There are 3.7 million titles in the Internet Movie Database (more commonly known as IMDb). Is there anything we can generalize about all those movies, other than that they usually (but not always!) involve dialogue?
In a recent theoretical article in the Psychonomic Bulletin & Review, James Cutting analyzed a large corpus of movies to promote a theory of popular movie form. Movies tell a story, and story-telling involves the reconstruction of our experiences in a narrative form that become the building blocks of our memories. That’s why a single word such as “rosebud” can evoke such strong, and sometimes emotional, memories and can elicit half a million Google hits, many of which are devoted to unraveling its mystery.
So what are movies really like?
Cutting approaches the problem by developing a system for describing the structure of movies as stimuli. He is not (yet) concerned with how humans process those stimuli, but he seeks to understand the stimulus landscape in a theoretically-coherent manner.
An important defining feature of this landscape is the distinction between the Fabula and the Syuzhet, Russian formalist terms that refer, respectively, to the “raw material” of a narrative and the way it is actually told. Thus, in Citizen Kane, the fabula is the actual story of Kane’s life, from birth to death in chronological order. The syuzhet, by contrast, is the way that story is actually told: starting with “rosebud” and Kane’s death, the film then interweaves flashbacks onto Kane’s earlier life with the present-day pursuit of this life’s story by a journalist.
With this distinction in mind, it becomes easier to analyze the structure of movies. Let’s begin with a very simple and seemingly trivial question: How many parts does a movie have? The answer is far from obvious. Whereas Aristotle believed that narratives had three parts—a beginning, middle, and an end—later literary traditions favoured narratives with five parts. For example, all of Shakespeare’s plays had five acts, and as the figure below shows, literary theorizing has frequently favoured a five-part structure of narratives:
Cutting shows that when one analyzes the density of transitions—such as cuts between scenes—in movies, they exhibit a statistical regularity that is suggestive offour acts. Cutting analyzed a large corpus of movies and standardized them (by considering the normalized density of events within each 1/100th of the movie’s duration) so that irrespective of the movies lengths, the speed of action could be compared across all movies in the corpus. The results are shown in the figure below, using a scale in which a lower point in the graph implies greater speed of the narrative:
In a nutshell, movies start out slowly with long scenes that introduce the actors, before speeding up slightly when introducing the complication. The complication is the part of the syuzhet that derails the protagonist’s intentions and entirely new circumstances are forced upon the character. Once past that hurdle, the movie slows again during development. This is because the narrative typically gets more complex, often dashing back and forth across parallel subplots, and longer shots are required to re-orient the viewer when subplots are crossed. During the climax, the movie speeds up and unfolds at a frenetic pace—no time for scene setting when the Fugitive is confronting the true culprit in the Psychonomics 2015 hotel in Chicago. Finally, the epilog is again a time for slower action.
Cutting reports several such studies involving a quantitative analysis of the corpus of movies.
For example, Cutting quantified the amount of motion depicted in a movie by tracing the extent to which pixels change across successive frames—more pixels change when motion is depicted than when the camera remains focused on a stationary object.
The results are shown in the figure below, with the abscissa again representing a standardized measure of time:
It is clear that motion declines during the setup, then remains relatively stable but picks up again during the climax, followed by a sharp return to general calm at the very end.
A mirror image results when luminance is examined:
Apparently movies are at their darkest just before the climax, and they are brightest at the very end. This systematicity is unlikely to be coincidental because brightness affects mood. The brighter the image, the better viewers generally feel—not surprisingly, therefore, movies are brightest at the (presumably happy) end.
All of these results support the notion that movies, for the past 75 years, have generally followed a four-act structure.
In his remaining analyses, Cutting focuses on the micro-structure of movies, such as the role of music and shot scales (i.e., whether a shot is a long shot or close-up) and functional variations in narrative structure across acts. The work is too extensive to permit a full summary here, but the way in which narrative shifts occur in a movie is intriguing and deserves to be highlighted.
The psychological consequences of physical continuity or discontinuity are very different from what one might think at first glance. Strikingly, cuts need not induce a sense of discontinuity even though they typically involve glaring and abrupt physical differences—indeed, our ability to perceive continuity is so strong that we sometimes miss the obvious:
But of course discontinuities do occur in narratives, and we are quite attuned to detecting a transition to another scene. There are three variables that can signal a transition to another scene: changes in location, changes of characters (arrival or departure of a person), and changes of time (flashbacks or dreams). Frequently, those variables are combined to signal a shift in narrative, for example when the setting moves from indoors to outdoors and involves a different set of people (perhaps even in a different time).
When those different means of introducing a new scene are quantified, the following pattern results:
Overall, there are fewer scene shifts during the complication and development phases, which reflects the fact that those segments involve more conversation—and conversational scenes tend to last longer than others, hence there are fewer transitions. Two of the variables, location and characters, mirror this overall trend perfectly. The third variable, time, presents a very different pattern: unlike the other variables, scene shifts based on time become less frequent as a movie heads towards its climax.
This decline might reflect the deadline that is embodied in most narratives across all movie genres: The protagonist(s) must attain a goal at a particular time. In consequence, as a movie approaches that deadline, there is a strong tendency to merge narrative time with real time—there is no time for flashbacks when Cary Grant must help Eva Marie Saint descend Mount Rushmore.
So what are movies really like?
Cutting’s work provides an in-depth answer at many levels—from the brightness of a movie to its macro structure. Far from being psychologically trivial, his work fills a void that is often left blank in cognitive psychology: We talk about “information processing” but frequently we seek to understand the “processing” without knowing much about the “information”.
Cutting’s work is of interest not just to cognitive psychologists but also to practitioners: Tomorrow we will find out what a screenwriter makes of Cutting’s analysis. Are movies really like this?
Article focused on in this post:
Cutting, J. E. (2016). Narrative theory and the dynamics of popular movies. Psychonomic Bulletin & Review, DOI: 10.3758/s13423-016-1051-4.