Voice editing lacks the precise caret placement of keyboard editing

What it is

A keyboard and touch can drop an insertion point on a character, even inside one. Voice has no equivalent precise caret: “after the third character” is unstable; “between ‘meeting’ and ‘notes’” must first recognize both anchors and may hit duplicates. Voice editing is therefore good at replacing chunks and bad at parking a caret in a specified gap.

Why it happens

A visuo-motor loop lets a finger or mouse aim at pixel-level position. Spoken reference is symbolic, depending on unique strings or indices in the text. Chinese has no spaces, so “the nth character” is often a different count for the user and for the tokenizer. Homophones and repeated words make anchors ambiguous. Even with perfect recognition, speaking a coordinate (row, column, offset) costs more memory than glancing and tapping. The voice channel therefore lacks a cheap “stop here” primitive. Scope decides which chunk to change; missing caret is the inability to put the insertion point in a seam inside the chunk. Together they force replace at word or sentence granularity.

Studying it

Compare three ways to locate—touch tap, spoken index, spoken anchor phrase—on tasks that insert one character or delete one punctuation mark. Use unspaced Chinese and sentences with repeated words. Dependent measures include whether the final caret sits in the intended seam. Do not extrapolate Chinese counts from English spaced text.

Where it stops holding

With a screen, voice for chunks and a finger for seams is a reasonable split; what is missing is precision in voice-only use, not a ban on mixing. Screen-reader navigation has its own granularity (character/word/line)—another caret that dictation must not pretend to provide. Monospaced code editing is stricter about column, and voice fits even worse.

Applying it

• Default voice edits to word or sentence granularity; do not lure people into “the nth character” for fine inserts.
• When a character-level change is needed, surface touch or keyboard immediately and park the caret on the just-recognized word, shortening the switch.
• If spoken locating is required, use “select this word” plus an on-screen highlight rather than an invisible index.

Related

• Same group: C7.13.1 Spoken correction depends on saying a replacement command rather than editing text directly · C7.13.2 The system must decide whether a correction targets recent text or an arbitrary earlier span · C7.13.4 Complex edits are less efficient to express in speech than by direct manipulation
• Adjacent: C7.17 Switching Between Voice and Keyboard · C6.13 Text Selection and Caret Placement
• Search: caret · voice editing granularity · insertion point