My name is James Sandford and I work in 成人快手 R&D鈥檚 User Experience team in the area of enhanced subtitles.
Current accessibility guidelines advise a maximum rate for subtitles (defined as the number of words per minute on screen). This blog post describes work at 成人快手 R&D that looked at why subtitles may exceed this rate and revisited the question of what that rate in the guidelines should be.
Recent work on the had shown subtitle rate varies widely and is often to be expected. For example, the speech on children's TV is often quite slow to allow easier understanding by children. In a similar way, it is often slow for nature documentaries where impressive imagery of beautiful animals and landscapes are the main focus.
At other times speech is often fast. Examples include weather presenters trying to get as much information into as short a time as possible or when a number of people have a heated discussion in a soap opera.
In addition, there are occasions when subtitles may be unintentionally fast or slow. This can be due to technical issues or as a result of the staff creating subtitles (subtitlers) trying to catch up after pausing to correct an error when subtitling live programming.
How fast should subtitles be?
We have guidelines on the maximum speed for subtitles. Ofcom say that subtitles should not exceed 160 to 180WPM (words per minute) and advise that subtitles over 200WPM would be difficult for many viewers to follow. We wondered why those numbers are what they are. What research led to them? Once in a while when you ask yourself these questions, you find things aren't quite as you might expect...
Little work had been done on the user experience of subtitles until the 1990s. Up until this point, the focus of much of the work had been on the technical challenges of producing, transmitting and displaying subtitles. The electronics that made subtitles possible were only just becoming cheap enough to see widespread use. The rules around how subtitles should be placed on the screen, when they should appear and what should be included in them were largely artistic in their creation and driven by technical constraints. They were based on audience feedback and the stylistic decisions of those creating the subtitles.
But as things settled down and the field of User Experience expanded to aid software design, researchers turned their focus to putting more scientific rigour into subtitle guidelines. It's obvious that if the words flash up on the screen too fast, you won't be able to read them.
So how fast is too fast?
In the 1990s and 2000s a number of studies were conducted focusing on the comprehension of subtitles at different speeds. These consisted of showing people videos with subtitles at different rates and asking questions related to the content of the subtitles to see how much viewers understood. These studies were designed such that subtitles were the only thing that could have a major effect on the answers. Some had video content such as a picture slowly moving on the screen and no audio. This meant that while there was something visually interesting on screen, it wouldn't influence enjoyment or understanding. Others saw this approach as un-natural. They used broadcasted programming, but to reduce the influences on answers to the subtitles they only asked questions unrelated to the imagery and were presented without sound. Most of these studies found subtitle rate had little effect on comprehension.
The only study we found that reported an effect used a small number of videos with high possibilities of influence from other factors. These videos used live subtitles (which are notoriously low in quality) and subjects were staff, students or subtitlers from the same university department as the author and who knew the intent of the study.
Other studies in this period looked at enjoyment of subtitles at various rates. These studies asked how much viewers enjoyed each video they'd seen, or if they thought the subtitles were too fast or too slow. Again, many of these studies used video content that wouldn't have an influence on answers and were presented without sound.
One study found a preferred rate of 145WPM and that 9% of viewers felt the subtitles too fast at 170WPM rising to 28% at 200WPM. Another study asked viewers to report their enjoyment and how much of the subtitles they felt they could read. While this study showed people enjoyed subtitles at a lower speed than they were capable of reading, it presented little data or details of the study conducted to support its conclusions.
We now knew that previous studies had shown subtitle rate had little effect on comprehension and that people enjoyed subtitles most at around 145WPM. We also had figures for when people felt they became too fast. But there were issues with these studies. Much of the video content was un-natural. The majority of subtitle users have some level of hearing and use this in conjunction with subtitles when viewing television. Others lip-read where possible in conjunction with the subtitles. Previous studies left subjects unable to use either of these sources of information and would provide little indication of enjoyment of subtitles in the real world. Furthermore, these studies didn't compare results from hearing impaired participants to those who had good hearing and viewed television without subtitles. In scientific terminology, there was no control.
A new study
It became obvious that to answer the question of how fast subtitles should be, we should conduct a new user study. We would need a control, would look at both block and word-at-a-time subtitles (the two most common forms of subtitling) and would use natural material where hearing and lip-reading were possible. To test enjoyment at various rates in a controlled way, realistic content would be created specifically for the study. To verify the findings in this section of the study, another section would use a range of broadcast videos with their broadcast subtitles.
