I live in China, work in tech, and operate in Mandarin and English daily. When I decided to attack real fluency in English — not "pass the TOEFL", but think, dream, and operate in English without friction — I needed a system. Not a course, not an app. A system.
This is the playbook I'm running. Based on Daniel Everett's framework, adapted for the specific interference patterns of Brazilian Portuguese, with weekly metrics and concrete tools.
The Everett Framework: Language as a Cultural Tool
Daniel Everett spent 30 years with the Pirahã in the Amazon. His book Language: The Cultural Tool (2012) destroyed the Chomskyan paradigm of "innate universal grammar." His thesis:
Language is not an instinct — it's a tool shaped by culture.
Practical implications for learners:
- Isolated grammar doesn't generate fluency. Cultural context does.
- Translation is an anti-pattern. It creates a layer of indirection that never goes away.
- Immersion isn't "nice to have" — it's the only mechanism that replicates how humans actually acquire language.
Everett learned Pirahã without a teacher, without a dictionary, without a reference grammar. Just exposure + attempt + feedback. He calls this the monolingual fieldwork method. I adapted his three core principles:
Principle 1: Immersion Islands
It's not about living abroad. It's about creating blocks of time where only English exists. Everett lived in cycles of total immersion followed by processing. The brain needs both.
Principle 2: Zero Translation (Monolingual Bootstrapping)
Everett pointed at objects and learned the word directly. No intermediary. When you translate, your processing is: thought → PT-BR → EN. When you stop translating: thought → EN. The latency difference is what separates "speaks English" from "is fluent."
Principle 3: Cultural Chunks
Natives don't speak in words — they speak in chunks (pre-fabricated pieces). "I'd be down for that", "let me circle back", "that tracks". Everett showed that language is fundamentally about cultural patterns, not grammatical rules. Chunks are the real unit of fluency.
Brazilian Portuguese Interference: The Specific Bugs
Before the plan, you need to understand exactly what PT-BR does to your English. It's not generic "accent" — these are mappable interference patterns:
Phonology — The 7 Main Problems
| # | PT-BR Interference | Error Example | Correct Sound (IPA) | Fix |
|---|---|---|---|---|
| 1 | Vowel epenthesis — adding a vowel after final consonants | "big" → "bigi", "stop" → "stopi" | Dry final consonant, no vowel | Train final consonants with hand on throat (no vibration after consonant) |
| 2 | θ/ð non-existent — "th" becomes "t/d" or "f/v" | "think" → "fink", "this" → "dis" | /θ/ (voiceless), /ð/ (voiced) | Tongue between teeth, blow. Minimal pairs: thin/fin, then/den |
| 3 | Palatalization of /t/ and /d/ — "tchi" and "dji" before /i/ | "team" → "tchim", "day" → "djay" | /tiːm/, /deɪ/ | Tongue tip on alveolar ridge, no palatalization. Drill: tip, teen, tea, dim, deep |
| 4 | Syllable-timed vs. stress-timed rhythm — PT-BR gives equal weight to each syllable | "comfortable" with 4 clear syllables | /ˈkʌmf.tə.bəl/ — 3 syllables, schwa dominant | Practice schwa /ə/ obsessively. It's the most common sound in English |
| 5 | Nasalization — nasal vowels where they don't exist | "can" with strong nasalization | /kæn/ — oral vowel | Awareness: hand on nose, feel vibration. Reduce |
| 6 | Brazilian /r/ — tap or uvular fricative | "red" with guttural R | /ɹɛd/ — alveolar approximant | Tongue curled back, touches nothing. Drill: red, run, right, around |
| 7 | Missing reduced vowels — pronouncing every vowel "full" | "banana" with 3 clear "a" sounds | /bəˈnæ.nə/ — only the middle "a" is full | Learn that English systematically reduces unstressed vowels to schwa /ə/ |
Prosody — The Invisible Problem
Even with perfect phonemes, Brazilians sound "foreign" because of rhythm. PT-BR is syllable-timed (each syllable lasts ~equally). English is stress-timed (stressed syllables are long, unstressed ones are compressed).
Exercise: say "I need to GO to the STORE to get some MILK". The capitalized words are long. All the others are fast, almost swallowed. "I-need-to" sounds almost like "I-nee-duh". That's stress-timing.
The 90-Day Plan: Week by Week
Phase 1: Foundation (Weeks 1-4) — Massive Input + Phonological Awareness
Goal: 3h+ of English input/day. Map your specific pronunciation errors.
Daily routine (Phase 1):
| Block | Activity | Time | Tool |
|---|---|---|---|
| Morning | Shadowing with transcript | 20 min | YouTube + script |
| Morning | Anki — pronunciation deck | 15 min | Anki (deck below) |
| Day | Passive podcast (commute, gym) | 60-90 min | Podcasts (list below) |
| Day | All media consumption in EN | ongoing | Netflix/YouTube in EN, EN subtitles |
| Evening | Journaling in English | 15 min | Notion/paper |
| Evening | Record 2 min speaking, compare | 10 min | Phone recorder |
Week 1-2: Diagnosis
- Record 5 min talking about any subject
- Compare with a native speaker talking about the same topic
- Identify your top 3 errors from the table above
- Set up Anki with minimal pairs for your specific errors
Week 3-4: Focused attack
- Daily shadowing focused on your errors (#1 priority)
- Minimal pairs: 50 pairs/day on Anki
- Goal: consistently produce /θ/, /ð/ and schwa /ə/
Week 4 milestone: Record the same 5 min from week 1. The difference should be audible in the pronunciation of your top 3 errors. If not, double the shadowing time.
Phase 2: Production (Weeks 5-8) — Forced Output + Chunks
Goal: Speak 30+ min/day in English. Accumulate 200+ active chunks.
