TOEIC Link Speaking — Filler Elimination and Hesitation-Marker Control Under Time Pressure: The Fluency Discriminator That Separates Band 24 From Band 28

Hesitation markers are the most audible fluency defect on the TOEIC Link speaking module, and they are the defect candidates least often train deliberately. A candidate who fills every planning gap with "um," "uh," "like," and "you know" can have accurate grammar, precise vocabulary, and clear pronunciation and still be capped in the 22-to-24 band, because the rubric scores fluency as a distinct discriminator and audible hesitation is the first thing the rater hears. The gap between band 24 and band 28 on the speaking module is rarely a grammar gap — it is a hesitation-control gap, and hesitation control is a trainable motor habit rather than a knowledge deficit.

The problem is sharpest under time pressure. On the extended-response and picture-description task types, the candidate must produce continuous speech against a countdown, and the cognitive load of planning-while-speaking is exactly what triggers filler. The instinct is to fill every gap with a vocalized marker because silence feels like failure. The rubric scores the opposite: a short, controlled, unfilled pause reads as deliberate, while a vocalized filler reads as a fluency breakdown. This guide maps the hesitation-marker inventory, the pause taxonomy the rubric treats differently, and the substitution protocol that converts filler into structured silence. For adjacent fluency skills, see the speaking fluency and hesitation recovery guide, the speaking hesitation-marker and filler reduction for fluency scoring guide, and the speaking strategic pausing and cognitive load distribution guide.

The six hesitation-marker categories

Not all hesitation markers cost the same. The rubric penalizes them in proportion to how much they disrupt the listener's processing, so the first step is to know which markers you produce and which cost the most.

Category 1 — Vocalized fillers

These are "um," "uh," "er," and "ah" — the pure vocalization markers with no lexical content. They are the most penalized because they carry zero information and signal planning failure directly. A response with more than two vocalized fillers per fifteen seconds of speech reads as effortful, and the fluency band drops accordingly.

Category 2 — Lexical fillers

These are "like," "you know," "I mean," "sort of," and "kind of" used as gap-fillers rather than as genuine hedges. They are less penalized than vocalized fillers because they carry surface lexical form, but a rater distinguishes a genuine hedge ("it was sort of a compromise") from a filler ("the report was, like, sort of, um, finished") quickly, and the filler use is penalized.

Category 3 — Repetition fillers

These are the involuntary repetition of a word or phrase while the next chunk loads — "the the report," "I think I think that." They read as a stutter-level fluency breakdown and are penalized near the level of vocalized fillers.

Category 4 — False starts

These are abandoned sentence beginnings — "The company decided— the decision was made to expand." A single controlled false start followed by a clean reformulation is tolerated; a chain of false starts signals that the candidate is planning at the sentence level in real time and cannot sustain a syntactic frame.

Category 5 — Drawled syllables

These are lengthened final syllables used to hold the floor while planning — "the reportuh is due Thursdayy." They are subtle and often invisible to the candidate, but a rater hears them as hesitation and scores them in the filler family.

Category 6 — Prefabricated stalling phrases

These are memorized floor-holders — "that's a good question," "let me think about that." Used once, sparingly, and early, they are neutral or mildly positive because they buy planning time without a fluency penalty. Overused, they read as evasion and delay the substantive response, which costs points under the content and task-completion categories.

The three pause types the rubric scores differently

The central insight of hesitation control is that silence is not uniformly bad. The rubric distinguishes three pause types, and the training goal is to convert penalized pauses into the rewarded type.

Pause type A — Juncture pause (rewarded)

A juncture pause falls at a grammatical boundary — between clauses, after a completed thought group, at a list boundary. It is short (under one second), it aligns with the syntactic structure, and it reads as deliberate phrasing rather than planning failure. Native and high-band speakers produce these constantly. They improve intelligibility and are scored positively.

Pause type B — Mid-constituent pause (penalized)

A mid-constituent pause falls inside a grammatical unit — between an article and its noun, between a verb and its object, mid-preposition-phrase. It signals that the candidate is retrieving the next word in real time, and it disrupts the listener's processing because it interrupts a unit the listener is still parsing. This is the pause type filler is used to mask, and the training goal is to eliminate the pause rather than fill it.

Pause type C — Filled pause (penalized)

A filled pause is any pause occupied by a hesitation marker from the six categories above. It is penalized whether it falls at a juncture or mid-constituent, because the vocalization itself is the defect. The counterintuitive rule: a silent mid-constituent pause is scored less harshly than the same pause filled with "um."

The substitution protocol — four weeks

The protocol replaces filler with structured silence and lexical bridging. It is a motor-habit rewrite, so it must be trained under recording, not just planned.

Week 1 — Awareness and baseline

Record five extended responses per day and transcribe them verbatim, marking every hesitation marker by category. Most candidates dramatically underestimate their filler rate; the transcription makes the invisible audible. Compute a filler-per-fifteen-seconds baseline. Do not attempt to reduce filler yet — the week-1 goal is only accurate self-perception, because you cannot suppress a habit you cannot hear.

Week 2 — Silent substitution

Record the same volume, but this time consciously replace every vocalized filler with a silent juncture pause. The response will feel slow and gappy to the candidate — this feeling is the correct target, because the perceived gap is far shorter to the listener than it feels to the speaker. The week-2 goal is to break the reflex that silence must be filled.

Week 3 — Lexical bridging

Introduce controlled lexical bridges to buy planning time at grammatical boundaries — discourse markers ("what's more," "on the other hand," "as a result") that carry structural meaning and simultaneously hold the floor. Unlike filler, a discourse marker advances the response structure while the next content chunk loads. The week-3 goal is to convert dead planning time into structural signposting.

Week 4 — Time-pressure integration

Run full timed responses under the module countdown, recording each one, and hold the filler-per-fifteen-seconds rate at or below the week-3 level despite the added pressure. The integration week proves that the habit survives cognitive load, which is the only condition that matters on test day.

Common failure modes during training

Failure — over-correction into robotic delivery

A candidate who eliminates all pausing produces a flat, rushed monotone that costs points under the intonation and naturalness categories. The target is not zero pauses — it is zero filled pauses and juncture-aligned silent pauses. Natural speech has rhythm, and rhythm requires pausing at boundaries.

Failure — lexical filler substituted for vocalized filler

A candidate suppresses "um" but replaces it with "like" or "you know," which lowers the audible-vocalization penalty only marginally. The protocol targets all six categories; substituting one filler family for another is not progress.

Failure — planning at the wrong grain

The deepest cause of mid-constituent pausing is planning at the word level rather than the chunk level. A candidate who retrieves words one at a time will always pause mid-constituent. The remediation is chunk-level planning — rehearsing prefabricated multi-word units so that a whole thought group loads at once. See the speaking thought-group chunking and pause placement for listener processing guide for the chunking drill that underlies this repair.

The scoring payoff

Filler control is one of the highest-leverage speaking interventions available, because it is closable in four weeks without expanding vocabulary or grammar range and it moves the fluency band directly. A candidate holding at band 24 on grammar and vocabulary who reduces filler-per-fifteen-seconds from six to under two will typically move the fluency sub-score by a full band, and because fluency is weighted heavily in the composite, the overall speaking band moves with it. The discriminator is audible, the fix is mechanical, and the training window is short — which makes hesitation control the first thing a plateaued speaker should train.