Tip of the Tongue: Why You Know a Word but Can’t Say It

The distinction matters more than it sounds, because the two accounts predict opposite things about how to get the word back.

What people actually know

Roger Brown and David McNeill ran the founding study in 1966 by reading participants definitions of uncommon words and catching them mid-struggle. What made the paper important was the discovery that people in this state are not simply blank.

They could report the number of syllables at above-chance rates. They could often report the first letter. They could produce words similar in sound and words similar in meaning, and distinguish which was which. Some could report the stress pattern. This is a great deal of information about a word you supposedly cannot retrieve.

That pattern is the central fact any theory has to explain. Retrieval is not all-or-nothing. The system has clearly reached the word — it is delivering fragments of its sound — and cannot assemble the whole thing.

The two-stage architecture

Models of word production separate semantic and syntactic information from phonological form. You first activate the concept and its grammatical properties, then the word's sound. A tip-of-the-tongue state is what it feels like when the first stage succeeds and the second stalls.

This explains the peculiar confidence. You know you know it, because the semantic representation — including grammatical detail like a word's gender in languages that mark it — is fully available. What is missing is the form. The knowing and the saying come apart, which is the same architecture behind feeling certain a moment is familiar with no memory to justify it.

The cross-linguistic evidence

The two-stage picture gets sharp support from languages that mark grammatical gender. Italian speakers in a tip-of-the-tongue state can report the gender of a noun they cannot produce, at rates well above chance — and gender is a syntactic property carried by the word's lexical entry, not something you could infer from its sound.

That is difficult to explain on any account where retrieval simply fails. The system has clearly reached the entry and extracted its grammatical properties while failing to deliver its phonological form, which is precisely what a two-stage model predicts.

The blocking explanation, and why it fell

The intuitive account is blocking: an interloper word — similar in sound, more accessible — captures the retrieval attempt and suppresses the target. It fits the phenomenology exactly. People stuck on a word frequently report an interloper arriving repeatedly and feeling like the obstruction. Diary studies find phonologically similar interlopers accompanying a large majority of naturally occurring states.

The key experimental support came from Jones and Langford in 1987, who presented phonological neighbours before the target and reported increased tip-of-the-tongue rates, while semantic neighbours produced no such effect.

It did not hold up. Follow-up work found the same pattern of results with the same materials when no blockers were presented at all — meaning the effect was an artefact of which words had been chosen rather than of the blocking manipulation. Subsequent studies have generally found that related primes help rather than hinder retrieval.

Why the feeling is misleading

This is a good case study in phenomenology being poor evidence about mechanism. The interloper genuinely does keep arriving, and it genuinely does feel like the problem. On the transmission account it is a symptom rather than a cause: partial phonological activation of the target also activates similar-sounding words, and those words are retrievable while the target is not, so they surface.

The transmission deficit account

The favoured alternative, developed by Deborah Burke, Donald MacKay and colleagues, holds that tip-of-the-tongue states arise from weakened connections between the semantic representation and the phonological one, so insufficient activation reaches the sound level to complete retrieval.

It predicts the age pattern well. Tip-of-the-tongue states become more frequent with age, and connections weaken with disuse — which explains why proper names are hit hardest. A name connects to a single person rather than to a web of semantic associations, so it has fewer routes for activation to travel and less frequent reinforcement.

It also predicts what helps. If the problem is insufficient activation rather than interference, then anything supplying phonological activation should assist — and studies find that related primes, and producing similar-sounding words, aid resolution rather than deepening the block.

The picture is not entirely closed. A recent analysis using diffusion imaging from a large ageing cohort found that measures of phonological network efficiency positively predicted tip-of-the-tongue rates, a result that complicates a simple weak-connections story. This remains an active question rather than a solved one.

What to do when it happens

Two things follow from the transmission account, and they run against common advice.

Stop trying is genuinely reasonable, but not because effort is counterproductive. Roughly half of these states resolve spontaneously within a couple of minutes, and the word often arrives once attention has moved on — consistent with an unresolved retrieval goal that stays active in the background, in much the same way an unfinished task keeps surfacing.

Running the alphabet is more defensible than it sounds. Producing candidate sounds supplies phonological activation, which is exactly what the deficit account says is missing. On a blocking account this should be actively harmful, since you would be generating competitors. That the technique works at all is a small point in the transmission account's favour.

And the state is not a sign of anything wrong. Young adults report one or two a week; the rate rises with age and with bilingualism, since maintaining two lexicons means each word is used less often. Frequency is a fact about how often you use a word, not about how well your memory works.