Jowonder (3)
| Posted: 2017 Apr 04
Gender bias in films laid bare by software:-
By Matt Reynolds in the New Scientist.
Machine learning is taking on Hollywood’s gender bias. Technology that automatically detects how often men and women appear on screen reveals that in recent popular films, men have had almost twice as much screen time as women.
The software uses algorithms for face and voice recognition that have been trained on annotated video to identify whether a character is male or female, and can measure how long they are on screen to a fraction of a second. It was developed by Shri Narayanan at the University of Southern California, Los Angeles, in partnership with the Geena Davis Institute on Gender in Media and Google.org, the search engine’s charitable arm.
In an analysis of the 100 highest-grossing live-action films from each of the past three years, the software found that women appear on average for just 36 per cent of the total time that characters are on screen. Oscar-winning films are even less representative, with women getting just 32 per cent of screen time and 27 per cent of speaking time in films that received an Academy Award.
Using algorithms for voice and face recognition that run in parallel, Narayanan’s program can analyse a feature film in less than 15 minutes. The software can also distinguish when one person is on screen but someone else is talking, says Narayanan. “Often when women are speaking, it’s actually the men that are shown on screen.â€
Of the 300 films analysed, the only genre in which women spent more time on screen than men was horror, with female characters receiving 53 per cent of screen time in such films. In dramas, women were only on screen for an average of a third of the film, and for crime films that figure dropped to less than a quarter. The system did not account for non-binary characters.
Quality time?
Calculating how often women appear on screen is one way to measure gender bias, says Ginette Vincendeau at King’s College London, but it’s also important to consider who they are portraying and what their characters are talking about. “It’s not just a question of quantity,†she says.
Narayanan is also using machine-learning algorithms to analyse film scripts, which could reveal further insights about the roles of female characters, such as the content of their dialogue. He has previously experimented with using natural language processing – AI that can process speech – to explore how the use of gendered language varies across films and genres.
Although the latest study indicates that Hollywood still has a long way to go until women are equally represented in films, it also suggests that audiences have an appetite for movies with female leads. Narayanan found that films with a female lead character performed better at the box office, earning more than $89 million on average, compared with $75 million for those with male leads.
The highest-grossing Best Picture nominee at this year’s Oscars to date was Hidden Figures, a biographical film about a group of female African American mathematicians who played a vital role at NASA in the early years of the US space programme. It has grossed over $158 million at the US box office so far.
But just because we’ve seen a few films with female leads do well at the box office doesn’t mean that Hollywood is changing its ways, says Vincendeau. The real agents of change in the film world are the production companies, she says, and it’s up to them to make sure that more women are placed in leading roles – both behind the camera and in front of it. Matt Reynolds
What scares me most is that we are used to be the victim in horror films and so we go up to a 53% on the screen there. For screaming I guess!
www.newscientist.com/article/2123926-machine-learning-reveals-lack-of-female-screen-time-in-top-films/
|
|
