Source
Ferdinand de Saussure (1916), Cours de linguistique générale, ed. Bally and Sechehaye, LIT-774; read in NOTE-594 from the third-edition text, Part Two, chapter IV (“La valeur linguistique”).
What was actually shown
An argument with examples, not an experiment. Every value, in and out of language, needs two things: something dissimilar it can be exchanged for (a coin for bread; a word for an idea) and similar things it can be compared with (other coins; other words). A word’s signification is the first; its value requires the second as well. The examples:
- French mouton may share a signification with English sheep but not its value, because English also has mutton for the meat.
- Within one language, redouter, craindre and avoir peur limit one another; “si redouter n’existait pas, tout son contenu irait à ses concurrents”.
- The Sanskrit plural is not worth the French plural, because Sanskrit also has a dual.
- French louer covers what German splits into mieten and vermieten; tense and aspect systems differ in the same way.
The same holds on the sound side: what matters in a word is the phonic differences that keep it apart from others, so phonemes are “oppositive, relative and negative”, which the book supports by free variation where no contrast is at stake (French uvular or rolled r) against its absence where one is (German r and ch; Russian t and t’).
What could have come out the other way: if terms carried their values independently of their neighbours, adding or removing a neighbour would leave a term’s range of use unchanged. The book offers cases in which it does not, and no test in which it might have.
What this does not say
- Not that a sign is nothing but differences. The book restricts “il n’y a que des différences sans termes positifs” to the signified and the signifier taken separately, and calls their combination in a sign “un fait positif”; between whole signs there is opposition, not difference. The common summary “meaning is purely differential” drops that qualification.
- Not that signification is unreal. The book keeps signification as one aspect of value, and says only that it does not fix the term.
- Not a claim about reference or truth. The account concerns the relations inside a language state; it says nothing about how terms relate to things, which the book sets aside with the nomenclature view.
- Not established beyond its examples. The examples are chosen; the book does not consider cases of semantic stability across changes in a field, and the text is the editors’ reconstruction from lecture notes.
- Not a statement about distributional models. That co-occurrence statistics recover word meaning is a later and different claim; this account may be read as its ancestor, but nothing here tests it.
Note of 2026-10-28: what holds it back
Kept Proposed by the sixth status review. Part Two ch. IV is read in full (NOTE-594), so reading is not the blocker. No study the record holds adds, loses or merges a term in a language’s inventory and measures how its neighbours’ use shifts, against terms outside the field. The embedding readings that cite this account (THEORY-182, THEORY-185, LIT-860) concern what a model learns from co-occurrence, and THEORY-518 concerns distributional neighbourhoods from four hand-picked words. ARG-033 and ARG-034 judge the record’s use of this reading, not the reading.
Note of 2026-10-30: what holds it back
Its promote_when is unmet: no study in the record changes a language’s inventory and measures the neighbouring terms’ use against a control. A split of what Saussure proposes, on THEORY-1110’s pattern, would be Active at once, since Part Two ch. IV is read in full (NOTE-594), but it frees nothing yet. CLAIM-082 asserts the account, and needs this entry promoted, not split. CLAIM-242 needs only the proposal: value as a concept, and that values do not cross between languages. The split is due when CLAIM-242 is otherwise promotable. CLAIM-242 then grounds on the split and drops this entry from its grounds.