🔎 A glimpse into our data...

To investigate the portrayal of women in movies, we used the CMU movie summary corpus which contains movie summaries from Wikipedia and their metadata extracted from Freebase [2]. For each movie, we will focus on its list of characters and its summary to explore this issue.

Data in numbers

Source: [2] David Bamman and al., Learning Latent Personas of Film Characters, ACL 2013, Sofia, Bulgaria, (2013)

Women are underrepresented in movies

We've seen that the large majority of movies include a higher fraction of male compared to female characters at least in the top grossing movies released in 2021. But, has it always been the case and is the trend changing? As we can observe on the graph, this inequality has always been present. Indeed, the fraction of female characters varies between 25% and 36%. Interestingly, we can see a slight increasing trend from the 80s coinciding with the slow improvement of women rights since the late 70s.

Is the underrepresentation of women the same for all movie genres?

In this plot, we observe that the top 10 genres featuring female or male characters are different. We notice that some genres are shared such as drama, thriller and romantic comedy but others are completely absent for the other gender. Additionally, some of the genres are very stereotypical. For example, 4 out of 10 genres for female characters are related to romance or family whereas male characters are much more present in action and science fiction movies.

Now that we know that the genres in which women are most featured in is different, let’s look at a few of the most common movies genres in the dataset and see if the fractions of male and female characters has evolved differently through time.

• As expected, in very stereotypical masculine movie genres such as Action/Adventure, the difference in fraction of male vs female characters is very large. Whereas, for a more stereotypically feminine movie genres, for example Romance or Musical films, the difference is smaller. Interestingly, it is still quite large, respectively about 15 and 24% . Indeed, one could expect that the fraction of women in movie genres, traditionally marketed to them, to be much higher than in more “masculine” movies. Additionally, no significant evolution is observed.


• Intriguingly, the difference in Horror film is relatively low compared to the other similar more violent movies genres. Nonetheless, this does not mean that women are more empowered in the Horror genre. Indeed, it is widely known that women are often the victim of atrocious violence in these movies.


• Finally, it seems that some of the genres evolved and other not. For example, the fraction in Comedy movies do not change significantly over time while the one of Drama seems to slowly decrease since the 80s.

In this initial exploratory analysis, we discovered that the fractions of female characters is much smaller than male and that it did not evolve a lot through time. So we were wondering, what are the typical roles and actions that female characters portray in movies? In order to research this question, we will analyze movie summaries by using natural language processing.

📝 Running NLP pipeline...

Here is an example of story that will be analyzed with the same process:

Above are the tables containing the extracted active and passive verbs and the number of mentions of each character. The NLP process first identifies the characters in the summary and then it counts the number of times each character or its references are mentioned and it extracts the verbs that are linked to them. Active and passive are defined by their grammatical tense. However, if a verb is extracted as active but figures in a list of descriptive verbs, as for example she seems sad, the verb is moved to the passive group. Next, the activeness score is computed by dividing the number of active verbs (after the potential shift of some verbs) by the total number of verbs extracted.
The first table is the result of the NLP process and the second one corresponds to what a human would extract. We can observe that the two tables are almost identical, only a few verbs (indicated by colours) are different which assures that the extraction is well done by our algorithm. Additionally, one can see that the missing/misplaced verbs are mostly present in difficult sentences structures, which is tolerable. Using the same process, all the summaries of the selected movies will be processed.

📊 Data Analysis

Are female characters as essential to plotlines as male characters?

Previously, we noticed that the fraction of female characters in movie metadata is much smaller compared to the fraction of male characters. Even though there seems to be an increase of the fraction of female characters between the 70s and 2010, this fraction was still around 35% in 2010 which is far from parity. Therefore, we wished to analyze if this difference is also present in the movie summaries. Indeed, a character may be present in the list of all characters of a movie, however it does not mean that the character is also mentioned in the summary of the movie as it could just be the housekeeper which appears only once in the movie and is not important for the story. Indeed, one can assume that the characters mentioned in the summaries are quite important for the story and have a certain contribution to its development and especially the ones mentioned several times.
As we can observe below, the fraction of mentions of characters in the summaries is around 35%. This value is very similar to the one computed in the CMU character metadata. Thus, it confirms that female characters are underrepresented and seem to have overall less importance compared to men. In addition, no statistically significant evolution has occurred between the 40s and 2010. In fact, the calculated fraction has remained relatively constant.

In order to analyze the activeness of the characters, we extracted all the verbs related to them and defined if the action was done by the character (active) or done to them (passive). At first glance of the graph, one might think that the men are more often described as doing things, e.g using the active voice. Indeed, the difference in activeness score among male and female characters is always positive through the years. However, the magnitude of this difference is very small (around 3-5%). Secondly, across the years, a decrease in activeness of male characters seems to appear. There is a statistical significant difference but its effect is very small, almost negligible.
By only doing this naive analysis, one might think that our initial hypothesis is incorrect. Indeed, this value seems to indicate that women and men in movies are portrayed to do active actions equally. However, it is important to realize that not all actions are equivalent. For example the two following sentences are both in the active voice, but have completely different level of “activeness” for the associated characters:
“Sophie fell in love with him at first sight.” or “Mark rushed to her rescue.”

What do male and female characters actually do?

As mentioned above, the context and underlying meaning of verbs is crucial to investigate the portrayal of characters. However our NLP pipeline does not keep the context of each verb. Therefore, it is impossible using computational methods to identify interesting clusters in the top occurring verbs without the context. Upon further analysis of the most frequent verbs, we identified three meaningful clusters. Indeed, it appears that a lot of the verbs are linked to violent actions, romantic relationships, and powerful actions. Moreover, these clusters are particularly meaningful as the propensity of each is different for female and male characters. Once these clusters were obtained, we counted the number of verbs and gave a score to each verb based on its number of occurrences. The higher its occurrences, the higher the score.

We can see that the stereotypical gender behaviour are indeed also present in the summaries. In fact, men have a high score in violence and power whereas women have a high score in love. Interestingly women have a relatively low score in violence in the active actions, female characters receive a lot of violence in the passive form. This is certainly due to the fact that in many plotlines women are victims of the violence and not the offenders. Here is a sample of our clusters:

Conclusion

We have seen that male characters outnumber female characters by about 2:1. However, the percentage of women characters seems to increase slowly. But how long would it actually take for the parity to be reached if we assume that it continues to grow with the same rate as now?

Place your bet. How long do you think it will take? Scroll to the right on the graph below to reveal the answer!

*Sigh* That’s a long time to wait for women to finally be represented… Moreover, as we have seen, it is not just about having more women characters in movies, but also about how they are portrayed. Indeed, more female and male characters going beyond gender stereotypes are needed in movies.
The very slow evolution as well as the stereotypical portrayal of women can probably be explained in part by the fact that most people involved in the movie industry are mainly men. Indeed, only 17% of directors and writer are women. [3] If we want the movies we watch to be more representative of women in today’s society and stop reinforcing dangerous stereotypical ideas, more women must be given the opportunity to write and tell their own narratives.

Gentlemen, step aside and make way for the damsels in power 👊

Source: [3] Katharina Buchholz, This is how female representation is rising across the film industry , Word Economic Forum (2022)