EPSO Translator cheat sheet

Translate meaning, not words: render the source's function and register in the target language; a literal calque that is grammatically correct but pragmatically odd is an error.

EU register conventions: institutional texts use consistent terminology (e.g. 'acquis communautaire', 'subsidiarity'); check the target-language style guide for fixed equivalents before inventing a rendering.

False friends and cognates: verify target-language sense independently (e.g. EN 'actual' ≠ FR 'actuel' = current; EN 'eventually' ≠ FR 'éventuellement' = possibly).

Sentence splitting and merging are legitimate translation strategies when target syntax (e.g. German verb-final subordinate clauses) cannot carry the source's clause structure.

Keep proper nouns, titles of legislation, and case names in the form prescribed by the target-language convention; do not translate names of institutions unless an established exonym exists.

Cohesion devices must be re-created: where the source uses pronouns or ellipsis, the target may need lexical repetition or a different anaphor to stay unambiguous.

Numbers, dates, and units: convert formats to target conventions (1,000.5 vs 1.000,5; 3/12/2024 vs 12/03/2024) and keep the value identical.

Punctuation and typography follow the target language (e.g. French spacing before ; : ! ?, German capitalisation of nouns, English serial comma policy of the style guide).

Self-revision test: back-translate a sample sentence mentally; if the meaning shifts or an ambiguity appears, the original rendering is not yet adequate.

Identify the text type and its function (legal, technical, journalistic, promotional) before translating; this determines terminology density and register choices.

Parse clause boundaries explicitly: locate finite verbs first, then subjects, then subordinate markers (that, which, weil, obwohl, aunque) to avoid mis-attaching modifiers.

Watch for scope ambiguities: quantifiers (all, some, only) and negation (not, no, non, kein) attach to specific constituents; misreading scope changes the proposition.

Distinguish given vs new information: theme-rheme structure signals what the sentence is really asserting, which guides emphasis in the target.

Lexical ambiguity resolution: use collocations and domain context, not the first dictionary sense (e.g. 'party' in a contract vs a news report).

Co-reference tracking: resolve pronouns and ellipsis to their antecedents before translating, especially in languages that drop subject pronouns.

Idioms and metaphors: identify fixed expressions as units; their meaning is non-compositional and must be matched by a target-language equivalent, not word-for-word.

Register markers: modality (shall, may, must, sollte, devra) and hedging carry legal or diplomatic weight; note them explicitly before rendering.

True/False/Cannot say: 'Cannot say' applies only when the passage neither states nor logically entails the claim; outside knowledge is never admissible.

Distinguish entailment from plausible inference: if the statement requires an assumption not in the text, the answer is 'Cannot say', not 'True'.

Scope words (all, some, most, only, always, never) are the most common trap; compare the quantifier in the statement with the quantifier in the passage.

Negation and double negation: rewrite the statement in positive form before comparing it to the passage to avoid polarity errors.

Temporal and causal claims: the passage may state sequence without causation; a statement asserting cause when only correlation is given is 'Cannot say' or 'False'.

Extreme wording in an option (only, entirely, impossible) is usually unsupported unless the passage uses the same strength.

For assumption questions, negate the candidate assumption: if the argument collapses, it is a necessary assumption; if not, it is not required.

For inference questions, the correct option must follow from the passage alone and add no new entities or relations.

Percentage change = (new − old) / old × 100; percentage points measure the arithmetic difference between two percentages, not the relative change.

Successive percentage changes multiply: +20% then −20% gives ×1.2 ×0.8 = 0.96, a net −4%, not 0%.

Reverse percentages: if a price after a 15% increase is 92, the original is 92 / 1.15 = 80, not 92 × 0.85.

Ratios and proportions: convert to a common unit before comparing; a ratio a:b scales as ka:kb, and part-to-whole = a/(a+b).

Averages: weighted mean = Σ(w·x)/Σw; the mean is sensitive to outliers, the median is not; check which the question asks for.

Rates and work: combined rate = sum of individual rates (1/a + 1/b); time together = ab/(a+b) for two workers.

Number series: test differences, then ratios, then alternating or second-difference patterns; confirm the rule on at least two consecutive steps.

Abstract matrices: check row-wise and column-wise rules for rotation (90°/180°), reflection, addition/subtraction of elements, and position shifts; a rule must hold for every row or column.

Data interpretation: read axis scales and units before computing; estimate first to eliminate options whose order of magnitude is wrong.