What the ear drops first
Under load, the interpreter's working memory sheds numbers, names and lists before anything else — which is why they are written down.
Under load the first things to go are numbers, names and lists, which is why they are written down rather than remembered.
The hierarchy of forgetting
Simultaneous interpreting is a dual-task operation in every sense: the interpreter is listening, analysing, reformulating and speaking at the same moment, in two different languages, while tracking where the speaker is going and monitoring the output already leaving their own mouth. Every cognitive resource is in use. When the load spikes — a fast speaker, an unexpected topic shift, a sentence structure that diverges sharply between source and target — something has to give. What gives is not random.
Research into working memory under interpreting conditions, drawing on the model elaborated by cognitive psychologist Alan Baddeley from the 1970s onward, consistently shows the same hierarchy of loss. Phonological working memory — the inner loop that holds sounds and words in a temporary buffer — collapses under dual-task pressure before semantic processing does. The interpreter continues to extract meaning from incoming speech well into a difficult passage; what slips first are the items that resist meaning-extraction: sequences of digits, proper nouns with no semantic content to anchor them, and lists whose internal logic is enumeration rather than argument. A minister's portfolio matters; the number of her ministry's budget line does not decode into anything beyond itself, and that kind of bare symbol is the first casualty.
A meeting day is longer than the speaking day, and preparation is most of the work. The whole shift
This is not a failure of training or attention. It is the architecture of memory under competition. The phonological loop can hold roughly seven items — the figure most associated with George Miller's 1956 paper, published in Psychological Review — and simultaneous production in a second language uses that loop heavily. There is simply no slack for a twelve-digit reference number delivered at conference pace.
What gets written down, and why
The professional response to this is a notepad and a specific discipline about what goes on it. Numbers go down in figures the instant they are heard; names are written phonetically if the spelling is unknown; lists are tallied with a vertical stroke for each item so the interpreter knows whether they have rendered all five of them. None of this is improvisation — it is the application of a principle that the booth has worked out empirically over decades: the voice can carry interpretation of meaning; the hand has to carry everything that meaning cannot hold.
The discipline differs depending on the mode. In consecutive interpreting, where the interpreter listens to a passage — sometimes several minutes long — before rendering it, note-taking is a complete system in itself. Consecutive notes are not transcription. They record logical relationships and rhetorical structure using symbols and abbreviations that are largely personal to the practitioner, with only the numbers, names and lists written in full, because those are the items that cannot be reconstructed from structure alone. The interpreter who forgets that a speaker said "we expect growth" can often reconstruct the sense from surrounding notes; the one who forgets whether the growth figure was 3.7 percent or 7.3 percent has lost something unretrievable.
In simultaneous interpreting the notepad serves a narrower but equally critical function, because there is no pause in which to review it. The interpreter writes a number or name mid-stream and must be able to glance down in the fraction of a second the booth allows while the reformulation of the surrounding sentence continues in their mouth. This is a trained skill and it takes time to develop: looking away from the glass while still tracking the speaker's gestures and the rhythm of delivery, absorbing the written item in a single glance and returning to the flow. Booths that are built to specification include a writing surface at exactly the right angle and height for this reason.
Numbers, names and lists: the specific failure modes
Each of the three categories fails in its own way, and knowing the failure mode matters because the consequences differ in severity.
Numbers are the most obvious risk. A currency figure, a vote count, a treaty article number, a date — all of these are content-dense symbols that carry no redundancy. If the interpreter renders 1.4 billion as 1.4 million, no surrounding context will correct the error; the two sentences are otherwise identical. The note habit is absolute here. AIIC's published guidance on conference preparation stresses that delegates should receive documents in advance precisely because the interpreter who has seen a figure in print before hearing it can pattern-match rather than cold-process it. An expected number is far easier to hold than a surprising one.
Proper names present a different problem. An unfamiliar name — a regional official, a scientific compound, a newly formed agency — arrives as a pure phonological event with nothing semantic to grip. The interpreter who has not encountered the name before must both transcribe it phonetically and continue interpreting the sentence it belongs to. If the name is misspelled or mispronounced in the output, this can matter enormously in legal or diplomatic contexts: the Court of Justice of the European Union treats precision in the names of parties, laws and precedents as a baseline requirement, and an interpreter rendering a party name incorrectly in a hearing creates a record problem, not merely an auditory one.
Lists are the subtlest failure. When a speaker enumerates — "there are four conditions: first… second… third… fourth…" — the interpreter must track position in the list while simultaneously producing the rendering of each item. Miss a count marker and the output list has three items where the source had four, and no listener can know. The tally stroke is the correct technique, and it is acquired early. What the inexperienced interpreter does instead is trust that the structure will carry itself, because the first two or three items feel manageable. It is the fifth and sixth that drop.
There is simply no slack for a twelve-digit reference number delivered at conference pace
Preparation as pre-loading
The asymmetry between expected and unexpected items points toward what preparation actually does. Working through a meeting agenda, reading background documents, extracting proper names and technical terms into a working glossary the night before — none of this primarily teaches the interpreter anything. What it does is move items from unknown to known, from cold-processing to recognition. A name encountered in print, even once, has a hook the phonological loop can hang it on. A number seen in a briefing document is expected and therefore far more likely to survive the load spike when the speaker says it.
This is why the Directorate-General for Translation and its interpreting counterpart at the European Commission treat document distribution before meetings not as courtesy to interpreters but as quality infrastructure. The relationship between preparation time and output accuracy is not linear, but it is real, and it operates precisely through the mechanism of pre-loading the categories that are most vulnerable under load. The interpreter who has seen the budget annexe is not reading from it; they are protected against the moment when the rapporteur rattles through twelve line items in forty-five seconds and the booth has to stay with them.
What the ear drops first, in the end, is not random and not recoverable by willpower. It is structural, predictable and partially preventable — by the hand on the notepad, and by the documents that arrive the evening before.