Modelling Gaze Behaviour in Subtitle Processing
The Effect of Structural and Lexical Properties
Keywords:audiovisual translation, reception, subtitles, eye-tracking, linear mixed-effects models
One of the main rules of subtitling states that subtitles should be formatted and timed so that viewers have enough time to read and understand the text but also to follow the picture. In this paper we examine the factors that influence the time viewers spend looking at subtitles. We concentrate on the lexical and structural properties of subtitles. The participant group (N = 14) watched a television documentary with Russian narration and Finnish subtitles (the participants’ native language), while their eye movements were tracked. Using a linear mixed-effects model, we identified significant effects of subtitle duration and character count on the time participants spent looking at the subtitles. The model also revealed significant inter-individual differences, despite the fact that the participant group was seemingly homogeneous. The findings underline the complexity of subtitled audiovisual material as a stimulus of cognitive processing. We provide a starting point for more comprehensive modelling of the factors involved in gaze behaviour when watching subtitled content.
Subtitles have become a popular method for watching foreign series and films even in countries that have traditionally used dubbing in this regard. Because subtitles are visible to the viewer a short, limited time, they should be composed so that they are easy to read, and that the viewer has time to also follow the image. Nevertheless, the factors that have impact the time it takes to read a subtitle is not very well known. We wanted to find out what makes people who are watching subtitled television shows spend more time gazing at the subtitles? To answer this question, we recorded the eye movements of 14 participants when they were watching a short, subtitled television documentary. We created a statistical model of the gaze behavior from the eye movement data and found that both the length of the subtitle and the time the subtitle is visible are separate contributing factors. We also discovered large differences between individual viewers. Our conclusion is that people process subtitled content in very different ways, but there are some common tendencies. Our model can be seen as solid starting point for comprehensive modelling of gaze behavior of people watching subtitled audiovisual material.