top of page

Toward Intelligent Subtitling

In my years of QC work, I have noticed an alarming trend — many otherwise good subtitlers will often follow their client’s guidelines too strictly, almost dogmatically, without ever stopping to think why those guidelines are the way they are and not knowing when to depart from them to ensure the audience’s viewing comfort. This leads to countless subtitling errors, because no matter how robust or well-thought-out a style guide is, there will always be gaps in it, and so there will be moments when you’ll need to make a judgement call based on your expertise and sense of what’s right rather than some written instruction.

Let me give you an example. In one of the Netflix original series, at the very beginning of the first episode, a man and a woman are presenting to a room full of investors. They’re about to unveil a groundbreaking technical innovation that will change the world. After a short conversation between the two, the woman reveals that the second presenter next to her, the male colleague she just talked to, is that innovation — not a human but an AI hologram designed to be used as a personal assistant. Everyone in the room is shocked and amazed. Everyone except for you, the viewer.

You see, according to the Netflix Timed Text Style Guide, all dialogue heard through electronic media, which that hologram undoubtedly is, must be italicized. So, this whole time the man’s speech was in italics, while the woman’s speech wasn’t — the viewer could tell that something was off minutes before the reveal, just by looking at the subtitles. The intended effect was ruined, thanks to rigid adherence to the rules.

This error could have been avoided by taking a sensible, thoughtful, open-minded approach to subtitling. So, in this new article series, I want to discuss this approach as well as issues caused by blind compliance, starting with perhaps their biggest source.

Part 1: Reading Speed


Subtitling norms have been a subject of hot debate for as long as I can remember. Timing, segmentation, formatting, positioning, you name it — nothing has escaped dispute and controversy. Among these norms, the most contested one is reading speed, with corporate guidelines, academic studies and national traditions often disagreeing on what exact number of characters per second (or words per minute) should be the limit for each particular language, audience, and so on.

At the same time, when given creative freedom, the most skilled and experienced subtitlers don’t obsess over that number as much as everyone else does, because they know just how unreliable it can often be, in multiple ways. First of all, as I wrote previously, CPS and WPM consider only the volume of subtitle text but not its other properties, like complexity or format. Unfamiliar words, puzzling dialogue, tricky syntax, italics and all caps, for example, will slow down your reading, and those two metrics do not reflect that.

slow r.png


Moreover, they don’t take into account the appearance of your subtitles — their font, size, border type, contrast, etc. These all can — and often do — affect one’s pace of reading, but our measurement units remain unaware of that influence.



Also, they disregard the mechanics of how our gaze travels when we watch subtitled content. It is a fairly complex process which depends on numerous factors, yet it gets reduced to a single quantity. As a result, the CPS/WPM formula doesn’t include many important parameters, such as, for instance, the time it takes for our eyes to jump from the image to the new sub. Because of that, the viewer has to read much faster than what our software is telling us, and more so for shorter subtitles.


​Finally, CPS and WPM ignore the other, non-verbal elements of AV content — the sound and the image. Along with subtitles, these two compete for the viewer’s attention, and their editing and intensity can be the deciding factor in how much display time and text condensation you want to give each individual sub. And yet, our way of measuring reading speed leaves them out of the equation.


This Zoom-call scene is visually intense, since you basically have to watch six images instead of one. It’s hard to keep up even without subs, let alone with ones near the cps limit. (From: Social Distance)

​(click on the arrows to go through examples)


With the CPS/WPM metrics so imperfect, it is no wonder that seasoned subtitlers choose to rely not just on the number their software shows them but also on their own intuition and sense of pace. Instead of tunnel-visioning on how many characters per second or words per minute a subtitle is, they get a bird’s-eye view on all the pieces at play and try to put themselves into the viewer’s shoes to strike a good balance between reading speed, synchrony and text accuracy while respecting the audio and the visuals.

So, I’d like to urge everyone to move away from myopic fixation on the reading speed limit given by the client and to adopt this holistic, nuanced, intelligent approach.

bottom of page