Source
Roman Jakobson (1959), “On Linguistic Aspects of Translation”, LIT-786, pp. 235–236, as read in NOTE-614. Jakobson credits the core observation to Boas (“Language”, 1938) and the chain image to Karcevski.
What was actually shown
Nothing was tested; the support is argument from worked examples.
- Expressibility. “All cognitive experience and its classification is conveyable in any existing language”: deficiencies are made good by loanwords, loan-translations, neologisms, semantic shifts and circumlocution, and a grammatical category the target lacks is rendered lexically (a dual as “two brothers”).
- Obligation. “Languages differ essentially in what they must convey and not in what they may convey.” Each verb “imperatively raises a set of specific yes-or-no questions”. Rendering “She has brothers” in a language with a dual forces a choice between two and more than two, or an explicit disjunction. Rendering “I hired a worker” in Russian forces a choice of perfective or imperfective aspect and of a masculine or feminine noun, while the Russian does not carry the English distinctions of “hired” against “have hired” or “a” against “the”.
- Chains. Because the two grammars pose different sets of two-choice situations, “a chain of translations of one and the same isolated sentence from English into Russian and vice versa could entirely deprive such a message of its initial content”. “The richer the context of a message, the smaller the loss of information.”
The thesis could come out wrong in two ways: if translators reliably leave obligatory target categories unmarked or recoverable from the source, or if round-trip chains of isolated sentences preserve the source’s distinctions as well as single steps do. The essay checks neither.
What this does not say
- Not linguistic relativity. Jakobson rejects Whorf’s inference from unlike grammars to unlike facts: cognitive content is translatable, and only what is obligatory differs. The thesis does not say speakers of different languages think differently, and the essay’s only evidence about thought (a 1915 test of personified weekdays) is cited without data and concerns “verbal mythology”, not cognition.
- Not that translation is lossy overall. The forced additions and omissions concern grammatically obligatory information; Jakobson holds that the message’s cognitive content can always be rendered, if need be lexically and at length. The loss is a default of fluent, grammatical rendering, not an impossibility.
- Not about pragmatic force. The examples are all of referential content (number, aspect, sex, tense, definiteness). The essay says nothing about stance, footing, register or illocution, so it gives no support to claims that those are or are not preserved.
- Not a measured rate. The chain claim is a “could”, stated through an analogy (currency exchanges at a loss), and the damping by context is asserted. Nothing fixes how fast a chain loses content, or when.
- Not for poetry. In poetry, jest, magic and “everyday verbal mythology” Jakobson holds that grammatical categories carry meaning themselves, and for poetry he says translation gives way to “creative transposition”. The thesis is stated for the cognitive function of language.
Later accounts of equivalence held in the record, Nida’s (LIT-791) and House’s (LIT-789), are unread here and are not evidence for or against this.
Note of 2026-10-29: what holds it back
No study read here measures, on the same items, that renderings between languages differing in an obligatory category mark it where the source left it open and drop distinctions the target does not oblige, nor that round-trip chains lose such distinctions faster than single steps. The essay argues from examples and tests neither.