For the controlled section of the study, we enlisted the help of our colleagues at 成人快手 North West Tonight (a local news programme). Together we produced 24 videos with subtitle rates at 20 WPM intervals between 90WPM and 230WPM, three at each speed. All were 30s long, accurately subtitled, and allowed for both lip-reading and hearing. The North West Tonight production crew and presenter, Annabel Tiffin, proved invaluable in giving the clips in this study a realism and consistency far beyond those in previous studies. We showed these videos to both hearing impaired subjects (with block and word-at-a-time subtitles) and hearing subjects (with no subtitles) who rated them for enjoyment and perceived speed on a scale of "Too fast" to "Too slow".
The results showed that both hearing and hearing-impaired viewers enjoyed subtitles/speech at a similar speed. Around 230WPM being 鈥渢oo fast鈥�, 115WPM 鈥渢oo slow鈥� and 175WPM about right. Furthermore, hearing viewers were more sensitive to extremely high or low rates than subtitle users. Essentially, when the rate of subtitles became an issue the rate of speech was already an issue for hearing viewers. As the majority of televised content is produced primarily for hearing viewers, rates that are hard to understand will either be avoided already or are intentionally hard to understand.
Now we had figures for preferred speed as well as those that are perceived as too fast and slow. But did these videos in which text and speech range from un-naturally slow to un-naturally fast provide any indication of how fast subtitles should be in the real world? To test this, we showed the subtitle users eight real-world videos with subtitles that were identified as being particularly fast. These clips all came from dramas, daytime TV, interviews, weather and quiz shows. Some clips had sections where lip-reading wasn't possible, some had music playing at the same time as speech and all were far faster than guidelines recommend.
All of these clips were rated better than predicted for their speed by the first section. Even more surprisingly, all were rated as being at the ideal rate or at least acceptable. This section showed that the "correct" speed for subtitles on a programme is dependent on the programme. These clips were rated as acceptable at higher rates because the narrative, style or number of people speaking in the clip demanded such a speed. Two people in conversation will speak faster than one person on their own. It makes sense for news to be presented slower than the weather. Two people having an argument will just seem strange if speaking at the same speed as a child's bedtime story. The best speed for subtitles is the speed of the speech. The best speed for speech is that at which it feels natural for that particular piece.
But we did see people rating clips as too fast and too slow in the first section. Why was that if the rate of the subtitles itself wasn't the issue? News is normally presented at about 175WPM, the preferred rate for the subtitles in the first section. If you go much above or below that the content feels un-natural. But what about the complaints we receive about the speed of televised content day-to-day? The truth is we don't actually get many complaints about the rate of subtitles. Most of those we do get can be attributed to other issues. Technical issues or limitations can cause subtitles to "bunch up" and be played out at high rates which people identify as "Too fast". If there is a delay in subtitles appearing after the words are spoken such as in live subtitling, subtitles are identified as "Too slow" in the sense that they are too slow at appearing. It may be that this use of these terms along with a sense that there must be a rate at which subtitles are too fast to read that has caused confusion in the past. This has resulted in the rate of subtitles being identified as a big issue for subtitle users. This study has shown that as long as the subtitles are of good quality and match the speech, rate isn't an issue.
Summary
So back to the original questions.
Q. How fast should subtitles be?
A. The answer appears to be as fast as they need to be.
Q. Where do we have issues with subtitle rate and why do they occur?
A. While rate isn't an issue for well-timed subtitles, it may be an issue where technical issues or limitations exist. But this is not an issue with rate, as such. But an issue with the subtitles matching the speech. Other work focused on reducing errors and lag will help mitigate these situations.
In conclusion, this study, along with , has shown that guidelines should not recommend subtitlers edit subtitles to a specific rate. Doing so will likely reduce viewer enjoyment of the content. Instead, we recommend that subtitlers should aim for well timed, accurate, verbatim subtitles to provide the best audience experience for subtitle users.
This work was published in the journal ''. Further work by 成人快手 R&D on subtitles is being presented on stand 8.G08 and in the Latest Advances in Assistive Technologies session at IBC 2015 and may be found on the .