each other.

A rough formula:

Communication effect ≈ your comprehensible output × listener tolerance × shared context × repair strategies.

Fluency ≈ active chunks × retrieval speed × pragmatic automation.

AI improves the first line. It does not automatically improve the second. You can lean on AI to understand you. You cannot lean on AI forever. You need to be understood by someone who is not trying to understand you.

Why Codex works and where it stops

Using Codex in English is a good drill. You force yourself to explain weird tasks in baby English. Codex recasts. You see “wire this into the pipeline.” You see “leave a handover note.” You see “use the carrier.” You see “rock the pitch angle.” The right side looks like primary school English. The left side was a fog. The drill exposes the gap between what you know and what you can produce.

The drill works because it is task-driven. You are trying to get something done. The words attach to a goal. The recast gives you immediate input. You learn old words in new slots.

The limit is that AI is too forgiving. It guesses. It fills. It smiles through your broken syntax. You need a listener who sometimes does not guess. You need to say it again, in another way, without the AI rescuing you.

A second limit: seeing a recast is not the same as owning it. If you read wire into and nod, you have receptive knowledge. If you use wire into three times the next day, in three different sentences, you begin to own it. If you never use it, it fades.

And AI English is not always your English. It can sound technical or oddly polished. In engineering, “rock the pitch angle” may be vivid, but oscillate, sweep, or modulate may be precise. Small words are powerful. They are not always exact.

Codex is a sparring partner. It gives you reps, feedback, and ideas. It does not run the automation for you.

From word lists to expression networks

If you already have 10,000 words, the next step is not another 10,000. The next step is to turn the words you have into a network.

Stop studying isolated words. Study chunks. Take decision and learn make a decision, reach a decision, regret a decision. Take handover and learn a handover note, do a handover, a smooth handover. Take wire and learn wire A into B, wire up, wire together.

Build scene scripts. Phone calls, meetings, apologies, arguments, small talk, asking for help, saying no. For each scene, prepare ten to twenty replaceable sentences. Conversation is script selection under pressure. You are choosing from prepared sentences.

Retell and back-translate. Read a paragraph in Chinese. Say it in English. Compare with a natural version. Then translate the natural version back into Chinese. Then say it again. Slow practice feeds fast speech.

Force reuse. If you learn wire into today, use it today. Three sentences. One about code. One about a team. One about a habit. If you do not use it, you lose it.

Train listening automation. Speaking is capped by listening. If you cannot hear connected speech, you cannot answer it. Shadowing, intensive listening, and retelling build the road.

Use English to do things. Write code comments. Search for answers. Argue with a chatbot. Leave a review. Take notes. Language is a tool. Tools get smooth from use.

Accept imperfection. Fluent speech is not error-free. Native speakers repeat, restart, pause, and change their minds. The goal is to keep moving with the errors, not to perform a perfect sentence.

Writing and speaking are different call systems

Writing is slow. You can search, edit, restructure, and change tone. A 10,000-word vocabulary can shine in writing. The enemy in writing is not retrieval speed. It is logic, structure, register, rhythm, and revision. You can write a long, complex sentence. That does not mean you can say it in a conversation.

Speaking is fast. It relies on chunks, listening, social scripts, and nerve. The enemy in speaking is freezing, translating, fearing mistakes, and losing the thread.

If you want to improve writing, read good essays. Imitate their structure. Revise your drafts. Collect formal expressions. If you want to improve speaking, shrink those expressions into sayable chunks. Then say them until they arrive without a search.

Next time you are in a meeting, you want to say, “Wire this into the pipeline.” The word does not hide. It arrives before the pause. You say it. The other person nods. You move to the next point.

Vocabulary is inventory. Fluency is the supply chain. The warehouse can be full. The trucks still have to move.