Daily routine (Phase 2):
| Block | Activity | Time | Tool |
|---|---|---|---|
| Morning | Shadowing (prosody/rhythm now) | 20 min | YouTube |
| Morning | Anki — chunks + pronunciation | 20 min | Anki |
| Day | Real conversation | 30 min | Cambly/iTalki/colleagues |
| Day | Active podcast (pause, repeat) | 30 min | Podcasts |
| Evening | Journaling + voice note | 20 min | Notion + recorder |
Week 5-6: Chunk Acquisition
- Write down 5 new chunks/day from podcasts and shows
- Anki format: front = context/situation, back = natural chunk
- Chunk examples by category:
Opinion: "here's the thing", "the way I see it", "that tracks", "I'd push back on that"
Agreement: "I'm on board", "that makes sense", "count me in", "I'm down"
Transition: "on that note", "speaking of which", "that said", "to your point"
Reaction: "no way", "that's wild", "I didn't see that coming", "fair enough"
Week 7-8: Conversational fluency
- 30-min conversations focused on using new chunks
- Ask for specific feedback from tutor: "flag my pronunciation errors"
- Practice stress-timing in long sentences
Week 8 milestone: Maintain a 30-min conversation on any topic without long pauses (>3s) to search for words. iTalki tutor should notice pronunciation improvement.
Phase 3: Refinement (Weeks 9-12) — Accent Polishing + Automatization
Goal: Reduce accent detection. Automate chunks. Operate in English without conscious effort.
Daily routine (Phase 3):
| Block | Activity | Time | Tool |
|---|---|---|---|
| Morning | Advanced shadowing (TED talks, debates) | 20 min | YouTube |
| Morning | Anki — maintenance | 10 min | Anki |
| Day | Operate in English (work, social) | max possible | Slack, calls, Twitter |
| Day | Podcast/audiobook (1.2x speed) | 60 min | Podcasts |
| Evening | 5-min recorded monologue | 10 min | Recorder |
Week 9-10: Accent Detail Work
- Focus on connected speech: linking ("an apple" → "anapple"), elision ("next day" → "nex day"), assimilation ("don't you" → "donchu")
- Shadowing focused on intonation and pitch patterns, not just words
- Practice intonation patterns: rising for questions, falling for statements, rise-fall for lists
Week 11-12: Automatization
- Reduce formal study, increase real usage
- Thinking in English should be the default (if you catch yourself thinking in PT-BR, switch)
- 100% consumption in English (including social media, Google searches)
Week 12 milestone: Record 5 min again. Compare with week 1 and week 4. Run the test: send the audio to 3 people who don't know you and ask "where do you think I'm from?" If they answer a specific country instead of "somewhere in South America", the accent still needs work. If they hesitate or get it wrong, it's working.
Tool Stack
Anki Decks (install TODAY)
- English Pronunciation Minimal Pairs — search on AnkiWeb, get one with audio
- Create a custom deck: front = sentence with problem word, back = native audio + IPA
- Settings: new cards 20/day, graduating interval 3 days, easy interval 7 days
Shadowing — How to Do It Right
1. Choose a 2-3 min video with subtitles (TED talks are ideal)
2. Listen once without pausing
3. Listen sentence by sentence, repeat imitating EVERYTHING: rhythm, pitch, pauses, emotion
4. Record yourself doing the same passage
5. Compare. Repeat the parts that diverge.
6. Do the entire passage together with the speaker (simultaneous)
It's not karaoke. It's muscular imitation. You're training the 100+ muscles of articulation.
Podcasts (by difficulty level)
- Intermediate: 6 Minute English (BBC), All Ears English, Espresso English
- Advanced: Lex Fridman, Huberman Lab, Acquired, My First Million
- Accent-specific: Rachel's English (YouTube), Pronunciation with Emma
YouTube Channels for Pronunciation
- Rachel's English — best channel for American English pronunciation, tongue position videos with intraoral camera
- Sounds American — minimal pairs and drills
- English with Lucy — British English, good for comparison
- Pronunciation with Emma — short practical exercises
Conversation Practice
- Cambly — native tutors on-demand, good for volume
- iTalki — professional tutors, better for structured pronunciation feedback
- Ask: "I want you to interrupt me every time my pronunciation is off. Be brutal."
Recording/Feedback Tools
- ELSA Speak — AI app that detects pronunciation errors phoneme by phoneme
- Forvo — pronunciation of any word by natives from different regions
- YouGlish — search any word and see YouTube videos with it in context
Metrics: How to Know If It's Working
Don't trust feelings. Measure.
| Metric | How to measure | Week 4 | Week 8 | Week 12 |
|---|---|---|---|---|
| Hours of input | Timer/app | 80h+ accumulated | 180h+ | 300h+ |
| Active chunks | Count in Anki (mature cards) | 50+ | 200+ | 400+ |
| Pronunciation errors | Record 5min, count errors | 50% reduction in top 3 | Top 3 resolved | <5 detectable errors |
| Response latency | Time until you start speaking in conversation | <3s | <2s | <1s |
| Accent test | Ask "where am I from?" | — | — | Can't identify easily |
| Thinking in EN | Self-report: % of day thinking in EN | 20%+ | 50%+ | 80%+ |
The Principle That Changes Everything
Everett's most counter-intuitive insight: language is social behavior, not academic knowledge. You don't "learn" English the way you learn math. You acquire English the way you acquired Portuguese — by living in it.
The entire plan above is reverse engineering of this principle. Create the environment. Force the output. Measure progress. Iterate.
There are no shortcuts. There is a system.
Phelipe Xavier — Brazilian in China, running this playbook in parallel with Mandarin and building startups. Current week: updated results in the next post.