Why writing good subtitles is no walk in the park

Subtitles have virtually become a part of everyday life. They guide us through informative or entertaining videos and reels on various social media platforms. Anyone with the right app can write subtitles, which raises the question of why anyone would have recourse to a professional subtitling service in the first place.

There is more to writing subtitles than just automatically converting speech into text. Particularly for professional applications in film and TV production or for online business content, such as image videos or an interview with the CEO, manual post-processing plays a crucial role. Nowadays, however, subtitles aren’t always optional: they are sometimes required by law. The Swiss Disability Discrimination Act goes further than merely stipulating that Swiss railway stations, for example, must provide barrier-free options for boarding trains; it also encourages more and more companies operating in the public sector to produce their videos and films in a format adapted to people with hearing difficulties.

The unprocessed output from AI-based tools isn’t of sufficient quality for these applications, as the tools can’t always convert spoken materials to the correct text in a way that is cast-iron guaranteed to be free from errors, particularly for passages spoken in accents, proper names and recordings where people speak quickly, or the sound is “contaminated” by background noise. This is where the pros come in: they are familiar with the technical and language specifications and requirements involved in subtitling, and know how to put them into practice.

Subtitles should error-free, right?

Of course. However, the language requirements for subtitles go far beyond correct spelling and grammar. Text – whether it’s spoken or written – is always composed of interconnected units of meaning. English is characterised by the fact that most sentences begin with the subject, followed by the verb that describes the action and ends with the object to which the action is being done. If we carve up these units unevenly or tear them to shreds in the subtitles, our brain has a much harder job extracting info from the subtitles than it would if they were well-formed and broken down into proper blocks. Below are two examples of this:

Did you notice a difference when reading the two examples? In the latter example, care was taken to ensure that:

• the articles weren’t separated from the nouns;
• the lines didn’t end with a conjunction;
• nouns weren’t separated from their adjectives;
• the second line in a subtitle block can be longer than the first line (pyramid form); and
• the lines don’t exceed a specific character limit.

These are only a few of the linguistic features that subtitles should typically follow to ensure they are as easy as possible for our brains to process.

What are the technical requirements for subtitles?

We’ve already seen two of the requirements: subtitles shouldn’t be too long ─ depending on client instructions, and lines shouldn’t exceed around 40 to 43 characters in length, including spaces. Ideally, subtitles should be no longer than two lines. If they are any longer, we won’t have enough time to read everything because the text we are reading will no longer match the image. The video will keep running and won’t wait for us to finish reading.

There is also an important technical requirement on precisely this point, called “speaking rate”. This measures how many lines need to be read per second. It’s calculated by dividing the number of lines of subtitles by the display time for the subtitles. This, in turn, is linked to the minimum and maximum time for which a subtitle can be displayed. These components are then affected by whether there is a scene change in the video, e.g. whether the scene starts with one speaker before switching to another.

That’s a lot to bear in mind, isn’t it?

The appropriate tools can help with some of the specifications involved in subtitling. They can use artificial intelligence to compile an automated machine transcript and divide that into two-line units with a maximum character length per line. However, this is a purely mathematical breakdown and takes no account of whether the units of meaning are carved up or not. This is where the real work begins, similar to in post-editing following machine translation – the raw transcript is reviewed by a professional translator and converted to a form in which the brain can most easily assimilate it.

In other words, a lot of work goes into good subtitles. And we must all have seen films with poor subtitles. Whenever this happens, we end up stumbling or pausing because we have to concentrate too much on the written text. There is a parallel to translations here, too, as translations that make us pore over the wording shouldn’t be considered translations at all. In just the same way, we should be able to watch a video or a film with as little effort as possible and be able to focus on the content shown in the video without having to re-read the subtitles until we understand them.

Still not convinced? Try the subtitling service from SwissGlobal. We’d be delighted to show you the difference between an unprocessed and a processed AI transcript. Contact our project management.