Typing in a Second Language

The dominant cost in second-language typing is finding the word, and it happens before any finger moves.

In a first language, the next word is available essentially instantly and often before you have finished the previous one — which is what allows a fluent typist's reading to run ahead of their hands. In a second language that pipeline is shorter and sometimes empty, so the hands wait.

The symptom is diagnostic: second-language typing produces pauses rather than errors. A typist who is slow because of retrieval types accurately in bursts separated by gaps, while a typist who is slow because of motor skill produces a steady, error-prone stream. If your second-language typing is accurate but halting, no amount of finger practice will address it — the constraint is vocabulary access.

Which is worth knowing because it saves effort spent in the wrong place. That distinction between pause-limited and error-limited typing also appears in the general case, in why typing speed plateaus.

This part is genuinely about the hands, and it is subtler than the retrieval problem.

Fluent typing is built on chunks — common letter pairs and word endings executed as single motions rather than as sequences of decisions. Those chunks are specific to a language, because letter frequencies are.

English gives you th, he, in, er and ing. German gives you ch, sch, ei and the compound-word habit of very long tokens. French gives you high frequencies of e with accents, plus qu and ou. Spanish gives you a much higher vowel ratio and different terminal patterns.

So a fluent English typist moving to German is not merely slower at German words; their automated chunks are the wrong chunks, and the sequences German actually produces are ones they have never drilled. Rebuilding that is real work and it is exactly the kind of work drilling addresses — the site's bigram and word-ending drills are built on English frequencies, which is a limitation worth stating rather than glossing.

Here the difficulty stops being cognitive and becomes purely mechanical, and every available solution is a compromise.

A dedicated national layout puts the accented characters on real keys, which is fast and correct — and it moves the other characters around, so anything you learned on a different layout partly breaks.

Dead keys put an accent on one key that combines with the next letter you press. This is compact and it introduces a genuinely awkward motor pattern: a keystroke that produces nothing, followed by one that produces two things.

A compose key or an international layout with AltGr gives you access to a wide range of characters through a modifier, at the cost of a three-key combination for characters that a national layout gives you in one.

And switching layouts per language means maintaining two positional maps and a switching action, with the associated errors when you type in the wrong one — which everyone who does this has done, usually into a password field.

The alternative-layout world assumes you type one language. Every optimised arrangement — Dvorak, Colemak, Workman and the rest — is built on a specific frequency table, and optimising for one language necessarily de-optimises for another.

There are national alternatives built the same way: Bépo for French, Neo for German. They are not translations of an English layout but separate designs from the same method, and each is optimised for its own language in the same way and with the same trade.

Which means a genuinely bilingual typist has three unappealing options: optimise for one language and accept the other is worse, use a compromise layout that is optimal for neither, or maintain two layouts and pay the switching cost forever.

That is a real and underdiscussed limitation of the whole optimised-layout project, and it is worth weighing before starting a transition, alongside the costs already set out in the real cost of switching layouts.

Software assistance behaves differently across languages, and mostly worse.

Autocorrect keyed to the wrong language is actively destructive: it will confidently replace correct second-language words with first-language ones, and it does so silently. Anyone who has typed a message in one language on a phone configured for another knows this failure mode well.

Multilingual correction has improved and is still weaker than single-language correction, because the model has to decide which language you are in from limited evidence, and short messages provide very little.

More importantly, the feedback loss described in autocorrect changed what accuracy means is worse here. A second-language typist has more genuine uncertainty about spelling, so the correction is doing more work — and therefore concealing more. Their sense of their own second-language spelling is even less calibrated than their sense of their first.

Diagnose first, because the two constraints call for opposite responses.

If your second-language typing is accurate but halting, the constraint is retrieval and typing practice will not touch it. Reading and writing in that language will; drilling will not. It is worth being clear about this because the effort is otherwise wasted.

If it is fluent but error-prone, the constraint is motor, and it responds to exactly the usual treatment — with one adjustment. Drill on text in that language rather than in English, because the chunks you need are the ones that language produces. General practice on English text builds English chunks, which is precisely what you already have.

And for the accented characters specifically, treat them as their own drill target. They are rare enough in ordinary practice that they never automate on their own, in exactly the way P, Q and Z do not automate in English — the mechanism is identical, and so is the remedy.