Unix’s spellchecker removed "id" because it kept sheltering a mistyped "is"

A real word lost its place in the Unix spellchecker’s dictionary for being too helpful to typos. “Id” kept appearing when people meant “is,” so the spelling list dropped it. Freud might have had something to say about that. The document’s author probably just wanted the right letter.

Douglas McIlroy explains the choice in Development of a Spelling List, published in 1982. His account describes a practical spelling checker built around real documents and the mistakes they produced. A dictionary entry could be linguistically valid and still make the tool less useful for that collection of writing.

On a QWERTY keyboard, the letters S and D sit beside each other. A finger arriving one key over can turn an ordinary “is” into an equally dictionary-shaped “id.” A lookup-only checker sees an accepted entry and moves along. The typo has acquired perfectly respectable paperwork, which is all the program asked to see.

McIlroy writes that the word “never appeared in its own right” in the relevant material. That observation gives the removal its rationale. The tradeoff is specific to the documents being checked; a psychology text would deserve a different conversation. Removing a legitimate term from every dictionary everywhere would be a much larger claim than his example supports.

The paper also discusses the opposite nuisance: good words flagged as suspicious because the list did not cover them. That puts the tiny deletion inside a larger balancing problem. Accepting too little interrupts writers unnecessarily. Accepting too much gives mistakes a place to hide. Maximizing the number of entries does not settle either question.

A spellchecker’s output is useful only when the writer can do something with it. Every harmless flag asks for attention, and every accepted typo avoids it. That makes the writer’s actual vocabulary part of the design, with the mundane documents doing more useful work than a promotional claim about an impressively large dictionary.

The example is especially clear because the mechanism fits under one fingertip. There is no need to guess whether the program understands the sentence. In this case, a membership check accepted the wrong input because the accepted set contained it. Adjusting that set changed what the program would question next time.

For anyone maintaining a custom dictionary today, the practical comparison is an entry added once to silence an annoying warning. If that entry is also a common mistake, the convenience may outlive the reason for adding it. Reviewing the actual occurrences would tell you whether the tool is helping a specialist vocabulary or simply offering cover to an adjacent key.

McIlroy’s choice leaves a genuine limit in place: a word-based checker can still accept the wrong valid word in a sentence. “Id” was a particularly tractable case in that workload. The rest of English had not agreed to become equally cooperative.

The id had to go.