Autocapitalization assumes English sentence boundaries and misfires in other settings

What it is

Mobile systems often force the next stroke to uppercase after a period, question mark, exclamation mark, or newline. That is autocapitalization. The rule assumes those marks equal an English sentence boundary and that what follows is a new subject. Chat ellipses, German nouns, Chinese (which has no case), and file.txt followed by a property do not satisfy that grammar. The effect is not substituting another word; it is flipping the case bit. Autocorrect changes word form; autocapitalization changes the letter box.

Why it happens

Detectors are mostly shallow: is the last inserted character in a terminal-punctuation set, sometimes with a check for abbreviations (Mr. should not fire). English abbreviations, decimal points, file extensions, and dots in URLs all look like periods. German capitalizes every noun, so a sentence-initial rule is both insufficient and liable to misfire on particles. A Chinese IME that autocaps Roman names turns iPad into Ipad, or rewrites lowercase pinyin drills. The Shift state can be popped up by the system while the user already aimed at lowercase. Unlike dictionary substitution, there is usually no candidate list—only a case bit—so the change is harder to notice.

Studying it

Collect real period contexts that are and are not sentence starts (abbreviations, decimals, filenames, chat ellipses), type them on a system keyboard with default autocaps, and score false uppers. Stratify language: English, German, mixed Chinese–English. Count “user later tapped back to lowercase” as failure; scoring only submitted text hides silent repair. Laboratory materials made of well-formed English declaratives badly underestimate.

Where it stops holding

In formal English prose, sentence-initial caps hit often enough that the rule is nearly adequate. All-caps forms (passport MRZ) should not use autocaps; they need forced caps, a different rule. Password and OTP fields are usually already off; usernames, invite codes, and coupon codes are often forgotten. Speech-to-text has its own sentence-initial logic and should not share a single switch with keyboard autocaps.

Applying it

• Turn autocaps off on fields whose semantics are code, username, email, filename, or coupon; do not inherit the “multiline text” default.
• For mixed Chinese–English and chat products, make period triggers more conservative, or capitalize only on an explicit new paragraph.
• Keep Shift state visible when autocaps fires so the next stroke is visibly going to be upper.
• Verify with real samples: abbreviations, v1.2, iPhone, Chinese with embedded English, on the target field, listing characters wrongly raised. Identifier fields must show none of these.

Related

• Same group: C6.24.2 Auto-formatting changes already-typed content without the user noticing · C6.24.3 Auto-formatting often backfires on proper names and code and needs an off switch · C6.24.4 Automatic changes must be visible and undoable in one step, or people will wonder whether they mistyped
• Adjacent: C6.11 Autocorrect · C6.25 IME state and switching
• Search: autocapitalization · sentence boundary · shift state