Viewers can keep up with fast subtitlesEvidence from eye movements

If subtitles flash by faster than the industry says viewers can handle, comprehension should drop, cognitive load should spike, and viewers should feel cheated. Follow that chain to its logical end — and then consider this: none of that happened. Seventy-four viewers watched films at speeds regulators call too fast, and they were fine. More than fine — they preferred it. To understand why that result matters, you need to know what the speed limit actually is. The subtitling industry measures presentation rate in characters per second, or cps, which is a more precise unit than words per minute because word length varies so much across languages. For decades, the dominant rule has been known as the six-seconds rule: a full two-line subtitle should stay on screen for six seconds. That works out to roughly 12 characters per second, or about 140 to 150 words per minute. It became the industry's golden standard, baked into textbooks and broadcast guidelines, even though its precise origins are difficult to trace. The world has moved on faster than the rule. Streaming platforms routinely push subtitle speeds to 17 or even 20 characters per second. Subtitles have gotten longer too — lines that once capped at 32 characters now stretch to 42. In Scandinavian television, the rate was historically around 12 characters per second; Central European broadcasters settled between 15 to 16. Global streaming just broke past 20. Meanwhile, broadcasters and subtitle professionals are still arguing about where the ceiling should be. That's the controversy Agnieszka Szarkowska and Olivia Gerber-Morón decided to test empirically. Their study recruited 74 volunteers — 27 English, 26 Spanish, and 21 Polish speakers, ages 19 to 42 — and showed them clips from Blue Jasmine, Inside Out, and Mad Men. Each clip ran four to six minutes, was dialogue-heavy, and was subtitled at one of three speeds: slow at 12 characters per second, medium at 16, and fast at 20. Translations into all three languages were matched for character count so that display times hit the target speeds precisely. The study ran as two separate experiments. In the first, the soundtrack was dubbed into Hungarian — a language deliberately chosen because it bears no resemblance to English, Polish, or Spanish. Every participant was screened to confirm zero Hungarian knowledge. The logic was clean: strip out audio comprehension entirely, and you're left with viewers who must rely on the subtitles. In the second experiment, the same clips were played in their original English. Now English speakers had full access to the audio, while the Spanish and Polish participants remained dependent on text. The key instrument was an eye tracker — an SMI RED 250 mobile device recording at 250 frames per second. This is where the study's real contribution lives. Comprehension questionnaires and self-report scales tell you what viewers think happened; eye-tracking tells you what their eyes actually did. Szarkowska and Gerber-Morón tracked five metrics for each subtitle: fixation count, mean fixation duration, absolute reading time, proportional reading time, and revisits. The last one means how many times a viewer's gaze returned to a subtitle after having already looked away. The comprehension results landed flat, which is itself a finding. Accuracy scores across the three subtitle speeds came in at 79.35 percent, 80.42 percent, and 80.55 percent — essentially identical. The statistical test confirmed it: no significant effect of speed on comprehension. So the fear that faster subtitles would leave viewers behind simply didn't materialize. What the eyes revealed was more interesting. Slow subtitles at 12 characters per second didn't help viewers — they provoked more work. English participants averaged 0.67 revisits per subtitle at the slowest speed, versus just 0.20 revisits at 20 characters per second. That means at 12 characters per second, viewers were returning to re-read roughly two out of every three subtitles. At 20 characters per second, only about one in five. Those revisits followed a consistent pattern: a viewer would read the subtitle, look away to the scene, then glance back at the same text again. That's not thoroughness. That's the eye-movement signature of inefficient processing — having to check something you didn't fully absorb the first time. Absolute reading time and fixation duration told the same story. English viewers spent 1,554 milliseconds on average reading subtitles at 12 characters per second, falling to 1,191 milliseconds at 20 characters per second. Mean fixation duration dropped from 206 milliseconds to 191. And here's the telling correlation: more time spent reading subtitles was associated with lower comprehension, with a Spearman correlation of negative 0.71. More revisits correlated with lower enjoyment, at negative 0.33. The viewers who lingered longest on the text understood the least and enjoyed it the least. Slow subtitles were not giving people more time to absorb — they were giving people more reasons to go back. Now here's where the second experiment flips everything. When the soundtrack was in Hungarian — unknown to everyone — viewers favored condensed, slower subtitles. Between 58 and 64 percent across the three speeds said they preferred subtitles with less text so they had more time to read. That makes sense. No audio to fall back on means every word in the subtitle counts. But switch the soundtrack to English, and the preference reverses completely. A majority — 55.3 percent — wanted full, unreduced subtitles even at the faster speed. The reason is a phenomenon Szarkowska and Gerber-Morón call mismatch. When subtitles are slowed down for a soundtrack viewers already understand, the text has to be condensed and retimed. This means subtitles can appear before the dialogue, linger after it ends, or omit lines entirely. Viewers notice. In the second experiment, slow subtitles at 12 characters per second scored a mismatch rating of 4.22 on a seven-point scale; fast subtitles at 20 characters per second scored 2.28. The statistical effect was enormous — a partial eta-squared of 0.54, meaning more than half the variance in perceived mismatch was explained by that single manipulation. One participant put it plainly: the subtitles annoyed me because they didn't match with what was being said, and it felt like a conflict. Another said, when I understand the language I want to read exactly what I'm hearing; when I don't understand it, I want minimal words. So the same speed that frustrates viewers in one context is the preferred speed in another. Language familiarity doesn't just modulate subtitle processing at the margins — it flips the preference entirely. Cultural background modulated things further. English participants — who historically have less experience with interlingual subtitles — reported the highest cognitive load across the study. Polish participants, from a subtitling culture, declared the lowest frustration and spent the least time looking at the subtitle area. Spanish participants, coming from a traditionally dubbing country, showed the highest mean fixation duration and the highest subtitle dwell time, and their comprehension scores were the lowest of the three groups. The differences weren't enormous, but they point toward a real effect of subtitle familiarity on processing strategy. What does this mean for the industry? Szarkowska and Gerber-Morón are clear. The evidence does not support defending 12 characters per second as the safe default for all viewing contexts. Their data show that for hearing viewers watching familiar-language content, fast and unreduced subtitles outperform slow and condensed ones on enjoyment, perceived quality, and eye-movement efficiency — even though comprehension scores stay flat across conditions. The six-seconds rule, inherited from an earlier television era, may be underselling what viewers can actually handle. The authors are careful on one boundary. Their participants were hearing viewers. Intralingual subtitles — captions designed for deaf and hard-of-hearing audiences — operate under different constraints, and these results don't automatically transfer. That distinction matters. A guideline change that helps hearing viewers with English-language streaming could be the wrong move for a deaf viewer relying entirely on captioning with no audio access at all. What remains open is a list worth sitting with: the interaction between subtitle speed and text editing hasn't been fully disentangled. Language proficiency, film genre, complexity of dialogue, and individual reading ability all need more systematic study. So does age. But Szarkowska and Gerber-Morón have done something important here. They took a controversy that was running on intuition and institutional inertia, and they put 74 pairs of eyes on it — literally — and let the data speak. The speed limit, it turns out, was set for a viewer who no longer exists. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.