A retro-definition of the term “word” In this talk, I propose a definition of the term word that can be applied to all languages using the same criteria. Roughly, a word is defined as a free morph or a clitic or a root plus affixes or a compound plus affixes. The paper relies on earlier definitions of the terms free, morph, affix, clitic, root, and compound, which are summarized here. I briefly compare the proposed definition with Bloomfield’s, and I say how word-forms differ from lexemes. This is a “retro-definition”, i.e. a definition of a term that is widely used but not widely defined. Many linguists seem to think that no definition of such a basic is needed or possible, but I will explain why I regard such a definition as very helpful. The definition is a “shared-core definition” (in the sense of Haspelmath 2021: §5), which means that it captures the core of what we think of as words in all languages, but it does not aim to be a substitute for language-particular categories that may be relevant in specific languages. Finally, I will explain why I think that an unnatural-seeming definition is better than a prototype definition or other options.