Can you “learn any language fast and never forget it” thanks to the Fluent Forever method? I’ve read Gabriel Wyner’s book and intend to find out. Instead of focusing on grammar, it emphasizes early mastery of pronunciation with deep immersion and intense acquisition of vocabulary with spaced repetition. I can vouch for the importance of pronunciation and spelling: since i’ve learned its phonology, i’ve been able to roughly understand written Dutch, because it sounds like a weird mashup of English and German. Conversely, the insane disconnect between its spelling and its pronunciation makes English a surprisingly hard language to master.

I’ve never used flashcards that much, however, even though i’ve read study after study extolling the virtues of spaced repetition. Learning Portuguese gave me the perfect excuse to reinstall Anki. Wyner offers a list of the first 625 words you should learn, which isn’t bad, but shows a heavy American bias (think cars, religion, and guns). Adding an image to each and every card is tedious, but as Wyner points out, the simple act of searching for the perfect picture is the first step in creating the neural networks that’ll ensure long-term memorization.

Adding audio cues is another story entirely: it’s far easier to find Brazilian Portuguese rather than European Portuguese examples, even on Forvo, the biggest “pronunciation dictionary”. I don’t mind it — my ultimate goal is to learn Galician, the language from north-western Spain that my grandparents spoke, whose pronunciation is closer to Spanish and Brazilian Portuguese than European Portuguese. But i’m living closer to Lisboa than to Brasília, which means i’d prefer my flashcards to contain European Portuguese samples.

A lot of people, Wyner included, swear by HyperTTS, a text-to-speech Anki add-on developed by VocabAI. The free version includes a few basic services, but Google Translate only “speaks” Brazilian Portuguese and the quality of Apple’s text-to-speech engine is mediocre at best. If you’ve already created your deck, VocabAI’s generous trial offer might be enough to generate all the pronunciations using a premium service. I’ve used less than a tenth of the 50,000 characters over 7 days quota to populate my 400+ flashcards.

If you need more, you can either subscribe to get a monthly allocation of characters, or – even better – buy a prepaid allowance of characters that doesn’t expire. HyperTTS includes OpenAI, ElevenLabs, Amazon and other text-to-speech services, but i was blown away by the quality of Microsoft Azure’s voices. If you told me that every word was recorded by a native European Portuguese speaker, i’d believe you. After only two weeks, it’s already done wonders for my ability to understand TV shows and “slow Portuguese” podcasts.

That being said, Wyner’s promise of quick and easy fluency is a sham. You can prepare for your next holidays in a few weeks; you should be able to read graded readers after a few months; but really mastering a language takes years of dedicated work. Fluent Forever can help you get there quicker, but it’s not a miracle method that’ll make you a polyglot in no time. I guess some nuance has been lost in the revised edition of Wyner’s book, which reads like a long advertisement for his company and his (expensive) learning app.

Books

This Way Up: When Maps Go Wrong (and Why It Matters) by Mark Cooper-Jones and Jay Foreman. They’re the men, those are the maps, this is the book. Being a terminally online history and geography major, i can’t say i learned a lot, but i guess this isn’t the point. This Way Up is fascinating in and of itself as the written adaptation of a YouTube channel.

For the most part, each chapter reads like a Map Men video, complete with weird tangents and absurd jokes. Other than the interminable “podcast transcript” about the Donner Party, each and every chapter is incredibly funny… almost to a fault. I made the mistake of binge-reading This Way Up, and it was overwhelming, even if i loved their bit about their dérive in London using a map of Paris. You should read a chapter here and there, not necessarily in order, almost like a little surprise from the YouTube algorithm.

Links

‌How AI assistance impacts the formation of coding skills by Anthropic. Good on Anthropic for not burying the lede: “using AI assistance led to a statistically significant decrease in mastery”. In their latest study, they found that junior Python developers using “AI coding assistance” didn’t work faster, understood far less, and tested far worse. This isn’t deskilling: this is unskilling, the birth of a generation that won’t develop advanced skills because it over-relied on LLMs, that’ll be able to produce code but not craft it. Anthropic might conclude that “the way we interact with AI while trying to be efficient affects how much we learn”, but it hasn’t changed the produce-now understand-never attitude of their products.