What a language you already speak is actually worth, in hours

"It helps" is the universal answer to whether knowing one language speeds up another. It is true and it is useless, because it is the same answer for Spanish→Portuguese and for Spanish→Japanese.

One of those is worth about a third of the project. The other is worth nothing.

The numbers

Our model applies a discount based on how the two languages are related. Portuguese to B2 with a tutor:

138 hours saved — about 30%. Spanish and Portuguese are both Romance languages, sharing vocabulary and grammatical machinery at close range.

Now Japanese to B2, from English plus Spanish: 1,939 hours either way. The discount is zero, and the tool says so in those words rather than finding something encouraging to report.

Why zero is the important output

A tool that always finds a benefit is not measuring anything.

The temptation is obvious: someone types in a language they worked hard at, and returning "no measurable head start" feels ungenerous. But the alternative is worse — a small invented discount on every pair, which would make the genuinely large ones look ordinary.

The 30% for Spanish→Portuguese only means something because Spanish→Japanese returns nothing.

The bit in between, which is where people get misled

The interesting cases are the small ones.

Knowing Russian makes German about 8% faster in our model. That is a real effect with a real mechanism — familiarity with a case system, with adjectives agreeing across gender and number, with grammar that lives in endings rather than word order. A Russian speaker meeting German cases has met the concept before.

And 8% should change nothing about your decision. It is worth roughly a month on a two-year project, which is smaller than the difference made by studying five days a week instead of four.

This distinction — real versus decision-relevant — is why the tool reports hours rather than a percentage. "8% faster" sounds like a reason. "About 40 hours off a 500-hour project" sounds like what it is.

It is also why our scenario pages only cross-link when the factor clears 10%. Below that, "you also speak Russian, so read the German page" is a true statement that wastes a click.

What the discount is built from

Four relationships, largest to smallest:

• Same branch (Spanish→Portuguese, Dutch→German) — the big one
• Same family, different branch (Russian→German, both Indo-European) — the 8% case
• Shared script with an unrelated language — small but real
• Typological similarity between unrelated languages — Japanese and Korean are unrelated by descent and structurally alike enough that the model credits it substantially

These coefficients are estimates, and marked as such. Transfer effects in third-language acquisition are well attested in the literature; no published coefficient exists to plug in. The values are set conservatively and the total is capped, so stacking several known languages cannot produce an implausible result.

That cap matters. Without it, someone entering five Romance languages would get a discount that says Italian is nearly free, which is not true and would be the model's arithmetic rather than a finding.

Using it

Run your own pair. If it says zero, that is the answer — what transfers between unrelated languages is the skill of having learned one, which is real and does not convert into hours.

How the factors were set, including which are published and which are ours.