Canonical URL
The address a page declares as its real one, so the same content reachable through several links is counted as a single page.
Duplicates appear without anyone creating them: a tracking parameter, a trailing slash, http next to https, the bare domain next to www. Each is a separate address holding identical content, and without a declared original the ranking is split between copies of the same page.
It is a hint, not an instruction. A search engine may pick a different address than the one declared, usually when the signals disagree — the sitemap listing one form, internal links pointing at another, a redirect leading somewhere third. Agreement between those matters more than the tag itself.
A page should name itself. Self-referencing canonicals on every page are the cheapest way to keep parameters from multiplying the catalogue, and on a bilingual site they sit alongside hreflang rather than instead of it: canonical picks the original among identical copies, hreflang pairs versions in different languages.