[brownbat]

There's something called the input hypothesis that's trending upwards in language learning. More input is basically a substitute for studying grammar, or even superior, because the brain is phenomenal at pattern matching. Bad grammar will sound "off" well before you understand the rule in your own language, same thing happens when learning new languages, given sufficient input.

It takes a leap of faith but learning grammar almost exclusively through example sentences is actually really efficient.

Krashen, Khatzumoto, Matt vs. Japan, and Antimoon are all worth reading if you want to go down this rabbit hole.

How you absorb the grammar of a language is a separate question from when you should do conversations though.

Some folks advocate conversation day one, but most pro-input folks do recommend waiting until you have a sense of how things work and can make fewer mistakes.

EDIT: One of the best articles explaining this anti-rule view: http://www.antimoon.com/how/input-gramrules.htm

[fenomas]

Thanks for this explanation. I'd not heard of the debate before, but now that you describe it I find I've been a fan of input learning for a long time.

With that said I still find GP's complaint pretty weird, since TFA strikes me as being in the "beginning stuff that even advocates of input learning would teach out of a book" category.

I mean, if I was teaching somebody Japanese then I'd probably cover how particles work on the first day. Surely only a madman would expect someone to absorb that organically.

[visarga]

Particles are the hardest parts of the language, often untranslatable by themselves. I have a 100 page booklet about these words which are so few they can sit on one or two rows.

[nullsense]

Can confirm. I studied grammar through Anki deck of about 2000 sentences and a tonne of native content. It took my about a year to truly internalize に to the point it became second nature and I could understand it intuitively.

Passed JLPT1 after a couple of years, and had the time of my life learning Japanese. Particles are the glue that holds it all together but they are deceptively simple on the surface with an ocean of complexity lying underneath.

[klipt]

> input hypothesis

Definitely seems to be the case in machine learning. Flexible models trained on huge amounts of data beat older hand tuned grammar rule models every time.