Difficulty is not one quantity, it is three, and they differ in when you pay. A writing system is front-loaded: a fortnight of unpleasantness and then it is over. Tone, case and agreement are distributed: a small tax on every sentence you build, for a year or more. And a formality system is back-loaded: it does not appear at all until you are good enough for it to matter, which is where it ambushes people.
Choosing a language by the cost you can see at the start therefore optimises for the one that matters least, and it is what almost everybody does, because a script is the only one of the three that is visible before you begin.
Eight facts about what makes a language expensive
- difficulty is distance, not complexity. No language is objectively harder; they are further from or closer to what you already have.
- Arabic, Chinese, Japanese and Korean are the standard hardest group for an English speaker, and they are hard for different reasons.
- the script is the smaller half. Learning the writing system feels like the obstacle and is usually a few months of a multi year project.
- diglossia doubles the work. Arabic requires a written standard and a spoken dialect, which are different enough to be two learning projects.
- tones are a listening problem first. Producing them is learnable; hearing them reliably in fast speech is what takes the time.
- Japanese has three scripts and a politeness system, and the second of those is the part that does not appear on any difficulty list.
- hard languages need more contact hours, roughly four times what a close language needs to reach the same level.
- the wall is at intermediate, not at the start. The early stages are exciting and structured; the middle is where the distance is actually paid.
A script is two weeks, and everybody overestimates it
An unfamiliar alphabet or syllabary looks like the reason a language is hard, and it is the cheapest difficulty in this whole subject. A script is finite, drillable, and has one right answer per symbol, and once it is done it never troubles you again.
The genuine exceptions are the systems that are not alphabets. A writing system where thousands of characters each have to be learned individually is not a front-loaded cost at all; it is a distributed one wearing a script's clothes, and it is why one category of language is expensive in a way the others are not.
The distributed costs are the ones that decide it
Tone, grammatical case, gender and verb agreement all share a shape: they are small individually and they appear in every sentence, so there is no point at which you have finished paying. A learner meets them on day one and is still paying attention to them in year two.
They also have a second property that makes them worse than their size suggests. A distributed cost consumes attention while you are trying to say something else, so it slows everything down rather than blocking anything, which is a much harder thing to notice and a much harder thing to practise away.
Where it stops: the cost that had not arrived yet
The back-loaded difficulty is the one nobody plans for, because nothing about it is visible while you are a beginner. A language with a developed formality or honorific system asks you to pick a register for every sentence based on who you are talking to, and beginner material sidesteps it by teaching one polite level and stopping there.
Then you get good enough to have a real conversation with somebody older, or more senior, or younger, and discover a whole system that was invisible for two years. The same shape appears where the written language and the spoken one have separated far enough to be two things, so a learner who worked from books arrives able to read a newspaper and unable to follow a family at dinner.
The stall is that this arrives at the point where the learner believed the hard part was behind them, which makes it feel like a betrayal rather than a stage.
Looking up the features before you commit
The World Atlas of Language Structures is free online and lets you look up, feature by feature, whether a language has tone, how many cases it uses, whether it marks gender, and where its verb goes. You can put a candidate language beside one you already speak and see which features are new rather than merely different.
It replaced a general reputation for difficulty, which circulates without ever saying where the cost falls and is usually a memory of somebody's first fortnight.
Its limitation is that it is a typological database written for linguists. It will tell you a language has eight cases and it has nothing to say about how long eight cases take to acquire, so it tells you the shape of the bill and never the size.
What it looks like when it goes wrong
The recognisable failure is a learner who did the script in a fortnight, concluded the language was overrated, and stalled six months later on something they never saw coming.
The second is avoiding a language entirely because of its writing system, which is avoiding the one part that would have been finished before the end of the first month.
Why the published totals cannot tell you any of this, and why the same number of hours produces very different experiences, is difficulty, usefulness and the honest reasons people pick, and the other axis that decides how far you get, which is how much material exists at all, is which languages an English speaker gets furthest in fastest.
The open question is what happens when a distributed cost from one language starts appearing in another, which is the specific risk of running two of them at the same time: when it works, and the pairs that interfere.