Spicing up Feature Film Credits with Classification — Part I

If you answered “no,” you’re normal.

Credits are boring.

But we take credits at the end of a movie for granted; after all, the credit sequence is factored into the running time of films.

Giving credits a second thought, it seems bizarre that we cap a film, an artform that uses dramatic visuals and rhetoric to evoke a visceral response, with such a dull format as the credit sequence.

Of course, the credits have earned their place in film as a traditional way of paying respect to its creators, sometimes even accompanied by thematic clips to make the last transition less stark.

Everything before the credits is meant to be enjoyed passively, while everything after hardly resembles the art that proceeds it — the traditional format suggests that you read a list of names and pay respect to each for the role they played in making a movie before it slides out of view.

Only with time and repetition does the average moviegoer learn to associate names (especially those below the line) with style, ethos, and privilege.

Actors have the better end of the stick.

Since their faces are projected in high detail while their voices project from surround sound speakers, you get familiar with the cast quickly; but, if you watched a movie for the second time, I bet you wouldn’t notice if all of the names of the crew roles were scrambled.

To try to close the “familiarity gap,” I thought that if I found the elements in names that help people recognize them, I could make credits more appealing for people.

Personally, I remember people’s names better when I share experiences with them.

Herein is the first complication: people will never meet most of the people whose names they see in front of them.

The next apparent solution would be to bombard people with the public profiles built around these people, e.


dynamically changing the videos of credit sequences and injecting IMDB biographies for each credited person.

This seemed effective but too burdensome.

“that” = reading the life story of every writer, sound designer, and costume writer of Transformers 7Next, I considered identity.

If occupation and expertise aren’t interesting enough to keep people around for the credits, I figured a common background would be.

Americans seem to light up when they meet someone who shares a common nationality, even if their ancestors immigrated 5 generations ago.

I ran with this truism and decided to label every credited person I could find with their nationality.

Then, I figured, I will have accounted for even the typical moviegoer who doesn’t care to look beyond the theater experience to “get to know” credited people.

“Oh my god, I’m Irish, too! “I figured, since the “typical moviegoer” can’t know the ethnic background of each member of the credits, the labeling process should reflect the same naive perspective and make an educated geographical estimate based on the name itself.

I figured wrong, but before I explain why, I’ll show you how far I took a hunch before it became glaringly obvious that my process was flawed.

I followed the intuition people use to assume the ethnicity and gender of someone they have never met and turned to machine learning models that associate names with places using survey, census, and poll data.

“Assume,” “ethnicity,” and “gender” have been used in the same sentence.

Exercise caution.

There are many tools to tack on identifiers to names, namely NamSor, which draws on massive datasets to provide a best guess of someone’s gender and ethnicity based on solely their name.

As sophisticated as NamSor has been engineered to work and as many onomasticians, linguists, and anthropologists as they consult, its (and any) AI classifier cannot operate with 100% accuracy.

It’s impossible for two reasons:The name sequence is not long enough.

 Take, for example, a similar problem of labeling a story with a genre.

If the story went, “The person walked,” you’d be equally unconfident about placing it into any genre.

But if the story went, “The girl forced trepid steps toward the dingey manor,” you’d presume this is the precursor of a horror story for a number of reasons, not the least of which is that manors only house evil.

Much more goes into a name than ethnicity, and much more goes into ethnicity than someone’s name.

 Just as someone wouldn’t look exclusively to their name to determine how they identify themself, assigning identifiers from the outside-in isn’t a valid means of understanding a population.

In the end, my plan to give more value to individual names than what appears on-screen caved under the responsibility to properly represent people.

I think I had the right intentions, but my approach was all wrong.

Rather than look for how people identified themselves, I thought, “this project calls for really good guessing,” and overlooked better solutions.

Classification, as objective as it seems, is problematic when applied to qualitative human characteristics.

In my next article, I’ll show how NamSor performed by augmenting personal data with fine-grained attributes, like ethnicity, in terms of diaspora (where movie creators most likely came to the United States from).

I’ll also show how this relates to base-truth information.

Stay tuned!.

. More details

Leave a Reply