English captions often run ahead of speech. A learner reads comfortable at the start of a subtitle line, watches the actor, and hears the word 1.8 seconds later without connecting the reduced spoken form to the spelling that appeared earlier. Briefly highlighting the word near its acoustic onset can pull the eyes and ears toward the same event.
This is a focused pronunciation activity for a few selected words inside a video. It is unsuitable as a permanent karaoke effect across every caption. The teacher chooses six or eight forms that learners know in print yet misrecognise or mispronounce, times the highlight, and checks perception and production afterward.
Captions and audio compete for timing
Captioned video combines a moving image, speech, and transient text. Learners distribute attention differently across those sources. Some read a whole subtitle line before the speaker reaches its final word. Others watch the face and glance down too late. The written and spoken forms may both be present while missing each other in time.
That timing gap matters for the mental lexicon. A learner may store vegetable with four evenly articulated syllables because spelling has dominated earlier study. Hearing a natural three-syllable version during a drama creates an opportunity to update the stored sound. The opportunity shrinks if the eyes have already moved to the next caption line.
One small visual change can mark the arrival. Turn vegetable from white to amber about 300 milliseconds before the sound begins, keep it highlighted through the word, then return it to white. No bouncing letters. No full-line colour wash. One word gets the cue.
An eye-tracking pronunciation study
Wisniewska and colleagues' 2023 study tested 58 first-year university students whose first languages were Spanish and Catalan and whose English was around B1 to B2. They watched two videos under five conditions: highlights beginning 300 or 500 milliseconds before the target sound, highlighting from caption-line onset, ordinary captions, and no captions. Eye tracking measured where and when learners looked.
In the adult study, the 300-millisecond and caption-onset enhancements increased attention to target words in different ways. Both synchronised conditions brought reading closer to the audio timing. On a later auditory lexical-decision task, every enhancement condition led to faster rejection of mispronounced target forms. Only the unsynchronised caption-onset condition improved accuracy. Average comprehension across clips was 87 percent, so participants were still following meaning rather than completing an isolated word drill.
Those results are promising and untidy. Faster lexical decisions can indicate an updated sound representation, yet the study did not show one timing condition winning every measure. It also assessed rejection of mispronunciations rather than a free conversation. The authors call the evidence initial, which is the right scale for a classroom recommendation.
A 2025 study with younger learners supplies a useful counterexample. Audio-synchronised keyword captions were compared with uncaptioned video among 33 children aged eight to nine in content and language integrated learning classes. Sixteen saw synchronised keyword captions and 17 watched without captions. Neither group improved significantly on phonological judgement, while both improved in oral production for target items and distractors. The timed captions showed no clear advantage over video alone in that small sample.
Another study broadens the subtitle comparison. Van der Kolk and colleagues compared first-language subtitles, English captions, and reverse subtitles, then measured meaning, written form, and pronunciation. Results differed by knowledge dimension, including gains in written form under reverse subtitles and English captions and mixed pronunciation findings. The format of screen text changes what learners attend to. “Subtitles worked” is too coarse a conclusion.
Pick words with a timing problem
Highlighting every unfamiliar word turns a film into a flashing worksheet. Select targets where print knowledge and sound recognition have separated. Good candidates include reduced syllables, weak vowels, silent letters, and stress patterns that hide a familiar spelling.
For a B1 office worker preparing for multinational meetings, a short clip might target comfortable, temperature, schedule, colleague, actually, and particularly. Six items are manageable. A proper noun, a plot detail, and two incidental new words can remain unmarked.
Run a quick pre-check before making the clip. Show each written word and ask the learner to pronounce it. Then play the word inside its sentence and ask them to select what they heard from two plausible forms. Highlight only words with a real mismatch. This prevents the teacher from spending 26 minutes timing a cue for a word learners already process accurately.
Build the viewing sequence
The first viewing should preserve the scene. Play 45 to 75 seconds with ordinary English captions and ask one meaning question. The learner needs to know who is speaking and what changed.
On the second viewing, add the timed highlights and pause after each target sentence so the learner can listen again, mark the stressed syllable, and repeat the whole phrase once. For I’m comfortable with Thursday, the practice unit includes the weak syllables and surrounding rhythm. An isolated recital of comfortable removes the speech pattern that caused the original miss.
On the third viewing, remove captions. Ask for the same meaning answer, then test the six words in fresh short sentences spoken by another voice. Choc's article on changing the voice during English ear training explains why one speaker can become part of the memorised cue. Keep the pronunciation target while changing the voice.
Production comes last. Give a two-minute role card that creates reasons to use three target words, perhaps rescheduling a Thursday meeting because a colleague has a temperature. Record the attempt. Score comprehensibility and the selected forms, leaving unrelated accent features alone.
Timing requires restraint
Automatic subtitle tools can provide word-level timestamps, though their alignment should be checked by ear. A highlight that arrives after the syllable teaches the learner to chase the audio. A cue that begins too early can pull attention toward spelling before the sound. For classroom materials, an approximate 300-millisecond lead is a reasonable starting point from the adult study, followed by local adjustment where subtitle timing drifts.
Keep the visual design quiet. One colour change, sufficient contrast, and no animation are enough. Learners with dyslexia, low vision, photosensitivity, or attention difficulties may prefer a static underline, a pause-and-replay version, or audio with a printed transcript. The timing idea can survive without flashing.
The activity also has a proficiency floor. A learner who understands only half the clip is already carrying a large meaning load. Use shorter, easier material or preteach the scene. Timed text cannot compensate for dense unknown vocabulary, rapid plot changes, and six simultaneous speakers.
Measure sound learning directly
A comprehension question checks the story. It says little about whether particularly acquired a clearer sound representation. Use three compact measures: hear the target among a correct and plausible mispronounced form, repeat it inside a phrase, and recognise it in a new speaker's sentence 72 hours later.
Record latency only if the tool makes it easy. A correct answer after 7.4 seconds may reflect deliberate spelling reconstruction, while a response at 1.6 seconds looks more like auditory recognition. Accuracy still comes first, and ordinary classroom timing will be rough.
For the next lesson, choose one 61-second clip and six known words that learners miss in speech, highlight each word once near its sound, remove the captions on the final replay, and listen for the moment comfortable stops needing its spelling.